Advanced Land Observing Satellite-4 (ALOS-4) is a spaceborne high-resolution and wide-swath synthetic aperture radar (HRWS-SAR) that uses a variable pulse repetition interval (VPRI) technique to achieve continuous wide imaging. In some ALOS-4 images, azimuth fractional ambiguity caused by the VPRI is observed, and it differs from the usual integer ambiguity resulting from inter-channel errors in that it occurs at smaller intervals. In this paper, we propose a sequential Doppler offset (SDO) method for locating the original target (OT) that causes azimuth fractional ambiguity. First, the ratio of the interval of integer ambiguity to that of fractional ambiguity is obtained, which is used to generate SAR images with different Doppler center frequencies. Second, the coherence between the sum image of the generated images and the image with a zero Doppler center frequency is calculated. Third, some points with coherence greater than a threshold are selected based on the coherence. Finally, the final OT is obtained by detecting the filtered selected points. Some experiments are conducted based on ALOS-4 L1.2 data, and the results demonstrate that the method locates the OT accurately. In short, the proposed method provides a starting point for fractional ambiguity suppression in HRWS-SAR.

Multiple-input-multiple-output synthetic aperture radar (MIMO-SAR) has potential for high-resolution and wide-swath (HRWS) imaging. Current research mainly focuses on its echo separation. However, how to achieve HRWS imaging based on separated echoes has been rarely discussed and remains as if all echoes are separated and reconstructed ideally so that some conventional imaging algorithms can properly focus. Thus, this paper proposes an HRWS imaging mode for spaceborne MIMO-SAR, named virtual delay-emission (VDE). First, the VDE delays the transmission time of some channels to reduce the effective phase center (EPC) redundancy of MIMO-SAR and increase the effective antenna area. Second, the VDE transmits orthogonal waveforms with the same autocorrelation function. Third, the VDE separates the mixed echoes and reconstructs the separated echoes in the frequency domain to achieve low azimuth ambiguity and high-resolution imaging. Some system simulations based on ALOS-2/-4-like system parameters are conducted, and the results demonstrate that the VDE achieves the 3m/900kmmode if the system hardware is implemented properly.

We propose a novel method to enhance the short-term memory of spin-wave-based reservoir computing (RC) systems by tilting the external magnetic field. The tilting of the field modifies the interaction between two spin-wave modes, enabling simultaneous propagation of both backward volume spin waves (BVSWs) and surface spin waves (SSWs). We present this system as an alternative realization of multiple-mode coexisting (MMC) structures. First, our numerical comparison of BVSWs and SSWs demonstrates distinct group velocities between these two modes. We then evaluate the spin-wave reservoir’s performance on a delay task under different magnetic field orientations, revealing that the MMC structure achieves a higher memory capacity than conventional single-mode configurations. Furthermore, in a temporal exclusive-OR (XOR) task, the MMC structure not only improves the performance accuracy but also supports longer delay times compared to single-mode setups. These findings underscore the critical role of field orientation in enhancing spin-wave RC performance and offer insights into practical implementations of spin-wave reservoir designs.

In microwave remote sensing, radio frequency interference (RFI) caused by other microwave systems has been a critical issue. RFI degrades the quality of synthetic aperture radar (SAR) images and disturbs phase and polarimetric information. Therefore, RFI must be detected and mitigated correctly to analyze SAR data accurately. In this paper, we propose a novel RFI detection method based on the polarization states of received signals in either the time or frequency domain. Our study demonstrates that RFI from artificial sources is often more polarized than radar echoes from natural scatterers. We use the degree of polarization (DoP) to measure the polarization states and detect highly-polarized RFI. The simulation results indicate that the proposed method was more accurate than the conventional power-spectrum-based methods and improved the detection limit by 4 dB compared to the conventional method based on the similarity between polarizations. We demonstrate the effectiveness of the proposed method by using data obtained by Phased Array type L-band Synthetic Aperture Radar (PAL-SAR) aboard Advanced Land Observing Satellite (ALOS).

In microwave remote sensing, radio frequency interference (RFI), which is caused by other systems such as radars, telecommunications, and navigation systems, has been a serious problem. RFI generates haze-like or linear-pattern artifacts in synthetic aperture radar (SAR) images and degrades their quality. Therefore, it is mandatory to detect and mitigate RFI properly. In this paper, we propose an RFI-detection method using degree of polarization (DoP) which focuses on the difference of polarization states between RFI and useful signals.

In recent years, the applications of complex-valued neural networks (CVNNs) have gained prominence in radar-related domains, notably in synthetic aperture radar and ground-penetrating radar. Unlike traditional real-valued neural networks, CVNNs encompass both amplitude and phase components, for inputs, weights, bias, and activations. In this paper, we adopt the inverse mapping architecture for neural network explainability to conduct feature importance analysis for enhancing the interpretability and transparency of CVNN’s decision. We propose this method for first identifying built-up areas, and then determining the prominent input features by inverse mapping. The feedforward CVNN achieved high training and validation accuracies of 97.09% and 99.25%, respectively. Additionally, the inverse mapping results identify the most significant contributing input features.

The Goldstein filter is one of the most commonly used methods to reduce noise in interferometric synthetic aperture radar data. However, the low filtering performance in highly noisy areas remains unsolved. In this paper, we propose an adaptive filter in the biquaternion Fourier domain (BQFD filter) to improve the filtering performance even in highly noisy regions. This method represents PolInSAR data in a biquaternion form to process the interferometric fringes of different polarimetric channels as a single signal. By using the biquaternion Fourier transform, the fringes can be effectively separated from the noise because of the high correlation of the fringes and the low correlation of the noise between the polarimetric channels. Experimental results show that digital elevation models generated with the BQFD filter are more accurate than those generated with the Goldstein filter.

Ground penetrating radar (GPR) based landmine detection has advantages such as high safety and high efficiency. There are various methods to process data acquired from GPR systems. A common method is the Riemannian quaternion self-organizing map (RQSOM), which effectively enables self-organization of polarization data in quaternion form for visualization. However, RQSOM does not take into account the phase information of scattering components. Studies have shown that phase information is correlated with polarization information and can be effectively integrated into a new form, phasor quaternion (PQ). To visualize PQ-type feature vectors, in this paper, we propose a novel algrithm, Riemannian phasor quaternion self-organizing map (RPQSOM). RPQSOM utilizes the geometric characteristics of the phasor part to effectively learn useful information in PQ, thereby achieving superior visualization performance. We conduct experiments for visualizing a mock landmine. The experimental results demonstrate that RPQSOM fulfills our design objectives and gets better visualization results compared to RQSOM.

Waveform diversity is an important technique to extend the observation coverage of synthetic aperture radar (SAR) without sacrificing spatial resolution and power consumption. Applying pseudorandom noise sequence (PN-sequence) to frequency modulation chirp (FM chirp) pulses is one of the possible solutions. In this paper, we propose to optimize the PN-sequence with a genetic algorithm (GA) to improve the range ambiguity suppression in SAR images of multiple swaths compared to the existing M-sequence. The imaging simulations show that the proposed method succeed in separating range ambiguous signals from multiple swaths and suppressing range ambiguity by 24.9 to 25.3 dB compared to the case using FM chirp pulse alone.

Recently, recurrent neural networks (RNNs) and the free energy principle (FEP) have been attracting much attention. FEP includes the so-called predictive coding (PC), in which the RNN input signals are coded as the difference between the RNN prediction output signals and original input signals. This paper proposes a new method for processing time-series data in RNNs, namely, time-series forecasting coding (TFC), developed by being based on the idea of PC. We consider multi-step ahead forecasting task with the following two schemes in our proposed TFC, i.e., smallest-delay feedback TFC and synchronous (temporally matching) feedback TFC. The latter is a natural extension of conventional PC. Experimental results show that the smallest-delay TFC presents the highest performance, while synchronous TFC shows comparable or lower performance, and then the conventional method indicates the lowest. This result suggests that the concept of PC is important even in time-series data processing, and that small delay is more important than synchronicity. This may have significant implications to the conceptual fundamentals of the PC and thus of the FEP.

We propose a novel spin-wave-based reservoir computing (RC) chip-and-antenna structure that utilizes two coexisting magnetostatic modes. In our structure, both backward volume spin waves (BVSW) and surface spin waves (SSW) simultaneously arise and propagate in a magnetic film, leading to enhanced nonlinearity. We name this type of setup multiple-mode coexisting (MMC) structures. First in numerical analysis of MMC structure, we examine the frequency spectrum of the output signals and quantify the degree of nonlinearity. We then perform a signal transformation task to evaluate the nonlinearity enhanced in our MMC structure. Our results show that the use of two modes significantly enhances the even and odd harmonics and achieves a high performance improvement index up to 25.4% compared to 5% in conventional single-mode cases. The results demonstrate the enhanced nonlinearity of the MMC structure, which is essential not only for improving the performance of spin-wave-based RC systems but also for enabling the practical implementation of spin-wave reservoir chips.

Suppression of range ambiguity is one of the most important issues in Synthetic Aperture Radar (SAR) observations. Range ambiguity is a phenomenon caused by scattered signals from outside of the desired observation range illuminated by preceding and/or following pulses, resulting in a blurred signal in the focused SAR image. Existing solutions such as waveform diversity-based methods and digital-beamforming (DBF) systems increase the complexity of SAR in both hardware and software domains. On the other hand, recent emergence of commercial SAR services, which apply a constellation of small satellites, demands a simpler solution. To meet this demand, we propose to solve the problem by use of the similarity between the range ambiguous signal and the observation swath signal of another acquisition. That is, we consider the case which two satellites observe the adjacent swaths independently so that an ambiguous swath for one satellite becomes the observation swath of the other one. Experimental results showed that the proposed approach makes it easy to detect and suppress strong range ambiguities.

