Defect Detection Of Photovoltaic Array Based On Visible And Infrared Images

The development and use of renewable energy have greatly reduced the global dependence on fossil fuels and greenhouse gas emissions.In the renewable energy,solar energy has the obvious advantages,so the effective utilization of solar energy has become an important development direction for the energy acquisition in the future.In order to ensure stable,efficient,and safe operation of photovoltaic power generation,it is necessary to timely detect the operation status of photovoltaic arrays and identify existing defects.This paper studies the visible and infrared images of photovoltaic arrays,and proposes an accurate and efficient defect detection method based on previous research using deep learning algorithms.The main research contents of this article are as follows:An infrared hot spot detection algorithm is proposed and a structure that can improve the small object detection is designed.During the feature extraction process of the object detection algorithm,two multiscale feature fusion modules are proposed,which perform the feature enhancement on the feature map scale and the receptive field scale,respectively.Using a two-stage dual branch detector and embedding it into the Faster R-CNN algorithm make the original algorithm more sensitive to small objects.A ablation experiment is conducted on the mouse dataset extracted from the COCO dataset for small object detection and comparison experiments are conducted with mainstream related algorithms to prove the effectiveness of the improved algorithm.Finally,more accurate hot spot detection is achieved on infrared images from photovoltaic arrays.The detection results show that the AP50 value iss 93.9%and the average AP value reaches 52.1%.Infrared and visible light images of photovoltaic arrays are used for the defect detection.Target detection algorithms and segmentation algorithms are combined through the region matching and secondary classification modules.Target frames of photovoltaic modules with abnormal regions of infrared images are matched,and a PoI Gathering module is proposed to achieve the aggregation of feature vectors at key points and perform the secondary classification on the detected photovoltaic modules.Finally,by judging whether there are temperature abnormal regions or foreign object occlusions,defective component locations are identified.Different networks are set up for testing.The proposed algorithm achieves the AP50 value of 88.4%for the component detection using lighter ResNetl8 and U2-Net lite networks and IoU index of 85.2%is reached for the segmentation of abnormal temperature region.In additon,the single detection time is about 20ms,enabling the detection of abnormal photovoltaic modules at a faster speed.