基于深度学习算法的川贝母、山楂及半夏饮片的智能鉴别  

Intelligent Identification of Fritillariae Cirrhosae Bulbus,Crataegi Fructus and Pinelliae Rhizoma Based on Deep Learning Algorithms

作　　者：吴冲 谭超群 黄永亮[3] 吴纯洁[2] 陈虎[1] WU Chong;TAN Chao-qun;HUANG Yong-liang;WU Chun-jie;CHEN Hu(Key Laboratory of Visula Synthesis Graphics and Image Technology for National Defense,Sichuan University,Chengdu 610065,China;Chengdu University of Traditional Chinese Medicine(TCM),Chengdu 611137,China;Affiliated Hospital of Chengdu University of TCM,Chengdu 610075,China)

机构地区：[1]四川大学视觉合成图形图像技术国防重点学科实验室,成都610065 [2]成都中医药大学,成都611137 [3]成都中医药大学附属医院,成都610075

出　　处：《中国实验方剂学杂志》2020年第21期195-201,共7页Chinese Journal of Experimental Traditional Medical Formulae

基　　金：四川省中医药管理局科学技术研究专项项目(2018QN003);四川省科技厅计划项目(2018JY0435)。

摘　　要：目的:利用人工智能和机器视觉技术,提出一种检测与评价中药材的新方法。方法:以川贝母、山楂及半夏饮片为研究对象,通过机器视觉采集图片大数据,建立图像数据库;通过对中药外在性状特征的智能分析,以深度学习为手段,研究建立深度卷积神经网络模型来实现定位检测、品种识别等功能,以显著提高中药快速鉴别的准确率。结果:测试的11种饮片(生山楂、炒山楂、焦山楂、山楂炭、松贝、青贝、炉贝、生半夏、姜半夏、法半夏、清半夏)图像品种分类准确度可达99%以上,具体类别的平均识别准确度可达到97%以上。结论:通过深度学习算法实现的中药饮片智能鉴别技术具有简洁、快速、精度高、可批量化检测的优点,可为中药质量检测与评价提供技术支持,并丰富了中药品质评价的研究思路。Objective: To propose a new method for detecting and evaluating traditional Chinese medicine(TCM)by artificial intelligence and machine vision technology. Method: Taking Fritillariae Cirrhosae Bulbus,Crataegi Fructus and Pinelliae Rhizoma as the research objects,big data of pictures was collected by machine vision and the image database was established. Through the intelligent analysis of the external characteristics of TCM,the deep convolutional neural network model was established to realize the functions of location detection and variety identification by means of deep learning,so as to significantly improve the accuracy of rapid identification of TCM. Result:The classification accuracy of 11 kinds of Chinese herbal pieces(raw,fried,parched and charred products of Crataegi Fructus,Pinelliae Rhizoma,Pinelliae Rhizoma Praeparatum Cum Zingibere et Alumine, Pinelliae Rhizoma Praeparatum, Pinelliae Rhizoma Praeparatum Cum Alumine and three products of Fritillariae Cirrhosae Bulbus)could be more than 99%,and the average recognition accuracy of specific categories could reach more than 97%. Conclusion: The intelligent identification technology of TCM decoction pieces realized by deep learning algorithms has the advantages of simplicity,rapidity,high precision and quantifiable detection,which can provide technical support for the quality detection and evaluation of TCM,and enrich the research ideas of quality evaluation of TCM.

关 键 词：川贝母 半夏 山楂 饮片 图像识别 深度学习 卷积神经网络 

分 类 号：R22[医药卫生—中医基础理论] R28[医药卫生—中医学]