Research On Aerial Image Recognition And Matching Technology In Photovoltaic Plants

With the rapid development in photovoltaic industry,the managers are facing increasing pressure to managing large sized PV plants.Inspection for PV plants by UAVs is a trend of future.It brings a new demand that utilizing computer image processing technology to process and analysis PV facilities' aerial images automatically.The thesis aims to inspect large sized PV plants' installed capacity and PV modules' faults.A strategy for inspecting those irregular and dispersed PV plants is proposed,and research on relative image processing technologies has been deeply carried out.In order to recognize PV modules in aerial images,the thesis suggests that PV strings' recognition is necessary before PV modules' recognition.The algorithms of image segmentation,morphologic processing,edge detection,contour extraction and selection are applied in recognizing all PV modules accurately.The method to achieve a lower missing rate under a light reflecting circumstance by adaptive image binarization and compensating undetected modules is also discussed.It's difficult to match PV strings in images for extreme similarity between strings,so the random color difference is utilized to solve the problem in this thesis.Code for every PV string based on the random color difference,and feature vectors are calculated in an improved LBP feature descriptor.Then the target to compare PV strings and find homonymy PV strings between two images can be realized.The thesis also extracts contours of four common PV modules faults' area respectively,and analyzes the contour feature and color feature of those faults' area,which can be used as criterions for classifying different types of PV modules' faults.