Real-time object detection and diagnosis of tomato quality using YOLO

Abstract

Tomato quality plays a critical role in both customer satisfaction and the efficiency of post-harvest processing. Traditional sorting and grading methods are labor-intensive, subjective, and unsuitable for high-throughput operations. This study proposes a smart AI-based vision system capable of real-time classification of tomato quality into three classes: fresh, damaged, and unripe. The system employs advanced deep learning techniques, specifically a custom-trained YOLO object-detection model, to analyze key visual attributes such as color, texture, and surface defects. A diverse, labeled custom dataset of tomato images was collected to train and evaluate the model. This dataset included tomato images with the 3 health conditions according to the output classes of the neural network. The results show that the system has achieved high accuracy and strong robustness across varying lighting conditions and backgrounds, making it suitable for deployment in real agricultural and industrial environments. By enabling fast, automated, and objective quality assessment, the proposed system significantly enhances the reliability and efficiency of tomato grading and contributes to improved food supply chain management. Received: 04 November 2025 Accepted: 16 December 2025 Published: 23 December 2025