Lightweight Image Dehazing via Physics-Guided Neural Networks

Images taken in smoggy weather have problems such as dim colors, low contrast, and target details being obscured by fog. Moreover, the degradation of image quality can reduce the accuracy and robustness of subsequent computer vision tasks. Therefore, the restoration of clear images through image enhancement or physical models is an important technology in the visual perception systems. In this paper, a lightweight image dehazing method combined with physical models is proposed, which aims to achieve a lighter and more effective image dehazing process. Specifically, this paper introduces a dehazing network structure based on a Physics-Guided Neural Network (PGNN). The network structure explicitly restores key intermediate processes in the Atmospheric Scattering Model (ASM) and enhances the physical consistency of the network learning process, thereby achieving the purpose of effective fog removal. In addition, this paper proposes a Physics-Guided Loss (PGL) function based on the physical model, which can guide the network toward the real physical fog removal process. In order to meet the deployment needs, especially the real-time operation requirements on embedded devices or edge platforms, this paper implements a preliminary lightweight design for the network structure and builds a complete experimental evaluation framework to validate the efficacy of the proposed method. Through quantitative analysis and visual comparison, the experimental results demonstrate the significant advantages of PGNN in improving image quality and optimizing deployment efficiency. The ablation study further supports the contribution of PGNN to model performance.