Infrared target detection supports the continuous observation of traffic participants and low-altitude targets across aerial and fixed-view imaging settings, particularly under weak or changing illumination. However, infrared targets are often small, weakly textured, and easily confused with thermal noise and background clutter. Large pretrained vision models offer strong representation and generalization capabilities, but their parameter counts and computational costs make direct deployment on edge platforms with limited resources impractical. To transfer these capabilities to lightweight models, this paper proposes knowledge distillation at the label level based on filtered teacher detections. A large teacher adapted to the infrared domain first generates candidate boxes. Candidate boxes are selected using confidence thresholds, class reliability, and spatial relationships with ground truth annotations. The retained teacher boxes and the original annotations jointly form the student training targets, while the student retains its standard detection loss and original inference structure. On an independent sequence-level test set, the YOLOv5n baseline obtains an mAP@0.5 of 0.2277 and an mAP@0.5:0.95 of 0.1070, whereas filtered label distillation obtains 0.2547 and 0.1153, respectively, over three matched seeds. The filtering rules and thresholds are fixed before this evaluation, and the fixed Epoch-30 checkpoint is used for every run. Both student architectures are deployed on RK3588.

The orthogonal time-frequency space (OTFS) modulation, operating in the delay-Doppler (DD) domain, possesses notable advantages in Doppler robustness, multipath suppression, and power efficiency, making it a compelling candidate for future sensing applications in highly dynamic environments. However, existing single-site or co-located multiple-input multiple-output (MIMO)-OTFS radar systems face limitations in wide-area surveillance, including insufficient detection performance under low signal-to-noise ratio (SNR) conditions, limited weak target detection capability, and constrained parameter estimation accuracy. To address these challenges, this article proposes a distributed MIMO-OTFS radar architecture based on DD-domain collaborative sensing. We construct a multistation collaborative topology using orthogonal waveforms and establish the end-to-end signal transmission model. Subsequently, we design a comprehensive collaborative sensing process, where each site first performs low-complexity coarse target detection via 2-D matched filtering, followed by multitarget registration and matching using the minimum distance error criterion. Based on this, a golden-section iterative off-grid refinement algorithm achieves fractional-level range-velocity estimation accuracy. Simulation results validate the effectiveness of the proposed framework, successfully achieving collaborative joint parameter estimation and registration for multiple targets. In the simulated 2 & times; 2 distributed configuration, the proposed system achieves an average peak SNR gain of 1.43 dB compared to single-station systems, and the off-grid estimation accuracy approaches the Cram & eacute;r-Rao lower bound. This study provides a feasible solution for practical distributed OTFS sensing.

To strengthen the security of large-scale space-air-ground integrated networks (SAGINs), this paper investigates covert communication against non-cooperative ground base stations (BSs) that detect satellite transmissions via signal power monitoring. In this scenario, numerous low-earth-orbit (LEO) satellites deployed across multiple orbital layers provide backhaul support for autonomous aerial vehicles (AAVs), thereby serving ground users. To reduce the probability of detection, the LEO network performs resource allocation to conceal transmission activities under co-channel interference. However, such a strategy may overlook fairness in resource optimization, potentially undermining cooperation between LEO satellites and terrestrial networks. To address this issue, we develop a two-stage hierarchical Stackelberg matching game to characterize the interaction between LEO satellites and non-cooperative ground BSs. At the upper stage, the LEO network acts as the leader and maximizes the communication rate through power allocation while satisfying covert constraints. At the lower stage, the non-cooperative ground BSs act as followers and competitively minimize their detection errors in response to the leader’s actions. To solve this problem efficiently, we integrate hierarchical game theory, the asynchronous Stackelberg decision transformer (ASDT), and multi-agent reinforcement learning (MARL) into a unified framework for multi-agent coordination. Numerical results demonstrate the effectiveness of the proposed hierarchical resource optimization strategy and provide useful insights for secure SAGIN deployment.

Due to the broadcasting characteristics of satellite-terrestrial integrated networks (STINs), security vulnerabilities have emerged as a critical concern requiring urgent mitigation strategies. Unlike traditional security methods, federated learning (FL) enables a large number of participants to collaborate without disclosing actual privacy data. Its potential as a framework that combines collaborative model training and covert payload transmission in STINs represents a significant research gap. This paper proposes FedSAT, a novel FL-based covert communication scheme for STINs, in which each participant in the FL process can utilize the shared learning protocol as a covert medium for transmitting arbitrary information in privacy-preserving framework. Our framework leverages the dual capabilities of FL for collaborative model training and covert payload embedding, utilizing Geostationary Earth Orbit (GEO) satellites and distributed terrestrial nodes to embed sensitive data within FL parameter updates. The system maintains model convergence accuracy while implementing strategic encryption to achieve robust sharing and transmission of payloads within the FL framework. Comprehensive simulation tests demonstrate the framework significant efficacy, achieving a 98.7% communication coverage for covert payload transmission under monitoring by low Earth orbit (LEO) surveillance satellites, with only a 0.8% decrease in model accuracy. This breakthrough achievement paves the way for a transformative paradigm in covert cross-domain communication for next-generation networks.

