Deep learning-based change detection (CD) is widely used for same-resolution images. However, high-resolution remote sensing images are often not continuously available over time, necessitating the ability to handle images with arbitrary resolution differences. Previous super-resolution (SR)-assisted CD methods are limited to processing bi-temporal images with fixed resolution differences and fail to adequately preserve edges. Moreover, existing CD methods ignore the potential of frequency domain analysis for difference features. To address these limitations, this article proposes a novel resolution-arbitrary CD network that enables CD with input images of arbitrary resolution differences. The framework is distinctive because it can solve resolution mismatch and boost accuracy by enabling high- and low-level tasks (i.e., CD and arbitrary scale SR) to learn from each other. A gradient-enhanced magnification-arbitrary module is designed to resize input images to a uniform size while preserving edge details better. Furthermore, unlike existing methods that solely consider difference features in spatial domain, a difference dual-domain learning module is designed to capture global contextual and local information from difference features in the spatial and frequency domains, improving the change representation. Experiments on four datasets including real and simulated data demonstrate that our method outperforms state-of-the-art methods in handling arbitrary resolution differences. The codes are available at https://github.com/LYT-max/RACDNet.

Distributed array radar(DAR)has the benefits including flexible deployment and excellent spatial resolution.However,the independent clock and oscillator configurations of each radar unit,along with the instability introduced by trigger signals transmitted over the feeder link,introduce errors in time and phase synchronization,thereby reducing the accuracy of coherent synthesis in a DAR.This paper proposes a time and phase synchronization error estimation method based on space-time spectral entropy.The distance-angle space-time two-dimensional spectrum is first constructed by establishing a space-time covariance matrix with time and phase synchronization faults.Then,based on entropy theory,the correspondence between synchronization errors and space-time spectral shape uncertainty is established.The time and phase synchronization errors are estimated by optimizing the space-time spectral entropy to minimize its value.Simulation experiments validate the accuracy of the proposed method,particularly exhibiting good estimation performance under low signal-to-noise ratio.

Distributed array radar (DAR) achieves large-aperture performance by combining small subarrays at different locations, which reducing system burden and manufacturing complexity while meeting low-cost and highresolution requirements in modern sensing. This paper proposes a 2-dimensional (2D) subarray optimization method for regular DARs with identical subarrays, aiming to minimize beamwidth (BW) and maximum sidelobe level (MSLL) under physical layout constraints. A directional expanded beam pattern (DiEBP) is introduced to reformulate non-analytic optimization objectives into a differentiable form, which enables gradient-descent updates within an alternating direction method of multipliers (ADMM) framework for this non-convex problem, with exponential smoothing and Monte Carlo tree search (MCTS) used for stabilization and initialization. Numerical experiments show that, compared with existing algorithms, the proposed method achieves about 10-15% narrower BW and 2 dB lower MSLL on average, along with a reduced Cramer-Rao Bound (CRB). Its effectiveness is further validated through DOA estimation and near-field imaging experiments.

Multimodal learning has played a pivotal role in survival prediction, particularly in integrating pathological images and genomic data for improving predictive performance. Pathological images provide macroscopic histological information about tumor morphology, while genomic data reveal molecular-level genetic characteristics. The integration of these two modalities enables a comprehensive characterization of tumor heterogeneity and disease progression mechanisms. Despite recent advances in multimodal integration that have significantly enhanced prognostic accuracy, challenges remain in effectively analyzing high-dimensional and heterogeneous whole-slide images and omics data. Current Transformer-based sequence modeling approaches suffer from limited computational efficiency when processing long feature sequences and capturing complex cross-modal interactions. To address these challenges, we propose an innovative cross-modal receptance weighted key-value (RWKV)-based framework, termed Surv-RWKV, for survival prediction. This framework integrates RWKV-based sequence modeling with advanced multimodal fusion strategies to enhance both predictive accuracy and model efficiency. Specifically, Surv-RWKV employs parallel RWKV-based encoders to model long-range dependencies in WSI tissue cluster patterns and genomic pathway activation profiles, achieving improved prognostic performance with optimized computational efficiency. Subsequently, a transport-based optimal cross-modal alignment module is introduced to establish semantic correspondences between histopathological and genomic feature spaces. Furthermore, a progressive feature fusion strategy is implemented to enable effective cross-modal interaction. An RWKV-based shallow fusion module is first developed to explore cross-modal dependencies through spatial-channel hybrid operations, thereby enhancing the representational quality of fused features. A cross-RWKV deep interaction module is then designed to further strengthen information synthesis via iterative cross-attention mechanisms, while simultaneously reinforcing intra-modal representation learning and cross-modal knowledge transfer. Surv-RWKV is expected to effectively capture such cross-modal correlations, thereby improving the accuracy and interpretability of survival predictions. Extensive validation across five TCGA cancer cohorts demonstrates that Surv-RWKV achieves state-of-the-art predictive performance with superior computational efficiency.

