Research on Power Inspection Image Defogging Method Based on Multilevel Feature Fusion Network

Observation of power equipment faces severe challenges under special foggy conditions, which seriously affects the performance of inspectors’ viewing and subsequent visual tasks. Existing image defogging algorithms use a single coding and decoding structure, which leads to problems such as incomplete fog removal and poor detail texture recovery in the defogged image. Aiming at the above problems, this paper constructs a power image defogging dataset and proposes a cross-dimensional power image defogging network, which achieves the defogging task by coupling different image features in high dimensions. Firstly, the high-performance feature extraction module (HPFEM) is utilised to extract the features of different dimensions adequately and efficiently, and then six HPFEM modules of different dimensions are used to form the main network of the whole enhancement network. Finally, the Multi-Level Feature Fusion Module (M-L FFM) is used to effectively integrate the local fine-grained features and global fine-grained features, which makes the image clearer. The experimental results show that the best de-fogging effect is exhibited in terms of fog concentration quality and non-uniform fog quality; the proposed HLFFM method improves the PSNR index and SSIM index by 18.2% and 15.1%, respectively, compared with the existing comparison algorithms.