EL Image Recognition Of Solar Cell Defects Based On Matlab

With the rapid development of photovoltaic industry, the quality of solar cells hasbecome more and more important. Good quality solar cells not only have good stabilityand use longer, but also have high photoelectric conversion efficiency. Quality hasbecome a very important indicator of market competitiveness of solar manufacturers.Defects seriously affect the solar cells’ quality, and most of the defects are hard to findwithout the aid of the other equipment. Therefore, it is necessary to find a fast andaccurate technology of identifying defects of solar cells.Electroluminescent detection technology can detect many kind defects in solar cellsby infrared imaging, including some hidden defects which are hard to find. It is widelyused as a very important detection method of solar cell defects. However, most ELimages are recognized by people. It is seriously affecting the recognition efficiency andaccuracy. With the widely used of digital image processing and pattern recognition, it isof great significance by using digital image processing to recognize solar defectsautomatic.In this paper we reprocess solar EL images by digital image processing technology:it uses polynomial model to correct EL images’ barrel distortion and uses Houghtransform and image rotation transform to correct images’ angle. It divides the solar cellsinto small solar cells. This article enhances small solar cells at the beginning. It includesstretching gray values, enhancing low gray values’ contrast and sharpening image. Itrecognizes normal solar cells, black center solar cells, dark solar cells and black solarcells by using GLM. Then using the features of the off grid and hidden crack and usingmorphological approaches, it classifies the defects and achieves good results. In order torecognize the off grid and hidden crack defects better, by extracting pixel value featuresof edge pixels neighborhood and using trained BP neural network it classifies the edgepixels, then it classifies the edges by statistical properties.