Distributed aperture radar (DAR) emulates an equivalent large-aperture performance by synthesizing multiple subarrays placed at different locations, offering a cost-effective solution for high-resolution radar systems. However, subarray spacing leads to spatially sparse sampling, resulting in grating lobes that degrade subsequent image interpretation. In this study, a near-field grating lobe suppression method based on subarray migration and spatial multiapodization (MA) is proposed for nonuniform 2-D sparse DAR imaging. In our method, an optimized subarray response with sharp main lobe and low sidelobe level is achieved via parallel subarray migration and MA processing, which is further utilized to mitigate the prominent grating lobes arising from sparse intersubarray placement. To address the spatial variation of array radiation characteristics under near-field effects, the proposed method is integrated into the back-projection algorithm (BPA) framework using an adaptive weighting strategy. Numerical experiments on extended targets demonstrate the superiority of the proposed method in grating lobe suppression, enabling high-fidelity, ambiguity-free near-field imaging with 2-D sparse DAR systems.

Polarization and intensity images fusion (PIF) has extracted extensive attentions as it can generate images with clear scene information and salient texture details of the object surface that are important for downstream applications. However, existing deep learning-based PIF methods usually lack interpretability and ignore the interactions among multi-modal features. To this end, we propose a novel interpretable low-rank sparse representation guided fusion network for polarization and intensity images (termed LSRNet). Specifically, a low-rank sparse representation deep unfolding module is designed to acquire the base and detail features of the source images, with the ability of improving the interpretability of the network. In addition, a cross-modal connection complementary feature extraction module is proposed, which aims to establish dependency among features of multi-modalities to fully extract complementary features of the source images. In order to demonstrate the validity of our LSRNet and take into account shortcomings of existing datasets for PIF, a multi-scene polarization and intensity image dataset, named MSPI dataset, is constructed, which includes 1034 high-resolution aligned image pairs. According to the best of our knowledge, this is the most comprehensive dataset for PIF that with a large number of image pairs, high resolution and multiple scene types. Extensive experiments on our MSPI dataset and two publicly available datasets (i.e., 12CFC and HCP) demonstrate the superior fusion performance, generalization ability, and desirable running efficiency of our LSRNet. Our codes and dataset will be publicly available at https://github.com/thebinyang/LSRNet.

Infrared and visible image fusion (IVF) encounters two challenges: 1) image unregistered, where parallax leads to blurred fusion results, and 2) adverse imaging conditions, which introduces outlier data (such as haze) that significantly degrade image registered and fusion performance. Existing IVF methods fail to account for the joint impact of unregistered and haze factor, often neglecting these challenges or addressing them in isolation. However, unregistered and haze are intrinsically coupled in fusion task. For example, haze can obscure structural details, leading to inaccurate registration and ultimately degrading fusion performance. To address the intrinsic coupling between unregistered and haze, we propose a progressive task-cooperative processing pipeline—dehazing, registration, and fusion—to achieve robust fusion of unregistered images under hazy conditions. To mitigate the challenges posed by architectural complexity and excessive parameter overhead in task progressive learning, we propose a lightweight model approximation paradigm through hierarchical knowledge distillation. The framework employs a stage-wise distillation optimization strategy that synergistically integrates: 1) primary task-specific distillation for modality-aware feature extraction, and 2) progressive task-cooperative distillation for fusion-oriented representation learning, which can improve fusion robustness for unregistered multi-modal inputs in hazy conditions. Extensive experiments demonstrate that our method achieves significantly superior performance compared to the State-Of-The-Art (SOTA) methods.

