Solar Cell Defect Detection Based On Deep Learning

As a new type of energy,solar power generation has attracted much attention,but in the production process of solar cells,defects such as cracks,broken and over-welding may occur,resulting in a problem of lowering the output power of the solar panel.Therefore,the defect detection of solar cells has become an important guarantee for the reliable operation of solar panels.Studying solar cell defect detection methods has important engineering practical significance.At present,in addition to visual inspection,the main method of industrial solar cell defect detection is machine vision image processing technology.However,due to the low efficiency of manual detection,the accuracy of machine vision detection needs to be improved,so it is urgent to upgrade the version of the detection program.Nowadays,computer storage and computing power are developing rapidly,and the deep learning theory system is gradually becoming mature.Because deep learning can describe the richness of data in the rich quantity of the amount of data to find the law of data distribution,accurate defect detection becomes possible.Defect detection based on deep learning will be expected to improve detection efficiency and improve the quality of solar cells.Therefore,this thesis introduces the deep learning algorithm into the solar cell defect detection task,and explores and studies the practice of deep learning in solar cell defect detection.The main research work and results of this thesis are as follows:(1)Generation of a solar cell defect detection data set.Based on the collection of solar panel images,pre-processing operations such as image enhancement,restoration,cropping,and uniform size were performed.After pre-processing,data sets for defect classification and positioning experiments were created according to the requirements of experimental tasks.(2)A method based on deep learning for defect classification of solar cells is studied.The experiment adopts the strategy of target classification,and explores the training strategies of various convolutional neural networks to achieve the purpose of separating defective solar cells.Furthermore,based on the research of the advantages and disadvantages of various convolutional neural network models and the adaptability of current tasks,the optimal training strategy is determined.The recognition rate of each network model has reached more than 93%,and the individual network model has reached 96%.the above.According to the training results of these network models,the model fusion operation is made in a targeted manner,which improves the robustness and generalization ability of the model.(3)A method based on deep learning for defect detection of solar cell chips is studied.The method uses the strategy of image target localization to train the convolutional neural network end-to-end,and realizes the function of identifying the defect and positioning the defect area with a rectangular frame to realize intelligent positioning detection.(4)A graphical user interface system for visual defect detection is developed.The system mainly has two main functions.These two functions correspond to two kinds of defect detection technologies,one is image classification detection,the system will display whether it is a defect image,and the other is defect location detection,the system will display defects in the image phase.Area and locate the defective area.The system is simple to operate and practical.