Efficient Classification of Rice Varieties Using Optimized Convolutional Neural Networks

Rice type classification is a crucial task in agri-cultural automation, aimed at improving quality control and ensuring market standards. This study presents a deep learning-based approach using a optimized Convolutional Neural Network (CNN) model implemented with PyTorch to classify five rice varieties: Arborio, Basmati, Ipsala, Jasmine, and Karacadag. Our proposed model diverges from conventional methods that rely heavily on complex machine learning algorithms or pretrained CNN architectures like MobileNetV2, VGG16, EfficientNetB0 etc., which require a large number of parameters and layers, re-sulting in increased computational complexity. Through proposed approach, we achieved a remarkable 99.81% validation accuracy and a minimal validation loss of 0.01 % with a significantly reduced model size 30.76 MB. The simplicity of the proposed CNN architecture not only enhances computational efficiency but also maintains high accuracy, making it a viable solution for rice type classification in practical applications. This study demonstrates that a fine-tuned CNN model can outperform more complex models, providing an efficient and effective alternative for agricultural classification tasks.