Integrated Sensing and Communication (ISAC) is recognized as a pivotal technology for the upcoming 6G era. In ISAC, communication-assisted sensing enhances the communication capabilities of multiple sensor nodes and strengthens their sensing abilities, especially within the domain of remote sensing target recognition, which is of great significance for accurately identifying targets from massive remote sensing data. However, the explosive growth of model size and data volume has placed a heavy burden on communication links. Motivated by this, this paper proposes a communication-assisted remote sensing target recognition framework based on hierarchical federated learning (HFL). Specifically, after the edge nodes complete cluster aggregation, the fully connected layer parameters are sent back to the client, while the parameters of the convolutional layer are uploaded and aggregated in the cloud server, and then sent back to the cluster server, which in turn sends them to each client. The cloud server only aggregates the convolution layer parameters and distributes them to the client. To further reduce communication overhead, an adaptive sparsity method is introduced. By combining error compensation and Top-k amplitude pruning, a very high sparsity is achieved at the expense of minimal accuracy. Simulation results illustrate that the presented algorithm surpasses the benchmark scheme in both communication overhead and average test accuracy in the task of remote sensing target recognition.

Traditional single neural network-based geo-localization methods for cross-view imagery primarily rely on polar coordinate transformations while suffering from limited global correlation modeling capabilities. To address these fundamental challenges of weak feature correlation and poor scene adaptation, we present a novel framework termed ICT-Net (Integrated CNN-Transformer Network) that synergistically combines convolutional neural networks with Transformer architectures. Our approach harnesses the complementary strengths of CNNs in capturing local geometric details and Transformers in establishing long-range dependencies, enabling comprehensive joint perception of both local and global visual patterns. Furthermore, capitalizing on the Transformer’s flexible input processing mechanism, we develop an attention-guided non-uniform cropping strategy that dynamically eliminates redundant image patches with minimal impact on localization accuracy, thereby achieving enhanced computational efficiency. To facilitate practical deployment, we propose a deep embedding clustering algorithm optimized for rapid parsing of geo-localization information. Extensive experiments demonstrate that ICT-Net establishes new state-of-the-art localization accuracy on the CVUSA benchmark, achieving a top-1 recall rate improvement of 8.6% over previous methods. Additional validation on a challenging real-world dataset collected at Beihang University (BUAA) further confirms the framework’s effectiveness and practical applicability in complex urban environments, particularly showing 23% higher robustness to vegetation variations.

In large-scale metrology (LSM), the transformation of the tracker base frame (TBF) is a predominant method to enlarge the field of view (FOV) of the tracking sensor for full-field 3D measurements. Nevertheless, such a process will introduce cumulative errors and significantly diminish the global point cloud alignment accuracy. To address this problem, we propose a novel tracker pose optimization method for TBF transformation. A pose graph optimization (PGO) model based on spatial distance constraints is implemented to improve the tracker pose accuracy. We also adopt a robust coefficient and a damping factor to simplify the experimental process and stabilize the convergence results. Simulations and experiments on high-speed train surfaces are conducted to validate our method's accuracy and effectiveness. The results indicate that our optimization method outperforms two existing methods in spatial positioning accuracy and point cloud alignment accuracy, which showcases its practical applicability and superiority in manufacturing scenarios.

Unmanned Aerial Vehicle (UAV) path planning is a critical task that directly affects the efficiency and safety of UAV operations in various fields. This paper proposes an improved Bi-RRT* algorithm to enhance the efficiency of path planning while ensuring feasible and smooth navigation. The proposed algorithm integrates a goal point switching mechanism and dynamic ellipsoid sampling with a goal-bias strategy to improve the sampling process, an enhanced expansion strategy combining goal-point-oriented growth with an improved artificial potential field (APF) mechanism to accelerate convergence, and a density-aware node rewiring strategy with greedy pruning and B-spline smoothing to generate an optimized path. Extensive simulations in five complex 3D environments demonstrate that the proposed method significantly outperforms five existing RRT* variants, achieving shorter path lengths, fewer iterations, reduced computation time, and a lower number of nodes while maintaining high robustness and adaptability in obstacle-dense environments.

