Research On Solar Cell Defect Detection System Based On Machine Vision

In recent years,with the acceleration of global industrialization,the energy and environmental crisis has become a problem to be solved.Faced with this current situation,clean energy solar energy has been widely used in various fields.As the core carrier of electric energy conversion,the quality of solar cells determines the efficiency of electric energy conversion.By analyzing the characteristics of defects in the solar cell images collected by electroluminescent detection platform,this paper designs the traditional image processing defect discrimination algorithm based on HALCON and the convolution neural network defect detection algorithm based on Keras,and compiles the image recognition system through C# combined with the automation using PLC as the controller to realize the defect detection of solar cells.The main research work of this paper is as follows:Battery image preprocessing.A weighted fusion filtering algorithm combining Gauss filtering and mean filtering is proposed.Compared with the single filtering algorithm,this algorithm can protect the local edge features of the image,reduce the noise of the image and meet the requirements of high-definition image for subsequent image processing.In addition,the priwitt edge detection algorithm is employed to separate the target area from the background area and extract the ROI of the image.Design of defect detection algorithm.Monocrystalline silicon cells: 1.As for the crack defect of solar cells,an improved gray threshold method is put forward.This method uses the variation coefficient method and the improved dynamic threshold method to construct the feature plane and locate the crack defect on the plane.2.Scratch defect.By analyzing the characteristics of scratch defect,a Gaussian line detection method is adopted.The Gaussian kernel function and lag threshold parameters are determined according to the scratch sample bank.3.For blocky defects such as black spots and broken grids,the location of defects is initially located by conventional local threshold method.Then small areas are further screened by morphological and area features.Finally,broken grids and black spots are classified according to shape features(rectangularity and roundness).Polycrystalline silicon batteries: On the basis of classical VGGNet,batch normalization layer and global pooling layer are added to standardize the model and reduce the occupancy of computer resources.Meanwhile,the improved VGGNet is implemented in the keras deep learning framework.Experiments show that the improved VGNet has better performance than the traditional convolution neural network in measuring quasi-curvature and convergence.System debugging and experimental analysis.The detection system is built by integrating the hardware module and developing the image recognition software.The experiment results show that the recognition rate of crack defects of single crystal batteries is 99.6%;the recognition rate of broken grids is 95.6%;all black spots and dirt are detected;and the recognition rate of scratches is 98%.In addition,the recognition rate of polycrystalline batteries is more than 91%.The time of single image recognition is between 250 ms and 300 ms.