Then, the network weights are used to identify and detect actual photovoltaic defects, thus providing a new concept for photovoltaic surface defect detection. For example, a convolutional neural network (CNN) can be used to extract defect features and help the network improve its ability to express defect feature information.
At present, the main methods for detecting surface dust on solar photovoltaic panels include object detection, image segmentation and instance segmentation, super-resolution image generation, multispectral and thermal infrared imaging, and deep learning methods.
Since manual detection of photovoltaic panel defects is relatively wasteful of time and cost, the current mainstream detection methods are machine vision and computer vision inspection.
When solar photovoltaic panel surface defect detection is applied to industrial inspection, the primary focus lies in achieving a highly accurate and precise model with exceptional localization capabilities, and the training model will basically not affect the detection speed.
Binhui et al. used electroluminescence images and GoogleNet to detect photovoltaic cell defects on the edge. Using electroluminescence images as defect datasets and GoogleNet as CNNs for defect detection networks.
Among them, algorithms such as YOLO [11, 12], Faster R-CNN , and RetinaNet [14, 15] in object detection methods can accurately mark the position and boundary of solar photovoltaic panels in the image, but due to the need for a large amount of computing resources, they have high requirements for hardware and environment.