Aircraft target detection in synthetic aperture radar (SAR) images is challenging due to discrete scattering points and complex environments. With neural network advancements, this task has become more feasible, although the existing networks often have too many parameters for performance-constrained devices like drones or sacrifice performance to reduce parameter count. To address this, we propose a lightweight multistage SAR aircraft target detection network, termed MSADNet, featuring a novel multistage asymmetric aggregation network (MAAN) module. The MAAN module enhances feature extraction through an improved asymmetrical bottleneck network and introduces a multiorder attention mechanism to focus on critical features. By carefully managing the shortest and longest gradient paths, the MAAN module achieves an organic integration of both, significantly improving network performance. The backbone network is constructed using the MAAN module, and the neck network employs a feature pyramid network (FPN) + PAN structure, integrating the MAAN module with a coordinate attention (CA) module. An anchor-free detection head is used for aircraft target detection. Testing on the SAR-AIRcraft-1.0 dataset demonstrates that MSADNet achieves a mean average precision (mAP) of 96.15% with only 4.88 M parameters. Compared to YOLOv8-s, MSADNet improves mAP by 0.61 percentage points while reducing the parameter count by 6.29 M, indicating high-performance detection with a low parameter count.

In contemporary security detection systems, the utilization of millimeter-wave radar has assumed a central role owing to its non-contact and innocuous nature. This study addresses the challenging issues of detecting low-resolution and small targets in active millimeter wave (AMMW) images concealed detection. We identify a prevalent drawback in existing detectors, specifically the adoption of strided convolution or pooling layers, leading to a loss of crucial details that adversely impacts the detection rate. To overcome this limitation, we introduce a novel convolutional structure, termed Wavelet-Conv, which maintains information integrity while segregating features from high and low frequency bands, effectively replacing the unfavorable design. Furthermore, we harness the wavelet transform to enhance channel and spatial attention mechanisms, enabling more effective utilization of frequency band features and ensuring interpretability in the computational process. In this pursuit, we integrate the proposed Wavelet-Conv and Wavelet-Attention modules into the YOLOv8 framework, culminating in a unified model, termed Wavelet-YOLO. Through rigorous experimental validation on two AMMW datasets, our approach exhibits superior performance by significantly enhancing the recall and mean average precision (mAP) of small targets in AMMW images, while maintaining competitive inference speed. Extensive experiments demonstrate the outperformance of our proposed method over existing state-of-the-art approaches.

Object detection in remote sensing images (RSIs) is crucial for ground observation applications such as land surveying, urban planning, and precision agriculture. With the advent of convolutional neural networks, the performance of object detection in RSIs has been significantly improved. However, most studies have focused exclusively on the feature extraction, processing, and representation of geographical objects. The lack of integration of relevant common-sense knowledge has led to some obvious missed detections and illogical false alarms in the detection results. This letter proposes a novel and widely applicable approach for updating and fusing object co-occurrence knowledge into a two-stage multi-class object detection framework, called Auto Learner of Co-occurrence. The method captures the correlation between different object categories during the training process with the Co-occurrence Knowledge Updating module and utilizes the established co-occurrence weight matrix to enhance classification with the Knowledge Fusing module. The effectiveness and adaptability of the proposed method were validated through experiments on two multi-class object detection datasets using four popular two-stage detectors. The experiments also indicate that the speed penalty and the parameter increase resulting from the method are negligible.

Azimuth multichannel synthetic aperture radar (SAR) working in beam steering modes can effectively obtain images with wide unambiguous swath and high azimuth resolution. However, azimuth phase center fluctuation (APCF) phenomenon can be induced by the beam rotation in azimuth. Consequently, addition phases will be introduced in echoes of different channels. Different beam steering laws result in different degrees of APCF phenomenon, which lead to different azimuth channel phase errors. These phase errors lead to paired azimuth ambiguities or targets offset in imaging results. In this letter, the effects of different beam steering laws on the imaging results are analyzed, and the total additional wave paths required by different beam steering laws are compared, aiming at providing a reference for the design of the practical beam steering SAR system. The impacts of different beam steering laws on the imaging results are verified by simulation experiments.

