A method for power grid line de-icing effect detection based on deep learning

To address the issues of low efficiency and insufficient accuracy in power grid line de-icing effect detection, a method for power grid line de-icing effect detection based on deep learning is proposed. This model incorporates advanced residual modules and attention mechanisms. Specifically, residual blocks with deformable convolutions and the MFEDSB module are introduced in the encoding stage, while a BAMB-based attention mechanism is applied in both the encoding and decoding stages. Additionally, the edge-aware and adaptively adjusted EEAUP upsampling module is utilized to enhance feature detail retention during the upsampling process. Experimental results show that compared to U-Net, the proposed model improves SEN by 6.75\% and F1 by $\mathbf{2. 3 8 \%}$. Compared to DRNet, SEN is increased by $\mathbf{4. 1 5 \%}$ and F1 by $2.77 \%$.