Tomato Classification Using Support Vector Machine

Abstract

Tomatoes are among the most widely cultivated vegetable crops and play a significant role in local cuisine and agriculture. However, there is limited knowledge about the different varieties of tomatoes, particularly the distinctions between Kurdish tomatoes (Khomali) and apple (Sewa) tomatoes. This study aims to automate the classification of these varieties using image-based machine learning techniques. A customized image dataset was created, and models based on Support Vector Machines (SVM) were developed for the classification task. Among the kernels tested, the linear kernel achieved the highest accuracy of 86%, especially after feature extraction using Principal Component Analysis (PCA), which improved performance from 83% to 86%. While Kurdish tomatoes were classified with high precision and recall, the recognition of apple tomatoes was hindered by class imbalance. Other kernels, including RBF, polynomial, and sigmoid, performed poorly due to overtuning and low recall rates. This research highlights the importance of kernel selection, data balance, and dimensionality reduction in enhancing classification results. Additionally, reducing image resolution is proposed to improve processing efficiency. By integrating PCA with linear SVM, this study offers a robust and scalable solution for tomato classification, contributing to better agricultural practices and more accurate recognition systems.