Sugar Content Prediction in Strawberries using SAM and YOLO

Department of Agriculture, Environmental, and Food Sciences, University of Molise, Campobasso, 86100, Italy

^* Corresponding author: This email address is being protected from spambots. You need JavaScript enabled to view it.

Abstract

The Brix° (TSS) level is one of the parameters to determine the quality of the fruits in terms of sugar content. The single measurement is time-consuming, while for commercial purposes, this measurement is necessary for consumer preference. The aim of this research is to utilize artificial intelligence (AI) to predict the brix (TSS) level in order to reduce time. The Brix levels were measured randomly on 90 fruits using a refractometer. These fruits were annotated with low (3–5.4), medium (5.46.6), and high (6.6–13.0) ranges according to the collected data using instance segmentation (SAM3) in Roboflow. These labeled images were split: 87% for training, 9% for validation, and 4% for testing using YOLOv8s. The 360 images were put into one folder for prediction. The training had an epoch of 300, a batch of 8, a resized image size of 640, and a patience of 50. According to our training result, the model achieved 100% recall, 32% precision, and 62% per-class AP@50, which is 49% for low classes, 69% for medium classes, and 67% for high classes. However, the estimation of the Brix level is 353 for medium and only 6 for the small Brix levels; one fruit was undetected. The model identifies all the fruit as being medium level since there is no varied chromatic variation. Further studies should identify the different ripeness levels for better prediction. Keywords: Brix° (TSS) prediction, chromatic variations, SAM3, deep learning, and YOLOv8s.