Rice Grains Classification and Quality Analysis Based on Morphological Features with Deep Learning Methods

Introduction: The main objective of this paper is to classify different categories of rice grains. In this research, it is focused on five types of rice: Basmati, Ponni, Kuthiraivali, IR 20, and Seeraga Samba. A total of fifteen features were extracted, including thirteen morphological features and two visual features—colour and texture. The rice grains were classified into three categories: Best, Good, and Average. Methods: To classify the images of rice grains, the aspect ratio and visual features were calculated. The ResNet50 and Xception deep learning models were employed for classification. The performance of these methods was evaluated using Accuracy, Precision, Recall, and F1-score metrics. Results: In this study, with respect to the F1-score, the ResNet50 and Xception models achieved classification rates of 99.51% and 99.67%, respectively. The experimental results demonstrate that these models significantly outperform existing methods for rice classification. Discussion: The Xception model excels at learning significant representations of rice grains by effectively separating spatial and channel-wise transformations. This characteristic allows it to generalize better with unknown and smaller datasets compared to ResNet50. However, it is important to note that the Xception model requires more computational time than ResNet50 due to its use of depth-wise separable convolution. Conclusion: Overall, this study concludes that the Xception model outperforms the ResNet50 model. Moreover, both models exceed the performance of other deep learning models in terms of Accuracy, Precision, Recall, and F1-score. conclusion: Overall, in this study, the Xception model performs better than the ResNet50; both ResNet50 and Xception models outperform the other deep learning models with respect to Accuracy, Precision, Recall, and F1-score.