基于MobileNet的10种清风藤植物石细胞图像智能鉴别研究  

Study on Intelligent Identification of 10 Species of Sabia Based on MobileNet and Stone Cell Images

作　　者：郭文凯 孙庆文 周宽文 GUO Wenkai;SUN Qingwen;ZHOU Kuanwen(Guizhou University of Traditional Chinese Medicine,Guiyang 550025,Guizhou,China)

机构地区：[1]贵州中医药大学,贵州贵阳550025

出　　处：《中华中医药学刊》2021年第1期160-162,I0041,I0042,共5页Chinese Archives of Traditional Chinese Medicine

基　　金：国家自然科学基金(81560707);贵州省国内一流学科建设项目(GNYL[2017]008);贵州省普通高等学校工程研究中心项目(黔教合KY字[2017]018);贵州省千层次创新型人才项目[贵中医(ZQ20150002)]。

摘　　要：目的建立基于MobileNet的10种清风藤植物石细胞图像智能分类研究方法。方法采用传统粉末制片方法制作临时装片并在相同倍数下拍摄图像,自建基于上述图像的10种清风藤植物石细胞图像数据集,对数据集图像进行归一化处理,采用MobileNet模型对数据集进行训练并改进,设置网络训练参数批次为18、优化器adam为5e-5、I2为0.05,得针对本研究的改进型MobileNet模型。结果经过15次迭代后得出MobileNet模型对10种清风藤植物石细胞图像的分类精度为89.66%。结论基于改进的MobileNet模型能够快速、准确、客观、无损的对10种清风藤植物石细胞图像进行分类,可以为后期基于石细胞的清风藤属或科的智能识别系统提供技术指导。Objective To establish a method based on MobileNet and stone cell images for the intelligent classification of 10 species of Sabia.Methods The traditional powder production method was used to make temporary loading and take images under the same multiples,and 10 species of Sabia stone cell image data sets based on the above images were constructed.The data set image was normalized,the data set was trained and improved by Mobilenet model,and the network training parameter batches were set to 18.The optimizer Adam was 5 e-5 and I2=0.05,and the improved Mobilenet model was designed for this study.Results After 15 iterations,it was concluded that the classification accuracy of the Mobilenet model to 10 kinds of Sabia stone cell images was 89.66%.Conclusion The improved Mobilenet model can classify 10 species of stone cell images of Sabia quickly,accurately and objectively,and provide technical guidance for the intelligent identification system of the genus of Sabia or subjects based on stone cells in the later stage.

关 键 词：清风藤属 MobileNet 石细胞 图像鉴别 

分 类 号：R282.5[医药卫生—中药学]