Research On Photovoltaic Image Registration And Photovoltaic Modules Fault Diagnosis Technology Based On Infrared And Visible Images

As the installed capacity of photovoltaic power stations continues to increase,the inspection and maintenance of photovoltaic power stations is becoming more and more important,and the advantages of PV(photovoltaic)module defect intelligent inspection technology by UAV(Unmanned Aerial Vehicle)in photovoltaic power stations are becoming more and more obvious.In this dissertation,aiming at realizing PV module defect diagnosis,combined with machine vision and deep learning,combined with the field scene of photovoltaic power station,a photovoltaic image registration technology and PV module defect diagnosis technology based on infrared and visible images are proposed.Among them,the main contents of PV module defect diagnosis technology include PV string segmentation,PV module segmentation and PV module defect identification.Since visible images are high-resolution images,direct defect recognition requires high hardware.Therefore,defect diagnosis is performed through the process of string segmentation,PV module segmentation and PV module defect classification.In this dissertation,three different methods are used to realize the segmentation of visible PV modules on the basis of realizing the segmentation of PV strings.And through the duallight image registration,the segmentation of the corresponding infrared modules of the visible modules is realized,so that the defects of the dual-light modules are classified,and the joint diagnosis of the defect information is achieved.The registration of infrared and visible dual-light photovoltaic images is mainly used to realize the coordinate correspondence between the pixels of the dual-light images.In this dissertation,two dual-light photovoltaic image registration techniques are proposed on the basis of string segmentation,including the perspective transformation registration method based on the four-corner coordinates of the string and the image registration method based on the mask image template matching of the dual-light string.By realizing the segmentation of visible modules and the registration of dual-light images,the segmentation of infrared PV modules can be indirectly achieved,and the one-to-one correspondence between infrared and visible PV modules images can be completed.