LFS-YOLO: A PV Panel Defect Detection Algorithm for Drone Infrared Sensors

Photovoltaic (PV) panel hot spot defects will reduce the power generation capacity of PV panels, which will seriously threaten the safe operation of the power station, and hot spot defect detection and timely repair is an important work of PV power plant operation and maintenance. As a key technology to improve the autonomous perception ability of drone infrared sensors, object detection has become the focus of drone inspection in PV power plants. In this article, a hot spot defect detection algorithm according to infrared images of aerial PV is proposed for practical engineering problems such as defects with different morphology, unclear boundary features, and small target defects. First, a multiscale lightweight convolution module C2f_LMSC is designed to realize multiscale feature extraction and reduce the number of parameters and computation amount of the model. Second, the Focal Nets are used to replace the original spatial pyramid pooling-fast (SPPF) module, which improves the model’s ability to capture the features of the defects with inconspicuous boundary features. Furthermore, in order to improve the model’s effectiveness in detecting small target defects, a feature pyramid small object resultful pyramid (SORP) is designed for small target detection. In addition, in order to improve the detection effect of the model on small target defects, a feature pyramid SORP for small target detection is designed. The test results show that the average accuracy of LFS-YOLO model detection is improved by 3.2%–91.9% compared with the pre-improvement period, the defect leakage phenomenon existed in the original model is improved, and the research results can provide technical support for the detection of PV defects by drones.