抄録
Electroluminescence (EL) imaging is a reliable technique for inspecting photovoltaic (PV) modules, as its high spatial resolution enables the detection of even minute flaws on the module surfaces. However, conventional EL image analysis, which is typically performed manually, is expensive, time-consuming, and requires expert knowledge of a wide variety of defects. In this study, we propose a method for the automatic detection of solar defects within a single EL image of a solar PV cell. Our approach utilizes a lightweight convolutional neural network (L-CNN) developed from scratch for feature extraction and a fully connected artificial neural network (ANN) for classifying functional and defective solar cells. The model is trained on 400 solar cell images extracted from high-resolution EL images of monocrystalline PV modules. The CNN model achieved an accuracy of 93.0% in detecting defective solar cells from the test dataset. Our model is light, making it easy to run on arbitrary hardware, and doesn’t require plenty of EL images for training. Our proposed method will facilitate continuous, fast, and accurate solar PV plant quality inspection. Proper maintenance of solar PV panels improves their efficiency and power output.
