An improved lightweight ConvNeXt for rice classification

Abstract

Under the pressure of climate change, the international food market is facing great uncertainty. Rice is widely grown as a major worldwide food crop, and different rice seeds often influence the future merit of a country's rice growth. As a major food crop widely grown around the world, the seed type of rice plays a key role in ensuring food security and optimizing agricultural productivity. Therefore, identifying the types of rice grains is an important task in rice breeding and cultivation. To this end, this study proposes a new model based on the ConvNeXt framework for detecting rice types, aiming to improve the identification efficiency. Our improved model achieved an average accuracy of 94.69 %. Compared to the baseline model ConvNeXt, the proposed network is more lightweight and more accurate. We conducted comprehensive experiments on the rice dataset from the GrainSpace public dataset to ensure the thoroughness and rigor of the study. Compared to existing models, our proposed model achieved the highest accuracy while maintaining lower FLOPs and parameters.

Keywords

• Rice classification
• Lightweight model
• Data augmentation
• Image classification