The spatial attention mechanism has been widely employed in the semantic segmentation of remote sensing images due to its exceptional capacity for modeling long-range dependencies. However, the analysis performance of remote sensing images can be reduced owing to their large intra-class variance and complex spatial structures. The vanilla spatial attention mechanism relies on the dense affine operations and a fixed scanning mechanism, which often introduces a large amount of redundant contextual semantic information and lacks consideration of cross-directional semantic connections. This paper proposes a new Cross-scan Semantic Cluster Network (CSCN) with integrated Semantic Filtering Contextual Cluster (SFCC) and Cross-scan Scene Coupling Attention (CSCA) modules to address these limitations. Specifically, the SFCC is designed to filter redundant information; feature tokens are clustered into semantically related regions, effectively identifying local features and reducing the impact of intra-class variance. CSCA effectively addresses the challenges of complex spatial geographic backgrounds by decomposing scene information into object distributions and global representations, using scene coupling and cross-scanning mechanisms and computing attention from different directions. Combining SFCC and CSCA, CSCN not only effectively segments various geographic spatial objects in complex scenes but also has low model complexity. The experimental results on three benchmark datasets demonstrate the outstanding performance of the attention model generated using this approach.

Long-tailed distribution is a common issue in remote sensing image classification (RSIC), and many datasets suffer from severe class imbalance. This imbalance often causes the classifier to focus on the head classes with more samples, neglecting the tail classes. As a result, the precision of the tail classes is reduced, which in turn affects the generalization ability of the classifier. To address this problem, a hybrid network based on Cross-Space Learning and Perception-Driven Mechanism (CLPM) is proposed to improve the classification accuracy of samples from the tail classes. The CLPM network consists of three components. The Cross-Space Representation Learning Branch (CS-RLB) is designed to enhance the representation capability of tail-class samples by integrating multi-scale and multi-region spatial features. In parallel, the Adaptive Perception Classification Branch (AP-CB) dynamically adjusts the receptive fields to improve generalization across different resolutions and challenging scenarios. Additionally, the CLPM innovatively applies the von Mises-Fisher (vMF) distribution to remote sensing images for high-dimensional inter-class feature modeling. Building on this, a vMF-based Contrastive Loss (vMF-CL) function is proposed. This approach effectively coordinates the learning processes of head and tail classes while enhancing the precision of feature representation. The effectiveness of CLPM is validated on datasets with varying balance ratios, including SIRI-WHU, CLRS, and NWPU-RESISC45. Results show that CLPM significantly improves tail classes accuracy while maintaining high recognition rates for head and middle classes. Compared to existing methods, CLPM has significant advantages in overall recognition accuracy, the long-tailed problem, and diversity adaptation.