The purpose of this paper is to propose a novel and general detection framework called Prior-Det, which is combined with prior knowledge for solar cells based on transformer, aiming at the problem of detection accuracy and reducing the false positive rate.
The prior knowledge of image features is introduced to position and preprocess, and the knowledge combining module (KCM) is designed to acquire the image position and presegmented features. And a Swin Transformer-based model as vision backbone which use the Mask R-CNN framework is used for defect detection for solar cells, thus it could benefit from the blooming development of transformer. Then, the improved Swin Transformer block (ISTB) which only needs to focus attention on the region between the windows, is mainly used to improve the speed of detection.
Some experimental results show that the Prior-Det can effectively detect the solar cell surface defects with higher accuracy and greater adaptability.
Solar cells defect detection is one of the important parts in solar cells manufacturing. The proposed methods enhance the understanding of physical phenomena by providing more accurate models for data analysis and prediction, advancing the development of measurement theory and methods.