Azimuth multichannel synthetic aperture radar (SAR) working in the sliding spotlight mode can effectively achieve ultrahigh resolution and wide swath imaging. However, multichannel sliding spotlight SAR echoes with large pulse bandwidth and azimuth steering angle are difficult to reconstruct and process due to the large range frequency-dependent Doppler centroid varying frequency support. To resolve this problem, two azimuth preprocessing approaches both based on the range frequency-dependent deramping strategy are proposed in this article. In both approaches, echoes are handled by a range frequency-dependent azimuth deramping function at first, which assign the signal measurements to unambiguous Doppler frequencies. Afterward, in the first preprocessing approach, azimuth multichannel raw data can be reconstructed and combined before azimuth upsampling in the conventional two-step preprocessing technique. The other way to handle the deramped multichannel raw data is the modified full aperture azimuth multichannel sliding spotlight reconstruction approach, which embeds azimuth multichannel reconstruction into azimuth convolution between the deramped raw data and the selected range frequency-dependent azimuth deramping function. Furthermore, the azimuth data in the second approach should be resampled to obtain a uniform azimuth sampling interval. Computational complexity and effects on the following imaging algorithm of both approaches are analyzed and compared. Simulation results validate the proposed two azimuth multichannel preprocessing approaches.

Staggered synthetic aperture radar (SAR) can effectively stagger range blind areas by periodically changing the pulse repetition interval (PRI), to achieve continuous imaging with an ultrawide swath. In conventional staggered SAR, azimuth samples are nonuniformly distributed due to the variable PRI, and complex resampling processing, such as best linear unbiased (BLU) interpolation and multichannel reconstruction processing, is required. In this article, a strict rule of PRI variation and phase center adjustment (PCA) is proposed for staggered SAR, to solve the problem of nonuniform azimuth sampling. In the proposed joint strategy of PRI variation and PCA, the relevant parameters for PCA are first determined. Then, the PRI sequence is designed according to PCA parameters to stagger blind ranges. Finally, the PCA rule is designed according to the designed PRI sequence and PCA parameters. During the data acquisition interval, the effective phase center is periodically adjusted pulse by pulse according to this rule, making the azimuth samples totally uniformly distributed. In addition, the discontinuously distributed missing samples can be estimated by signal estimation approaches. With the proposed joint rule of PRI variation and PCA, the emergence of false targets in imaging results can be avoided, and the complex resampling processing can be relieved to reduce the computational complexity. Simulation experiments on imaging results verify the advantages of this strategy.

Radio frequency interference (RFI) sources pose threats to wideband synthetic aperture radar (SAR) systems and accurate SAR image interpretation. Since most of RFI sources are narrowband, notch filtering is a simple but effective method for RFI suppression. In this letter, a modified two-step notch filtering approach combined with linear prediction is proposed to improve the SAR image quality. The notch filtering is used to mitigate narrowband RFI energy, while the linear prediction is introduced to recover the missing range spectral component of the SAR raw data from the desired scene, which is removed together with RFI sources by the notch filter. Because of the Gibbs phenomenon in Fourier series, the small residual RFI energy after notch filtering is enough to cause image visual disturbances and affect the accuracy of the following missing range spectrum extrapolation. The two-step notch filtering with a limited bandwidth is applied for better RFI sources mitigation. Simulation results on both simulated targets and real SAR raw data validate the proposed approach.

Array SAR can overcome the problems such as overlay and shadow caused by the terrain that the traditional SAR observes with large changes in slope. However, compared with traditional SAR, array SAR inevitably has multi-channel amplitude and phase errors due to its large number of channels. If these errors are not corrected, radar imaging will be degraded. This paper proposes a method based on time-domain processing to compensate for possible errors in the time domain. Finally, the effectiveness of this method is verified by CS algorithm experimental simulation.

In this paper, attention was paid to the impacts of RFI (Radio Frequency Interference) on SAR (Synthetic Aperture Radar) image ship detection with a proposed net structure based on deep learning. The RFI of different intensities can destroy the quality of SAR images, and pose difficulty for ship detection in marine environment. We acquired the dataset that contains numerous sample slices from two satellites, and took different ranges of SIR (Signal to Interference Ratio) to evaluate the robustness of the framework on the conditions. The experimental results show that the SAR images corrupted by RFI can affect the detection performance greatly. And it may provide suggest to researchers to conduct the interference properly when using the deep learning algorithm to monitor the ships in SAR images.

According to the characteristic of different air bubbles and sediments content in the Yellow River ice (YRI), the YRI is classified into three types: thermal ice (TI), frazil ice (FI) and consolidated ice (CI). In this paper, we are modeling YRI firstly, and the backscattering coefficient of the ice cover based on this model is deduced on the basis of the integral equation model (IEM) and the vector radiation transfer theory (VRT). Due to the characteristics of different types of YRI corresponding to different microwave electromagnetic scattering, a method of classification of YRI based on decision tree is proposed. The method uses horizontal polarization backscattering coefficient, correlation coefficient and cross polarization power to classify different types of YRI. Finally, the classification accuracy is evaluated by using the confusion matrix (CM). The results show that the classifier is effective for the YRI classification.

