Publisher Copyright: © 2026 IEEE. EC/HE/101071179/EU//SUSTAIN EC/HE/101099491/EU//HOLDEN

Remote sensing image super resolution is crucial for enhancing geospatial recognition accuracy, yet existing methods based on deep learning present significant bottlenecks such as low computational efficiency, insufficient preservation for local details and global semantics. To address these challenges, we propose a novel CSMNet framework, which pioneers the integration of Mamba and implicit neural representation within UNet. We model long range dependencies of spatial and channel information based on the visual state space (VSS) model. In order to promote multi-scale feature interaction and preservation, we propose a fusion strategy to merge the upsampled features from bottom to top. Experiments on UCMerced and AID datasets demonstrate that our method averagely achieves prominent PSNR(dB)/SSIM values of 30.81/0.8535 and 33.30/0.8666, and 40.2% inference speed increasement than recent continuous SR methods.

As the Internet of Things (IoT) and Sixth Generation (6G) technologies advance rapidly, the cryptographic identification of electronic devices has become a critical issue in information security. Radio frequency fingerprint (RFF)-based specific emitter identification (SEI) has emerged as a prominent physical-layer authentication technique. To enhance the stability and accuracy of multi-target recognition in complex electromagnetic environments, a novel technique for individual specific emitter identification based on time-frequency similarity contrastive learning (TFSCL) is proposed. In this study, we present a novel pre-training method utilizing a deep complex-valued pyramid network (DCPN) to enhance the extraction and reconstruction of time series and frequency domain sequences. The DCPN enables contrastive learning of signal features in both the temporal and frequency domains, significantly reducing computational complexity and improving pretraining performance. Additionally, we first introduce the Time-Frequency Synchronization Data Added (TFS-DA), a Time-Frequency Hybrid Data Added Technique that employs Gray code to generate random sequences, effectively improving feature representation in both domains. Empirical results demonstrate that the proposed method achieves an accuracy rate of 97.12% on an automatic-dependent surveillance-broadcast (ADS-B) dataset that contains 10 categories with only data labeled 10%. On a LoRa dataset containing 30 categories with only data labeled 10%, the accuracy rate reaches 77.06%.

With the rapid development in the field of the Internet of Things (IoT), location information of people has become an indispensable part of achieving intelligent and digital management. Ultra-wideband (UWB) technology is widely used in indoor active positioning due to the characteristics of its pulse signals, which can achieve high-precision positioning within a range of a few centimeters. However, in device-free localization (DFL), the UWB tomography localization technique is highly disturbed by artifacts and requires a priori information about the number of known people to solve for the location of the target. In order to overcome this challenge, a training-free method is proposed in this paper to suppress radio tomographic imaging (RTI) artifacts in UWB systems. This method uses a single anchor L-shaped movement (LSM) to optimize the link layout and utilizes an improved elliptical model for precise linkpixel mapping. Finally, image reconstruction is achieved through guided filtering technology, effectively suppressing artifact interference. Experiments show that the method can achieve high-accuracy localization of multiple targets based on unknown a priori information about the number of people using received signal strength (RSS). When the number of positioning targets is 4, the quantity detection accuracy can reach 97%, and the system positioning error is less than 0.19 m.

With the rapid development of the economy and technology, radio frequency identification (RFID) technology is growing rapidly. However, significant multipath effects in complex environments reduce the accuracy of traditional methods, limiting practical applications. To address these challenges, our work presents a novel synthetic aperture radar RFID (SAR RFID) localization method leveraging dual rotating antennas. By utilizing received signal strength indication (RSSI) and establishing its functional relationship with angular position, the horizontally and vertically rotating antennas improve coverage and reduce multipath errors. Additionally, a multivariate nonlinear regression algorithm integrated with particle filtering is introduced to enhance adaptability to nonlinear data, significantly improving localization performance. The experimental results show that compared with the traditional fixed antenna system, the positioning error of the proposed system in 3-D environments is less than 13 cm. Our system achieves a favorable tradeoff among performance, hardware cost, and deployment complexity, due to the flexible configuration enabled by rotating antennas and the SAR mechanism. The proposed method can provide accurate and stable state estimation in complex nonlinear scenes, and offers a robust and practical RFID positioning solution for practical applications.

