Authors

Abstract

Traditional rice seed classification methods rely on manual observation of morphological features, which are inefficient and limited in accuracy. To improve the efficiency and accuracy of rice seed classification, this paper proposes a deep learning-based rice seed classification method using the SE-ResNet network architecture. This architecture integrates SENet into ResNet, enabling the model to capture and learn sensitive differential features among rice seeds. Through comparative experiments, the classification accuracies of SE-ResNet, ResNet, and AlexNet on the rice seed dataset were 89.58%, 72.97%, and 76.35%, respectively. The results demonstrate that SE-ResNet significantly outperforms ResNet and AlexNet in classification accuracy, validating its superiority in rice seed classification tasks.

Abstract in Chinese

传统的稻种分类方法依赖人工经验对形态特征的观察，效率低且准确性有限。为提高稻种分类的效率及精度，本文提出了一种基于深度学习的稻种分类方法，采用SE-ResNet网络结构，该结构将SENet集成到ResNet中，使其能够捕捉并学习稻种之间的敏感性差异特征。通过对比实验，SE-ResNet、ResNet和AlexNet三种网络结构在稻种数据集上的分类准确率分别为89.58%、72.97%和76.35%。结果表明，SE-ResNet的分类准确率明显高于ResNet和AlexNet，验证了其在稻种分类任务中的优越性。