• 首页
  • 期刊简介
  • 编委会
  • 投稿指南
  • 收录情况
  • 杂志订阅
  • 联系我们
引用本文:葛 琪,吴丽丽,康立军.基于改进ResNet50的中药材分类识别[J].软件工程,2025,28(4):16-20.【点击复制】
【打印本页】   【下载PDF全文】   【查看/发表评论】  【下载PDF阅读器】  
←前一篇|后一篇→ 过刊浏览
分享到: 微信 更多
基于改进ResNet50的中药材分类识别
葛 琪,吴丽丽,康立军
(甘肃农业大学信息科学技术学院,甘肃 兰州 730070)
geqi1103953519@163.com; wull@gsau.edu.cn; klj@gsau.edu.cn
摘 要: 为了提升中药材图片分类的准确率,提出了一种基于改进ResNet50的中药材分类识别方法。首先,引入了卷积块注意力模块(Convolutional Block Attention Module,CBAM),增强了模型对中药材特定特征的识别能力。其次,对标准的ResNet50中的卷积快捷连接进行了优化,减少了特征图的信息损失。最后,在模型后端集成了金字塔池化模块(Pyramid Pooling Module,PPM),该模块能整合多尺度的上下文信息,显著增强了模型捕获全局特征的能力。实验结果表明,相较于原模型及VGG16,改进后的模型在中药材识别上达到了94.75%的准确率,为中药材分类领域的后续研究工作提供了支持及优化的方向。
关键词: 中药材图像分类;ResNet50;CBAM注意力模块;PPM金字塔池化
中图分类号: TP391    文献标识码: A
Classification and Recognition of Chinese Medicinal Materials Based on Improved ResNet50
GE Qi, WU Lili, KANG Lijun
(College of Information Science and Technology, Gansu Agricultural University, Lanzhou 730070, China)
geqi1103953519@163.com; wull@gsau.edu.cn; klj@gsau.edu.cn
Abstract: To improve the accuracy of Chinese medicinal materials image classification, an enhanced ResNet50based classification method is proposed. Firstly, the Convolutional Block Attention Module (CBAM) is introduced to refine the discriminative features of Chinese medicinal materials. Secondly, the convolutional shortcut connections in standard ResNet50 are structurally optimized to mitigate feature map information loss. Finally, a Pyramid Pooling Module (PPM) is integrated at the backend of the model to aggregate multi-scale contextual information, significantly enhancing global feature representation capabilities. The experimental results show that, compared to the original model and VGG16, the improved model achieves an accuracy of 94. 75% in Chinese medicinal materials recognition,providing support and optimization directions for the subsequent research in Chinese medicinal materials classification.
Keywords: image classification of Chinese medicinal materials; ResNet50; Convolutional Block Attention Module(CBAM); Pyramid Pooling Module (PPM)


版权所有:软件工程杂志社
地址:辽宁省沈阳市浑南区新秀街2号 邮政编码:110179
电话:0411-84767887 传真:0411-84835089 Email:semagazine@neusoft.edu.cn
备案号:辽ICP备17007376号-1
技术支持:北京勤云科技发展有限公司

用微信扫一扫

用微信扫一扫