Abstract Aiming to detect novel objects from only a few annotated samples, few-shot object detection (FSOD) has undergone remarkable development. Previous works rarely pay attention to the perspective of gradient propagation to optimize existing methods, therefore failing to make full use of information for novel objects in gradient propagation. We propose a method to solve this problem based on two-stage fine-tuning. A domain adaptation module with multi-constraints is used to promote the spread of gradients, a classification promotion network is used to improve the effect of classification, and a multi-path mask head is added to enrich RoI features. Experiments on PASCAL VOC and COCO datasets show that our model significantly raises the performance compared with previous methods (up to 1–5 $$\%$$ % in average).

Instance segmentation can help vehicles or robots enhance their understanding of a scene through the pixel-level segmentation of different objects. However, occlusion and boundary blur, especially in cases with similar colors or textures, are still challenges encountered in real-time robust segmentation tasks. To segment a complete instance boundary, the existing 2D approaches fuse local and abstract semantic features derived from the color domain, which leads to homogeneous semantic information, and efficiently separating different objects is difficult in some cases. To address these complicated scenes, inspired by a human prediction processing strategy, where “the brain fills in missing information in advance to help make better decisions”, this study proposes a real-time asymmetric dual-stream instance segmentation algorithm embedding a depth-predictive architecture that provides the covisible depth information of objects. Furthermore, a cross-domain data fusion method and an enhancement-decoupling loss are designed to complement RGB data by utilizing the rich foreground and boundary details of the predicted depth map. In addition, our model can be fine-tuned to integrate it with real depth domain data provided by different input devices. Extensive experiments conducted on the COCO, OCHuman and CityScapes datasets demonstrate the effectiveness of our method. We further deployed our DSDP method on a UAV platform for validation purposes and qualitatively confirmed its validity.

In the predictive maintenance of modern industries, accurate fault diagnosis under complex conditions is now a major research focus. Recent research has demonstrated the effectiveness of deep learning in advancing bearing fault diagnosis. However, due to the scarcity of industrial failure data, achieving robust generalization in complex working conditions remains a challenge. To address this, we propose the causal disentanglement-based hidden Markov model (CDHM), which is designed to recognize the underlying causality in bearing vibration signals, capturing essential fault patterns for a more accurate and generalizable fault representation. Compared to signal-processing methods, deep learning approaches bypass the complex signal analysis, yet overlook the significance of signal theories in precise fault diagnosis. Nevertheless, the bearing vibration mechanism sheds light on the fact that the vibration induced by a certain type of fault has a consistent pattern across different system conditions, while the fault-irrelevant vibration such as noise and interference varies. Therefore, the CDHM constructs a time-series structural causal model (SCM), offering a new perspective on the interconnections of bearing vibration signals. Based on the SCM, a hidden Markovian variational autoencoder (VAE) is designed to progressively disentangle the vibration signal into two parts: a fault-relevant representation capturing essential bearing fault characteristics, and a fault-irrelevant representation capturing system and environmental interference. While unsupervised causal disentanglement typically presents optimization challenges, the CDHM benefits from cross-domain fault diagnosis tasks by leveraging the cross-domain consistency of the fault-relevant representation and the domain sensitivity of the fault-irrelevant representation. This design aligns the optimization objectives of causal disentanglement learning and cross-domain transfer learning, enabling mutually reinforcing optimization and ensuring robust generalization across diverse operating conditions. We validate the CDHM through experiments on the Case Western Reserve University (CWRU), Intelligent Maintenance System (IMS), and Paderborn University (PU) datasets, demonstrating its strong potential for industrial applications.

