Infrared Image Fault Detection of Photovoltaic Modules Based on Vision Transformer

Solar photovoltaic panels are prone to failure and damage due to manufacturing, transportation, installation and environmental factors, resulting in energy loss.By detecting infrared images of the panels, loss of power production can be estimated and operation and maintenance costs reduced.Based on this, this paper designed a method of detecting photovoltaic abnormal infrared images based on Vision Transformer, and achieved the purpose of classifying different fault types by detecting abnormal infrared images.Vision Transformer firstly divides the input images into image blocks every certain area size, then combines the segmented image blocks into sequences, and feeds the combined results into Transformer's special multi-head self-attention for feature extraction.Finally, Cls Token is used for classification.Experimental results show that the accuracy of infrared image detection based on the proposed method reaches 95.787%, which is 11.9%higher than Xception model and 17.74%higher than VGG16 model.