Multispectral attention-based network for single image deraining

In challenging weather conditions, rain streaks can significantly degrade captured scenes and negatively impact other low-level tasks. Therefore, researching rain removal tasks is of utmost importance. The diversity of rain streak characteristics, such as different orientations, densities, and magnitudes, makes capturing rain information challenging. To address this issue, this paper introduces a multispectral attention-based network called MSABN, which exhibits exceptional performance in the deraining task. To effectively extract rain information, we design a multispectral attention module (MSAM). Moreover, in complex rainy scenarios, image background information is often distorted, leading to the loss of essential details. To address this, we propose a feature enhancement module (FEM) that extracts background features to compensate for the lost original details. Additionally, we employ a context aggregation mechanism (CAM) to enhance the information flow, thereby optimizing the proposed algorithm's performance. Our MSABN demonstrates superior performance compared to state-of-the-art methods for both synthetic and real-world scenes. Moreover, experiments on semantic segmentation and object detection tasks also show that our method outperforms other comparative algorithms.