With the rapid development of the Internet of Things (IoT), an increasing number of industrial demands have become urgent. Radio-frequency identification (RFID) system plays a crucial role in addressing these challenges. It has now become a significant direction for the development of industrial automatic identification and data collection technology. However, the existing 3-D high-precision absolute localization methods have issues that need to be resolved. For example, the antenna sampling position needs to be used as known prior information, and the timestamp information of the samples is not fully utilized. Additionally, relying on multiple antennas or reference tags as auxiliary measures reduces the system flexibility. To overcome these challenges, we propose a single-antenna, multitarget, and 3-D localization method based on an attention mechanism and neural network regression model inspired by the synthetic aperture (SA) method. This method suggests using a single antenna moving uniformly on a slider for continuous motion and sampling. By utilizing sample phase and time-domain information, it achieves 3-D localization of cargo boxes in intelligent warehousing environments. In comparison with the current state-of-the-art localization solutions, our method does not give the use of reference tags and utilizes a reduced number of antennas. The experimental results prove the feasibility of using only mobile antennas without reference tags to perform 3-D multitarget positioning tasks. The combined dimensional average positioning error of 3-D indoor positioning achieves a high-precision positioning, reaching 2.04 cm and exhibits good robustness.

The four stages (first heart sound (S1), systole, second heart sound (S2), and diastole) of heartbeat sounds recorded by contact seismocardiogram (SCG) reflect the health of the heart, but these stages are challenging to measure by noncontact millimeter wave radar. If the sampling rate of millimeter wave radar is increased, this will increase the amount of data storage needed for the long-term monitoring of human vital signs. This article presents an algorithm for reconstructing the envelope of high-frequency heart sound signals using low-frequency millimeter wave radar signals, as well as a heart sound envelope segmentation algorithm based on peak points. Its design principle is a combination of signal processing and a transformer network, which is called T-HSER. This technique maps the low-frequency radar signal into a high-frequency heart sound envelope signal through the transformer network and determines the four different stages of the heart sound using appropriate thresholds. Based on the training of more than 30000 heartbeats of 25 healthy subjects and the prediction evaluation of six subjects, the T-HSER algorithm is shown to reconstruct the high-frequency heart sound envelope signal with high correlation. Moreover, the mean correlation can reach 0.85 on one minute of data, which is higher than that of the bidirectional long short-term memory algorithm, and can effectively distinguish the four stages of the heart sound so that the mean absolute error (MAE) between the predicted value and the ground truth of S1 and S2 is within a tolerable range (70 ms). At the same time, the algorithm is suitable for low sampling rate radar, which greatly reduces the amount of data storage required.

Basic life activities of the human body can be classified as either static or dynamic. Millimeter wave radars can monitor physiological signals in the static state and posture in the dynamic state. These biometrics can be used for identity recognition to address matching vital sign monitoring with the user identity information. However, existing studies are based on single-modal information for identity recognition using either static or dynamic information. This study proposes a framework called MN-UIV for enabling user identify verification using static and dynamic information via millimeter wave radar. The design method entails a combination of signal preprocessing and multimodal neural networks. First, the algorithm uses two radars: Radar 1 obtains physiological signals (in the time and frequency domains of breathing and heartbeat) in the static state and Radar 2 obtains the dynamic range angle image (DRAI) of the posture in the dynamic state. Multimodal information is later formed to extract features from multiple aspects. Here, the user identity information is verified using two designed multimodal neural networks (CNN+LSTM_Encoder (CLE), ResNet_Encoder (RNE)). Finally, based on the predictive evaluation of five users, the average recognition rates of the two multimodal neural networks were 96% and 96.4%, respectively. Compared to the average recognition accuracy of single-modal neural networks, it has a maximum improvement of 45.2% and 31%, respectively. It can continuously monitor the signs and posture of the body, thereby reducing system idle time and improving time utilization.

In passive Internet of Things, existing synthetic aperture-based 3D localization methods face many challenges, such as high computational load, a large aperture of a virtual antenna array, and sensitivity to noise. To address these challenges, this paper develops a synthetic aperture scheme for integrated localization and navigation, which implements the localization algorithm with a trajectory generated by the navigation algorithm. The localization problem is formulated by multidimensional scaling, which exploits phase differences involving the spatial information between target tags and a virtual antenna array. The new formulation allows the system to provide an accurate location estimate with a large moving step and sparse virtual antenna array of narrow apertures. The navigation problem is formulated to decrease errors of distance differences. Moreover, a navigation criterion is established to determine the feasibility of a virtual antenna position based on phase measurements, and an efficient navigation algorithm is proposed to find such a feasible point. Extensive numerical results validate our theoretical analysis and the performance of the proposed scheme.

