Learning Rate Regularized EfficientNet based Rice Grain Classification

One of the most popular grain items cultivated across the globe is rice. Rice seeds may be recognized by their appearance, form, and ethnicity. Rice types may be distinguished owing to their various traits. This rice category assessment relates to the seed efficiency, which includes various hereditary pairings. Along with population development, there is a rise in the demand for rice, thus farmers require high-quality rice grains. Due to the variety of rice grain instances currently available on the market, rice grain classification becomes crucial. A significant amount of time and effort might be consumed on manual classification of grains. The technological innovations could be utilized to automatically identify rice grain images and provide grain categorization based on grain features. This paper suggests Learning Rate Regularized Efficient Net (LRR-EN) to classify the rice grain more efficiently. The 30,000 Rice images of Rice Image Dataset from Kaggle is used in the suggested model. The examination of image distribution is the first step in the LRR-EN model. The best CNN model is selected by fitting the rice images with existing CNN models after image augmentation. When compared to other CNN models, the Efficient Net demonstrated an accuracy of 92.41%. The learning rate parameter was now fine-tuned to increase accuracy using the EfficientNet CNN. The augmented rice images are fitted to LRR-EN using several learning rates, such as 0.0001, 0.001, 0.01, and 0.1. The experimental findings reveal that, when compared to different learning rates, the suggested LRR-EN with a 0.001 learning rate demonstrates the greatest Accuracy of 99.01%.