Study On Surface Stain Detection And Color Binning Of Solar Cell Base On Machine Vision

Solar power has the attention of the world as its non-pollution and inexhaustible.The defect inspection of silicon-based solar cell,which is the main device to transform solar power to electricity,is getting more and more important.And as the differences in the production process,solar cell will show different colors.For the beauty,it must be classified by color.At present,this work is mainly done manually.In this paper,factors and improvement methods that affect the fingerprint detection and accuracy of color' binning based on machine vision has been studied.First,hardware of machine vision system has been set up and the illumination condition can be adjusted in many dimensions.A resolution test program for lens has been programmed which is based on ISO 12233-2003 Resolution Chart.The program can read resolution from color picture got in monochromatic lighting and noise can be shield which cannot be implemented in existed software.In fingerprint stain inspection,the definition of fingerprint in different lighting condition has been studied.We find fingerprint show different definition in different lighting spectrum.The fingerprint in red lighting has the best definition.And the fingerprint has the same definition in different lighting angle.According this,we find the best lighting condition for fingerprint detection which is 450 illumination of red light.In this condition,the gray of finger,fingerprint and silicon background has the highest contrast.And then we finish the fingerprint detection program which is based on threshold segmentation.As the R signal obtained through color camera in white illumination is so weak that it is susceptible to noise,four factors and methods to improve accuracy of classification has been studied:a)the noise of camera;b)the effect of the len vignetting on the dynamic range of the image;c)pixel data output formats(8 bits and 12 bits);d)enhance the R signal by increase the lighting intensity of red.We get the colorimetric characterization model of camera by using the ColorChecker DC chart.Tristimulus value XYZ of solar cell,which is device-independent,is converted from RGB signal by using the model,and then it is converted into CIEL*a*b*color space so that we can classify the solar cell based on the color difference.Experimental result has shown that the color classification system is effective and the effect of noise on R channel values can be reduced from about 20%to about 5%by increasing the intensity of red light.