Radio-frequency identification (RFID) and computer vision (CV) technology are widely employed in intelligent storage systems for sensing, locating, identifying, and monitoring storage cargo. However, both of them have applicability scenarios and limitations. In this article, we propose a system for spatial perception of storage cargo based on the fusion of RFID and CV data. Specifically, we employ a mobile robot carrying an RFID reader and an RGB camera to move between shelves. The reader is connected to two vertically deployed antennas that receive phase values from the tags on the cargo to perform phase unwrapping. Then, we construct a linear system of equations to solve synthetic aperture radar (SAR)-based 3-D localization to obtain the location of the target tag efficiently and accurately. For the images of the shelves captured by the RGB camera, we use image localization techniques based on color and texture features to obtain the pixel coordinates of the cargo in the image. Since the sampling range of the RFID reader antenna is larger than the shooting range of the RGB camera, we use the coherent point drift (CPD) algorithm with threshold judgment to match the localization results of the two subsystems, enabling us to display the inherent information of the target cargo on the plane rendering. Our proposed system is evaluated through various experiments, and the results show that our localization algorithm achieves high accuracy in 3-D space, and the matching algorithm has high robustness to the number of cargo and missing cargo or tags.

Weighted multidimensional scaling (WMDS), an algorithm extending multidimensional scaling (MDS), has been utilized in a broad spectrum of localization. However, there are still two unsolved theoretical questions: 1) The estimator provided by MDS depicts a relative placement of targets which require further Procrustes analysis to recover the actual placement, referred to as the absolute placement. This fact produces a question: Does the estimator of WMDS is still relative? If not, what underlying mechanisms assert the absoluteness? 2) It has been proved that WMDS attains the Cramér–Rao lower bound (CRLB) when the ranging distribution is Gaussian. Does it hold for a general distribution? If not, what restrictions on distributions are required? Motivated by such theoretical incompleteness, this article offers an in-depth theoretical analysis on WMDS in the scheme of range-based localization. With regard to question 1), we reveal the mechanisms that assert the estimator of WMDS always represents exactly the actual locations and prove that the absoluteness is introduced at the moment of variable separation and all the subsequent matrix equation transformations preserve the absoluteness. As for question 2), the functional behaviors and maximum condition of CRLB are examined under a general ranging distribution. Then, via the comparison to the variance of the WMDS estimator, the stationary condition on ranging distributions for WMDS to attain the CRLB is provided. Extensive simulations are performed to validate the theoretical conclusions numerically. And there exists conformity between numerical and theoretical results.

In fully-connected passive RFID backscatter networks, it is challenging to provide accurate range estimations due to complex channels. Facing this problem, we propose a differential analysis based anti-multipath technique by introducing a reference tag. Through the linear difference between target tag received power measurements with the reference tag activated and not, the power with respect to the reader-reference-target link is separated out from the mixed measurements, achieving the mitigation of multipath and accurately ranging. Under distributed localization schemes, after breaking down the full network into a series of fragments, it is vital but challenging to determine how to assemble them satisfactorily. VIABLE, virtual-actual assembling algorithm, is proposed to achieve distributed and cooperative localization. The virtual assembling phase rectifies the fragments over and over until their errors converge, which enables the mining of a satisfactory and adaptive assembling order. Subsequently, the actual phase assembles the rectified fragments together with that order and accomplishes the overall localization accurately. The theoretical analysis of performance limit is presented via the derivation of Cramér-Rao lower bound approximated by a particle approach. Extensive simulations demonstrate that our proposed framework outperforms existing algorithms for cooperative localization.

