Research Of Silicon Solar Cell Based On Visual Detection Method

With the energy crisis, solar energy plays an important role in the development andutilization of energy, such as the recent developed photovoltaic power generation whichhas great potential of new energy technologies, and solar cells are part of the core. Thedefects in production process of solar cell, may result in the decrease of the photoelectricconversion efficiency and therefore the products will need to be detected before it finished.Solar cell detection depends on the labor, but there are many disadvantage, such as theartificial observation is easy to appear deviation and not easy to control, high labor costsand so on. For silicon material plays an very important role in the raw materials of solarcell, we propose the detection algorithm based on machine vision which applies in eachstage of silicon solar cell production. In turn, the defect detection algorithm include basedon silicon solar wafer,on solar cells and solar panel.For detecting the defect of solar wafer, we uses NIR(near infrared) LED as lightsource and NIRCCD camera as image acquisition device, to get the surface and internalmicro-crack images of solar wafer. In the detection algorithm design, the traditional edgedetection method or the binarization method,which based on the micro-cracks showed thecharacteristics of low gray level and high gradient magnitude on the silicon wafer, is notsuitable for low contrast image. So in this paper,we uses the anisotropic diffusionalgorithm for micro crack detection, which based on the gradient distribution of differentimages,sharpening the high gradient values which are defect area,and smoothing the lowgradient values which free area, in other words the algorithm can sharpen the defective andsuppress the noise. But in the algorithm, numerical diffusion function, sharpeningparameters change directly affects the defect detection results, and there is lack of auniform diffusion function expression. So this paper proposes utilizing the improvedanisotropic diffusion algorithm as the operator image edge extraction, according to the seedpixel is determined by region growing algorithm will be hidden flaws is separated from thebackground, the test algorithm is high precision, and can meet the online solar siliconwafer defect inspection requirements.For detecting the defect of solar cell,we proposes an improved Otsu algorithm afterreserached the general image segmentation method.The online solar cell defect inspectionis divided two cases:the first is detecting whether the solar cell defect or not,the second ispositioning the defect and analyzing the reason when there is a defect solar cell.For the first case,we use the spatial domain detection method after researching the typical defectfeatures.For the second case,we use the image reconstruction method in frequencydomain,and operate the difference bewteen the orignial image and reconstructed image,tolocation the defect.According to the diversity surface texture of monocrystalline siliconand polycrystalline silicon solar cell,the wavelet transform is used for the regular texturecharacteristics one,and the other one is reconstructed by foruier reconstruction algorithm.For detecting the defect of solar panel,the solar panel is consists of some solarcells,but the defect won’t appear in each solar cell.Using the previous imagereconstruction method to search each solar cell whether it is free one or defected one willcause data redundancy and low efficiency.To solve the disadvantage,we uses theindependent component analysis (ICA) separation matrix to reconstruct the inspectedimages from the training set,in oreder to enhance the defect information and remove theregular texture of solar panel image. The FastICA as the commonly used ICA method hasthe advantage of rapid convergence,but it cannot convergence when the initial point awayfrom extreme point. Therefore,we proposes the Particle Swarm Optimization(PSO) intoFascICA algorithm,the global best position as the best classification matrix and theIC(independent component) are obtained by PSO method,after that we reconstruct theimages to detect the defect. The research on solar cell defect classificaiton method,weproposes to use the support vector machine(SVM)algorithm based on AdaBoost classifierto train the sample,and classfy the input image using the SVM classifier,finally output theresult. To solve the disadvantages of beforehand training samples in solar panel defectdetection and classification procedure,we propose a detection and classification methodwhich can not rely on reference sample because the partition threshold is obtained by thesample itself.The speed and result of using this method are all haved the satisfactory results,and it has a great application prospect for currently rely on artifical detection of solar cellindustry.