Research On Cloud Detection Algorithm Of Remote Sensing Images Based On Feature Enhancement Networ

Remote sensing image cloud detection has always been a research hotspot in the field of remote sensing image processing.With the rapid increase of remote sensing data and application demand,cloud detection technology is also developing and improving in order to improve the application effect of remote sensing technology.For complex clouds or mixed scenes of clouds and ground objects,the detection accuracy of traditional methods is low,and it is difficult to accurately distinguish clouds and ground objects.In recent years,cloud detection methods based on deep learning have gradually become a research hotspot,usually through the use of deep learning models such as convolution neural networks.Although the deep learning method has achieved good results,there are still some shortcomings and room for improvement.This paper proposes some improvements according to the shortcomings of the existing methods.To solve the problems of difficult recognition of edge features and low accuracy of feature extractor in remote sensing image cloud detection,this paper proposes a cloud shadow detection algorithm based on multi-level feature enhancement network for remote sensing image.The algorithm uses feature enhancement module to make feature extraction better,and enhances the segmentation ability of edge details through multi-scale feature fusion module and upsampling feature guidance module.Aiming at the low efficiency of global context information acquisition and the loss of deep and important local information in the convolutional neural network,this paper proposes a cloud shadow detection algorithm based on the multi-level fusion of transformer and CNN semantic features in remote sensing images.The algorithm can effectively capture low-level spatial features and high-level semantic context information through the Transformer branch,complete the coding from local features to global features through the CNN branch,and use the fusion module to integrate the advantages of CNN and Transformer.The above problems have been effectively improved.Finally,the effectiveness of the proposed algorithm is proved by experiments.