期刊论文

Research and Application of Chinese Herbal Medicine Plant Image Recognition Based on Deep Learning

中文

信息与电脑

1003-9767

2022

34

12

32-34

10.3969/j.issn.1003-9767.2022.12.011

湖南省教育厅科学研究项目（项目编号：20C1435）； 长沙市自然科学基金项目（项目编号：kq2202260）；

本校为第一机构

为提高中草药植物图像识别的准确率，使用深度学习技术对常见的中草药植物图像进行识别与分类。首先，通过拍照采集15种常见的中草药图像并进行数据预处理，构建中草药植物图像数据集；其次，选取VGG、GoogLeNet、ResNet、MobileNetV2、MobileNetV3共5种网络模型对图像进行识别并开展对比实验；再次，通过迁移学习优化网络结构，同时通过数据增强扩充现有的数据集；最后，对实验结果的准确率和损失率进行对比分析。MobileNetV3网络表现最为优异，其识别度高达99.64%，选取其作为开发“智能中草药识别App”的网络模型。实验结果表明，基于深度学习的卷积神经网络可以极大提高中草药植物图像识别的...

To improve the accuracy of herbal plant image recognition, the common herbal plant images are recognized and classified using deep learning techniques. Firstly, 15 common herbal images were collected by photographing and data preprocessing to build a herbal plant image dataset;secondly, a total of five network models, VGG, GoogLeNet, ResNet, MobileNetV2, MobileNetV3, were selected to recognize the images and conduct comparison experiments;again, the network structure was optimized by migration learning, while Finally,the accuracy and loss rate of the experimental results were compared and anal...