Rice Varieties Classification Using Neural Network and Transfer Learning with MobileNetV2

Rice plays a vital role in global food security, feeding over 8 billion people worldwide. The classification of rice varieties based on visual characteristics, such as color, shape, and texture, is crucial for quality inspection and efficient farming practices. However, existing classification methods heavily rely on subjective and time-consuming human visual inspection, leading to errors and inefficiencies. This research paper explores the potential of deep learning approaches, specifically neural network and transfer learning with MobileNetV2, for accurate and efficient rice variety classification. The paper builds upon previous studies that have demonstrated the superiority of deep learning over traditional machine learning methods. Experimental results using a publicly available dataset of 75,000 rice grain images, labeled with five rice varieties, show that both the neural network and MobileNetV2 models achieve high accuracy rates. The neural network model is trained using stochastic gradient descent with 250 epochs, while the MobileNetV2 model leverages pretraining on the ImageNet dataset and is trained with the Adam optimizer for 10 epochs. The comparative analysis reveals that the MobileNetV2 model performs slightly better, benefiting from its ability to extract features from images learned from the large-scale ImageNet dataset. The findings highlight the potential of deep learning methods for rice variety classification, offering improved accuracy and efficiency in rice quality inspection. The choice between the neural network and MobileNetV2 models depends on the specific requirements and priorities of the application, with the neural network offering flexibility and transparency, and MobileNetV2 providing high accuracy through pre-trained feature extraction.