Building footprint extraction is a fundamental task in remote sensing image analysis, with wide-ranging applications in smart city development, urban planning, and disaster response. Despite recent advances, existing methods still struggle to simultaneously capture fine-grained boundary details, corresponding to high-frequency information such as sharp edges and textures, and maintain the integrity of overall building shapes, which depends on low-frequency global structural information. These limitations often lead to blurred or inaccurate boundaries and incomplete building representations, especially in complex urban environments with densely packed or irregularly shaped buildings. This paper presents a novel Frequency-Guided Dual-Encoder Network (FDENet) for building footprint extraction, which combines a graph-aware wavelet decomposition with a frequency-controlled visual state-space model within a dual-encoder attention framework. Specifically, the proposed framework incorporates a Graph-Constrained Wavelet Transformer Block (GAWTB) to establish a dual-encoder structure that operates across both spatial and frequency domains. The high-frequency branch utilizes residual convolution blocks to enhance texture and boundary feature representation, while the low-frequency branch integrates a Frequency-Selective Visual State-Space Model (FSVM) to capture global structural information. Additionally, a Mixed-Frequency Attention Mechanism (MFAM) is introduced to facilitate the deep fusion of high- and low-frequency features, improving their interaction and complementarity. To address the persistent issue of blurred boundaries, a cross-stage adaptive fusion loss function is designed, incorporating an Edge-Aware Loss to further enhance boundary precision. Comprehensive experiments on multiple public building extraction datasets demonstrate that FDENet achieves state-of-the-art (SOTA) performance, significantly improving both boundary sharpness and structural consistency. The source code is available at https://github.com/darkseid-arch/BF-Seg-FDENet.

Semantic segmentation of forward-looking sonar (FLS) images is crucial for enhancing situational awareness in marine environments. However, FLS images are often degraded by environmental noise, similarity noise, and shading effects, which result in low resolution, poor signal-to-noise ratio, and suboptimal image quality. These issues significantly hinder the accuracy of semantic segmentation in FLS images. To address these challenges, we propose a novel method called MFSonar, which is based on the Transformer-Mamba architecture. MFSonar incorporates a Context Channel Denoising Module that exploits the similarity characteristics of local and global features to effectively suppress similarity noise and enhance target features. Additionally, the Multiscale Frequency-Domain Decoding Module integrates multiscale frequency-domain convolution with visual state space blocks, leveraging frequency-domain characteristics to mitigate environmental noise and occlusion shadows. Furthermore, our approach prioritizes local features before global features to achieve effective fusion and enhancement of global semantic features and multiscale local visual information. Extensive comparative experiments across multiple datasets demonstrate that MFSonar achieves state-of-the-art performance. Moreover, ablation studies and visual comparisons on a primary dataset validate the superiority, effectiveness, and uniqueness of our approach. Our implementation is available at https://github.com/NWPUFranklee/PVSonar.git.

Road extraction from remote sensing imagery is crucial for a variety of applications, including transportation monitoring, disaster response, and urban planning. However, existing methods often fail to accurately delineate sparse, curvilinear, and boundary-blurred road structures in high-resolution images, leading to incomplete detail preservation and inadequate contextual understanding. To address these challenges, we propose a novel Frequency-Driven Dual-Branch Mamba Network (FDMamba) for precise road extraction from remote sensing imagery. The proposed FDMamba integrates frequency-aware modeling with a dual-branch architecture, enabling collaborative learning of fine-grained edge details and global spatial dependencies. Specifically, FDMamba comprises three key modules: a Fourier Reconstruction Attention Mechanism (FRAM) to enhance high-frequency boundary information and low-frequency structural representation; a Rotation-aware Mamba Module (RAMamba) that leverages multi-path state space modeling for robust directional perception of road structures; and a Phase-guided Feature Fusion Module (PFFM) for effective cross-scale alignment and fusion of high- and low-frequency features. Furthermore, to mitigate the issue of blurred or ambiguous boundaries, we introduce a hybrid loss function that combines binary cross-entropy, focal loss, and frequency-aware loss, explicitly guiding the model to focus on edge structure and multi-frequency complementary information. Extensive experiments on three benchmark datasets, CHN6-CUG, DeepGlobe, and Massachusetts, demonstrate that FDMamba consistently outperforms state-of-the-art methods in terms of F1-score and IoU, achieving superior boundary clarity and structural continuity while preserving overall geometric integrity. The code is available at https://github.com/darkseid-arch/RE-FDMamba.