The Goldstein-Werner (GW) filter is one of the most widely used methods to suppress noise in interferometric synthetic aperture radar (InSAR) data. However, the insufficient filtering performance in highly noisy areas remains unsolved. In this article, we propose a biquaternion Fourier domain (BQFD) filter to improve the filtering performance even in highly noisy areas. This method represents polarimetric synthetic aperture radar interferometry (PolInSAR) data in a biquaternion form to process interferometric fringes of different polarimetric channels as a single signal. By using the biquaternion Fourier transform (BQFT), the fringes can be effectively separated from noise due to the high correlation of the fringes and the low correlation of noise between the polarimetric channels. In the experiments, we find that the BQFD filter produces more accurate digital elevation models (DEMs) than those produced with the GW filter. We also find that the accuracy of DEMs can be further improved by choosing an axis of the BQFT adaptively in each filtering window.

Quaternion neural networks (QNNs) form a class of neural networks constructed with quaternion numbers. They are suitable for processing 3-D features with fewer trainable free parameters than real-valued neural networks (RVNNs). This article proposes symbol detection in wireless polarization-shift-keying (PolSK) communications by employing QNNs. We demonstrate that quaternion plays a crucial role in the symbol detection of PolSK signals. Existing artificial-intelligence communication studies mainly focus on RVNN-based symbol detection in digital modulations having constellations in complex plane. However, in PolSK, information symbols are represented as the state of polarization, which can be mapped on the Poincare sphere and thus its symbols have a 3-D data structure. Quaternion algebra offers a unified representation to process 3-D data with rotational invariance and, therefore, it keeps the internal relationship among three components of a PolSK symbol. Hence, we can expect that QNNs learn the distribution of received symbols on the Poincare sphere with higher consistency to detect the transmitted symbols more efficiently than RVNNs. We compare PolSK symbol detection accuracy of two types of QNNs, RVNN, existing methods such as least-square and minimum-mean-square-error channel estimations, as well as detection knowing perfect channel state information (CSI). Simulation results including symbol error rate show that the proposed QNNs outperform the existing estimation methods and that they reach better results with two to three times fewer free parameters than the RVNN. We find that QNN processing will bring practical use of PolSK communications.

We propose a non-destructive method to estimate dielectric constant distribution based on radar image focusing. Water leaks from pipes and deterioration of materials in buildings have a significant impact on building users. However, they are often covered with materials, and periodic inspection is time-consuming. Electromagnetic radar is capable of detecting material changes since the dielectric constant significantly changes where the material is deteriorated or flooded. Several methods have been proposed to estimate a uniform dielectric constant of materials by using electromagnetic radar. In this paper, we obtain spatially-changing constant distribution by using buried objects such as pipes and rebars in such a manner that their radar image obtain a better focus, which means that the objects are clearly visualized.

We propose a feature synthesis and extraction method by focusing on target phase for weakly scattering objects. This method is capable of distinguishing weakly scattering landmines from clutter. Landmines left in various regions in the world have been a serious problem, and mine detection for their clearance is an important issue. Along with techniques for detecting buried objects, determining whether a detected object is a landmine or not is also crucial for efficient clearance. Various methods for distinction of landmines have been proposed, but it has been a difficult task. In this paper, we propose a method to extract features of weakly scattering buried objects such as plastic landmines by synthesizing complex-valued scattering information and excluding phase rotation due to propagation.

Precise classification of glacier surfaces is crucial for glacial health monitoring. Conventionally, in multispectral optical remote sensing, normalized indices have been used for glacial surface classification. These indices are obtained empirically by observing the difference in spectral reflectance of the target in specific bands. However, as the number of satellites, sensors and observation bands increase, there is a need for a more automated method for determining application-specific normalized indices. In this paper, we propose the use of all the bands of Sentinel-2 optical sensor for generating multiple normalized indices and the determination of application-specific significant indices by using inverse mapping. We use the normalized indices for pixel-by-pixel classification of a glacial region and observe an overall 84.81% accuracy compared to the ground truth data. Then, we apply inverse mapping dynamics to the classification results to discover new indices useful for glacier classification.

In this paper, we propose the attention-based reservoir computing and conduct an experiment of InSAR phase filtering. We show that the proposed method can estimate the wrapped phase values more accurately compared with a conventional method using only reservoir computing networks.

As the resolution of polarimetric synthetic aperture radar (PolSAR) and polarimetric interferometric SAR (PolInSAR) data increases, coherent processing of polarization and phase information becomes more important in wide applications including land surface classification and digital elevation model (DEM) generation. In this paper, we propose biquaternion neural networks (BQNNs) to process polarization and phase information pixel by pixel. We first propose a biquaternion neuron model. Then, we derive the backpropagation of BQNNs using Wirtinger derivatives. Finally, we show that our proposed neuron model can directly transform physical scattering mechanisms into different scattering mechanisms in a coherent manner, which is essential for feature extraction and their effective combinations.

Ground penetrating radar (GPR) based landmine detection has advantages such as high safety and high efficiency. There are various methods to process the data obtained from GPR systems. One of the common methods is Riemannian quaternion self-organizing map (RQSOM), which can effectively make the polarization data self-organize for visualization. However, RQSOM cannot take into account the spatial degree of polarization (DoP) of the extracted data. Spatial DoP contains useful information for landmine visualization. To overcome the limitation, in this paper, we propose a novel algorithm, generalized Riemannian quaternion self-organizing map (GRQSOM), which utilize both polarization and spatial DoP during self-organization. Thus, better visualization performance can be obtained. We conduct experiments for the visualization of a mock landmine. The experimental results show that GRQSOM achieves our design and gets better visualization results compared with RQSOM.

To extend the observation coverage of synthetic aperture radar (SAR), existing methods cost spatial resolution and/or power consumption. In this paper, we propose to separate range ambiguous signals by applying Pseudo Noise sequence (PN-sequence) to Frequency Modulation chirp (FM-chirp) pulses and observe multiple observation swaths. Experimental results show that the proposed method successfully separates multiple range ambiguous signals. Compared with the cases using the traditional FM-chirp pulses, the proposed method suppresses range ambiguities 21.6 to 23.8 dB.

This paper proposes a physical-device-based time-domain fading channel prediction scheme using spin-wave reservoir computing. We numerically construct a spin-wave reservoir chip that adopts new spin-wave transducers named Film-penetrating transducers (FPTs). We arrange three FPTs as exciters and forty-nine FPTs as detectors connected to the reservoir input and readout, respectively. We feed communication channel information collected in an actual fading environment to evaluate the prediction performance. We calculate the average root mean squared errors (RMSEs) and obtain the symbol error rates (SERs) for various forecasting time lengths. We find that our proposed scheme can achieve accurate channel prediction without any frequency-domain conversion. We obtain robust communication performance up to a forecasting time length of 8 ms. These results suggest a high capability of spin-wave reservoir computing in the application of channel prediction and its promising potential in other possible time-sequential computational tasks.

This paper proposes quaternion-neural-network (QNN) based robust symbol detection in wireless polarization-shift-keying (PolSK) communications. In these years, many ma-chine learning approaches based on real-valued neural-networks (RVNNs) outperform conventional methods for symbol detection in wireless fading channels. However, existing studies mainly focuses on digital modulations having constellation diagrams in complex planes, such as phase shift keying and amplitude phase shift keying. In PolSK, information symbols are represented as the state of polarization (SOP) of the propagating wave. Since a SOP can be described by Stokes parameters and mapped on the Poincare sphere, its symbols have a three-dimensional (3-D) data structure. Quaternion-algebra expressions offer a unified representation to process 3-D data with rotational invariance and thus it keeps the internal relationship among three components of a PolSK symbol. Hence, QNNs learn the distribution of received symbols on the Poincare sphere with higher consistency and detect the transmitted symbols more efficiently than RVNNs. We compare PolSK symbol detection accuracy of two types of QNNs, RVNN and two conventional channel estimation methods, namely least square and minimum mean square error. Simulation results such as symbol error rates show that the proposed QNNs outperform other adaptive symbol detection methods robustly. We find that the QNN processing brings practical use of PolSK communications.

Non-line-of-sight (NLOS) optical wireless communication (OWC) based on atmospheric scattering attracts more and more attention due to their mitigation of pointing, acquisition, and tracking requirement. However, the channel relies on scattering radiation and thus causes significant dispersion and inter-symbol interference (ISI), which limits high data rate communications. Reservoir computing (RC) is a computational framework derived from recurrent neural networks (RNNs). It is suitable for sequential data processing with low training cost. In contrast to multilayer perceptron (MLP)-based symbolwise detectors that treat ISI as noise, a RC performs sequence detection that takes ISI into account. In this paper, we propose to employ the RC approach to detect transmitted symbols of NLOS OWC systems without knowing channel state information. Our simulations show that the RC can perform symbol detection in various channel conditions and it brings lower bit error rate results than the MLP-based detection.

Ground penetrating radar (GPR) based landmine detection has advantages such as high safety and high efficiency. There are various methods to process the data obtained from GPR systems. One of the commonly used methods is to visualize the quaternion-type polarization data by quaternion self-organizing map (QSOM). However, QSOM can not take into account the geometric property of the polarization data. Then, degree-of-polarization (DoP) error is introduced in the self-organization process of QSOM, which leads to unsatisfactory visualization results. To overcome the limitation, in this paper, we propose a novel processing, Riemannian quaternion self-organizing map (RQSOM), which takes into account the geometric property to eliminate the DoP error with the help of logarithmic and exponential maps in Riemannian geometry. We analyze its basic dynamics and compare it with that of QSOM. We conduct experiments to visualize a mock plastic landmine with QSOM and RQSOM. The experimental results show that RQSOM realizes expected self-organization dynamics, thereby achieving better visualization results compared to QSOM.

This paper proposes model-based homogeneity (MBH) to extend compressed sensing (CS) for detection of subsurface objects. The scattered waves at landmines have features spe-cific to their material and structure. The features spread over a landmine, showing its shape. We calculate the MBH value from spatial distribution of scattering feature vectors by using a model having the landmine shape. We can utilize CS with MBH because the result of MBH calculation is sparse in general. This method eliminates clutter and reduces mea-surement points. Experiments demonstrate that this method makes the time for detection one-twentieth of that in a conventional method.

