Research On Solar Cell Color Sorting Algorithm Based On Deep Learning

In the actual production,there will inevitably be color differences within a single solar cell and color differences between multiple solar cells,which will not only affect the overall aesthetics of the solar cell,but also affect the photovoltaic power generation efficiency of the solar cell.Therefore,it is of practical significance to detect the color difference within solar cells and classify the color grade between cells.This paper takes solar cell as the research object,first analyzes the causes of chromatic aberration of solar cell,outlines the environment and process of solar cell image acquisition,then carries out image pre-processing,and finally completes the detection of intra-cell chromatic aberration and inter-cell color grade classification of solar cell on this basis,and the main research contents are as follows.(1)Image pre-processing.In order to obtain high quality solar cell image data,the paper selects the maximum inter-class variance method to obtain the best threshold to separate the foreground and background for the problem of prominent background in some of the collected images;selects linear transformation to enhance the brightness of the images for the problem of darkness in some of the collected images;uses flip,rotate and add noise to expand the data set for the problem of insufficient data set,thus The generalization and robustness of the model are increased.(2)A neural network based on the fusion of ResNet50 and VGG16 is proposed to detect the in-sheet color difference in solar cells.In order to improve the accuracy of solar cell in-cell color difference detection,this paper analyzes the basic principles and structures of typical convolutional neural network models,and then uses two methods of adaptive weight adjustment and result-weighted average to adjust the weights of VGG16 and ResNet50 fusion networks.The experiments show that the best results for solar cell color difference detection are achieved when the two models are fused with 4 to 6 weights,and the accuracy can reach 97.8%,which is better than the traditional convolutional neural network model.(3)A ResNet50+Transformer fusion model for solar cell inter-cell color classification is proposed.In order to improve the accuracy of solar cell inter-cell color class classification,this paper firstly performs feature extraction by ResNet50 to improve the local feature expression capability of the network,and then uses Transformer to obtain the location information and category information of key regions of the image,and finally completes the solar cell cell classification by MLP Head layer.The experiments show that the model successfully applies the Transformer model,which is mostly used in the field of natural language,to image classification,and achieves the classification of solar cell color classes by changing the network structure,and the classification accuracy of the model can reach 98.5%,which is higher than the respective individual network models and the traditional convolutional neural network model.