Defects Detection Based On Electroluminescence Imaging Of Silicon Solar Cells And Automatic Machine Recognition

With the development of solar industry, the efficiency and reliability of solarcells are increasingly important. Defects have the most and direct impacts on theperformance of solar cells. Most of the defects are difficult to be discovered andidentified as they are hidden inside. So it is valuable to find a fast and preciselymethod for detecting defects in solar cells.Electroluminescence imaging technology, as a new method to detect defects insolar cells, has gradually been applied to large-scale industrial production process. Itcan discover hidden defects through observation of solar cells’ fluorescence images.However, most of the fluorescence images are analyzed by human eyes. With theincrease of capacity of solar cells, the recognition speed of human eyes significantlylimits the production throughput. Therefore, automatic detection technology has asignificant bright prospect.In this paper, two types of image processing algorithms are applied to typicaldefects in solar cells, including black cell, busbar break, finger interruption andmicro-crack. The algorithm based on local threshold applies a rectangle neighborhoodand a contrast-stretching transformation function to images. The defects then can beeasily located by morphological image processing. The algorithm based on ThresholdUniform Local Binary Pattern (TULBP) and BP neural network uses uniform patternsto reduce quantities of characteristic values. Combined with the BP neural network todetect defects, it transforms the complex defects detection problem into neuralnetwork pattern recognition problem. The experimental results show that bothalgorithms are suitable for inline defect detection, including single-crystal andmulti-crystal solar cell, with the rapid speed and high accuracy rate.