A Single Image Rain Removal Algorithm Based on Deep Learning

This paper proposes a single image rain removal algorithm based on deep learning to address the long-standing issues of incomplete rain removal, background blur, and detail loss in image rain removal tasks. Firstly, this article constructs a deep feature extraction network based on the encoder decoder architecture to extract deep level detailed feature information. Secondly, a lightweight feature extraction subnet is proposed, which is first used for shallow feature extraction and then embedded into the deep feature extraction network to fully mine and capture deep level rain pattern information, thus achieving better image restoration. Finally, we tested the network performance on multiple public datasets, and the results showed that our network outperformed other rain removal algorithms in terms of PSNR and SSIM metrics across multiple datasets, with an average metric improvement of O.S3dB compared to the current excellent algorithm MSPFN. This proves that the proposed rain removal network can effectively remove rain streaks and restore images while preserving detailed information.