Power Grid Field Violation Recognition Algorithm Based on Enhanced YOLOv5

Abstract Aiming at the problem that Yolov5 is difficult to accurately detect backlighting and small samples in national grid power transmission distribution operations, studies illegal behavior technology of “two wear one wear” operation, proposes a violation recognition algorithm based on enhanced YOLOv5. First, add new detection layer use BiFPN (bi-directional feature pyramid network) for fusion, so layers different scales can better learn weight enhance fusion ability. Secondly, CBAM (Convolutional Block Attention Module) module before output map make full channel spatial information achieve model accuracy recall. The experimental results six self-made data sets show mAP this method 91.0%, which 5.6% higher than original algorithm, indicating has stronger predictive ability robustness identification personnel safety tools scenarios.