Understanding the formation and development mechanisms of soil heavy metal(loid)s (HMs) is essential for targeted prevention and sustainable soil management. However, these mechanisms remain incompletely resolved because pollution reflects diverse inputs and complex environmental interactions. A deep learning model (HMNet) was developed to adaptively integrate multi-source open-access datasets and delineate contamination patterns at fine spatial resolution, thereby reflecting the integrated effects of diverse inputs on contamination distribution. The result demonstrates that HMNet has robust performance in both monitoring (average Kappa of 0.777, Recall of 0.815, and overall accuracy of 82.82%) and generalization (average Kappa of 0.762, Recall of 0.799, and overall accuracy of 81.61%). Driver attribution using DeepSHAP, combined with correlation analysis, revealed a hierarchical structure of controls, with anthropogenic activities acting as the dominant external inputs, soil physicochemical properties governing HM retention and mobility, and vegetation, topography and climate modulating spatial heterogeneity. Structural equation modeling was further employed to quantify direct and indirect interaction pathways, indicating that anthropogenic activities elevate HM accumulation risk primarily by weakening soil-vegetation buffering capacity, whereas climate-weathering processes indirectly enhance HM stabilization by promoting parent-material alteration and vegetation recovery. Building on these findings, management strategies were formulated that prioritize interventions according to regulatory risk class and dominant process controls. Recommended strategies include source identification and industrial or traffic abatement, fertilizer-quality control and precision nutrient management, erosion control, sediment interception and drainage design, and site-specific soil amendments and vegetation restoration, implemented through pilot testing, targeted monitoring, and adaptive evaluation. Overall, this study establishes an integrated framework linking monitoring, mechanism analysis, and management guidance, providing a process-based paradigm for refined understanding and control of soil HM pollution.

Natural disasters pose a threat to the safety of human life and buildings. Rapid and accurate building damage assessment (BDA) on remote sensing images is crucial for disaster response and recovery. However, most methods are constructed on the ideal co-registered bitemporal remote sensing images, neglecting the misalignment in practice. In this paper, we propose a novel building damage assessment method, termed FlowMamba, which can effectively handle the offset between the pre- and post-disaster images in BDA task. Specifically, a vision mamba backbone with four stages is utilized to extract multi scale features from the pre- and post-disaster images. Then, a differential optical flow alignment module is designed to estimate shift matrix to align pre- and post-disaster features. Furthermore, a category distance-aware loss function is tailored for the BDA task, which replaces fixed binary values of the penalty factors in to soft values of inter-class distance. Extensive experiments on the xBD dataset, the BRIGHT dataset and four out-of-distribution disaster scenarios validate the robustness and effectiveness of the proposed FlowMamba. Our code is available at https://github.com/flying318/FlowMamba.

Deep learning-based change detection (CD) has gained popularity in the field of remote sensing. However, most deep learning-based methods share a common limitation on the availability of massive high-quality labeled training data, where noisy labels are inevitably introduced during the annotation process. Noisy labels induce a surge in epistemic uncertainty and aleatoric uncertainty, thereby compromising the CD performance. Existing CD methods overlook the problem of noisy labels and fail to handle inherent uncertainties. To address this limitation, we propose a novel fuzzy logic-integrated network for robust CD with noisy labels, termed FuzzyCD, which integrates fuzzy logic to effectively model the two types of uncertainties (i.e., epistemic uncertainty and aleatoric uncertainty), mitigating the interference of noisy labels. Specifically, an epistemic uncertainty estimation (EUEU) unit is proposed, which treats xor as fuzzy logic for feature fuzzification, thereby enhancing the model's cognitive capacity. In addition, an aleatoric uncertainty estimation unit (AUEU) is designed, which characterizes aleatoric uncertainty through probabilistic inference. By transforming the input feature space into a fuzzy feature space, the two fuzzy logic units significantly alleviate the negative impact of noisy labels, achieving improved CD performance and enhanced model robustness to noisy labels. Comprehensive experiments on three public CD datasets (e.g., LEVIR-CD, CDD, and WHU-CD) validate the effectiveness and superiority of the proposed FuzzyCD compared to other state-of-the-art methods.

Infrared and visible image fusion aims to integrate complementary multi-modal information into a single fused result. However, existing methods 1) fail to account for the degradation visible images under adverse weather conditions, thereby compromising fusion performance; and 2) rely on fixed network architectures, limiting their adaptability to diverse degradation scenarios. To address these issues, we propose a one-stop degradation-aware image fusion framework for multi-degradation scenarios driven by a large language model (MdaIF). Given the distinct scattering characteristics of different degradation scenarios (e.g., haze, rain, and snow) in atmospheric transmission, a mixture-of-experts (MoE) system is introduced to tackle image fusion across multiple degradation scenarios. To adaptively extract diverse weather-aware degradation knowledge and scene feature representations, collectively referred to as the semantic prior, we employ a pre-trained vision-language model (VLM) in our framework. Guided by the semantic prior, we propose degradation-aware channel attention module (DCAM), which employ degradation prototype decomposition to facilitate multi-modal feature interaction in channel domain. In addition, to achieve effective expert routing, the semantic prior and channel-domain modulated features are utilized to guide the MoE, enabling robust image fusion in complex degradation scenarios. Extensive experiments validate the effectiveness of our MdaIF, demonstrating superior performance over SOTA methods.

