Research On Defect Anomaly Detection Of Photovoltaic Panels Based On Deep Learning

Energy is an essential and important factor for the development of human society,but the fossil energy in nature is limited and non-renewable.The excessive use of fossil energy can exacerbate global energy imbalance and depletion of energy reserves,as well as lead to serious environmental crises and climate change.As a clean and renewable alternative energy,light energy has received attention from countries around the world.Photovoltaic power generation is a technology that converts solar energy into electricity,and photovoltaic panels are devices that use this technology for energy conversion.Among them,the energy conversion efficiency of photovoltaic panel is one of the important indicators to measure its performance,but the surface defects of photovoltaic panel often lead to reduced performance and shortened service life,which is not conducive to the popularization of optoelectronics.Therefore,the detection of abnormal defects on the surface of photovoltaic panels is very important.For the detection of defects on the surface of photovoltaic panels,th e current detection technology includes manual detection,machine vision detection,infrared detection and so on,In China,manual detection is the main method and machine vision inspection is used less.In foreign countries,it mainly uses machine vision detection,infraree detection and other high-tech means to detect surface defects.In contrast,there is still a lot of room for progress in China.However,with the expansion of the scale of photovoltaic power station,the defects of photovoltaic panel surface are detected by manual detection methods.The progress of computer technology,on the one hand,promotes the development of computer vision technology in surface defect detection,on the other hand,provides more advanced and efficient technical methods,such as the use of artificial neural network for surface defect detection.This paper is based on deep learning of photovoltaic panel defect anomaly detection research.Before the network training,it is necessary to preprocess the image data in the input dataset to reduce the impact of noise signals on the image data.In response to the interference of noise signals,this paper compares several commonly used filtering methods and ultimately chooses to use wavelet denoising technology to denoise photovoltaic panel images.In addition,in order to determine the type of wavelet basis used for processing photovoltaic panel images,this paper proposes the concept of relative mean square error,and chooses to use db 4 wavelet and 5-layer decomposition layers to compare the results.Aiming at the Angle deflection problem of photovoltaic panel image,this paper designs and realizes the function of image angle correction.The Angle correction algorithm is designed by using Radon transform to calculate the projection,Hough transform to discover the line and linear fitting to obtain the slope of the fitted line.Then compare the correction results,this paper uses the Hough transform method design algorithm to realize the Angle correction,and in order to improve the execution speed of the algorithm,this paper also improved the Angle correction method,the improved Angle correction algorithm can ensure high correction accuracy at the same time,but also can increase the execution speed to more than 3 times the original.For photovoltaic panel images that have completed angle correction and noise reduction processing,it is necessary to identify and obtain their main contour positions.Because the size and position of the main contours of photovoltaic panels are different in a dataset composed of 1500 photovoltaic panel images,in order to accelerate the speed of neural network training and improve its accuracy,this paper needs to perform edge recognition on photovoltaic panel images to filter out the main contour position information of photovoltaic panels.Then,the pre-processed data set can be input into the neural network for training after defect labeling.In order to select the optimal neural network to realize the detection of photovoltaic panel defects,this paper uses YOLOv3 and the improved Yolov3-SPP3 network to train and compare the labeled data set,and the comparison results show that the improved Yolov3-SPP3 network has a better performance.In order to improve the speed of network training and the accuracy of defect detection,the K-means clustering algorithm is used to calculate the most priority box check through the data set after defect labeling.The experimental results show that,compared with the original prior box,the highest priority box can achieve higher accuracy of defect detection.Finally,the trained network model is used to design and implement the photovoltaic panel defect detection system,and finally complete the detection of surface defect types of photovoltaic panel images.Experiments are designed to test the defect detection system.The test results show that the photovoltaic panel defect detection system can effectively realize the detection of photovoltaic panel defect types.To sum up,this paper uses deep learning to detect surface defects of photovoltaic panel images.Firstly,image preprocessing methods such as noise reduction processing,Angle correction and contour screening are used to preprocess the original image data set.Then,the preprocessed data set is marked with defects.Finally,the data set is input into the neural network for training,and the trained network model is used to design the photovoltaic panel image defect detection system.Specifically,the study of this paper not only provides a new idea for the abnormal detection of photovoltaic panel image defects,but also can use the defect detection system for photovoltaic panel defects detection,timely discovery and replacement of defective photovoltaic panels,so as to improve its use efficiency.