Research on Remote Sensing Image Segmentation Algorithm Based on Res2Net and SANet

In remote sensing intelligent interpretation, the pivotal role of image segmentation is to enable precise delineation and analysis of ground objects. Addressing inherent challenges posed by the unique imaging angles and scopes in remote sensing imagery, this study introduces an advanced segmentation network leveraging Res2Net and SANet architectures. This innovative network adeptly extracts critical attributes of ground objects—such as size, shape, color, and texture—utilizing a sophisticated multi-level integration approach. This methodology ensures effective differentiation of various ground object types. Furthermore, the network integrates a novel channel weight distribution aligned with global features and an attention mechanism, optimizing the retention of pixel-level information and bolstering the capacity to capture essential details. Empirical validation across ten distinct ground object categories has established the algorithm's dominance in remote sensing image segmentation, marked by an impressive overall accuracy of 84.8%. This achievement underscores the practical applicability and efficiency of the proposed solution in the domain of remote sensing analysis.