定制化AI训练平台EasyDL在青葙子及其混伪品图像分类中的应用  

Application of the Customized AI Training Platform EasyDL in the Classification of Semen Celosiae and Its Adulterants

作　　者：谭新宁 吴文如[1] 梁婉晴 陈浩文 赵凯莹 张睿 TAN Xinning;WU Wenru;LIANG Wanqing;CHEN Haowen;ZHAO Kaiying;ZHANG Rui(School of Pharmaceutical Sciences,Guangzhou University of Chinese Medicine,Guangzhou 510006,China)

机构地区：[1]广州中医药大学中药学院,广东广州510006

出　　处：《中国民族民间医药》2022年第15期40-45,共6页Chinese Journal of Ethnomedicine and Ethnopharmacy

基　　金：中央本级重大增减支项目(2060302);国家中医药管理局全国中医药创新骨干人才培训项目(国中医药人教函〔2019〕128号);广州中医药大学省级大学生创新创业训练计划项目(S202010572075)。

摘　　要：目的:利用人工智能和机器视觉技术,建立一种基于深度学习的青葙子及其混伪品图像分类方法。方法:通过定制化AI训练平台EasyDL,以青葙子药材及其混伪品的微性状图片为训练数据,对青葙子及其混伪品图像分类模型进行训练,并将该应用导入微信小程序,以便推广。结果:利用EasyDL构建的深度学习模型,青葙子及其混伪品图像分类准确率可以达到93.7%~94.8%。对于本系统所采集的药材图像,微信小程序识别率达到了80%~100%。在相同的饮片图像采集环境下,该模型能够准确识别出图像中药材的种类,具有比较稳定且比较好的识别效果。结论:利用人工智能和机器视觉技术,建立了基于深度学习的青葙子及其混伪品图像分类方法,拓宽了中药品质评价的研究思路,为人工智能在中药鉴定领域的普及提供了参考。Objective To establish an image classification method of Celosiae semen and its adulterants based on deep learning by using artificial intelligence and machine vision technology.Methods The image classification model of Celosiae semen and its adulterants was built and trained with the micro character images of Celosia argentea and its adulterants as training data on Baidu s EasyDL AI platform.Then the application was imported into WeChat applet for promotion.Results The classification accuracy of the deep learning model constructed by EasyDL can reach 93.7%-94.8%.For the images of medicinal materials collected by this system,the recognition rate of WeChat applet reaches 80%-100%.Under the same image acquisition environment,the model can accurately identify the type of Chinese herbal medicines in the image,which is relatively stable and has good recognition effect.Conclusions Using artificial intelligence and machine vision technology,an image classification method of Celosiae semen and its dulterants based on deep learning was established.It broadens the research ideas of quality evaluation of traditional Chinese medicine,and provides a reference for the popularization of artificial intelligence in the field of identification of traditional Chinese medicine.

关 键 词：EasyDL 青葙子 图像识别 人工智能 

分 类 号：R284[医药卫生—中药学]