It is difficult to classify Hunshandake sandy land due to complex terrain, an improved classification method is proposed in this paper to promote the effect of classification. Firstly, scattering characteristics of ground objects are extracted. Then, the extracted features is combined to Jensen-Bregman logDet(JBLD) for classification. Finally, Hunshandake sandy land obtained by Radarsat-2 satellite is classified to verify the effectiveness of the method.

Since the echo characteristics of moving targets are different from the stationary scene, the traditional reconstruction algorithm, i.e., the reconstruction filter algorithm, is not applicable. In this paper, a novel reconstruction approach of the moving target for a multichannel in azimuth high-resolution wide-swath (HRWS) synthetic aperture radar (SAR) system, modified from the traditional reconstruction filter algorithm, is proposed. The core of this method is to solve the problem of reconstruction filter mismatch by adapting the motion parameters of the moving target. Compared with the reconstruction filter algorithm, this method can not only achieve the accurate reconstruction of the stationary scene, but also reconstruct the spectrum of the moving target with nice performance. The simulated results of point targets are shown to confirm the effectiveness of the proposed method.

For the next generation of spaceborne synthetic aperture radar remote sensing satellites, high resolution and wide coverage are important goals. Digital beam-forming (DBF) with multichannels in elevation is a great and promising candidate to cover wide swaths. In this letter, we focus on onboard digital processing for DBF in elevation. It is generally believed that the onboard processing for DBF is challenging due to the limitations in flight-hardware availability, the heavy computational load, and the high resource occupation. In order to reduce the computational load of DBF, one novel processing scheme is proposed. This proposed scheme performs a modified time-variant weighting on a real intermediate frequency signal of each subchannel, and the weighted signals of all channels are summed to two real data streams. To obtain a correct DBF output, a modified quadrature demodulation process is presented. Then, the scheme is extended to apply FIR filters for overcoming pulse extension loss. Furthermore, improved time-variant weighting coefficients are derived to compensate the phase errors brought by the FIR filtering process. Compared with the present processing flow, the proposed scheme could significantly reduce the computational load and resource occupation.

The novel displaced phase center antenna (DPCA) azimuth multichannel technique is well suited to improve the coarser azimuth resolution of the terrain observation by progressive scans (TOPS) mode for a fixed total receive antenna length. However, both the azimuth multichannel nonuniform sampling and the aliased Doppler spectrum in TOPS raise the azimuth data processing difficulty. This letter proposes an innovative full-aperture processing approach, which adds an azimuth derotation step for each azimuth channel before azimuth multichannel data reconstruction and utilizes the prefiltering operation in the two-step focusing technique to resolve both azimuth ambiguity problems caused by the DPCA technique and azimuth beam progressive scanning. In addition to presenting the proposed azimuth preprocessing approach, the impact of the approach on the azimuth ambiguity-to-signal ratio is analyzed in detail. Simulation results validate the proposed multichannel processing approach for the TOPS mode.

This letter extends the “two-step” focusing approach to sliding spotlight (SS) synthetic aperture radar (SAR) with stepped frequency chirps, where two major points should be looked out. One is the slant-range variation among one group of subpulses introduced by the radar platform position changing. In the stripmap mode, an effective approach to compensate this variation is to multiply each subband echo with a phase ramp in the Doppler domain. Herein, we apply this approach to the SS SAR. The other is the residual Doppler bandwidth after azimuth deramping with a fixed reference function. To remove this residual bandwidth, a range-frequency-dependent reference function (FDRF) combined with azimuth zero-padding was developed in a spotlight SAR with a single subband. In this letter, we extend the FDRF to the multiple-subband SS SAR, where the FDRF also depends on the subband carrier frequency. The main contribution of this letter is that the chirp z transform is employed to acquire an identical azimuth output pixel interval for different range frequencies and different subbands. All the aforementioned processing and the bandwidth synthesis are imbedded into the “two-step” focusing approach, which is validated by simulation results.

