AUTOMATED TOMATO QUALITY ASSESSMENT USING TRANSFER LEARNING AND MACHINE LEARNING CLASSIFIERS

Abstract

The growing demand for high-quality tomatoes and large-scale production has highlighted the need for efficient inline quality grading systems. Manual grading is labor-intensive and costly, prompting the development of automated solutions. This study presents a hybrid approach that combines pre-trained convolutional neural networks (CNNs) for feature extraction with traditional machine learning algorithms like support vector machines (SVM), random forest (RF), and k-nearest neighbors (KNN) for classification. Using the NVIDIA Jetson TX1, a tomato image dataset was created, and preprocessing techniques were applied to enhance feature learning. The CNN-SVM model stood out, achieving 97.50% accuracy in binary classification (healthy vs. rejected) and 96.67% in multiclass classification (ripe, unripe, rejected). On a public dataset, the CNN-SVM model reached 97.54% accuracy, outperforming other hybrid models. Key performance metrics, including accuracy, recall, precision, specificity, and F1-score, were evaluated..