TOMATO QUALITY CLASSIFICATION

Automated tomato quality classification is essential for improving post-harvest handling and ensuring consistent delivery of fresh produce.This study presents a deep learning-based approach to categorize tomatoes into four distinct quality classes: Unripe, Ripe, Old, and Damaged.Using a labeled image dataset from Kaggle, we investigate the effectiveness of advanced neural architectures in tackling this classification task.While traditional convolutional neural networks (CNNs) have been widely used for similar image-based assessments, we propose an enhanced methodology that integrates two state-of-the-art models: ResNet, known for its ability to extract detailed visual features through residual learning, and a hybrid model combining the Swin Transformer with a Random Forest classifier.The Swin Transformer captures both local and global image features using a hierarchical attention mechanism, while the Random Forest improves decision-making through ensemble learning.This combined approach addresses challenges such as inconsistent lighting, variety differences, and diverse ripeness levels across samples.To validate performance, we evaluate all models using accuracy, precision, recall, and F1-score, aiming to offer a reliable and scalable solution for automated tomato quality assessment, with potential applications in smart farming and supply chain optimization.