Research On Detection Technology Of Weak Boundary Defects Of Solar Cells Based On Deep Learning

Solar cells are the core of photovoltaic power generation systems.The detection of cell defects is particularly important for improving the quality of solar products during the production process.However,there are many production processes for solar cells.Defects will be covered and diffused after production,which results in blurred edges and poor contrast.In view of this,this paper studies the hardware and software algorithms under the condition of photoluminescence(PL)imaging.In terms of hardware,the multi-source LED-PL imaging scheme is used to replace the original laser imaging scheme.In terms of software algorithm,the traditional t detection algorithm and cell defect segmentation method based on deep network are studied respectively.On this basis,an improved defect segmentation method is proposed based on U-net network.The improved method achieves better segmentation results.In addition,aiming at the more serious crack detection in solar cell component.A convolutional method is proposed for crack detection based on screening and locating the detected area of PL image.This method achieves better detection results.The main research contents of this paper include:Aiming at the problems of high noise,low defect contrast,and uneven gray-scale distribution in the PL image of the cell.Considering the production process and economic factors,the cell defect imaging scheme is improved.A multi-light source LED-PL imaging scheme is proposed to replace the original laser imaging solution,which improves the imaging accuracy.At the same time,the scheme solves the problems of short service life,easy loss and high price of high-energy lasers in the current PL technology.Aiming at the problems of blur boundaries,large scale spans,and small battery slice image data sets,the traditional segmentation method and the deep learning segmentation method using U-net network are compared.Here,an improved U-net network is proposed based on deep learning segmentation method.The cell defect segmentation method of net network uses a variety of segmentation result evaluation indicators to analyze the results.The results verifies the segmentation effect of the above method on the blur boundary defects of this type of cell.In view of the large size of the cell component and the small proportion of the defect area,the method of locating the ROI first and then classifying is used to detect the cracks of the cell module.Firstly,the cell image is preprocessed,and then the method based on cluster-based is used to screen and locate the target area.Finally,three different convolutional neural network models are used to detect defects in the cell.The accuracy is compared to make the optimal recognition accuracy rate reaches 99.25%.The experimental results verify that the above method can accurately detect the crack defect of the solar cell component.