GSHC-Net: Image Restoration Network Based on Gated-Supervised and Hierarchical Cascaded for Transmission System in Adverse Weather

During the inspection of the transmission system, adverse weather conditions may result in visual occlusion, rendering the condition of transmission lines and power equipment components difficult to identify. To comprehensively address this issue, the study proposes a multi-stage image restoration network based on gated-supervised and hierarchical cascaded feature fusion scheme (GSHC-Net) to mitigate degradation effects in inspection images caused by adverse weather conditions. The main contribution is on two aspects: Firstly, to tackle the blurred characteristic details of power equipment components like transmission lines under adverse weather conditions, the gated-supervised attention network (GSAN) is proposed. This method employs a gated supervised attention mechanism and an adaptive convolution kernel to restore the fuzzy target details in recovered images so as to improve the overall visual quality. Secondly, a hierarchical cascaded feature fusion scheme (HCFFS) is presented to utilize the contextual semantic feature information efficiently. This scheme minimizes information loss during recovery processes for enhanced authenticity and accuracy of the recovery results. Finally, experiments reveal that GSHC-Net can improve the visual quality indexes of degraded images under different adverse weather, such as PSNR, by 4% on average.