Radio-frequency identification (RFID) electronic license plate (RF-ELP) has been widely used to enable various automatic vehicle identification applications. Endowing RF-ELP with mobile vehicle sensing capabilities, such as localization and speed measurement is of practical importance, yet there is no solution on the shelf. Moreover, the position information is essential for accurate speed measurement, while the related RFID-based vehicular localization and indoor mobile localization methods suffer from at least one of the following major limitations: 1) difficult to deploy in practice; 2) requiring moving speed in advance; 3) only working for indoor-speed vehicles; and 4) not well compatible to frequency-hopping mechanism. To overcome the above limitations, this article proposes an RF-ELP-based mobile vehicle sensing (RESensing) system. RESensing conducts a new signal phase collection strategy to ensure the phase coupling in road-speed cases and converts phases of each interrogation to the relative speed to make it immune to frequency hopping and interinterrogation phase fluctuation. Then, the speed measurement and longitudinal localization are simultaneously performed by solving a nonlinear optimization model. Furthermore, the propagation model and antenna radiation pattern are investigated to facilitate the received signal strength index (RSSI)-based accurate lane-level lateral localization. To our knowledge, RESensing is the first RF-ELP-based speed measurement and localization system for mobile vehicles. The performance of RESensing is evaluated by real experiments under specifications of GB/T 37987 and EPC C1G2, which shows that RESensing achieves the mean speed error ratio of 4.34%, the longitudinal localization error of submeter level, and the lane estimation accuracy of nearly 100%.

In this paper, MUSE, an algorithm enabled by backscattering tag-to-tag network (BTTN) is presented to accomplish simultaneous 2-D localization of large-scale (10 m × 10 m) massive (20$\sim$∼50) passive UHF RFIDs. In BTTNs, the most intractable problem is the high-frequency loss of range measurements. In a particular case of 30 tags to be located with maximum communication range being 3 m, the rate is nearly up to 85.75 percent. In the proposed framework, we utilize relevant knowledge in the theory of graphs to obtain underlying subsets in which tags can communicate with each other and then assemble them stage by stage to achieve overall localization. Theoretical analysis shows that multistage assembly imparts extraordinary characteristics to MUSE: Assembling rectifies fragment maps given some condition, and in later stages prevents errors flowing down into the next stage. Experimental analysis shows that the condition is easy to satisfy. Furthermore, an analytical expression for the Cramér-Rao lower bound is also derived as a benchmark to evaluate the localization performance. Extensive simulations demonstrate that MUSE outperforms existing algorithms for simultaneous localization.

The device-free localization (DFL) has promising application prospects in intrusion detection, emergency rescue, and smart homes, because it does not require the target to carry any auxiliary positioning equipment. Radio tomographic imaging (RTI) is one of the most potential DFL techniques and has many advantages over other methods. However, in passive ultrahigh frequency radio frequency identification scenario, there are few researches and many problems to be solved. The difficult but urgent matter is how to identify the locations of multiple targets from many false targets and artifacts. This paper proposes a novel method based on cross-sectional scan (CSS), gray value distribution analysis (GVDA), and naive Bayes classifier to solve this problem. The CSS obtains the gray value distributions of the local maximum pixel in an RTI reconstructed image. Then, the GVDA extracts several characteristic parameters from gray value distributions, such as the size, height, and shape of the peak. Finally, the naive Bayes classifier utilizes these series of characteristics to judge whether local maximum pixels are false targets or real targets. The method can also recognize the number of targets that are very close to each other. Simulation and experimental results show that this method can accurately determine the locations and the number of targets.

Radio frequency identification (RFID) technology, which is one of the important implementation ways of Internet-of-Things (IoT), has achieved much attention in indoor localization areas. Passive ultrahigh frequency (UHF) RFID tag localization has a great development recently. Most of traditional passive UHF RFID localization algorithms can only achieve the position of one tag at a time while multitag localization is desired in many RFID applications. In this paper, we proposed the weighted multidimensional scaling (WMDS) based on received signal strength (RSS) method and the tag-to-tag communication system to achieve multitag cooperative localization. The targets are marked with passive UHF RFID tags. RSS method is used to determine the distance between readers and target tags through channel models. We can also obtain the estimated distances between target tags by the tag-to-tag communication system. The estimated locations of the tags are determined by the calculation of Euclidean distance matrix through a few iterations. Simulation results show that the WMDS algorithm achieves higher localization accuracy than traditional algorithms and the cooperation between tags improves the localization accuracy.