To utilize a unified waveform to realize radar and covert communication functions simultaneously, we propose a novel high speed covert dual function radar communication waveform using phase-reduced shift keying and code-domain index modulation (HCDFRC-PRSK-CDIM). First, the spherical codes are selected for code-domain index modulation to improve the communication rate. In addition, the binary phase shift keying scheme with a reduced phase are introduced to embed the information in linear frequency modulation (LFM) signals with covertness. Simulation results show that the designed waveform has good covertness and bit error rate (BER) performance without affecting radar detection performance.

The spectral reflectance measured in situ is often regarded as the “truth”. However, its limited coverage and large spatial heterogeneity often make the ground-based reflectance unable to represent the remote sensing images. Since the spatial scale mismatch between ground-based, airborne, and spaceborne measurements, the applications of geological exploration, metallogenic prognosis and mine monitoring are facing severe challenges. In order to explore the influence of spatial scale effect on rock spectra, spectral reflectance with uncertainty caused by differences in illumination view geometry and spatial heterogeneity is introduced into the Bayesian Maximum Entropy (BME) method. Then, the rock spectra are upscaled from the point-scale to meter-scale and to 10 m-scale, respectively. Finally, the influence of spatial scale effect is evaluated based on the reflectance value, spectral shape, and spectral characteristic parameters. The results indicate that the BME model shows better upscaling accuracy and stability than Ordinary Kriging and Ordinary Least Squares model. The maximum Euclidean Distance of rock spectra caused by spatial resolution change is 6.271, and the Spectral Angle Mapper can reach 0.370. The spectral absorption position, absorption depth, and spectral absorption index are less affected by scale effect. For the area with similar spatial heterogeneity to the Huangshan Copper–Nickel Ore District, when the spatial resolution of the image is greater than 10 m, the rock’s spectrum is less influenced by the change in spatial resolution. Otherwise, the influence of spatial scale effect should be considered in applications. In addition, this work puts forward a set of processes to evaluate the influence of spatial scale effect in the study area and carry out the upscaling.

This paper investigates an intelligent trajectory planning algorithm for unmanned aerial vehicles (UAVs) that utilizes track deception techniques against radar networks. The main objective of the algorithm is to minimize the flight distance of UAVs while generating a coherent phantom track to deceive radar networks. Firstly, we analyze the geometric coupling relationship among the false target, the UAV, and the radar, deriving the motion control equations for UAV trajectory planning. Subsequently, we mathematically formulate the problem of intelligent trajectory planning for UAVs based on this coupling relationship. The optimization model aims to reduce the flight distance of the UAVs as much as possible, considering the strict dynamic constraints of UAV platforms. Furthermore, by leveraging the pre-designed phantom track and prior knowledge of radar network locations, we employ the particle swarm optimization (PSO) method to solve the resulting optimization problem. Finally, through simulation results, we validate the effectiveness and feasibility of the proposed algorithm.

Remote healthcare has been an important application of 6G and is increasingly attracting attention. In the field of medical image analysis, federated learning (FL) is widely considered for medical data collected by Internet of medical things (loMT) devices from different hospitals. In FL, communication overhead and local data valuation are two inevitable issues that urgently need to be addressed. The communication overhead between local clients and a central server can assist in enhancing the perception ability of FL. The data quality of local clients should be positively correlated with the contribution of their aggregation model. As such, this article discusses an efficient causal learning-based compression scheme of local data to reduce the amount of data for communication, namely, feature-heterogeneity-aware model, which will lower communication overhead without affecting the performance. Additionally, local client evaluation based on a model-driven and artificial intelligence (AI)-driven mode is analyzed respectively with the assistance of blockchain due to the data sensitivity of medical images in this article. As a result, the client weight during global aggregation can be adjusted adaptively according to their objective contribution. The simulations are made, and the results validate the effectiveness of proposed schemes.

This paper studies a coalition game theoretic power allocation algorithm for multi-target detection in radar networks based on low probability of intercept (LPI). The main goal of the algorithm is to reduce the total radiated power of the radar networks while satisfying the predetermined target detection performance of each radar. Firstly, a utility function that comprehensively considers both target detection performance and the radiated power of the radar networks is designed with LPI performance as the guiding principle. Secondly, it causes a coalition to form between cooperating radars, and radars within the same coalition share information. On this basis, a mathematical model for power allocation in radar networks based on coalition game theory is established. The model takes the given target detection performance as a constraint and maximizing system energy efficiency and optimal power allocation as the optimization objective. Furthermore, this paper proposes a game algorithm for joint coalition formation and power allocation in a multi-target detection scenario. Finally, the existence and uniqueness of the Nash equilibrium (NE) solution are proven through strict mathematical deduction. Simulation results validate the effectiveness and feasibility of the proposed algorithm.

