Evaluation of Time-Series Color Change in Peach Fruits Using Deep Learning

Abstract

In this study, we propose a method to automatically obtain the Color Development Index (CDI) from orchard images to evaluate peach peel color non destructively. The CDI is a numerical index that increases with the progress of peel coloration, allowing the degree of peel coloration to be intuitively determined. The proposed method uses Mask R-CNN to detect fruit regions to obtain the fruit peel color on the image, and a model is constructed to automatically detect fruit regions that are not hidden by leaves, bags, or other objects. The model was constructed by automatically creating 1,000 composite images using cutout images of leaves, bags, and fruits, and training them. Using the model, we automatically obtained daily CDI values from peach orchard observation images during the harvest season and found that the error rate was less than 10% compared to the CDI manually obtained from the images, suggesting the effectiveness of the proposed method.