With the rise in the number of remote-sensing satellites in the past decade, interest has been drawn towards the fusion of satellite data. This study investigates the fusion of features obtained from the L-band ALOS-2 Synthetic aperture radar (SAR) and the Sentinel-2 optical satellite by the use of neural networks. It identifies the prominent features by using an inverse-mapping algorithm. The fusion results show an increased classification accuracy compared to that of independent sensors. The algorithms have been tested on the data of the 2018 earthquake in Sulawesi, Indonesia.

In this paper, we propose an interferometric synthetic aperture radar (InSAR) phase unwrapping method by use of dynamics of the Kuramoto model that is one of phase oscil-latory networks. The network generates a quality map which induces a proper unwrapping path by solving singular points (SPs) or residues. We show that our proposal unwraps in-terferometric phase with higher accuracy compared to other model-algorithm-based unwrapping methods.

Quaternion convolutional neural networks (QCNNs) are in-herently useful for image processing and PolSAR land classification, since they can learn the relationship between the components of input vectors with quaternionic rotations. However, conventional QCNNs fix their rotation axes represented by quaternion weights, resulting in reduction of the degree of freedom (DoF) and the lose of the expression ability. In this paper, we propose QCNNs which learn all the four parameters of the quaternion weights by backpropagation. They perform learning with the maximum of DoF so that they take full advantages of quaternion learning. In addition, we use two totally different features, namely, Pauli RGB features and normalized Stokes vectors. We experimentally found that Pauli RGB features are suitable for discrimination between town and forest, and Stokes vectors between water and grass. Combining their two results complementarily improves classification results. Our proposed QCNNs show the best classification performance compared with real-valued convolutional neural networks and fixed-axis QCNN. These results demonstrate the strength of the proposed QCNN in adaptive polarization processing in multimodal data in the PolSAR field.

Synthetic Aperture Radar (SAR) has been sharing its radio band with other systems. Similar to recent Dynamic Spec-trum Sharing (DSS) in switching from 4G to 5G teleCommunication system, future SAR will be required to share the spectrum with more systems than ever. However, current state-of-the-art Radio Frequency Interference (RFI) detection and suppression methods are mostly based on the spec-trum analysis, which survey the amplitude of the signals in the frequency domain and extract anomalistic signals from backscattered echoes. In order to enable further dynamic and coherent RFI detection and suppression, we performed a polarimetric analysis of SAR raw data and classification of the interfering signals in polarimetric domain.

Mobile communications are often affected by various fading phenomena. Since channel state information (CSI), important to overcome channel fading, is always changing over time, channel prediction is necessary to know the correct CSI. Current channel prediction methods are mainly based on the amplitude and phase of the received signal. However, because the polarization state of the propagating wave also changes with time, the polarization mismatch problem occurs between an arriving wave and a receiving antenna, resulting in lower communication quality. In this paper, we propose the prediction of polarization states by using a quaternion neural network (QNN), which makes full use of the geometric properties of polarization states represented on the Poincare sphere to achieve a high-precision prediction. The experimental results show that, by predicting the polarization state, we can obtain more accurate CSI and a lower bit error rate.

We propose a spin-wave antenna structure that penetrates a garnet film, which we named film-penetrating transducers (FPTs). FPTs possess a zero-dimensional feature that allows flexible placements and sufficient numbers of input/output electrodes for spin-wave reservoir computing. We first explain the structure and operation of FPTs. Then, we numerically construct and analyze a basic spin-wave reservoir chip model. We obtain intricate patterns of spin-wave propagation and interference that reflect the high dimensionality of the system. We also demonstrate the nonlinearity of the output electrical signal from an FPT detector of the system. Our results strongly suggest that FPTs preserve the important properties of a physical reservoir while ensuring large degrees of freedom for both the arrangements and amounts of spin-wave exciters and detectors.

In this paper, we propose reconstructive reservoir computing (RRC) to detect anomaly in time-series signals. In the RRC, an echo state network (ESN) learns to reconstruct normal input signals fed to its input terminals. Since it fails to reconstruct abnormal signals, RRC can detect anomaly based on its reconstruction error. It is shown for the first time that an ESN reconstructs time-series signals effectively for anomaly detection while we already know that it can realize anomaly detection by forecasting. Experiments demonstrate that the RRC works for anomaly detection effectively. Though forecasting errors are used in conventional methods for anomaly detection working for time-series signals, we find experimentally that the reconstruction method has an advantage in its larger margin between normal and abnormal errors. We also find that a smaller leaking rate enhances the ability of anomaly detection. In general, reservoir computing has merits of fast training and, consequently, less energy consumption. We can also make reservoir computing work with the use of physical phenomena. Utilizing reservoir computing is really meaningful for these aspects which layered or other types of recurrent neural networks do not have. Anomaly detection by RRC will be an important application of micro physical-reservoir devices in the near future.

This paper proposes model-based homogeneity (MBH) to extend compressed sensing (CS) for landmine-detection ground penetrating radar (GPR). Conventional CS methods have difficulty in distinguishing landmines from clutter since it principally pays attention to signal magnitude. In contrast, our method visualizes landmines based on homogeneity of high-dimensional scattering features in a spatial model. It realizes both the exclusion of clutter and the reduction of measurement points. Experiments demonstrate that the total measurement and processing time is reduced to one-twentieth of a conventional dense measurement case. We also investigate the influence of model size and number of landmines on the performance. The proposed method is capable of visualizing any objects having respective shapes by configuring corresponding models.

Visualization by ground penetrating radar (GPR) systems has a wide demand in various application fields. Our conventional visualization method mainly extracts feature vectors based on the frequency and spatial correlation of scattering parameters and then uses complex-valued self-organizing maps (CSOM) in unsupervised grouping to achieve visualization. However, this method sometimes shows problems of unclear boundary and deformed shape of a detected object. In this article, we propose a novel algorithm, phasor-quaternion self-organizing map (PQSOM), based on the Poincare vectors combined with phase, which is organized in phasor-quaternion (PQ) form. By considering both the polarization state and the phase information as respective counterparts, we achieve a more accurate grouping result. Through visualization experiments of underground landmines, we find that, compared with the previous methods, the boundary clarity and the shape of objects are improved.

With the penetration of artificial intelligence (AI) technology into industrial applications, not only computational effectiveness but also computational efficiency in machine learning (ML) methods has been increasingly demanded. Reservoir computing (RC) is an ML framework leveraging a dynamicreservoirfor a nonlinear transformation of sequential inputs and areadoutfor mapping the reservoir state to a desired output. Since only the readout is trained with a simple learning algorithm, RC has attracted much attention as a promising approach to enhance compatibility between high computational performance and low learning cost. In addition, recent studies on physical reservoirs implemented with various physical substrates have boosted the potential of RC in the development of effective and efficient AI hardware. Therefore, it is time to further explore the new frontiers in extremely efficient RC.

In this paper, we propose complex-valued reservoir computing (CVRC) to deal with complex-valued images in interferometric SAR (InSAR). We conducted classification of land forms by dealing with the interferogram data as sequential pixel values with nonlinearity in the complex domain.

This paper proposes a quaternion convolutional neural network (QCNN) for PolSAR land classification. The QCNN learns spatial features of polarization with quaternion convo-lutional layers. The QCNN learns relationship between components of input three-dimensional vectors. This property also makes the QCNN map inputs to space of higher dimension with the same number of parameters than a real-valued CNN (RVCNN). In our experiments, the QCNN shows better classification performance than conventional networks. We also present visually that the quaternion kernels extract spatial features by quaternionic convolution.

Previously we proposed a ground penetrating radar (GPR) system employing complex-valued self-organizing maps (CSOM) to deal with feature vectors based on the frequency and spatial correlation of scattering parameters. It performs unsupervised grouping to achieve visualization. However, this method sometimes shows problems of unclear boundary and shape distortion of a detected object. In this paper, we propose a novel method, phasor quaternion self-organizing map (PQSOM), based on the Poincare vectors combined with phase, which are organized in phasor quaternion (PQ) form. By considering both the polarization state and the phase information as respective counterparts, PQSOM is able to self-organize more effectively, which in turn leads to better grouping performance. Through visualization experiments of landmines, we find that PQSOM has better grouping results compared with CSOM and QSOM.

Complex-valued convolutional neural networks discover and/or adaptively classify local features in interferograms very effectively in interferometric synthetic aperture radar (InSAR). In this paper, we investigate the influence of label errors in teacher images on the classification performance. We find that performance is not affected so much from teacher-label errors as much as 10% or more. We also analyze the error characteristics experimentally.

This paper investigates the spatial distribution of information effective for function learning in a spin-wave reservoir-computing garnet chip. We map the neural weights of a readout neuron virtually connected massively and densely to the reservoir chip. We find that the spatial weight distribution shows wavefront-like lines, suggesting the importance of concurrent and time-different interferences of the spin waves. We also estimate the size of reservoir output electrodes required for the proper information extraction. These results are significantly useful for designing spin reservoir chips in the near future energy efficient devices.

In a complex-valued convolutional neural network, its elementary unit consists of a complex-valued convolution layer and a complex pooling layer. The pooling layer has a variety in its dynamics. In this paper, we propose complex absolute-value max pooling to extract complex-amplitude feature patterns meaningful for discovery and/or adaptive classification of land form in interferometric synthetic aperture radar (InSAR). Experimental examination into amplitude and phase values in convolutional kernels reveals that useful land-shape features emerge through self-organization in high-magnitude kernels, which suggests that the proposed dynamics is successful in extracting important features.

This letter proposes a data arrangement for fully polarimetric synthetic aperture radar (PolSAR). It is an essential novel method in the use of the rotation transformation in data interpretation. The key point of the proposal is employing a single pixel-based and selective rotation transformation for each pixel before the speckle filtering. The experimental results with ALOS2-PALSAR2 data show that the proposed data arrangement has much higher performance in recognizing double-bounce scattering in the man-made target area. At the same time, it is effective in avoiding the overestimation of double-bounce and/or surface scattering in natural target areas.

