Automated Tomato Quality Grading using Resnet50 Based Deep Learning and Digital Image Processing

Abstract

Accurate tomato grading is essential for maintaining quality standards in agricultural supply chains. The paper presents an automated tomato quality grading system using digital image processing and Convolutional Neural Networks (CNNs). Tomato images are captured under controlled conditions and preprocessed using noise reduction, illumination normalization, and segmentation techniques. A deep learning model utilizing the ResNet50 framework captures distinguishing features such as colour, texture, and surface defects, and classifies tomatoes, including unripe, ripe, damaged, and defective. Experimental results demonstrate high classification accuracy and consistent performance. The system reduces reliance on manual inspection and enables fast grading in real-time, making it appropriate for scalable agricultural and supply chain applications. The suggested system attains an accuracy of 95.4% and shows real-time capability with an average prediction duration of under one second, making it suitable for practical agricultural deployment.