This paper develops a collaborative trajectory planning and resource allocation (CTPRA) strategy for multi-target tracking (MTT) in a spectral coexistence environment utilizing airborne radar networks. The key mechanism of the proposed strategy is to jointly design the flight trajectory and optimize the radar assignment, transmit power, dwell time, and signal effective bandwidth allocation of multiple airborne radars, aiming to enhance the MTT performance under the constraints of the tolerable threshold of interference energy, platform kinematic limitations, and given illumination resource budgets. The closed-form expression for the Bayesian Cramér–Rao lower bound (BCRLB) under the consideration of spectral coexistence is calculated and adopted as the optimization criterion of the CTPRA strategy. It is shown that the formulated CTPRA problem is a mixed-integer programming, non-linear, non-convex optimization model owing to its highly coupled Boolean and continuous parameters. By incorporating semi-definite programming (SDP), particle swarm optimization (PSO), and the cyclic minimization technique, an iterative four-stage solution methodology is proposed to tackle the formulated optimization problem efficiently. The numerical results validate the effectiveness and the MTT performance improvement of the proposed CTPRA strategy in comparison with other benchmarks.

在传统多假设跟踪(MHT)算法中通常会假设杂波强度先验已知,当观测场景中杂波未知且空变时,该假设将会导致跟踪算法性能急剧下降.针对这一问题,本文提出一种基于自适应高斯混合模型(GMM)在线估计未知杂波的改进MHT算法.首先利用自适应GMM拟合未知杂波空间分布,并自适应地估计出波门内的杂波强度;然后将其应用于MHT处理中,有效改善航迹得分计算和最优假设航迹估计的准确性,进而实现在杂波未知场景中的稳定跟踪.仿真结果表明,在未知杂波观测场景中,所提算法相比传统MHT算法和MHT-GMM算法获得了更好的数据关联准确性和航迹维持性能.

The spectral reflectance measured in situ is often regarded as the “truth” of objects, which plays an important role in Earth observation applications. However, in situ measurements are influenced by several factors such as atmospheric conditions, illumination and view geometry (I&VG), cloud coverage, and adjacency effects. In order to avoid the influence of these factors, in situ measurements are usually carried out under sunny days and close to noon. However, the impact of I&VG is still present in most cases. At present, people still know little about the influence mechanism of I&VG. Moreover, correcting the impact of I&VG is also a problem that needs to be urgently solved in reflectance spectroscopy. In this work, experiments are carried out using the multi-directional hyperspectral remote sensing simulation facility (MHSRS2F), which allows adjustment and control of the I&VG parameters. This paper proposes an uncertainty evaluation model for I&VG and quantifies the uncertainty caused by different I&VG parameters. Then, the sensitivity of reflectance to I&VG at different wavelengths is explored based on uncertainty models. Finally, a correction model for reflectance under different I&VG conditions is proposed. The results reveal that the uncertainty and sensitivity caused by observation height are relatively high, regardless of the surface heterogeneity. It directly affects the size of the field of view and the physicochemical characteristics of the object. For objects that approximate the Lambertian surface, more attention should be paid to the selection and variation of solar and view zenith angles and view azimuth angles. For objects with surface heterogeneity, the selection and variation of solar azimuth angle, view azimuth angle, and solar zenith angle are more crucial. The correction model proposed in this paper has a 41.25% correction effect on different view zenith angles, but the correction effect on other environmental factors is not significant.

交互式多模型算法(IMM)和基于模糊控制的交互多模型算法(FIMM)是实际中常用的目标跟踪算法,然而其模型集合固定,当需要大量模型覆盖目标机动时,会导致计算量激增,且过多模型可能带来不必要的模型竞争,降低跟踪性能.针对这一缺陷,提出了一种基于模糊控制的改进自适应IMM算法(FAIMM),采用一种模型概率的非线性映射处理方法实时筛选模型子集,剔除无用模型,增加有用模型的权重,并通过模糊推理机制自动调整过程噪声水平,使得算法对不同的目标机动模式具有更强的自适应能力.仿真结果表明,提出的算法跟踪性能优于IMM算法以及FIMM算法,能够更好地匹配目标的机动模式.

传统多假设跟踪(Multiple Hypothesis Tracker,MHT)算法假定杂波强度先验已知,在未知杂波的观测场景中,杂波强度误差将导致数据关联的准确性急剧下降.针对这一问题,本文提出一种基于核密度估计(Kernel Density Estimation,KDE)的在线杂波估计MHT算法.首先利用核密度函数拟合未知的杂波密度函数,并自适应地估计出该时刻波门内的杂波强度;然后利用杂波强度估计值计算假设航迹的得分函数,提高了数据关联的准确性和目标跟踪的稳定性.仿真结果表明,在未知杂波观测场景中,MHT-KDE算法有效改善了航迹的连续性,减少了虚假航迹数.