欢迎访问《中国粮油学报》编辑部网站！

陈卫东,范冰冰,刘超,丁俊丹,何为.基于改进ConvNeXt的水稻种子品种分类识别研究[J].中国粮油学报,2025,40(4):197-204

基于改进ConvNeXt的水稻种子品种分类识别研究

Study on Rice Seed Variety Classification and Recognition Based on Improved

投稿时间：2024-05-14 修订日期：2024-08-13

DOI：

英文关键词:rice seed ConvNeXt Deep learning Variety identification Machine vision

基金项目:财政部和农业农村部国家现代农业产业技术体系资助项目(CARS-03)

作者	单位	邮编
陈卫东^*	河南工业大学，粮食储运国家工程研究中心	450001
范冰冰	河南工业大学
刘超	河南工业大学
丁俊丹	河南工业大学
何为	河南工业大学

摘要点击次数: 0

全文下载次数: 0

中文摘要:

传统的水稻种子品种识别主要依靠人工完成,存在主观性强、识别效率低等问题,本研究提出了基于改进的ConvNeXt的水稻种子品种检测模型CS-ConvNeXt。该模型通过引入shuffNetV2 Unit降低模型碎片化程度的同时增加了不同分支通道之间的信息通信与共享,其次,引入Channel Shuffle以增强不同尺度水稻种子图像的多层特征之间的跨通道信息交互。选用5类水稻种子为测试对象,并与现有的卷积神经网络模型ResNet50、InceptionV2、MoblienetV3、ConvNeXt进行比较,结果表明,本研究所提出的CS-ConvNeXt模型在准确率、精确率、召回率及F1值分别达到了98.22%、98.04%、98.10、98.06%,消融实验结果验证了本研究提出的方法对模型性能的提升。提出了一种无损、高效、准确的水稻种子品种识别检测方法,为水稻种子品种识别研究提供了依据。

英文摘要:

The traditional rice seed variety identification mainly relies on manual work, which has the problems of strong subjectivity and low identification efficiency, etc. In this study, we propose a rice seed variety detection model CS-ConvNeXt based on the improved ConvNeXt, which reduces the fragmentation degree of the model through the introduction of the shuffNetV2 Unit and increases the information communication and sharing between different branch channels. Secondly, Channel Shuffle is introduced to enhance the cross-channel information interaction between multi-layer features of rice seed images at different scales. Five types of rice seeds were selected as test objects and compared with the existing convolutional neural network models ResNet50, InceptionV2, MoblienetV3, and ConvNeXt. The results show that the CS-ConvNeXt model proposed in this study achieves 98.22% in terms of accuracy, precision, recall, and F1 value, respectively, 98.04%, 98.10, and 98.06%, and the results of the ablation experiments verified the enhancement of the model performance by the method proposed in this study. A non-destructive, efficient and accurate detection method for rice seed variety identification is proposed, which provides a basis for rice seed variety identification research.

关闭