Solar Cell Defect Detection Based On Convolutional Neural Network

Solar cell,as a key component of solar power generation system,its quality is an important factor affecting the power generation efficiency of battery modules.In the production of batteries,there may be some problems that affect the normal use,such as stains,cracks,etc.,so it is essential to carry out defect detection.In recent years,the Convolutional Neural Network(CNN)has a very outstanding performance in the field of image processing,Solar cells can also be examined in the form of images.Therefore,this paper takes the solar cells as the research object and deeply studies the promotion effect of convolutional neural network on the defect detection of solar cells.Firstly,in view of the problems of high cost,single function and low efficiency existing in traditional solar cell detection methods,a solar cell defect detection framework based on CNN was proposed to realize the rapid detection of cell defects.The test results show that the detection time of solar cell defects using this framework is only 0.15 ms for each solar cell,and the detection accuracy can reach about 65%.However,due to insufficient training data sets,model is prone to over-fitting.Then,aiming at the problem of model overfitting caused by insufficient amount of training data,a data enhancement algorithm based on Deep Convolution Generative Adversarial Networks(DCGAN)and random image fusion,TFRF＿NET,is proposed.In this method,sufficient "false" defect data is generated by DCGAN,and then it is randomly fused with "true" defect data.The results show that after the training set is expanded by TFRF＿NET,the test accuracy of the model is improved by about 3% and7%,respectively,compared with the generation adversation network data enhancement method and the traditional data enhancement method.It is shown that TFRF＿NET can effectively expand the training data set and increase the diversity of training samples,thus alleviating the problem of model overfitting.Finally,in order to promote the application of CNN in industrial testing,an improved lightweight network VGG16＿Light is proposed.In addition,combined with TFRF＿NET and VGG16＿LIGHT,a lightweight solar cell defect detection system based on TFRF＿NET is proposed.In the experiment of detecting whether there are defects in the battery,the test accuracy of the proposed algorithm is nearly 7% higher than that of the traditional algorithm.In the experiment of battery defect classification,the test accuracy of the proposed algorithm is nearly 11% higher than that of the traditional algorithm.Meanwhile,after the lightweight treatment,the number of model parameters was reduced to about 1/2 of the previous one,and the test time for each image was shortened from 42 ms to 19 ms.It is proved that the proposed algorithm can effectively alleviate the problem of model overfitting.And compress the model to speed up testing.