We propose an adaptive subsurface 3-D visualization system based on a complex-valued self-organizing map (CSOM). Conventionally buried things can be detected in the so-called B-scan images obtained by a ground penetrating radar. In contrast, our proposed method is able not only to detect their presence but also to classify the targets by the self-organizing dynamics in the CSOM. Instead of utilizing only the amplitude information in the time domain, we use both the amplitude and the phase information to obtain the scattering coefficients of scatterers by use of the phase retrieval method.

The development of hardware neural networks, including neuromorphic hardware, has been accelerated over the past few years. However, it is challenging to operate very large-scale neural networks with low-power hardware devices, partly due to signal transmissions through a massive number of interconnections. Our aim is to deal with the issue of communication cost from an algorithmic viewpoint and study learning algorithms for energy-efficient information processing. Here, we consider two approaches to finding spatially arranged sparse recurrent neural networks with the high cost-performance ratio for associative memory. In the first approach following classical methods, we focus on sparse modular network structures inspired by biological brain networks and examine their storage capacity under an iterative learning rule. We show that incorporating long-range intermodule connections into purely modular networks can enhance the cost-performance ratio. In the second approach, we formulate for the first time an optimization problem where the network sparsity is maximized under the constraints imposed by a pattern embedding condition. We show that there is a tradeoff between the interconnection cost and the computational performance in the optimized networks. We demonstrate that the optimized networks can achieve a better cost-performance ratio compared with those considered in the first approach. We show the effectiveness of the optimization approach mainly using binary patterns and apply it also to gray-scale image restoration. Our results suggest that the presented approaches are useful in seeking more sparse and less costly connectivity of neural networks for the enhancement of energy efficiency in hardware neural networks.

In this paper, we report the results of the experimental analysis observing the actual pixel variation properties in PolSAR data having various resolutions. Present PolSAR has reached a decimeter-level high resolution. In general, the resolution of PolSAR data is lowered down to 10m-20m in the real space by performing the multi-look process to reduce noise in the pixel values widely for land classification. However, lowering resolution prevents us from discovering new land classes potentially enabled by the resolution enhancement. Through our experiments, we aim to confirm whether it is really meaningful to utilize the respective pixel signals of high-resolution PolSAR data for land classification without any lowering resolution. Although the main cause of the pixel variation occurring in the actual PolSAR data has not been elucidated yet in this letter, we would like to show this experimental results as a material for discussion.

In this paper, we propose a three-port dielectric waveguide probe system for human blood glucose level monitoring. The probe has one transmission port, one detection port and one reference port. The frequency of the millimeter wave used in numerical analysis is ranged from 55 to 85 GHz. We establish a simple thin tissue model representing an earlobe. The results show the potential of using three-port waveguide probe to estimate human blood glucose level by showing more linearlized phase values than two-ports probes by utilizing the reference values. Our proposed three-port dielectric waveguide probe presents the feasibility of using millimeter wave to detect the slight changes in human blood glucose at a realistic concentration level.

Search-and-rescue radar systems are expected to become more powerful and wider applicable in many disasters such as earthquakes and tsunamis in these days. They should realize not only the detection but also the separation of vital signals from multiple lives under debris. This paper proposes a radar system based on complex-valued frequency-domain independent component analysis (CF-ICA) to distinguish respective victims. Its microwave part working as a continuous-wave single-input multiple-output Doppler radar consists of switched multiple antennas and vector network analyzer for orthogonal detection to generate complex-valued baseband signals. Its processing part is a CF-ICA to realize signal source separation. A preliminary pendulum target experiments demonstrate that the system detects and separates multiple target signals successfully.

When we process PolSAR data by machine learning to analyze vegetation, it is necessary to know how polarization changes by scattering. In order to clarify the physical mechanisms of scattering in PolInSAR (polarimetric interferometric synthetic aperture radar), we examine how polarization changes by PPO-BD (Pixel-by-Pixel Optimization considering Baseline Difference), which is an adaptive filter of InSAR images. By using scattering sphere we propose, we find that the features of the scattering mechanisms are enhanced after polarization is optimized by PPO-BD. It means that PPO-BD performs the phase optimization by taking into account the differences in the scattering mechanisms among pixels. We find that, when PPO-BD optimizes the phase to reduce singular points by utilizing polarimetric data, it also emphasizes the polarimetric mechanism information pixel by pixel.

We propose an adaptive subsurface three-dimensional visualization system based on a complex-valued self-organizing map (CSOM) to deal with time-domain phase information. Conventionally phase information of time domain is regarded as meaningless in radar processing. We think that amplitude information of time domain presents the position of targets and phase information depends on kinds of scatterers. In this paper, we show time-domain phase information is very valid for clustering of targets. We succeeded in clustering adaptively targets by employing complex-valued self-organizing map.

Reservoir computing is a computational framework which is originally based on software recurrent neural networks and recently achieved with physical systems as well. In our previous paper [Nakane et al., IEEE ACCESS vol. 6, p. 4462, 2018], we have proposed a spin-wave-based reservoir computing device with multiple input/output electrodes, and have demonstrated its high generalization ability in the estimation of input-signal parameters performed by the spin-wave-based reservoir computing. To successfully execute many types of estimation tasks with machine learning, it is necessary to investigate fundamental properties of spin-wave-based reservoir computing, particularly the relation between its input and output. From this background, the purposes of this work are to demonstrate a different estimation task with pulse input signals and to analyze the properties of spin waves which have important roles in the task. We first describe our approach to obtain spin waves with the features useful for reservoir computing, by considering the fundamental properties of spin waves and feasible device technologies. Then, we investigate detailed characteristics of locally-excited spin waves in a garnet film by micromagnetics simulation. Using the resultant spin waves, we demonstrate a pulse interval estimation task, and achieve a high diversity in the time-sequential signals generated by the spin-wave-based reservoir. The spin-wave-based device is a highly promising hardware for next-generation machine-learning electronics.

Reservoir computing is a computational framework suited for temporal/sequential data processing. It is derived from several recurrent neural network models, including echo state networks and liquid state machines. A reservoir computing system consists of a reservoir for mapping inputs into a high-dimensional space and a readout for pattern analysis from the high-dimensional states in the reservoir. The reservoir is fixed and only the readout is trained with a simple method such as linear regression and classification. Thus, the major advantage of reservoir computing compared to other recurrent neural networks is fast learning, resulting in low training cost. Another advantage is that the reservoir without adaptive updating is amenable to hardware implementation using a variety of physical systems, substrates, and devices. In fact, such physical reservoir computing has attracted increasing attention in diverse fields of research. The purpose of this review is to provide an overview of recent advances in physical reservoir computing by classifying them according to the type of the reservoir. We discuss the current issues and perspectives related to physical reservoir computing, in order to further expand its practical applications and develop next-generation machine learning systems.

In this paper, we propose a hierarchical polarization feature generation using a self-organizing codebook to realize unsupervised land classification that fully utilizes the detailed polarization information contained in high-resolution polarimetric synthetic aperture radar (PolSAR) data. PolSAR has reached a decimeter-level high resolution. In general, conventional methods lower the resolution of the PolSAR data to 10-20 m in the real-space distance to classify observation regions into land classes such as farm, forest, and town. However, lowering resolution prevents us from discovering new land classes potentially enabled by the resolution enhancement. The hierarchical method we propose here not only classifies observation regions successfully into land classes such as farm, forest, and town that humans can naturally distinguish but also discovers new land subclasses findable only in high-resolution PolSAR data. We explain these two types of our achievements (classification/discovery) through experimental results for Japan Aerospace Exploration Agency's polarimetric and interferometric airborne SAR-L2 data having decimeter resolution.

Interferograms obtained by synthetic aperture radar often include many singular points (SPs), which makes it difficult to generate an accurate digital elevation model. This paper proposes a filtering method to compensate SPs adaptively by using polarization and phase information around the SPs. Phase value is essentially related to polarization changes in scattering as well as propagation. In order to handle the polarization and phase information simultaneously in a consistent manner, we define a new number, phasor quaternion (PQ), by combining quaternion and complex amplitude, with which we construct the theory of PQ neural networks (PQNNs). Experiments demonstrate that the proposed PQNN filter compensates SPs very effectively. Even in the situations where the conventional methods deteriorate in their performance, it realizes accurate compensation, thanks to its good generalization characteristics in integrated Poincare-sphere polarization space and the complex-amplitude space. We find that PQNN is an excellent framework to deal with the polarization and phase of electromagnetic wave adaptively and consistently.

This paper proposes a novel data filtering algorithm for fully polarimetric synthetic aperture radar (PolSAR) based on the degree of polarization (DoP) information. First, we define the homogeneity degree and polarization independence degree using the DoP information, and propose a feature plane to characterize the target feature. Second, employing the feature plane, we categorize the targets into three types and assign specific filtering policy for each type to estimate the optimal filtering window sizes. Finally, the $T$ -matrices of fully PolSAR data are filtered using the windows with estimated optimal sizes. Compared with boxcar filter, refined Lee filter, scattering model-based filter, and improved sigma filter in processing ALOS2-PALSAR2 data, the proposed DoP-based algorithm presents the best filtering performance.

This paper proposes a complex-valued convolutional neural network for land form classification and discovery in interferometric synthetic aperture radar (InSAR). Since the amount of satellite-borne SAR data has been increasing drastically, it is necessary to structurize the local features contained in observation data prior to utilization in the so-called big data framework for higher usability. Convolutional neural networks have such potential in general. However, there exists no network that can deal with complex amplitude data obtained in InSAR consistently. In this paper, we propose a complex-valued convolutional neural network to deal with InSAR. We demonstrate that the network classifies slopes and plains adaptively and, moreover, indicates small volcanos similar to a sample volcano (Omuroyama) included in the InSAR data. We also find their characteristic features emerging in the kernels in the convolution layers. These results reveal that the proposed complex-valued convolutional neural network is capable of successfully discovering unidentified lands similar to a prepared sample, which is highly useful for the InSAR data structurization.

As you open this January issue of the IEEE Transactions on Neural Networks and Learning Systems (TNNLS), I hope everyone enjoyed a great holiday season and is excited for the new year of 2019. I am very delighted and honored to report several key metrics of IEEE TNNLS to the community.