The fusion of low-resolution hyperspectral images (LR-HSI) with high-resolution images (HR-MSI) can significantly enhance the spatial resolution of hyperspectral images (HSI), thereby improving its applicability in HSI tasks. Current convolutional neural network (CNN)-based HR-MSI and LR-HSI fusion networks primarily focus on local neighbourhood relations, often overlooking global feature mappings. Some approaches integrate transformers to explore global intrinsic feature relationships, albeit at the cost of requiring extensive data. Moreover, existing fusion methods predominantly concentrate on learning modality-shared features. To address these limitations, this study introduces a local adaptive HR-MSI and LR-HSI feature pyramid fusion network based on N-Gram transformer architecture. Inspired by N-Gram models in natural language processing, N-Gram models are incorporated with sliding window self-attention (WSA) within transformers to enhance the interaction of feature domain information. Group convolutions are employed to reduce channels and enhance computational efficiency. Additionally, a novel local adaptive fusion strategy is proposed to effectively integrate spatial and spectral information from both modalities while mitigating cross-modal disparities. This strategy carefully balances local specificity and global context through a designed global bias mechanism. Furthermore, the adoption of a feature pyramid structure as the core of the HSI-MSI fusion model facilitates multi-scale feature extraction and fusion. Extensive experimental validation demonstrates that our proposed framework significantly outperforms existing methods in enhancing the quality of fused images and in classification tasks.

Existing mainstream infrared and visible image fusion methods mainly focus on enhancing local features or global features, lacking hierarchical interaction between different scales. Recently, the Mamba architecture based on the state space model has demonstrated powerful capabilities in natural language tasks, enabling efficient extraction of global features. Based on this, this paper proposes a state space-based multi-scale feature fusion network framework, termed HSFusion. First, we propose a Multi-Scale Feature Extraction Network, its integrates the advantages of global and local feature extraction from the State Space Model (SSM) and convolutional network through the Residue Multi-Head SSM Encoder. This effectively preserves both the global and local information of features while significantly reducing noise interference. Second, a Cross-Level Global-Local Attention Model is designed to model full-scale global-local feature hierarchical relationships at different feature scales. It simulates the relationships between global and local hierarchies across scales, promoting high-level semantic associations of images. Finally, an adaptive feature fusion module is adopted to dynamically realize cross-modal information interaction. Through a gating mechanism, it adaptively models the global feature relationships of different modalities. Experimental results on three distinct datasets demonstrate that the proposed multi-scale state-space framework excels in image fusion tasks, offering significant advantages in both performance and efficiency compared to existing methods, while also demonstrating robust generalization ability in advanced visual tasks.

Change detection (CD) with synthetic aperture radar (SAR) images remains pivotal for environmental monitoring and disaster management. Deep learning has powerful feature extraction capabilities for CD, but suffers from complex architectures and limited interpretability. While BLSs demonstrate advantages in structural simplicity and interpretability, their feature representation capacity remains constrained. In high-precision CD with SAR images, strong feature representation capability is required, along with an uncomplicated framework and high interpretability. Therefore, a novel paradigm named PC-BiBL is proposed which achieves seamless integration of deep learning and broad learning. On the one hand, it employs a hierarchical cross-convolutional encoding (HCCE) module that uses pseudo-random cross-convolution (PCConv) for hierarchical cross-feature representation, aggregating contextual information. PCConv is an untrained convolution layer, which can utilize specialized pseudo-random kernels to extract features from bitemporal SAR images. On the other hand, since back-propagation algorithms are not required, the features can be directly fed into the bifurcated broad learning (BiBL) module for node expansion and direct parameter computation. BiBL constructs dual-branch nodes and computes their difference nodes, explicitly fusing bitemporal features while highlighting change information—an advancement over traditional BLS. Experiments on five SAR datasets demonstrate the state-of-the-art performance of PC-BiBL, surpassing existing methods in accuracy and robustness. Quantitative metrics and visual analyses confirm its superiority in handling speckle noise and preserving boundary information.