This paper analyzes the signal properties of spaceborne squinted sliding-spotlight synthetic aperture radar (SAR). Both the squint angle and the azimuth beam steering during the whole acquisition interval will lead to the Doppler spectrum back-folding. According to the special signal properties of this mode, a new full-aperture focusing approach, which includes three major processing steps, is proposed. In this approach, the first processing step introduces an azimuth convolution and azimuth data mosaic to resolve the azimuth spectral folding phenomenon by generalizing the existing two-step focusing technique for conventional sliding-spotlight SAR data focusing. Afterward, the modified range migration algorithm is adopted to process the resulting raw data. As the azimuth time duration of the raw data is obviously reduced in the first processing step, the obtained SAR image may be back-folded in azimuth. The final azimuth postfiltering step is to resolve the possible aliased SAR image without any azimuth data extension. The proposed full-aperture focusing processor is efficient since only a limited azimuth data extension is required to resolve the back-folded Doppler spectrum and SAR image. Imaging results on simulated raw data validate the proposed imaging approach.

The inherent limitation between azimuth resolution and range swath width in conventional spaceborne synthetic aperture radar (SAR) systems can be overcome by introducing the displaced phase center antenna technique. In these SAR systems, echoes from all subapertures should be reconstructed and combined together before single-channel SAR processors. However, in the multichannel sliding spotlight and Terrain Observation by Progressive Scans (TOPS) modes, azimuth beam progressive sweeping during the whole acquisition interval leads to that the total Doppler bandwidth spans over several N ·PRF intervals, where N is the number of azimuth channels and PRF is the pulse repetition frequency. As a result, conventional azimuth multichannel reconstruction algorithms for the stripmap mode are not directly suitable for sliding spotlight and TOPS modes. This paper proposes a novel imaging processor for both modes according to their azimuth echo properties. The key point of the proposed focusing processor is the first processing step of multichannel azimuth data reconstruction, which extends the two-step focusing technique to process raw data of the azimuth multichannel case. In addition to azimuth data preprocessing and accurate range cell migration correction steps, the final postprocessing step is added to correct the possible back-folded SAR images. Imaging results on simulated raw data validate the proposed imaging approach.

A synthetic aperture radar (SAR) raw data simulator of distributed target scenes is an important tool for testing the system parameters, imaging algorithms, and so on, particularly for these recently proposed novel imaging modes. The Terrain Observation by Progressive Scans (TOPS) mode achieves the wide-swath imaging coverage and overcomes major drawbacks in ScanSAR by electronically steering the azimuth beam along track during the whole acquisition time. In this letter, a novel TOPS mode raw data generation approach from conventional wide-beam SAR imaging modes is proposed, by invoking the fact that the TOPS raw data can be considered as a part of the stripmap/spotlight raw data with the exploited wide azimuth beam interval. In this approach, an azimuth-varying filter is introduced to obtain the desired TOPS raw data and set others to zero. Moreover, the conventional stripmap and spotlight mode raw data can be efficiently generated in the 2-D Fourier transformed domain. Therefore, a fast raw data simulator can be easily achieved by taking this approach. Simulation results validate the proposed simulation approach.

In order to reap the potential benefits that waveform diversity can provide for spaceborne synthetic aperture radar remote sensing, echoes from different subpulses constituting a complete transmit waveform should be effectively separated at first. This paper presents a new separation approach implemented by an advanced null-steering beamformer on satellite. Compared with common null-steering beamforming, our approach will take into account the characteristics of echo signal from the scene and accordingly embed the finite-impulse response (FIR) filtering process into the modified null-steering beamformer to deal with the issue of pulse extension. In this paper, echo signals generated by multidimensional encoded waveform will be analyzed in detail; based on this analysis, FIR filter and the new null-steering beamformer are derived. Simulation results show that much better separation performance can be obtained by our approach than by conventional null-steering beamforming.

The Terrain Observation by Progressive Scans (TOPS) mode is a novel spaceborne imaging mode which can be used to obtain wide-swath coverage and overcome major drawbacks in conventional ScanSAR. An efficient full-aperture imaging approach, which takes advantage of the two-step focusing technique and the azimuth baseband scaling operation, is presented for processing the TOPS-mode synthetic aperture radar (SAR) data. First, the proposed two-step focusing technique for spotlight and sliding spotlight SAR data focusing is adopted to resolve the aliased Doppler spectrum. Afterward, the following extended chirp scaling processing procedure with azimuth scaling factors is used to implement the residual TOPS raw-data focusing. Since the use of subapertures is avoided and only a limited azimuth-data extension is required, this algorithm is highly efficient. Simulation results validate the proposed imaging approach.