This paper proposes an adaptive glucose-concentration estimation system. Diabetes prophylaxis and monitoring require long-term, frequent and accurate observation of blood sugar levels. Electromagnetic measurement is ideal in its non-invasiveness and the possibility of high accuracy. Debye relaxation model indicates that millimeter-wave is promising for this purpose. In this frequency band, however, both phase and magnitude changes have significant meaning. Then, we employ a complex-valued neural network that deals with phase and amplitude information adaptively in a consistent manner. Experiments demonstrate effective estimation of glucose concentration of a realistic range of blood sugar.

We propose an unsupervised PolSAR land classification system consisting of quaternion auto-encoder and quaternion self-organizing map (SOM). Most of the conventional methods extract features necessary for the land classification based on a few of scattering models predefined by human beings. However, we can not expect classification into a large number of land categories recognizable to humans by using such restricted features. In this paper, we propose a method employing quaternion auto-encoder and quaternion SOM for feature extraction and classification, respectively. As a result, we succeed in discovering new and more detailed land categories. For example, town areas are divided into residential areas and factory sites.

We propose a codebook-based hierarchical polarization feature vector generation to realize an unsupervised land classification with high-resolution PolSAR data. PolSAR has reached the high-resolution of decimeter level. Conventional methods perform the spatial averaging in 10m to 20m square real-space area to classify observed land into categories such as farm, forest, and town. However, with this averaging, we can not expect to discover new detailed land classes by resolution improvement, since the resolution of the PolSAR data is lowered in the averaging process. Our proposal in this paper generates feature vectors useful for classifying the land pieces into categories while preserving the detailed polarization features in respective pixels of the high-resolution PolSAR data. Then, the method discovers the new detailed land classes that can be available only in the high-resolution PolSAR data.

Synthetic aperture radar (SAR) shares its radio frequency with the other systems and thus, inevitably receives their signals. Such cross-talk, radio frequency interference (RFI) has been traditionally eliminated by spectrum filters. However, a latest L-band SAR, ALOS-2, receives frequency modulated wideband signals that cannot be detected by traditional spectrum filters. These RFIs have the same or wider bandwidth than SAR and thus, appears in SAR images. They also interrupt interferometric (InSAR) and polarimetric (PoISAR) analysis because it appears in every observations and in full-polarimetric mode images. We report the investigation results of the profiles of such RFI in Tokyo bay, Japan.

In this paper, we examine the potential of interferometric analysis using L-band synthetic aperture radar (SAR) for flood monitoring. The integration of amplitude and interferometric coherence is one of the novel methods for flood monitoring in urban area. However, its accuracy has not been evaluated especially in case of L-band SAR data. Here, we used the 2015 September flood in Joso city, Ibaraki prefecture, Japan for the evaluation of the latest L-band SAR satellite ALOS-2 PALSAR-2.

This paper proposes a filtering method to compensate phase singular points (SPs) adaptively by using phase and polarization information around the SPs simultaneously. Phase value is essentially related to polarization changes in scattering as well as propagation. To handle phase and polarization information simultaneously in a consistent manner, we define a new number, phasor quaternion (PQ), by combining complex amplitude and quaternion, with which we construct a theory of phasor quaternion neural networks (PQNNs). Experiments demonstrate that the proposed PQNN filter compensates SPs very effectively.

Recently we proposed a singular-unit restoration filter based on complex-valued neural networks (CVNN) that deal with spatial spectrum in interferometric synthetic aperture radar. We named it entire-spectrum-processing CVNN (ESP-CVNN) filter. This filter utilizes more neural generalization ability than other conventional methods. It shows a higher performance with a smaller or almost the same processing time. In this paper, we analyze the relationship between its performance and geographic conditions to be processed. Experiments reveal that the ESP-CVNN filter is superior to conventional filters in particular when an observation area has higher density of phase singular points.

Quaternion neural networks (QNNs) achieve high accuracy in polarimetric synthetic aperture radar classification for various observation data by working in Poincare-sphere-parameter space. The high performance arises from the good generalization characteristics realized by a QNN as 3-D rotation as well as amplification/attenuation, which is in good consistency with the isotropy in the polarization-state representation it deals with. However, there are still two anisotropic factors so far which lead to a classification capability degraded from its ideal performance. In this letter, we propose an isotropic variation vector and an isotropic activation function to improve the classification ability. Experiments demonstrate the enhancement of the QNN ability.

Interferograms obtained in interferometric synthetic aperture radar (InSAR) often suffer from decorrelation and singular points (SPs) originating from thermal noise and interference. To analyze the phenomenon, first, this paper presents the results of scaled optical experiment free from thermal noise, where the SP origin is interference. We find that the amplitude of the SP-constructing pixels, namely, singular unit, and of nearby pixels is lower than that of other pixels. This amplitude reduction is enhanced by multilooking process. These results suggest that the number of effective scatterers in a single pixel has reduced to such an extent that individual interference has become visible. We also conduct the same analysis on the SAR data. We find that plain areas show the same features as the optical experiment, implying the same mechanisms of SP generation. In contrast, sea areas present no localization, indicating thermal noise in electronics as the major reason. It is widely known that interference among many incoherent scattered waves presents Rayleigh or similar distribution in its amplitude as a result of central limit theorem. As the number of scatterers reduces, the amplitude becomes log-normal or other distribution. However, no analysis was reported on the local properties in such a case that the central limit theorem does not hold. Investigation of such local properties will also be useful in designing SP filters. The significance of noncentral-limit-theorem situations will increase its importance in the use of SAR data, of which resolution becomes further higher in the near future.

We propose an unsupervised polarimetric synthetic aperture radar (PolSAR) land classification system consisting of a series of two unsupervised neural networks, namely, a quaternion autoencoder and a quaternion self-organizing map (SOM). Most of the existing PolSAR land classification systems use a set of feature information that humans designed beforehand. However, such methods will face limitations in the near future when we expect classification into a large number of land categories recognizable to humans. By using a quaternion autoencoder, our proposed system extracts feature information based on the natural distribution of PolSAR features. In this paper, we confirm that the information necessary for land classification is extracted as the features while noise is filtered. Then, we show that the extracted features are classified by the quaternion SOM in an unsupervised manner. As a result, we can discover even new and more detailed land categories. For example, town areas are divided into residential areas and factory sites, and grass areas are subcategorized into furrowed farmlands and flat grass areas. We also examine the realization of topographic mapping of the features in the SOM space.

In this paper, we propose two methods to optimize scattering mechanism vectors in polarimetric interferometric synthetic aperture radar (InSAR) dealing with high-resolution SAR data. The methods optimize an interferogram by focusing on the pixel-by-pixel variation of the scattering mechanisms. We show the efficacy of the methods using fully polarimetric interferometric radar data obtained from the Advanced Land Observing Satellite-2 PALSAR-2 to generate accurate digital elevation models successfully. Finally, we confirm the importance of considering the polarization in such InSAR systems by comparing parameters optimized in the proposed methods with those calculated from the well-known Pauli coherency matrix.

In this paper, we first determine the optimal window size for ALOS2-PALSAR2 data is 7 × 7. To preserve the accuracy of incoherent interpretation and the high resolution of original data, simultaneously, we proposed the combination use of various window sizes in Stokes vector based data interpretation. The experimental results show that the proposed method can provide interpretation results in success and can preserve much more target details than conventional fixed window size method.

We propose a method for generating precise digital elevation models from high resolution SAR data in PolInSAR, focusing on the pixel-by-pixel variety of scattering mechanisms. We show the effectiveness of the proposed method by using ALOS-2 SAR data.

Interferograms obtained by interferometric synthetic aperture radar generally include many singular points (SPs) originating from interference distortion and noise in the measurement. The filtering process is one of the key techniques in the generation of an accurate digital elevation model (DEM). In this paper, we propose a filter to remove SPs and related distortion using a complex-valued neural network in the spectral domain. It removes SPs nonlinearly and adaptively by referring to the statistics in neighboring windows in the 2-D frequency domain, where the textural features are represented in a more continuous manner than in the real space domain. Experiments demonstrate that the proposed method removes SPs and the distortion in SP-constructing four pixels, namely, the singular unit, more effectively than the conventional filters, resulting in the generation of a more accurate DEM.

Filtering process plays significant roles in the generation of a digital elevation model (DEM) from interferogram obtained by interferometric synthetic aperture radar (InSAR). In order to remove the distortion of a so-called singular unit (SU), this paper proposes two novel filtering techniques which both exhibit strong nonlinearity. The first method attempts to remove the distortion by focusing on amplitude peaks in the spectrum. The second method eliminates the distortion by using a complex-valued neural network, a network that learns the correlation between the spectrum in the window with no SU and those of shifted ones around it. Both filters are so effective in the removal of the distortion at the SU, that it allows us to generate a highly accurate DEM.

Previously, we have proposed a successful land classification method using a quaternion neural network (QNN) to process parameters based on Stokes vector representation. In this method, the activation function used in the QNN is anisotropic and is applied to input quaternions of which elements are separately and independently processed. In this paper, considering the isotropy of Poincare-sphere space, we propose a new isotropic activation function. Experimental results show that the QNN with the proposed activation function achieves more accurate land classification.

In this paper, we propose an effective wet snow mapping method with focus on the incident angle of microwave. Surface scattering is dominant for both wet snow and bare ground. However, it is expected that the characteristic of the wet snow scattering is different from the bare ground one according to the variation of dielectric constant. At the same time, surface scattering characteristics, especially depolarization, also depend on the incident angle. First, we evaluate numerically the degree of polarization of horizontal incident wave as an example with a simplified integral equation model (IEM). We also examine real data of full polarimetric synthetic aperture radar (PolSAR). The results shows that the degree of polarization depends on the difference of incident angles rather that of dielectric constants. Then we conduct wet-snow mapping by supervised learning with teacher areas for large / small incident angles and snow / bare ground. The mapping result agrees well with the estimation by optical data. It is found important to take into account the incident angle in snow mapping.

