AI-Enable Rice Image Classification Using Hybrid Convolutional Neural Network Models

Abstract

Rice is the most preferred grain worldwide, leading to the development of an automated method using convolutional neural networks (CNNs) for classifying rice types. This study evaluates the effectiveness of hybrid CNN models, including AlexNet, ResNet50, and EfficientNet-b1, in distinguishing five major rice varieties grown in Turkey: Arborio, Basmati, Ipsala, Jasmine, and Karacadag. It is estimated that there are 75,000 photographs of grains, with 15,000 images corresponding to each type. The training is improved by the use of preprocessing and optimization approaches. The performance of the model was assessed based on sensitivity, specificity, precision, F1 score, and confusion matrix analysis. The results show that EfficientNet-b1 achieved an accuracy of 99.87%, which is higher than the accuracy achieved by AlexNet (96.00%) and ResNet50 (99.00%). This study shows that EfficientNet-b1 is superior to other models that have emerged as state-of-the-art automated classification models for rice varieties. This indicates that there is a balance between the computational efficiency and the accuracy of EfficientNet-b1. These results exemplify the potential of CNN models for agriculture by reducing the restrictions associated with conventional classification approaches. These limitations include subjectivity and inconsistency regarding categorization.