ASABE Technical Library - Abstract

Click on “Download PDF” for the PDF version or on the title for the HTML version.
If you are not an ASABE member or if your employer has not arranged for access to the full-text, view access and subscription options

A Fundamental Study on Automatic Garlic Cultivar Identification and Mass Prediction Based on Image Processing and Machine Learning

Published by the American Society of Agricultural and Biological Engineers, St. Joseph, Michigan www.asabe.org

Citation:  2026 ASABE Annual International Meeting  2600293.(doi:10.13031/aim.202600293)
Authors:   Wonyeol Choi, Jinho Son, Juhee Lee, Bumseok Park, Hyunggyu Park, Yushin Ha
Keywords:   Computer vision, Cultivar identification, Garlic, Machine learning, Mass prediction, Postharvest sorting.

Abstract. Garlic (Allium sativum L.) is an important upland crop in South Korea, where postharvest sorting and quality grading directly influence market value. Conventional mechanical sorters are efficient for size-based grading; however, dried stems, roots, and foreign materials can affect sorting accuracy, and cultivar-specific settings may be required. This study evaluated the feasibility of a vision- and machine learning-based framework for garlic cultivar identification and mass prediction. Two cultivars, Changnyeong Daeseo and Uiseong, were used. After removing one outlier, 200 Changnyeong Daeseo samples and 299 Uiseong samples were analyzed. Five morphological variables were measured using digital calipers,and bulb mass was measured using an electronic balance. and cultivar-specific mass prediction models were developed using Linear Regression, Random Forest, and k-Nearest Neighbors with 5-fold cross-validation. Bulb width and thickness were the dominant predictors of mass in both cultivars. Linear Regression showed the best performance, with R² = 0.871 for Changnyeong Daeseo and R² = 0.933 for Uiseong. A YOLOv11 model achieved mAP50 = 0.990 with no inter-cultivar misclassification. In the fully automated scenario using PLY-extracted width and thickness, R² values were 0.727 and 0.797 for Changnyeong Daeseo and Uiseong, respectively. These results suggest that the proposed framework can support flexible, cultivar-specific, vision-based garlic sorting.

(Download PDF)    (Export to EndNotes)