Pre-equalization and adaptive diversity schemes are available for channel compensation. Such techniques require channel prediction because a channel changes in time. Previously, we reported a channel prediction technique to increase the accuracy by focusing on polarization state and combine multiple polarization components estimated in the nearest past signal set. However, when the polarization state is changing as well, this method fails. In this paper, we propose a channel prediction combined with polarization prediction on the unity-radius Poincare sphere by using a quaternion neural network. We demonstrate a higher accuracy in channel prediction, resulting in a low bit error rate.

Polarimetric synthetic aperture radar is expected to distinguish wet snow from bare ground. However, since both of them show surface scattering, which is sensitive to incidence angle, it often fails in the distinction in mountainous areas. In this letter, we propose an adaptive distinction method using quaternion neural networks. In the ALOS-2 data, we find a monotonic and nonlinear dependence of the degree of polarization on the incidence angle. Then, we feed multiple-incidence-angle teacher information in the learning process. The distinction results of the proposal present higher accuracy than those of the conventional Wishart distinction and a quaternion neural network without the incidence angle information.

In principle, we can construct highly precise digital elevation model (DEM) with interferometric synthetic aperture radar (InSAR) observations. However, in actual observed images, we find so many rotaional points, namely, phase singular points (SPs), that we cannot determine the height accurately. It is significantly important to investigate the SP phenomenon. Presently the origin of the SPs is not entirely clear. In this paper, we design a scaled experimental setup in optics to elucidate the mechanisms of the SP generation. From the results, we confirm that low amplitude pixels (dark speckles) are emphasized by multilook process, and that many of these pixels are singular unit pixels. The multilook process is equivalent to interference of multiple complex-amplitude values. In addition, a similar interference is considered to occur in a single pixel of a single-look SAR image. It is suggested that interference of waves from multiple scattering sources in a single pixel can be one of the causes of SP generation.

In this work, we discuss the effect of coordinate rotation on the proposed Stokes vector based PolSAR data interpretation algorithms. The core work for making clear the effects is finding the change regulations for the zero aperture/orientation routes and aperture/orientation triangles with coordinate rotations. We mathematically analyze and summarize the regulations. The analysis results have shown that such regulations are predicable. With these regulations, we can possibly further improve the Stokes vector based interpretation algorithms.

This paper presents a method for removing spikes in digital elevation models (DEMs) caused by residues in interferometric synthetic aperture radar (InSAR) phase image. We consider that the scattering mechanism is properly modeled by the small perturbation method for fractal surfaces and present a model that relates the phase and magnitude in InSAR image. This data model provides the regularization term of the method, without directly enforcing smooth phase or magnitude. Noise models are given by additive Gaussian for the phase and multiplicative non-unit-mean gamma for the magnitude. Experiments with simulated and real L-band data show that the proposed method considerably improves DEM accuracy and simultaneously suppresses speckle and phase noise.

This paper presents a new phase unwrapping method for synthetic aperture radar interferometry (InSAR). Compared with other phase observations, InSAR data are unique due to the foreshortening effect in SAR images. This effect makes the interferogram phase anisotropic, i.e., the statistics of phase gradients along the ground-range axis are different from those in the azimuth direction. Furthermore, the distribution of phase gradients in the ground-range direction is not symmetric. The proposed method targets the most likely unwrapped phase by considering the foreshortening effect in SAR observation, noise characteristics in InSAR phase data, and fractional Brownian surface as a suitable model for natural topography. Experiments with simulated terrains and real InSAR data show that the method gives comparable or better digital elevation model results than the other tested methods.

In this paper, we propose a new polarimetric synthetic aperture radar (SAR) data interpretation method based on a locally averaged Stokes vector. We first propose a method to extract discriminators from all three components of the averaged Stokes vector. Based on the extracted discriminators, we build four physical interpretation layers with ascending priorities, i.e., the basic structure layer, the low-coherence targets layer, the man-made targets layer, and the low-backscattering targets layer. An intuitive final image can be generated by simply stacking the four layers in the priority order. We test the performance of the proposed method over Advanced Land Observing Satellite Phased Array type L-band SAR (ALOS-PALSAR) data. Experimental results show that the proposed method has high interpretation performance, particularly for skew-aligned or randomly distributed buildings and isolated man-made targets such as bridges.

In Synthetic Aperture Radar (SAR) interferometry(InSAR), we generally expect that there is only one unique dominant scatterer in a pixel. To make a SAR interferogram, we have to observe every place twice. Between the observations, the dominant scatterer is expected to unchange. However, in actual situations, there are multiple scatterers in one pixel, and a dominant one may change observation by observation. Our main idea in this paper is that this change happens more frequently than we have expected, and that this phenomenon can generate singular points (SPs) in the SAR interferogram which prevent us from accurate phase unwrapping. Here, we present the results of the preliminary experiments using real aperture radar system, which suggest one of the mechanisms of the singular point generation.

In this paper, we elaborate several considerations on the possible factors restricting the accuracy of covariance/coherency (C/T) matrix-based unsupervised classification. Then we make an exploration on constructing Stokes vector-based unsupervised classification. The experimental results for Fujisusono area show that Stokes vector-based method can distinguish building and farmland targets more correctly. It also shows potential to distinguish vegetations with different height or thickness.

This paper reports experimental results of a wideband acoustic imaging method based on power-inversion adaptive array (PIAA) scheme realized by a complex-valued spatio-temporal neural network (CVSTNN). For acoustic imaging with a high resolution in the range direction (direction of propagation), used pulse should be short. A short pulse has a wide frequency band, which is also favorable for avoidance of target breaking through acoustic resonance. However, because of the wide bandwidth, conventional adaptive arrays often fail in beamforming or null steering. We combine a CVSTNN and PIAA to realize a precise null steering. Experiments demonstrate that the CVSTNN-PIAA method presents a higher resolution than conventional methods.

We propose a quaternion neural-network-based land classification in Poincare-sphere-parameter space. By representing the Stokes vector on/in the Poincare sphere geometrically, we construct two analysis parameters, namely, the position vector and the variation vector, to describe the feature of a pixel in test area. Then, by employing a quaternion feedforward neural network, we generate successful classification results for detecting lake, grass, forest, and town areas. In comparison with the conventional C-matrix-based methods, the proposed method has higher classification performance, especially in detecting forest and town areas. Moreover, the classification result of the proposed method is not influenced by height information. This fact suggests that the proposed classification method can be used for complicated terrains.

Channel prediction is an important process for channel compensation in a fading environment. If a future channel characteristic is predicted, adaptive techniques, such as pre-equalization and transmission power control, are applicable before transmission in order to avoid degradation of communications quality. Previously, we proposed channel prediction methods employing the chirp z-transform (CZT) with a linear extrapolation as well as a Lagrange extrapolation of frequency-domain parameters. This paper presents a highly accurate method for predicting time-varying channels by combining a multilayer complex-valued neural network (CVNN) with the CZT. We demonstrate that the channel prediction accuracy of the proposed CVNN-based prediction is better than those of the conventional prediction methods in a series of simulations and experiments.

The fifteen papers in this special issue focus on complex and hyper-complex neural network applications. Complex-valued neural networks (CVNNs) exhibit very desirable characteristics in their learning, self-organizing, and processing dynamics, which makes them attractive for applications in various areas in science and technology. For example, they are perfectly suited to deal with complex amplitude, composed of amplitude and phase, which is one of the core concepts in physical systems dealing with electromagnetic, light, sonic/ultrasonic, and quantum waves.

We propose a direction of arrival (DoA) estimation method using a complex-valued neural network (CVNN) for ultrawideband (UWB) systems. We combine a complex-valued spatiotemporal neural network with power-inversion adaptive-array scheme for null-steering DoA estimation. Simulation and experiments demonstrate that the proposed method shows an estimation accuracy higher than that of conventional multiple signal classification method and a spectrum floor lower than that of real-valued neural network. These results suggest that the CVNN deals with signals more properly as wave information in the null synthesis in UWB systems.

SAR interferogram requires a robust filtering method for generation of appropriate digital elevation model (DEM).We previously proposed a filtering method based on complex-valued Markov random field (CMRF) model. The CMRF filter estimates the pixel values which generate the singular point (SP) and by use of the complex correlation between neighbor pixels. Since a complex number is represented by two real numbers, the complex-correlation learning (i.e., CMRF filter) is similar to the real-imaginary separate double-dimensional correlation learning (RI-MRF filter). In this paper, we compare the bases of their processing dynamics and the filtering accuracy between the CMRF and RI-MRF filters in experiment for ALOS data.

We propose the use of Poincare sphere parameters for a fast supervised PolSAR land classification. The scattering matrix is represented by a point which indicates the polarization states on/in Poincare sphere. Then, by analyzing the distribution features of the points, the test area is classified into, for example, four types of targets: lake, grass, town and forest. This analyzing process can be implemented by employing a neural network. The experimental result shows that the Poincare sphere parameters are highly useful for classification. It is possible that the method will contribute to reduce the computational complexity of PolSAR classification process and provide higher accuracy.

This paper presents a fine coregistration method for synthetic aperture radar (SAR) image processing, such as in InSAR interferogram generation. Under the assumption that SAR images are properly modeled as fractional Brownian motion, relative subpixel offsets between two images can be derived from the statistics of their increments. The method does not require upsampling or cross-correlation, thus allowing for an accurate offset estimation with less computational load. Implemented as a local coregistration procedure, it also provides a nonrigid geometric alignment that nicely follows the topography of the area. Experimental results show that the method gives comparable results to the conventional method, in terms of the accuracy of the generated digital elevation models. In particular, it displays superior accuracy for images with near homogeneous fractal behavior.

We evaluated our co-registration method which uses the number of SPs and amplitude information for ALOS data. Experimental results showed that our proposed method improves the quality of the interferogram as much as the case of JERS-1 data.

We propose a landmine-visualization system consisting of multiple complex-valued self-organizing maps (CSOMs), in which we pay attention to mutual information among them for integrating multimodal information. In particular, we focus on the use of similarity indices, which we can obtain with a small calculation cost. We demonstrate that the trends in the similarity indices are almost identical with that of mutual information. Consequently we can integrate multimodal information with a realistically small calculation cost.

