Research On Infrared Photovoltaic Inspection Image Fault Detection Based On Computer Vision

With the improvement of photovoltaic power generation efficiency and the continuous expansion of photovoltaic scale,a series of problems in operation and maintenance are gradually emerging.The most common is module failure caused by the installation environment and aging of equipment.With the rapid development of UAV technology and artificial intelligence,the use of UAV equipped with dual light camera for patrol inspection can effectively meet the needs of large-scale photovoltaic installation environment,and combined with computer vision technology can quickly and accurately detect various fault points.In this thesis,according to the characteristics of UAV infrared patrol image,the problem of photovoltaic module fault target detection is deeply studied,and the following aspects are mainly completed.Failure mode analysis and patrol strategy formulation of photovoltaic modules.Starting from the structure of photovoltaic modules,the formation mechanism and consequences of various faults are analyzed,and then the detection object is selected according to the characteristics of UAV patrol inspection,and the UAV patrol inspection scheme is developed according to the characteristics of photovoltaic field distribution.Fault detection method based on corner detection and line scanning.The preprocessing part is used to enhance the image contrast and preliminarily segment the photovoltaic region;In order to detect the fault location,first extract the PV module area,use morphological processing,flood filling,contour fitting and other methods to extract each contour in the image,and distinguish the module area and background based on the improved Harris corner detection principle;In the fault extraction part,the line scan fault extraction method is used,and different judgment criteria shall be formulated according to the fault type.Target detection method based on deep learning.Based on Faster RCNN and YOLOv5 detection models,the former is optimized from the perspective of backbone feature extraction network and anchor initialization;The latter improves the prediction ability of the model in many ways.First,Fuzzy Cmeans clustering algorithm is used to preliminarily cluster the anchor,and the cluster center is used as the initial cluster center of K-means clustering algorithm to obtain the anchor that is more consistent with the target data size.Then the loss function is optimized,the original regression loss calculation method is changed from GIOU to EIOU loss function which is more powerful,and the confidence loss balance coefficient is adaptively adjusted to improve the model training effect.Finally,the In Re feature enhancement module is added in front of each detection layer to enhance the target feature extraction capability by enriching feature representations;In the experiment,the infrared photovoltaic data set created was used for comparative ablation experiments to verify the effectiveness of each algorithm.This study provides an efficient and convenient scheme for large-scale photovoltaic field inspection,which is of great significance for improving the efficiency of photovoltaic power generation and eliminating potential safety hazards in time.