SAR and Multispectral Image Fusion Based on Dual-Channel Hybrid Attention Block and Dilated Convolution

The fundamental task of image fusion is to extract image features. Due to the difference in channel dimensions between SAR (Synthetic Aperture Radar, SAR) images and Multi spectral (MS) images, it is difficult to fully extract and utilize the high frequency detail information of SAR images and the low frequency spectral information of MS images for existing algorithms to alleviate the detail loss and color distortion problems of fused images. This image fusion algorithm is based on two-channel hybrid attention and dilated convolution. Firstly, a two-channel network is used to extract high-frequency detail features and low-frequency spectral features of SAR and MS images, and then a dilated convolution is introduced to capture the image's context and channel context information, followed by mapping to the hybrid attention module for feature enhancement, and finally, the image is reconstructed by decoding network to fuse the images. And the loss function based on Spectral Angle Distance (SAD) is constructed, which can further alleviate the problems of spectral distortion and detail loss. Experiments show that the fusion network proposed in this paper has a more powerful feature extraction ability for images than other methods and is better than existing algorithms in subjective and objective evaluations.