Applications of complex-valued neural networks (CVNNs) have expanded widely in recent years-in particular in radar and coherent imaging systems. In general, the most important merit of neural networks lies in their generalization ability. This paper compares the generalization characteristics of complex-valued and real-valued feedforward neural networks in terms of the coherence of the signals to be dealt with. We assume a task of function approximation such as interpolation of temporal signals. Simulation and real-world experiments demonstrate that CVNNs with amplitude-phase-type activation function show smaller generalization error than real-valued networks, such as bivariate and dual-univariate real-valued neural networks. Based on the results, we discuss how the generalization characteristics are influenced by the coherence of the signals depending on the degree of freedom in the learning and on the circularity in neural dynamics.

Fractional Brownian motion (fBm) is a useful model to represent various natural phenomena and its synthesis has been a topic of interest in literature. This paper presents an algorithm to generate two-dimensional fBm based on circulant embedding method. Although this method has been proven in simulating one-dimensional fBm, its extension into two-dimensional is not straightforward. To solve the problem, we use circulant embedding as an exact method to synthesize the second-order increments of fBm, and recover the desired fBm via integration in frequency domain. The advantage of using second-order increments is its direct extensibility to any dimension. Experimental results show that the proposed method works efficiently fast and offers acceptable results when compared to the theoretical statistics of fBm.

Correcting the phase ambiguity of synthetic aperture radar (SAR) interferogram is an important process to create an accurate digital elevation model (DEM) for analyzing the geometry. The ambiguity often appears as the rotational point in the phase map which are called singular points (SPs). One of the origins of the SPs is the local distortion of the master and the slave, the source of interferogram. To solve this problem, we previously proposed a local co-registration method which employs the SPs as the evaluation criterion(SPEC method). In this paper, we propose an improved version of this co registration method. This version refers the amplitude in formation for shape-from-shading technique, which estimates the divergence of the range direction. We demonstrate the effectiveness of the improvement by comparing the DEMs generated from those interferograms.

There is increasing demand for landscape acquisition using interferometric synthetic aperture radar for frequent Earth observation. Making an accurate digital elevation model (DEM) from an interferogram is often seriously affected by a mass of singular points (SPs) included in the interferogram. The origin of a type of SPs lies in inaccurate coregistration in making the interferogram. However, there have not been any method more effective than affine transformation based on cross correlation. Numerous research works have aimed at how to process the SPs resulting in the coregistration. In this paper, contrarily, we propose an additional method of local and nonlinear coregistration which employs the number of SPs as the evaluation criterion. This method provides us with an interferogram to generate a more accurate DEM. Experiments demonstrate that, with the proposed method, we can obtain a DEM having higher signal-to-noise ratios.

We propose a new numerical-reconstruction method of an object image from its digital hologram. The proposed matching pursuit on a pair of domains (MPPD) method employs a spatial-domain basis and its (Fresnel) transform-domain pair. The transform domain basis is used to decompose the hologram, which yields a set of coefficients. Then, these coefficients are used to reconstruct the spatial-domain object image using the predefined spatial basis. We show the robustness of the proposed method against noise on a simulated hologram of spherical particles. By employing spatial-domain gaussian basis and its transform pair, the image of these particles are recovered successfully. The effectiveness of the proposed method is also demonstrated to a real microscopic-hologram of silica gel spherical particles.

Millimeter-wave imaging is suitable for security systems because of its ability to detect nonmetallic threats concealed under clothes. In this paper, we propose an active millimeter-wave imaging system using a complex-valued self-organizing map (CSOM) that realizes adaptive clustering of complex-valued texture in space and frequency domain to visualize targets. In the choice of winners in the CSOM, we also propose complex-valued inner-product metric to enhance the coherent merit in active imaging as well as to reduce the glaring and speckle distortion. We compare the performance obtained for the proposed method with that of Euclidean method.

Interferometric synthetic aperture radar (InSAR) is a key technology in geoscience. In the generation of a digital elevation map (DEM), the elimination of singular points (SPs) is the most important process besides the phase unwrapping (PU). A SP means a point where the phase rotation is not zero in the interferogram obtained by InSAR. What yields the SPs? One reason is a big cliff actually existing in the observation region. Empirically, such cliff-generated SP pairs (positive and negative SPs) are located at a distance from each other. Contrarily, other SP pairs, which make up the majority of the SPs, emerge closely to each other. Such close pairs arise from the autointerference caused by the diffraction in electromagnetic-wave propagation including the local permittivity fluctuation effect related to moisture vapor density in the air and other effects. We call the former the global SPs, while we do the latter the local SPs.

This paper discusses what the merits of complex-valued neural networks (CVNNs) arise from. First we look back the mathematical history to elucidate the features of complex numbers, in particular to confirm the importance of the phase-and-amplitude viewpoint for designing and constructing CVNNs to enhance the features. The viewpoint is essential in general to deal with waves such as electromagnetic-wave and lightwave. Then we point out that, although we represent a complex number as an ordered pair of real numbers for example, we can reduce ineffective degree of freedom in learning or self-organization in CVNNs to achieve better generalization characteristics. This wave-oriented merit is useful widely for general signal processing with Fourier synthesis or in frequency-domain treatment through Fourier transform.

In generating a digital elevation map from an interferogram obtained by interferometric synthetic aperture radar, the filtering process is as important as the phase unwrapping process. Before unwrapping, we usually have to restore the data image by reducing so-called singular points (SPs) by filtering process without destroying or smearing delicate fringes. Previously, an effective SP restoration method was proposed based on the complex-valued Markov random field (CMRF) model. However, there is still room for improvement in the definition of SPs and in the formulation of CMRF parameter estimation. In this letter, we propose a novel scheme by introducing a new concept, namely, the ldquosingular unit.rdquo We also estimate CMRF parameters locally as a weighted sum by taking the distance and SP numbers in sample sites into account. By using this restoration method, we demonstrate a high-performance removal of SPs. We also find that delicate landscape features, which are often lost in conventional filtering, are preserved appropriately. We confirm the quality in higher signal-to-noise ratios obtained against actual height data.

In this paper, we propose a method to distinguish between handwritten and machine-printed characters with no need to locate character or text-line positions. We transform a local region in a document image into frequency domain to extract feature values including fluctuations caused by handwriting. We feed the feature values to an optimized multilayer perceptron (MLP) to get likelihood of handwriting. We call this method the spectrum-domain local fluctuation detection (SDLFD) method. Experimental results show that our method distinguishes handwritten characters from machine-printed ones with no need of text-line position information. We also found that the scheme is robust against the change in scanning resolution.

The complex-valued Markov random field (CMRF) model is a powerful basis in complex-amplitude image processing. In this paper, we propose a method to reduce singular points (SPs) included in interferograms based on the CMRF model by focusing on the local pixel-value correlation. We deal with the SP-forming four pixels as a set, namely the singular unit (SU), in the CMRF-based compensation of data values distorted in electromagnetic-wave propagation with interference. We find that the method reduces the SP number with less processing distortion.

This paper reports recent progress in an adaptive ground penetrating radar imaging system based on a complex-valued neural network (CVNN), i.e., a complex-valued self-organizing map (CSOM). In the CSOM processing, we deal with feature vectors that represent complex-amplitude texture in space and frequency domains. We developed a switched walled linearly tapered slot antenna (walled-LTSA) array for the front-end. A higher resolution results in a better classification quality. To realize a high resolution in range and azimuth directions, we utilize a wide frequency bandwidth in frequency stepping operation, and a special switching scheme for the walled LTSA. We conducted experiments in Cambodia. In this paper, we report successful plastic landmine visualization, not only for targets buried in normal sand but also for those in wet laterite soil at the Siem Reap test site. Adaptive coherent radar imaging is one of the most potential application fields of the CVNNs.

In this paper, we propose a method for distinction between handwritten and machine-printed characters with no need to locate positions of characters or text lines. We call the proposed method psilaspectrum-based local fluctuation detection method. The method transforms local regions in document images into power spectrum to extract feature values which represent fluctuations caused by handwriting. We employ a multilayer perceptron for the distinction. We feed the obtained feature values to a preliminarily optimized multilayer perceptron (MLP), and the MLP yields likelihood of handwriting. We prepare a document image which has randomly aligned characters for an experiment. The experimental result shows that our method can distinguish handwritten and machine-printed characters with no need to locate positions of characters or text lines.

Lightwave has attractive characteristics such as spatial parallelism, temporal rapidity in signal processing, and frequency band vastness. In particular, the vast carrier frequency bandwidth promises novel information processing. In this paper, we propose a novel optical logic gate that learns multiple functions at frequencies different from one another, and analyze the frequency-domain multiplexing ability in the learning based on complex-valued Hebbian rule. We evaluate the averaged error function values in the learning process and the error probabilities in the realized logic functions. We investigate optimal learning parameters as well as performance dependence on the number of learning iterations and the number of parallel paths per neuron. Results show a trade-off among the learning parameters such as learning time constant and learning gain. We also find that when we prepare 10 optical path differences and conduct 200 learning iterations, the error probability completely decreases to zero in a three-function multiplexing case. However, at the same time, the error probability is tolerant of the path number. That is, even if the path number is reduced by half, error probability is found almost zero. The results can be useful to determine neural parameters for future optical neural network systems and devices that utilize the vast frequency bandwidth for frequency-domain multiplexing.

We propose a resolution improvement technique for a high-density array of walled linearly tapered slot antennas (LTSAs) in ground-penetrating radar imaging. In the system, we use the walled-LTSA array in a near-field measurement configuration and select various pairs of transmitter and receiver antennas out of the array elements. We prepare four selection modes so that we can practically enhance the imaging resolution. We remove the direct-coupling effect and compensate the path-length fluctuation caused by element individuality and element selection. Experiments demonstrate that the technique realizes higher resolution image acquisition to yield a clearer segmentation of antipersonnel plastic landmines in combination with adaptive image processing, based on complex-amplitude texture, even on a wet laterite soil condition.

