Research On Color Classification And Intelligent Splicing Of Solar Cells

In recent years,the photovoltaic industry has been developing rapidly.As renewable energy,the recycling of solar cells is increasingly prominent.Due to the small color difference and uneven distribution on the surface of the waste solar cells,the color classification of solar cells is a problem.In this thesis,deep learning and image processing technology are used to study and realize the color classification of waste solar cells.And in order to solve the problems of low efficiency and low precision in the process of image splicing of large-scale solar cell module,this thesis uses image splicing technology to research and realize the intelligent splicing of large-scale solar cell module image.The main research is as follows:(1)By using the image augmentation technology,the color classification data set of solar cells based on multi-color space is constructed,which provides data for the deep learning model of color classification of solar cell.Aiming at the problem that the amount of solar cell training data is small,the solar cell training data set is expanded by using image flipping,rotation and adding noise.(2)The color classification of solar cells is studied,and an intelligent classification algorithm based on multi-color space is proposed.Based on the LeNet-5 convolution neural network,a deep learning model of color classification of solar cell is constructed,which optimizes the traditional LeNet-5 network structure to improve the classification performance.Then,based on the analysis of different color spaces of solar cells,a multi-color space classification fusion algorithm is proposed.The experimental results show that the accuracy of the algorithm is higher than that of the traditional LeNet-5 network,BP neural network and SVM algorithm.The combined classification effect of the algorithm in RGB + lab + HSV color space is the best,and the classification accuracy is 94.56%.(3)A data acquisition design scheme of distributed multi camera is proposed to provide a splicing platform for large-scale solar cell modules.In order to avoid the distortion of single camera,the local image of large size solar cell module is collected by multi camera.According to the size of the workpiece,the common camera is used to construct the camera array,which can replace the traditional industrial camera for image acquisition and reduce the cost.(4)The intelligent splicing of large-scale solar cell modules is studied,and a fast splicing algorithm of low-cost solar cell modules is proposed.By customizing the threshold of Harris corner detection and using adaptive non maximum suppression,the algorithm solves the problem of false corner and corner cluster produced by traditional Harris corner detection algorithm,and improves the splicing speed and accuracy.Through the splicing experiment of large-scale solar cell modules,it is verified that the algorithm in this thesis is better than the traditional Harris splicing algorithm in running time and feature point mismatch,and can quickly and accurately achieve seamless splicing of industrial images.