Radio frequency identification (RFID) is a promising technology in indoor localization. However, multipath cannot be avoided in indoor passive ultrahigh frequency RFID localization environment, which results in poor localization accuracy. In this paper, a range-based localization scheme has been developed to mitigate the effect of multipath errors and improve the localization accuracy. The proposed localization methods are based on semi-definite programming and second-order cone programming and do not require any statistics of multipath error. First, all the phase information of multifrequency is utilized. Dual-frequency phase difference of all dual-frequency combinations can be obtained to range. The ranging error is assumed to follow the Gaussian distribution. Then, convex relaxation methods with limiting the mean ranging error to one standard deviation of the Gaussian distribution are proposed. Finally, in order to solve the feasibility problem, the constraint of the mean ranging error is relaxed. Simulation results demonstrate that the proposed methods outperform some existing localization algorithms in multipath environment.

RFID can play great roles in future context-aware applications such as ambient intelligence and internet of things. Ranging technology will be one of interesting topics for RFID application. In the paper we compare the phase of arrival (POA) and time of arrival (TOA) based ranging behavior for RFID application. We characterize the ranging behaviors from the view of Cramer-Rao lower bounds (CRLB), ray tracing and empirical measurements. First, we analyze and compare the CRLB for POA and TOA based ranging in AWGN environment theoretically. Second, we use ray tracing method to model the distance and bandwidth influence on POA and TOA based RFID ranging in multipath environment. Third, we perform measurement to validate the noise and multipath influence on POA and TOA based RFID ranging. The parameters in ray tracing and measurements are also provided to further support the behavior comparison between TOA and POA ranging. The results show that in short range application such as RFID, POA based ranging has a comparatively better performance than TOA. It will benefit the future selection of POA or TOA technique, narrowband or wideband ranging method regarding specific scenarios for RFID application.

Structured data representing entity descriptions often lacks precise type information. That is, it is not known to which type an entity belongs to, or the type is too general to be useful. In this work, we propose to deal with this novel problem of inferring the type semantics of structured data, called typification. We formulate it as a clustering problem and discuss the features needed to obtain several solutions based on existing clustering solutions. Because schema features perform best, but are not abundantly available, we propose an approach to automatically derive them from data. Optimized for the use of schema features, we present TYPifier, a novel clustering algorithm that in experiments, yields better typification results than the baseline clustering solutions.

Instance matching and blocking, a preprocessing step used for selecting candidate matches, require determining the most representative attributes of instances called keys, based on which similarities between instances are computed. We show that for the problem of learning blocking keys and key values, both generic techniques that do not exploit type information and supervised learning techniques optimized for one single predefined type of instances do not perform well on heterogeneous Web data capturing instances for which the predefined type is too general. That is, they actually belong to some subtypes that are not explicitly specified in the data. We propose an unsupervised approach for learning these subtypes and the subtype-specific blocking keys and key values. Compared to state-of-the-art supervised and unsupervised learning approaches that are optimized for one single type, our approach improves efficiency as well as result quality. In particular, we show that the proposed strategy of learning subtype-specific blocking keys and key values improves both blocking and instance matching results.

In this article, a new Neuro-Space mapping method is presented aimed at using neural networks to automatically enhance nonlinear device models, such as FET models. Compared with previously published space mapping methods, our proposed method produces better modeling accuracy and provides more effective combinations of mapping structure with existing coarse model. In our proposed models, separate mappings for voltage and current at gate and drain are used as the mapping structure. Training methods for mapping neural networks are also proposed. Application examples on modeling MESFET devices and the use of new models in DC, S-parameter and combined DC and S-parameter simulation demonstrate that our proposed Neuro-Space mapping model matches more closely with the device data than that by the traditional Neuro-Space mapping method for modeling nonlinear microwave devices.

This paper discusses the implementation mechanism and its application of an interpreter for a framed temporal logic programming language called framed tempura. Firstly, the basic approach based on the normal form is presented. Then, the structure of the interpreter is illustrated and each of its modules is explained. The work flow of the reduction of programs is given in detail. In particular, the implementation approaches of several important new constructs including frame, await, projection, pointer are presented. As an application, the interpreter is used as a simulator of service models of OWL-S for the Web service composition.