Search

Automatic defect classification in photovoltaic (PV) modules is gaining significant attention due to the limited application of manual/visual inspection. However, the automatic classification of defects in crystalline silicon solar cells is a challenging task due to the inhomogeneous intensity of cell cracks and complex background. The present study is carried out for automatic defects classification of PV cells in electroluminescence images. Two machine learning approaches, features extraction-based support vector machine (SVM) and convolutional neural network (CNN) are used for the solar cell defect classifications. Suitable hyperparameters, algorithm optimisers, and loss functions are used to achieve the best performance. Solar cell defects are divided into seven classes such as one non-defective and six defective classes. Feature extraction algorithms such as histograms of oriented gradients (HOG), KAZE, Scale-Invariant Feature Transform (SIFT) and speeded-up-robust features (SURF) are used to train the SVM classifier. Finally, the performance results are compared. It is concluded that CNN's accuracy for solar cell defect classification is 91.58% which outperforms the state-of-the-art methods. With features extraction-based SVM, accuracies of 69.95, 71.04, 68.90, and 72.74% are obtained for HOG, KAZE, SIFT, and SURF, respectively. The present study may contribute to making a PV system more efficient for classifying defects to improve the power system efficiency.