Due to the significant decrease in the recognition performance of monitoring images caused by rain streaks, it delays the perception and judgment of navigation situations by ship drivers. Existing methods cannot adapt to rain streaks of different densities and sizes, and cannot effectively preserve background details. To solve this issue, a deep learning based image rain removal network is proposed, which can present residuals in the detail layer between rainy and clean images. The detail layer can propagate information from the lower layer to the upper layer, which is composed of a cascaded network. The experimental results and data analysis show that the proposed method can perform well in removing rain on both artificially synthesized images and natural real images, with subjective and objective evaluations superior to other comparative algorithms.
维普
万方数据