As the main component of photovoltaic power station, photovoltaic cells have defects, such as hidden cracks, scratches, hot spots, and broken gates, which affect the performance of photovoltaic cells and the operation status of photovoltaic power stations, so it is very important to carry out defect detection of photovoltaic cells. A pulsed electric infrared thermography (PEIT) experimental system is established, and the system is used to carry out detection experiments of photovoltaic cells with different types of defects and to collect infrared thermography sequences. Two kinds of supervised learning algorithms, linear discriminant analysis (LDA) and quadratic discriminant analysis (QDA), are used to process thermography sequences, and compared with principal component analysis (PCA) and fitting correlation coefficient (FCC). The experimental results show that the PEIT algorithm can effectively detect the defects of photovoltaic cells, and the QDA algorithm is better than LDA, PCA, and FCC algorithms in signal-to-noise ratio, information entropy, and mean square error, and it can effectively identify all kinds of defects in photovoltaic cells.