Abstract Accurate identification of surface defects of power equipment is a key link to ensure the safe operation of the power grid. However, problems include a high missed detection rate of small targets and blurred positioning boundaries. To this end, this study proposes an improved Faster R-CNN (Faster Region Convolutional Neural Network) framework based on dual-path attention enhancement. First, a coordinate attention module is embedded in the feature extraction stage, and the feature response of small defects is enhanced by establishing a channel-space collaborative perception mechanism. Secondly, a bidirectional feature pyramid network is designed, and cross-layer jump connections are used to fuse shallow detail features with deep semantic information. Finally, a dynamic intersection-over-union loss function is introduced to construct a scale-adaptive bounding box regression mechanism. In the experimental conclusion, the recall rate of the research model in target detection reached 88.7%, an increase of 27.9% compared with the 60.8% of the baseline model. This improvement has significantly improved the model’s defect location and recognition capabilities in complex backgrounds, which can effectively change the robustness of surface defects of power equipment and other performance.
Yang et al. (Fri,) studied this question.