This study uses the detection model's confidence as a score for the image. It emphasizes building a proper scoring model and performing quantitative evidence analysis. The goal is to improve the interpretability and scientific reliability of detection results in judicial contexts. - View it on GitHub