摘要
在图像处理和深度学习的中药材鉴技术的研究中,针对识别速度与精度的提升问题,提出一种基于改进TCM-Net 的鉴别方法。先是创建了规范化的中药图像数据集为鉴别技术提供了数据支持,同时引入注意力机制的概念,设计了相关的系统模型,即Attention-T CM-Net,应用于中药识别的实际操作中。首先提升了关注每种药材特征的性能,然后改进移动倒置瓶颈卷积模块,此设计既保证了轻量级网络的实现,又加强了中药识别的准确性。结果表明:无迁移学习中模型top-1准确率为97.48%,宏准确率98.26%;引入注意力机制后模型的top-1准确率和宏准确率分别是98.15%和98.62%。实验证明此系统模型适用于图像处理和深度学习的中药材鉴别。
In the research of image processing and deep learning for TCM identification technology,aiming at the improvement of background features,recognition speed and accuracy,a lightweight TCM identification network system Attention-T CM-Net model is proposed based on the Attention mechanism.Through the channel attention mechanism and the introduction of the mechanism of spatial attention,strengthen the attention of the characteristics of traditional Chinese medicine(TCM),and for mobile inverted bottleneck convolution module is improved,the lightweight network model on the basis of guarantee the high accuracy of traditional Chinese medicine identification,the final experiment,this method accuracy and top-1 macro accuracy were 99.25%and 99.52%,respectively,The parameter quantity was 4.03M and FLOP was 0.39B,indicating that it is suitable for image processing and deep learning identification of Chinese medicinal materials.
作者
王青青
董能峰
杨怡
刘盼
WANG Qingqing;DONG Nengfeng;YANG Yi;LIU Pan(Baoji Vocational&Technical College,Baoji Shanxi 721000,China)
出处
《自动化与仪器仪表》
2023年第1期30-35,共6页
Automation & Instrumentation
基金
陕西省教育厅科研计划项目《“太白七药”种质资源的收集与保存》(19JK0047)。
关键词
中药材鉴别
图像处理
深度学习
注意力机制
identification of Chinese medicinal materials
image processing
deep learning
attentional mechanism
