Smart Farming: CNN-Based Rice Variety Classification for Global Impact

The cultured variety is Oryza sativa and it is consumed daily by over 50% of the global populace, with differences between the rice varieties made on grain size, aroma, texture and ease of cooking. Some of the famous types include Karacadag, Basmati, Jasmine, Arborio, and Ipsala. The processes used previously in rice classification involved visual methods and were tedious, inconclusive, and imprecise. As the world continues to produce more rice, there is a strong need to employ innovative classification techniques that can serve the purpose as needed. The current research introduces a deep learning algorithm called Convolutional Neural Networks (CNNs) as a useful tool for rice variety classification. CNNs perform exceptionally well when training images for classification where features from images are learned to determine different patterns, which include but are not limited to texture and shape. The CNN model training involves use of dataset of 80,000 images of five rice variety images with each image resized to 256 x 256 pixels. To improve the classification accuracy and avoid overfitting, the authors use four approaches including data augmentation and convolutional and pooling layers. The outcomes of the proposed model show a high prediction accuracy level, 98%, which proves that CNNs can be applied to rice sorting and grading automation. This presents the findings of this study alongside a unique solution all farmers could apply to improve rice quality control and marketability, which in turn offer cheaper ways of improving its quality hence benefiting the global rice industry.