Small non-coding molecules known as microRNAs (miRNAs) play a critical role in disease diagnosis, treatment, and prognosis evaluation. Traditional wet-lab methods for validating miRNA–disease associations are often time-consuming and inefficient. With the advancement of high-throughput sequencing technologies, deep learning methods have become effective tools for uncovering potential patterns in miRNA–disease associations and revealing novel biological insights. Most of the existing approaches focus primarily on individual molecular behavior, overlooking interactions at the multi-molecular level. Conventional graph neural network (GNN) models struggle to generalize to heterogeneous graphs, and as network depth increases, node representations become indistinguishable due to over-smoothing, resulting in reduced predictive performance. GONNMDA first integrates similarity features from multiple data sources and applies noise reduction to obtain a reconstructed, comprehensive similarity representation. It then constructs heterogeneous graphs and applies a root–tree hierarchical alignment, along with an ordered gating message-passing mechanism, effectively addressing the challenges of heterogeneity and over-smoothing. Finally, a multilayer perceptron is employed to produce the final association predictions. To evaluate the effectiveness of GONNMDA, we conducted extensive experiments where the model achieved an AUC of 95.49% and an AUPR of 95.32%. The results demonstrate that GONNMDA outperforms several recent state-of-the-art methods. In addition, case studies and survival analyses on three common human cancers—breast cancer, rectal cancer, and lung cancer—further validate the effectiveness and reliability of GONNMDA in predicting miRNA–disease associations.

High-precision semantic segmentation of multimodal remote sensing imagery is critical for applications such as urban planning and land monitoring. However, existing methods often suffer from cross-modal semantic misalignment, geometric discontinuity, and loss of frequency information, resulting in blurred boundaries and incomplete object representations. To overcome these challenges, we present GLFRNet, a novel Global-Local Alignment and Frequency Reconstruction Fusion Network for multimodal semantic segmentation. GLFRNet introduces three core innovations. First, a Dual-Stage Structure-Aware Encoder (DSAE) is designed to combine convolutional and state space modeling, thereby enhancing both local texture extraction and global geometric feature representation. Second, a Global-Local Feature Projection Alignment Adapter (FPAA) is incorporated to explicitly align semantic and structural information between optical and DSM modalities. Third, a Phase-Guided Frequency Reconstruction Module (PGFRM) leverages the Fourier transform to jointly reconstruct low-frequency structures from DSM and high-frequency textures from optical images. Extensive experiments on the ISPRS Vaihingen and Potsdam datasets demonstrate that GLFRNet achieves state-of-the-art performance, with mIoU, OA, and mF1 scores of 87.03%, 91.93%, and 92.83% on Potsdam, and 85.35%, 93.27%, and 92.33% on Vaihingen, respectively. GLFRNet significantly outperforms previous methods, particularly in segmenting elongated objects, occluded regions, and complex terrain. All codes will be available on https://github.com/darkseid-arch/GLFRNet.

Accurate semantic segmentation of high-resolution remote sensing images (RSIs) requires strong structural awareness and precise boundary delineation, which remain challenging due to complex spatial patterns and semantic ambiguities. To address these issues, we propose ReSMamba, a relational structure-aware segmentation framework based on linear state space modeling. Specifically, we introduce the Structure-Aware Topology Mamba Block (STMamba), which incorporates topological priors into the state space to capture global context and maintain directional consistency across spatial dimensions. A Structure-Aware Semantic Assembly Module (SASA) is further developed, leveraging graph-based feature reconstruction and class-guided semantic prototypes to mitigate feature distortion and reduce inter-class confusion. Additionally, we design a Directional Connectivity Prediction Module (DCPM) to explicitly model and enhance pixel-level connectivity in multiple directions, significantly improving boundary continuity and the recovery of fine structural details. Extensive experiments demonstrate that ReSMamba consistently outperforms state-of-the-art methods, achieving mIoUs of 79.74$\%$, 79.10$\%$, and 60.79$\%$ on the ISPRS Potsdam, ISPRS Vaihingen, and LoveDA datasets, respectively. Ablation studies further validate the unique effectiveness and complementarity of each proposed component. These results highlight the strong structural modeling capabilities and practical value of ReSMamba for intelligent remote sensing image interpretation. The code and segmentation results will be released at https://github.com/darkseid-arch/ReSMamba.