Defect Detection And Sorting System For Photovoltaic Cells Based On Deep Learning

As the core component of photovoltaic modules,the quality of photovoltaic cells directly affects the power generation efficiency and safety of modules.However,the production of photovoltaic cells is prone to many defects caused by factors such as processing technology and materials.Therefore,it is of great engineering value and significance to carry out research on intelligent defect detection of photovoltaic cells.Although machine vision algorithms have been used in photovoltaic cell defect detection,the detection accuracy and adaptability are generally low due to the complex background texture and large feature scale differences of photovoltaic cell EL images.At the same time,there are many kinds of defects in the battery sheet,and the uncertainty of the defect position is high,which further increases the difficulty of detection.Therefore,this paper studies and designs two detection algorithms based on deep learning,which are applied to the different requirements of photovoltaic cell defect detection.The main research work and innovative achievements of this paper are as follows:(1)In this paper,image acquisition is completed based on electroluminescent imaging technology.Based on this,image characteristics are analyzed,and image processing methods are used to complete image redundant information cropping,data enhancement and other operations.According to the different algorithm requirements in this paper,the processed data are respectively labeled and made into data sets for defect classification and specific defect identification and positioning.(2)In order to solve the problem of poor defect discrimination performance,a lightweight classification recognition model was designed in this study to achieve the purpose of separating defective photovoltaic cells.Optimize Dense Net by introducing depth-separable convolution to reduce the amount of model parameters.At the same time,the Leaky Re LU activation function is used to improve the generalization ability and robustness of the model,and the SE module is embedded behind the Dense Block module to dynamically adjust the weight value of each channel.Strengthen the ability to express model features.It is verified by experiments that the network complexity is lower,the convergence speed is faster,and the accuracy rate reaches97.88%,which is 3.17% higher than that before the improvement.Moreover,the optimized model has significantly reduced its parameter size,allowing the network to undergo a lightweight processing without compromising the detection accuracy.(3)In view of the complex background texture and large difference in feature scales of cell EL images,this study proposes an optimized RFPN-CBAM-Faster R-CNN target detection model for specific defect classification and positioning tasks.The model employed a deep residual network Res Net50 with a feature pyramid network(FPN)to enhance the feature extraction capability and adaptability to multi-scale defects.The region proposal network(RPN)was improved by replacing ROI Pooling with ROI Align to improve defect recognition and localization accuracy,and a convolutional block attention module(CBAM)was embedded to enhance the attention and feature expression ability of the model on defect information.Experimental results showed that the proposed model achieved high detection accuracy and excellent recognition performance for multi-scale defects,which can effectively meet the needs of intelligent defect localization and detection.(4)In order to make the algorithm model of the above design research more efficient in the actual production line,a software system for the classification and detection of photovoltaic cell defects was developed,and an experimental platform was built for running tests.The system has an accuracy rate of 98.05% in the classification test of the physical defects of photovoltaic cells,has good performance advantages,and can better meet the actual production needs of photovoltaic cells.