Color Classification And Defect Detection Of Photovoltaic Cells Based On Machine Learning

For the color classification and defect detection of photovoltaic cells,most manufacturers still use the manual detection methods,This method can easily cause fatigue and misjudgement.This thesis use PSO-SVM method and improved convolutional neural network method to realize color classification and defect identification of photovoltaic cells.The specific work is as follows:In the color classification of the photovoltaic cells,the image is firstly rotated and corrected by Radon transform and edge detection.Then,use the edge distance calculation method to extract effective region of the image,next use the median filter method doing image denoising,we use this methods will complete the image preprocessing of photovoltaic cells.Secondly,extract the image RGB color moments and grayscale features.Use SVM classifier to complete the color classification of the photovoltaic cells.Aiming at the problem that SVM parameters are difficult to choose,use particle swarm optimization method to optimize the parameters of SVM.Simulation results show that compared with Gaussian mixture model algorithm and KNN algorithm,PSO-SVM method has the highest classification accuracy.In the defect detection of photovoltaic cells,first of all,use the maximum connected region method to extract the image region,and then use the perspective transformation method to correct the image which are in perspective phenomena,after using edge distance calculation method to extract the standard image,finally the standard image is segmented,and obtain 60 small-sized cells with the same size of the image.we use this methods will complete the module image preprocessing of photovoltaic cells.On this basis,this thesis use multi-channel convolutional neural network method and random forest method to complete defect detection,through three independent CNN channel,it can extract the image feature from different scales,and output is performed at the full connection layer.Use the random forest classifier replaces the output layer of the convolutional layer to realize image recognition and classification.Simulation results show that the multi-channel convolutional neural network combined with random forest has high recognition rate and short recognition time,which is superior to the traditional defect detection algorithm.Finally,we design the photovoltaic cells color classification and photovoltaic cellsdefect detection through the MATLAB graphical user interface and Py Qt5 graphical user interface.Users can achieve photovoltaic cells color classification and defect identification through simple operations,providing operators with a quick and effective operation platform.