Automated Shallot Classification on Conveyor Belts Using Watershed Transform and Otsu Thresholding

Manual classification of shallots often lacks precision, especially when bulbs overlap or have irregular shapes on conveyor systems. This study presents an automated computer vision approach combining Otsu thresholding and the watershed transform to detect and measure shallot diameters. Data were collected from video recordings under varying conveyor speeds and camera distances. The processing pipeline—consisting of preprocessing, segmentation, and feature extraction achieved a segmentation accuracy of 99.57% and a diameter estimation Mean Absolute Percentage Error (MAPE) as low as 2.84%. These results demonstrate a significant improvement in separating overlapping objects and measuring irregular shapes. This system contributes to the literature on post-harvest classification automation and shows strong potential for implementation in industrial shallot processing lines to enhance efficiency, ensure consistent quality, and significantly reduce reliance on manual labor. However, the system's performance may vary under extreme lighting conditions, indicating opportunities for further improvement.