Solar Cell Defect Recognition based on Orthogonal Learning Strategy

The quality of silicon wafers is an important factor restricting the efficiency and service life of photovoltaic power generation. In order to inspect the quality of silicon wafers, a defect recognition method based on orthogonal learning strategy is proposed where support vector machine is combined with binary tree for multi-class classification. Firstly, the adaptive threshold is set to remove the raster lines in the original image, and Fourier reconstruction image is used to enhance the defect. After that, we extract the image features. With the help of orthogonal learning strategy, an orthogonal array of feature data is established to implement the initial defect classification, and the classification results are analyzed by factor analysis. The extracted features are sorted according to their influence degree, and the improved support vector machine is used to classify the feature data accumulated one by one. Finally, genetic algorithm and grid search are introduced to optimize the parameters. The recognition accuracy of the designed classifier is up to 96.6%. The experimental results indicate the effectiveness of the proposed method.