LAUSR

ABSTRACT Paeoniae Radix Alba slices, widely used in traditional Chinese medicine and as a functional food, play a vital role in health maintenance. However, their quality evaluation still depends on manual sensory inspection, which is subjective, labor-intensive, and difficult to standardize, leading to inconsistent grading and reduced consumer trust. To address this issue, we developed an intelligent grading system integrating deep learning and image processing techniques, focusing on diameter measurement and defect detection. A dataset of 1100 defect-containing images was used to train and evaluate YOLOv10 and Faster R-CNN. YOLOv10 achieved superior performance with a [email protected] of 96.4%, Recall of 90.1%, and Precision of 89%, outperforming Faster R-CNN in both accuracy and speed. Additionally, OpenCV was employed to calculate slice diameter and area. Combining defect detection and dimensional analysis, the system demonstrated high robustness and reliability. The method was deployed into a smartphone application, PRA-detect, allowing users to evaluate slice grades using common reference objects (e.g. ID cards) without specialized hardware. This approach provides a rapid, nondestructive, and portable solution for standardized grading of Paeoniae Radix Alba slices, with strong potential for real-world application in quality control and consumer decision-making.