Classification of Manufacturing Defects in Multicrystalline Solar Cells With Novel Feature Descriptor

The automatic defect recognition for near-infrared electroluminescence images is a challenging task, due to the random shape of the crystal grains and intensity variation in the appearance of multicrystalline solar cell. However, the automatic defect detection systems can meet the growing demand for high-quality products during the intelligent manufacture. Thus, in order to obtain more discriminable defect features under heterogeneous background disturbance, a novel feature descriptor (FD) named as center pixel gradient information to center-symmetric local binary pattern (CPICS-LBP) are proposed, which can fuse the CPICS-LBP by thresholding each pixel of the image into binary code. The defect classification in the heterogeneous background can be dramatically enhanced. Next, the feature description ability of the CPICS-LBP descriptor is analyzed by the gradient map and the experimental results. Furthermore, in order to realize robust extraction of the defect feature, a method named as bag of CPICS-LBP is presented as a global concentrated feature extraction scheme by employing the similarity analysis and clustering of the local image patch features from the image sample. Moreover, the classification accuracy and time efficiency are verified by some experimental results and analysis of solar cells' image data set. The experimental results show that the proposed methods achieve state-of-the-art classification results.