Objective Non-Destructive Estimation of Rachis Browning in Table Grapes Using Image Segmentation and Color Distribution
Objective Non-Destructive Estimation of Rachis Browning in Table Grapes Using Image Segmentation and Color Distribution
Yehonatan Bar Moshe, M.Sc. student at School of Industrial & Intelligent Systems Engineering
Advisors: Prof. Noam Koenigstein, Dr. Yael Salzer
Abstract:
Rachis browning is a key post-harvest quality indicator in table grapes, significantly impacting marketability and consumer perception of freshness. Understanding the factors contributing to rachis browning is essential for maintaining product quality and minimizing economic losses in the supply chain. Effective management relies on accurate assessments of browning levels. Previous studies have used either subjective expert evaluations or objective methods that require removing fruit from the rachis, an approach that destroys the crop. However, advancements in image processing now offer a non-destructive, automated alternative for quality assessment.
In this study, we propose a pipeline for objective color analysis in images, consisting of two main procedures. First, a fine-tuned Mask R-CNN model performs image segmentation to isolate the rachis from a grape bunch without requiring berry removal. Second, the extracted rachis undergoes color distribution analysis, where we estimate multivariate Gaussian parameters for RGB color distribution to quantify browning levels. Initial results indicate that the fine-tuned Mask R-CNN model effectively segments the rachis with high accuracy. Additionally, a simple linear regression analysis shows a correlation between the estimated color distribution parameters and domain expert color grading. These findings highlight the potential of our approach for automated, objective quality control in agriculture and beyond
Bio:
Yehonatan Bar Moshe, is an MSc student in industrial engineering and a deep learning researcher at the DELTA lab, under the supervision of Prof. Noam Koenigstein and Dr.Yael Salzer.
