Remote sensing image semantic segmentation method based on multi-scale feature fusion and SAM

The invention provides a remote sensing image semantic segmentation method based on multi-scale fusion and a spatial attention module, and aims to more accurately predict a pixel point relationship between object boundaries and enhance a semantic segmentation effect of a remote sensing image by fusing multi-scale features of the image and enhancing a position mapping relationship between feature points. The method comprises the implementation steps of firstly constructing a Swinin-Transfomer module, a multi-scale feature fusion module, a double-path space attention decoder module and a processing module, training data sequentially passes through the four modules to train a SwinDSA-meige network, and finally performing semantic segmentation on a predicted image. According to the method, the problems of position mapping relation and context relation in image segmentation can be improved, and the method is better than most methods in small object segmentation. 本发明提出一种基于多尺度融合和空间注意力模块的遥感图像语义分割方法，旨在通过对图像的多尺度特征进行融合和