A Lightweight Real-Time Image Deraining Network for Maritime Port Scenes

With the rapid development of the marine economy, maritime ports, as global trade hubs, face severe challenges in their safety supervision. Detecting and tracking ship targets in rainy weather environments is one of the key issues. Rainfall can lead to a decrease in image quality, which in turn affects the performance of detection and tracking. Considering that port monitoring equipment usually has limited computing resources, designing a high-performance, lightweight, and real-time processing capability de training algorithm is of great significance for promoting the widespread application of intelligent maritime monitoring technology. This article proposes an improved algorithm based on LPNet to achieve triple optimization for port scenes. A lightweight real-time image rain removal network based on LPNet is proposed to solve the problem of incorrect elimination of background structure and residual rain streaks in lightweight rain removal models. The main improvements include: adjusting the configuration of the number of channels in each layer, effectively controlling the number of parameters, and improving the ability to extract high-frequency components of rain bands; Introducing GhostConv to reduce computational complexity; Design a rain band adaptive feature enhancement module to enhance the response of the rain band area; And a multi-scale enhanced upsampling module (MEUM) is adopted to solve the edge blurring problem caused by upsampling. The experimental results show that compared with other lightweight de training models, the improved model exhibits excellent performance: SSIM index reaches 0.9087, PSNR index reaches 30.58dB, the number of parameters is only 64K, and the inference time is only 37ms, providing efficient and reliable real-time de training technical support for resource limited devices.