UV-Induced Fluorescence Imaging Combined with YOLO26-Based Classification for the Potato Surface Condition Assessment

Potato surface feature identification accuracy is a crucial parameter in determining the effectiveness of post-harvest classification systems and product quality assurance. Conventional color imaging systems often face limitations in distinguishing surface defects due to overlapping reflectance characteristics in the visible spectrum. This study proposes a quantitative classification framework that combines ultraviolet (UV)-induced fluorescence imaging with a YOLO26s-based deep learning classification model to identify three potato tuber surface conditions: normal, greening, and skinning injury. The YOLO26s classification model was trained using fluorescence images to learn distinguishing surface features. The proposed model achieved a training accuracy of 0.871, a validation accuracy of 0.871, and a testing accuracy of 0.875, demonstrating consistent generalization performance across the dataset. These results show that combining fluorescence images with deep learning classification provides a reliable and scalable approach for automatic potato surface quality assessment. This framework demonstrates strong potential for practical implementation in industrial grading systems.