CONVOLUTIONAL NEURAL NETWORK-INTEGRATED IMAGE PROCESSING METHOD FOR THE POST-HARVEST REMOVAL PROCESS OF GARLIC CLOVE (ALLIUMSATIVUM)

초록

This study proposes a convolutional neural network (CNN)-integrated approach for automating the basal plate removal process in garlic (Allium sativum) cloves. The irregular shapes, sizes, and orientations ofgarlic cloves pose challenges for mechanized post-harvest processing. To address this, a CNN-integrated image processing pipeline was developed to identify the basal plate's coordinates [x, y] and cutting angles from orthoimages ofgarlic cloves. The preprocessing steps involved RGB-to-HSV color space conversion and Otsu's thresholding for clove segmentation, followed by principal component analysis (PCA) for horizontal alignment of the garlic images. A five-layer CNN was designed to classify the basal plate position as left or right, achieving an accuracy of 98.25%. Subsequently, quadratic regression was applied to predict optimal cutting lines. The proposed algorithm demonstrated a root mean square error (RMSE) of 7.78 degrees and an R2 value of 0.91 in static detection tasks. The results indicate the method's potential for integration into high-speed conveyor systems for real-time processing, reducing manual labor and improving efficiency in garlic post-harvest handling.

키워드