Semantic Segmentation of High-Resolution Remote Sensing Images Based on RS³Mamba: An Investigation on the Extraction Algorithm of Rural Compound Utilization Status

In this study, we utilize Gaofen-2 satellite remote sensing images to optimize and enhance the extraction of feature information from rural compounds. This is achieved through the implementation of the RS³Mamba+ deep learning model and the construction of Mamba-assisted branching with the aid of multi-directional selective scanning (SS2D) and the STEM network framework. The primary objective of this approach is to capture the long-distance spatial dependence of the compounds in high-resolution remote sensing images. Additionally, it aims to minimize computational loss. The introduction of a multiscale attention feature fusion mechanism is an important development in this field. This new mechanism has been demonstrated to optimize feature extraction and fusion, enhance edge contour extraction accuracy in courtyards, and improve the recognition and differentiation ability of the courtyard and complex texture regions. The feature information of courtyard utilization status is finally extracted using empirical methods. A typical rural area in Weifang City, Shandong Province, is selected as the experimental sample area. The results show that the extraction accuracy reaches an average intersection and merger ratio (mIoU) of 79.64% and a Kappa coefficient of 0.7889. This improves the F1 score by at least 8.35% compared with models such as U-Net, ResNet, Transformer, and so on. Furthermore, the mIoU improves by 7.41%. The mIoU has been enhanced by 7.41%. The efficacy of the algorithm in suppressing false alarms triggered by shadows and intricate textures is noteworthy. It is a valuable instrument for the extraction of compounds from rural compounds by leveraging condition feature information.