Classification of rice product using deep learning techniques

Rice is a basic foodstuff and an economic agricultural product of strategic importance both globally and in Turkey. In addition to the production activities concentrated especially in the Asian continent, rice production is also widely carried out in regions such as Edirne, Samsun and Manisa in Turkey. There are many different types of rice in terms of morphology, color and shape in the world and in Turkey. However, manual classification processes carried out with traditional methods lead to inefficiencies in terms of time and cost, and this situation reveals the need for more innovative and automated approaches in agriculture. In this context, a comprehensive dataset consisting of 75 000 images in total, 15 000 for each type, was used to classify Arborio, Basmati, Jasmine, Ipsala and Karacadag rice varieties commonly grown in Türkiye. The images were evaluated based on both morphological features and RGB and HSV color spaces; classification operations were performed using deep learning-based Convolutional Neural Networks (CNN). In addition to widely used transfer learning models such as VGG-16, InceptionV3, EfficientNetB0, ResNet50 and DenseNet, seven different new model architectures developed with layered improvements were tested within the scope of the study. Data preprocessing and data augmentation techniques applied in this process positively affected model performances and significant increases were achieved in classification accuracies. As a result of the comparative evaluations, it was observed that all tested models exhibited high success rates; especially the developed Model 7 stood out with an accuracy rate of 99,6% and reached a level that could compete with transfer learning-based models. The findings obtained show that convolutional neural networks and transfer learning methods can work with high accuracy in the classification of rice varieties. This study demonstrates the effectiveness of artificial intelligence-supported automation systems in the classification of agricultural products and shows that these technologies have a wide potential for use in applications aimed at increasing efficiency in the agricultural sector.