An Intelligent Machine Learning Framework for Automated Tomato Maturity Detection and Optimal Harvest Scheduling

Abstract

Crop maturity classification is crucial for reducing post-harvest waste and supporting precision agriculture. This research presents an automatic machine learning system for tomato maturity detection and scheduling for optimal harvesting. A dataset of 5,551 tomato images, divided into Unripe, Ripe and Old maturity levels, is used for training and testing. Transfer learning on a deep learning architecture based on MobileNetV2 is used to learn spatial and textural cues. Particular attention is given to differentiating visually similar ripeness levels, such as ripe and old tomatoes, by capturing subtle texture features and signs of dehydration. The model exhibits a high validation accuracy of 93.96% in the early stages of training, showing efficiency and effectiveness. The proposed system also successfully detects the commonly neglected "Old" class that is easily misclassified in traditional approaches. The proposed approach offers a lightweight and efficient system for real-time application, facilitating accurate harvest planning and promoting the adoption of smart automation in agriculture.