Micro-crack detection of multicrystalline solar cells featuring shape analysis and support vector machines

This paper presents a strategy for detecting micro-crack in the multicrystalline solar cells. This detection goal is very challenging because micro-crack defects occur inside the cell and can only be visualized with the technique such as electroluminescence (EL) procedure. EL images of solar cell are segmented and analyzed by means of advanced image segmentation technique and shape analysis. The output from these procedures is the dataset of shape features that represent crack and non-crack pixels. The classification of the shapes is achieved by the implementation of the artificial classifier based on the support vector machines (SVM). A number of SVM algorithms are considered in this study to address the issues of the non-linear separation and the imbalanced samples between classes in the dataset. The result indicates that the SVM with penalty parameter weighting is more accurate, resulting in the sensitivity, specificity and accuracy of 91.8% 97.2 % and 97.0 % respectively.