Automatic Segmentation Algorithm of Solar Cells Based on Grid Corner Extraction

Defect detection in photovoltaic (PV) panels based on electroluminescence (EL) images has been a prominent approach in the field. These approaches rely on pre-segmented PV cells as input. However, practical EL images suffer from background interference and image deformation caused by variations in perspective, making the segmentation and extraction of PV panel cells challenging. To overcome these issues, algorithm ASSC(Automatic Segmentation of Solar Cells) is proposed. Concretely, a combination of Image processing techniques and convolutional network is employed to address these problem. Additionally, image deformation is tackled by implementing perspective correction, which transforms the PV panel into a front view. Moreover, the semantic segmentation network is leveraged to extract the grid corner points of PV panels. Compared to traditional image techniques, ASSC offers a more accurate solution without prior knowledge of PV panel specifications. Experimental results on a dataset collected on site demonstrate the effectiveness of ASSC which achieves an IoU(Intersection over Union) metric of up to 97.25%.