76 / 2022-03-10 20:11:08

Defect Detection of Monocrystalline Photovoltaic Modules with Electroluminescence Images Based on a YOLOX Model

Monocrystalline photovoltaic module,electroluminescence image,defect detection,YOLOX

Photovoltaic (PV) module defects directly cause power loss of cells, and a large number of defects in a single module may lead to module scrap. Traditional manual inspection of PV modules is time-consuming. Since the defects can be observed clearly in the electroluminescence (EL) images, it is feasible to achieve automatic detection of the defects by image recognition methods. This paper presents a method to detect defects in EL images of monocrystalline PV modules using YOLOX algorithm. An EL image dataset of monocrystalline PV modules was constructed, consisting of 792 images with 4 types of defects. The stochastic gradient descent (SGD) optimizer with cosine annealing algorithm was applied to the YOLOX object detection model to improve the training effect of the network, and the weight decay method was utilized to reduce the effect of overfitting phenomena. During model training, the Mosaic data augmentation method was used to enrich the background of the detected objects. The trained YOLOX model was applied for defect detection of 238 test sample EL images, and the model reaches a mean average precision (mAP) of 95.81%. This study is efficient to automatically detect defects in EL images of monocrystalline PV modules, which can provide reference for quality inspection of PV modules.