An Improved AOD-Net Dehazing Algorithm for Image Detail Restoration
Jindong Chen, Luo Ping, Xiafu Lv · 2024
The advent of industrialization and the development of automation technology have gradually integrated machine vision technology into our production and daily lives. However, some production operations generate fog, which reduces the quality of images and interferes with subsequent advanced visual tasks. To meet the real-time and lightweight requirements of the production process, ADD-Net has been widely applied. However, during the training process, there are issues such as loss of image details, limited generalization capability, and weak dehazing ability under low-light conditions. In this paper, we propose an improved ADD-Net algorithm specifically targeting image detail restoration. We incorporate a CA attention module into the feature extraction module to enhance the network's feature representation capability. Additionally, we employ a multi-scale model to enhance the ability to restore image details. Finally, our algorithm utilizes a combined loss function of L1 and MS-SSIM, and optimizes the search to obtain the optimal parameters for the model. Experimental results demonstrate that our algorithm not only achieves high SSIM and PSNR values but also exhibits excellent ability to restore image details.