We propose a walled linearly tapered slot antenna (LTSA) array to visualize plastic landmines. Previously, we reported an adaptive nonlinear visualization system based on a complex-valued self-organizing map (CSOM) that deals with complex amplitude texture in reflection images at multiple frequencies. The system distinguishes landmines from clutter by paying attention to textural features obtained by high spatial resolution and wideband reflection measurement. Because the system employed a mechanical scan of a pair of horn antennas, the measurement required a long time. An array antenna can reduce the time. The antenna element to be used there should therefore be compact and wideband. This paper reports the design and fabrication of a walled LTSA array visualization system. The antenna element has a 14 times 28 mm aperture size, and works at the 8-12 GHz frequency band. Because the structure is a simple combination of glass epoxy substrates and metal plates, we can easily fabricate low-cost and lightweight arrays. Electrical switches realize a high-speed scanning of 12 times 12 = 144 elements in total. We also report the results of a visualization experiment, in which plastic landmines are clearly visualized with the array in combination with the adaptive CSOM processing. Detection of landmines at frequencies of 10 GHz is only likely to be possible for targets buried a few centimeter deep or where the soil attenuation is very low. This might be a severe limitation of applicability of the method, as in field conditions soil attenuations of 10 dB or considerably more are commonly encountered, requiring the radar to operate at frequencies below 2-3 GHz. The best solution may be a multisensor system comprising these complementary high- and low-frequency radars.

How to process phase singular points (SPs), or residues, is a difficult problem in a 2-D phase-unwrapping process to generate digital elevation maps (DEMs). Although the minimum-cost network-flow method is an effective and widely used technique, some problems still remain. That is, the method often generates spikes in high SP-density areas and long clifflike artifacts for isolated SPs. It also takes a long time to unwrap phase data that contain many SPs. In this paper, we propose a new unwrapping method, namely, singularity-spreading phase unwrapping (SSPU), which solves these problems. In this method, we spread the singularity at SPs around to make the closely located positive and negative SPs combine gently with each other or make the isolated SPs fade away. Experiments demonstrate that the spreading process compensates the distortion in phase values in the vicinities of SPs appropriately for landscape reconstruction. The SSPU generates high-quality DEMs with smaller calculation costs than the conventional method. Besides the simple SSPU, we also present weighted SSPU where we utilize amplitude information to improve the performance further. In addition, we discuss the relationship between landscape characteristics and SSPU performance.

We propose a novel phase unwrapping method where we spread the singularity in phase map with fractional phase compensators. We find that the obtained digital elevation maps have higher quality than those obtained in conventional network programming method. In addition, the calculation cost is very small. We present the basic idea and the processing procedure.

We analyze the developmental-learning dynamics with which a motion-control system learns multiple tasks similar to each other or advanced ones incrementally and efficiently by tuning its behavioral mode. The system is based on a coherent neural network whose carrier frequency functions as a mode-tuning parameter. We consider two tasks related to bicycle riding, i.e., to ride as temporally long as the system can (task 1) and to ride as far as possible in a certain direction (task 2) which is an advanced one. We compare developmental learning to learn task 2 after task 1 with the direct learning of task 2. We also examine the effect of the mode tuning by comparing variable-mode learning (VML), where the carrier frequency is set free to move, with fixed-mode learning (FML), where the frequency is unchanged. We find that VML developmental learning results in the most efficient learning among the possible combinations.

Computer-generated hologram (CGH) is expanding the application fields. However, the CGH generation requires a long calculation time. In particular, the generation of a CGH stream for three-dimensional movies takes a huge amount of calculation cost. This paper proposes a small-calculation-cost method to generate a CGH stream based on coherent neural networks that deal with complex-amplitude information with generalization ability in the frequency domain. After carrier-frequency-dependent learning, we can generate a CGH stream by sweeping the carrier frequency with neural interpolation thanks to frequency-domain generalization.

This paper presents a snake algorithm for boundary detection of objects in phase images. Since phase image is a modulo-2 field, the proper gradient vector field (GVF) for snake dynamics should be taken from an unwrapped phase. This paper proposes a procedure to avoid such unwrapping by applying modulo-2 gradient estimation. Performance assessment is conducted by comparing a boundary detection result of non-modulo-2 GVF snake estimate with a modulo-2 regularized GVF snake. It is shown that the proposed method converges to the expected boundary, while the non-regularized procedure does not.

We propose a developmental learning architecture with which a motion-control system learns multiple tasks similar to each other or advanced ones incrementally and efficiently by tuning its behavioral mode. The system is based on a coherent neural network whose carrier frequency works as a mode-tuning parameter. In our experiments, we consider two tasks related to bicycle riding. The first is to ride as temporally long as the system can before it falls down (task 1). The second is an advanced one, i.e., to ride as far as possible in a certain direction (task 2). We compare developmental learning to learn task 2 after task 1 with the direct learning of task 2. We also examine the effect of the mode tuning by comparing variable-mode learning (VML), where the carrier frequency is set free to move, with fixed-mode learning (FML), where the frequency is unchanged. We find that VML developmental learning results in the most efficient learning among the possible combinations. We discuss the effects of the incremental task assignment as well as the behavioral mode tuning in developmental learning.

We present a neuro-beam former of ultra-wideband (UWB) pulses employing complex-valued spatio-temporal multilayer neural network, where complex-valued back propagation through time (CV-BPTT) is used as a learning algorithm. The system performance is evaluated with a UWB mono-cycle pulse. Simulation results in suppressing multiple UWB interferes and in steering to multiple desired UWB pulses, demonstrates the applicability of the proposed system.

We propose a spectral gradient method that is a novel method to extract character/text regions from general scene images. We obtain the distribution of the degree of likelihood of character/text regions by calculating the spatial variation of texture. We evaluate the texture variation by the gradient of local spatial spectra. A characteristic Fourier transform process, named hybrid spectral gradient method, is also developed to achieve a high extraction performance. This method is based on human foveation and can be applied for a wide range of languages and letters.

This paper presents a novel method to reduce the phase-unwrapping (PU) distortion by being based on two-dimensional fractional Brownian motion (fBm) theory. The method incorporates fractal geometry estimation to the result of conventional global-transform PU (GTPU). For the spatial-frequency spectrum of an observed phase image, we estimate the fractal dimension, by assuming an almost constant dimension over the image, and compensate the distorted spectrum of a GTPU result. It is demonstrated that the proposed method increases the signal-to-noise ratio of PU results for simulated data with various noise levels. Evaluations on an actual InSAR phase image also show that the method significantly improves the quality of the conventional GTPU result in its fine structure in particular.

We present an improved transform-based phase unwrapping (PU) system that employs a recursive structure. Each stage, which is identical with others, performs PU by FFT method and gives a solution as well as a residual (phase) error. The residual error is then reprocessed by the following stages. Experimental results for simulated and real InSAR phase images show significant improvement over conventional results of a single stage system.

This paper reports experimental results showing that the recall dynamics of analog associative memories is largely influenced by signal-voltage symmetry of synaptic weights and inverse-noninverse delay-time equality of neurons. We propose a highly symmetric synapse and an equi-delaying neuron. We fabricated an association chip comprised of them to demonstrate a high association performance. In comparison experiments, we also observe large performance degradations when the symmetry or delay equality is deteriorated. We analyze the dynamics based on the statistics of recall results. The proposals and the analysis results are widely applicable to analog recurrent convergence circuits.

This paper proposes a predictive self-organizing map (P-SOM) that performs an adaptive vector quantization of migratory time-sequential signals whose stochastic properties such as average values of signals in each cluster are varying continuously. The P-SOM possesses not only the weight corresponding to the signal values themselves but also those related to the time-derivative information. All the weights self-organize to predict appropriate future reference vectors. The prediction using the time-derivative weights enables the separation of continuously varying components form random noise components, resulting in a better performance of the adaptive vector quantization. That is to say, the stationary random noise components are captured by the ordinary weights, whereas the migrating components are captured by the first (and higher) order time-derivative ones. An application to a mobile communication receiver using quasi-coherent detection is presented. By utilizing both the ordinary and time-derivative weights consistently, the P-SOM generates a predictive reference vectors and quantizes the migratory signals adaptively. Simulation experiments on the bit-error rates (BERs) demonstrate that a P-SOM adaptive demodulator has a superior capability to track phase rotations caused by the Doppler effect. A theoretical noise analysis is also reported for the conventional SOM and the P-SOM. It is found that the calculation results are approximately in good agreement with the experimental ones.

We propose a new adaptive noise reduction method for interferometric synthetic aperture radar (InSAR) complex-amplitude images. In the proposed method, we detect residues (singular points) in the phase image as well as their neighbors at first. Normal areas that contain no residue are used for the estimation of correct pixel values at the marked residues according to 5th order non-causal complex-valued Markov random field (CMRF) model. The process is performed block-wise with the assumption of a locally stationary condition of statistics. Using a CMRF lattice complex-valued neural-network, the error energy defined as the squared norm of distance between signal and estimated values is minimized by LMS steepest descent algorithm. Eventually, the number of residues is decreased. An application is also presented. An InSAR image around Mt. Fuji is processed by the proposed technique and then phase-unwrapped by the branch-cut method. It is found that after the application of the proposed method, a better phase unwrapped image can be obtained successfully.

We propose complex-valued region-based-coupling segmentation neural networks that divide complex-amplitude images into regions and attach labels on them. A neurodynamics is obtained by an energy reduction rule in combination with phase-singularity elimination. A millimetre wave radar system is constructed using the network as a preprocessor for image recognition and reconstruction. Experiments demonstrate that the network works as a variable segmentation processor. The relation between network parameters and segmentation functions in the experiments is also reported.

Coherent-type artificial neural networks whose behavior is controlled by carrier-frequency modulation are proposed. The network learns teacher signals associated with an information-carrier frequency as a network parameter. The total network system forms a self-homodyne circuit. The learning process is realized by adjusting delay time and conductance of neural connections. Experiments demonstrate that the network behavior is successfully controlled by the carrier-frequency modulation. This result will be applicable not only to signal processing but also to frequency-multiplexed optical neural computing and quantum neural devices such as carrier-energy-controlled neurons in the future.