Cultivated land recognition from remote sensing images based on improved deeplabv3 model

With the development of remote sensing image technology and semantic segmentation technology, using remote sensing image to segment cultivated land area has become an important and challenging task, The current semantic segmentation algorithm used in remote sensing image cultivated land segmentation has some problems, such as low detection accuracy, unable to distinguish narrow roads and so on. This paper is based on DeepLab v3 algorithm. In order to improve its ability to distinguish narrow targets, we improve the network structure, use ResNeSt network, and introduce feature pyramid structure; By using the CCNet self-attention module, the network's ability to obtain image context information is improved, so as to improve the segmentation accuracy; Adding post-processing module in reasoning process. Reduce the false detection rate of the model. The experimental results show that compared with the original DeepLab v3 and other commonly used segmentation models, the improved DeepLab v3-RFCT model improves the detection accuracy and enhances the ability to distinguish narrow roads.