Morphological features for multi-model rice grain classification

Abstract

In the realm of agriculture and food processing, the automated classification of rice grains holds significant importance. The diverse varieties of rice available demand a systematic approach to categorization. This study tackles this challenge by employing diverse machine learning models, including support vector machine (SVM), random forest (RF), logistic regression (LR), decision tree (DT), Gaussian naive Bayes (GNB), and k-nearest neighbors (K-NN). The dataset, sourced from Kaggle, features five distinct rice types: Arborio, Basmati, Ipsala, Jasmine, and Karacadag. After the images undergo preprocessing, a set of 13 distinct morphological features is extracted. These features ensure a comprehensive representation of rice grains for accurate classification. This study aims to create an intelligent system for efficient and precise rice grain classification, contributing to optimizing agricultural and food industry processes. Among the models, K-NN demonstrated the highest classification accuracy at 97.80%, surpassing random forest (97.51%), DT (97.48%), GNB (96.99%), SVM (96.85%), and LR (96.05%). Our proposed K-NN-based classification model achieves an accuracy of 97.8%, demonstrating competitive performance and outclassing several state-of-the-art methods such as artificial neural network (ANN) and modified visual geometry group16 (VGG16) while maintaining simplicity and computational efficiency. This underscores the effectiveness of K-NN and RF in enhancing the precision of rice variety classification.