Tomato Ripening Stage Prediction Using a Cascade Model of YOLOv8 and ViT

Precise fruit or vegetable maturity assessment is crucial in agriculture to optimize harvest time and product quality. However, this process faces several challenges. Traditionally, maturity prediction was done visually, which is both time-consuming and labour-intensive. Therefore, there is a need of advanced methods for ripeness prediction using computer vision and deep learning techniques while considering environmental factors. This study shows a cascade model that combines the YOLOv8 (You Only Look Once Version-8) object detection algorithm and Vision Transformer (ViT) classification algorithm architectures to measure tomato maturity accurately. The cascade model first uses YOLOv8n to locate and detect tomatoes within the images and then uses ViT to classify the fine-grained maturity stage. The model was trained, validated and tested on a dataset of 1460 images which consisted of mature, immature and partially mature classes. The cascade approach shows promising performance, achieving 93.1% accuracy in tomato maturity assessment. This is a significant improvement over the YOLOv8x classification model, which scored 87.93% accuracy on the same dataset. The proposed cascade model enables automated assessment of tomato maturity, which could be highly beneficial for enhancing performance in optimizing crop harvesting and minimizing post-harvest losses.