Hyperspectral and multispectral image fusion (HMIF) enhances the representational capability of remote sensing imagery by jointly exploiting spectral and spatial information. However, existing methods still struggle to jointly preserve local spatial details and global spectral consistency. Specifically, CNN-based methods capture local textures well but struggle with global spectral consistency, while Transformer-based methods leverage long-range dependencies but lack local structural inductive bias, often causing blurred fine details. In this paper, we propose a graph-based spatial–frequency learning framework (GSFL) that integrates a scene-adaptive cross-modal graph (SACG) and channel-aware frequency modulation (CAFM), simultaneously enhancing spatial details and spectral consistency. Specifically, SACG introduces a cross-modal graph with graph attention, explicitly capturing structure-aware long-range dependencies for preserving spectral integrity; unlike existing GAT-based methods, it adaptively determines neighborhood size using global gradients and inter-band Pearson correlations, enabling robust performance across diverse scenes. Meanwhile, CAFM designs a cross-domain spatial-guided frequency modulation paradigm, leveraging spatial activation intensities as content-aware priors to dynamically construct spectral gates; unlike content-agnostic static filtering or purely frequency-driven attention, it enables adaptive enhancement of critical high-frequency components while suppressing noise, effectively restoring sharp structural details. To further improve performance, we propose a complexity-aware expert transformation (CAET) built upon a differentiable mixture-of-experts (MoE) framework to refine joint representations. In contrast to existing methods with uniform fixed transformations, CAET adaptively reweights multiple isomorphic experts based on local scene complexity, facilitating deep semantic enhancement and fine-grained feature recalibration. Comprehensive experiments on several benchmark datasets demonstrate that the proposed GSFL method achieves superior performance, surpassing several state-of-the-art methods in both quantitative metrics and qualitative evaluations, with ablation studies validating the contribution of each component. The source code and datasets are available at https://github.com/weiweiyaobaoshou/GSFL.

Pastures are the main feed source for cattle in Brazil, and pasture-based production systems require frequent measurements and monitoring of forage productivity and quality to optimize animal performance. In this context, this study aimed to develop and evaluate machine learning models for the quantitative estimation of dry matter (DM), crude protein (CP), and neutral detergent fiber (NDF) in tropical pastures using multispectral and photogrammetric data acquired by an airborne sensor mounted on a remotely piloted aircraft (RPA), as well as to generate maps of the spatiotemporal variability of these parameters. The assessment was conducted in a commercial pasture area of approximately 200 hectares, divided into 19 paddocks, cultivated with Urochloa brizantha cv. Marandu and managed under rotational grazing during the year 2023. A total of 190 field samples were collected and associated with multispectral reflectance data, vegetation indices, digital surface models, and pasture management information. Regression models based on random forest (RF), combined with recursive feature elimination (RFE), were developed to estimate forage productivity and nutritional attributes. The general DM prediction model showed good performance (R² = 0.74), indicating strong potential for estimating pasture productivity. Moderate and low predictive performances were obtained for NDF (R² = 0.56) and CP (R² = 0.48), respectively. External validation revealed reduced model transferability (R² < 0.2), highlighting the influence of temporal variability and the intrinsic complexity of forage nutritional parameters. Spatial analyses revealed pronounced spatiotemporal variability in DM, CP, and NDF across the paddocks. Overall, the results demonstrate that airborne multispectral sensing combined with machine learning is a promising approach for large-scale monitoring of pasture productivity and quality, supporting decision-making in pasture-based livestock systems.

Owing to the limitations of imaging sensors, hyperspectral image (HSI) typically suffer from low spatial resolution. To obtain HSI with high spatial resolution, HSI-MSI fusion has become an effective and widely adopted technique. However, existing deep learning-based HSI-MSI fusion methods often struggle to capture local details and global context, especially when features span multiple scales. To address these issues, we propose a novel Transformer-based multiscale cross-attention fusion network (MCA-Net). MCA-Net integrates three key innovations to overcome these challenges. Firstly, the heterogeneous convolution parallel attention enhancement module (HCPAEM) combines dilated depthwise separable convolutions with parallel attention mechanisms to effectively enhance the representation of both local and global features. Secondly, the multiscale local-global feature extraction module (MLGFEM) integrates convolutional neural networks(CNNs), Transformers, and multiscale feature extraction strategies, modeling non-local and complementary information at multiple scales. Finally, the deep cross-attention fusion module (DCAFM) employs deep cross-attention mechanism to model the correlation between HSI and MSI, promoting the comprehensive fusion of spatial-spectral features. To validate the effectiveness and superiority of MCA-Net, we conducted comparative experiments on six datasets, including Pavia Center, Pavia University, Washington DC, Botswana, Chikusei and WHU-MHF. Experimental results demonstrate significant improvements over state-of-the-art fusion methods. For instance, on the Washington DC dataset, compared with the state-of-the-art method among the comparison algorithms, our method improves PSNR by 11.47