Multiscale defect detection of photovoltaic infrared image based on feature enhancement

The use of drones equipped with infrared cameras to inspect photovoltaic (PV) power stations has gradually become the mainstream. Aiming at the problem that the infrared image of PV modules obtained by aerial photography has complex background, various defect sizes, and especially difficult detection of small size defects, a multi-scale defect detection method based on feature enhancement is proposed. Firstly, the receptive field block (RFB) is introduced into the backbone network to replace the original SPPF structure, so as to enhance the feature extraction ability of the model in complex background. Secondly, the weighted bidirectional feature pyramid network (BiFPN) is used to replace the traditional PANet structure in the neck network, and the efficient fusion of multi-scale features is realized through the bidirectional connection and weighted feature fusion mechanism. Finally, through the experimental verification on the infrared image defect dataset of PV modules, the proposed model is significantly better than the classical detection algorithms such as YOLOv5, YOLOX, and YOLOv7 in detection accuracy, and can effectively adapt to the defect detection tasks of PV power stations in different scenarios.