YOLO-based classification algorithm for irregular and occluded Bonnet bellflower roots (Codonopsis lanceolata): Optimizing thermal processing through image extraction, clustering, and CFD simulation

초록

This study proposes a novel system for optimizing the steaming process of Bonnet balloon flower (Codonopsis lanceolata) roots by integrating YOLO-based object detection, image analysis, K-means clustering, and computational fluid dynamics (CFD) simulations. It focuses on size classification for determining optimal thermal processing conditions, with YOLO-based object detection offering an innovative approach to accurately, nondestructively sorting irregularly shaped roots into distinct classes. Roots were categorized into three classes (0, 1, and 2) based on morphology, with average lengths of 297.64 f 31.34, 152.67 f 29.87, and 189.17 f 23.45 mm, and thicknesses of 37.73 f 2.24, 28.89 f 2.37, and 29.67 f 2.77 mm, respectively. YOLOv12 achieved high accuracy, with Intersection over Union (IoU) values of 0.999 (well-arranged) and 0.953 (occluded), even under complex conditions. Classified roots, particularly in classes 1 and 2, exhibited lower internal crack rates (47.36 f 2.59 % and 32.87 f 2.45 %) and more uniform reductions in total dissolved solids (TDS) compared to non-classified roots. Steaming for 40 min at 100 degrees C ensured microbial safety, while smaller roots (class 1 and 2) reached thermal equilibrium faster (45.78-51.22 min) due to efficient heat transfer. The system's integration of YOLO-based classification presents a major advancement in optimizing thermal processing of irregularly shaped plant roots by enabling rapid sorting and improving product consistency.

키워드