牛膝和川牛膝药材的特征提取与图像识别  

Feature Extraction and Image Recognition of Achyranthes bidentata and Cyathula officinalis

作　　者：王耐[1] 卢文彪[1] 凌秀华 梁丽金[1] 李熙灿[1] 李睿[1] 

机构地区：[1]广州中医药大学中药学院,广州510006

出　　处：《中国药房》2017年第12期1670-1673,共4页China Pharmacy

基　　金：国家自然科学基金资助项目(No.81573558)

摘　　要：目的:提取牛膝和川牛膝药材的特征,并建立其图像识别方法。方法:采用MATLAB软件编程拼接牛膝和川牛膝药材的横切面显微图像,提取颜色、不变矩、纹理和横切面维管束组织特征;将数据整理成数据矩阵,通过Zscore函数对数据矩阵进行标准化,通过Princomp函数进行主成分分析;采用BP神经网络识别模式。结果:药材样品显微图像拼接处的组织细节保持完整;测得每组药材样品图像的27个特征参数,经主成分分析,选取11个主成分参数建立BP神经网络,两种近缘药材样本(n=50)的BP神经网络平均识别率为100%。结论:该方法可用于中药材显微图像自动拼接,及牛膝和川牛膝药材的图像识别。OBJECTIVE： To extract the feature ofAchyranthes bidentata and Cyathula oJficinalis, and to establish image recognition method. METHODS： The microscopic image stitching ofA. bidentata and C. oJficinalis was implemented by MATLAB. The color, invariant moment, stripes and the features of vascular bundle in cross section were extracted. The data was organized into data matrix, and then data matrix was standardized by Zscore function; principal components were analyzed through Princomp function. BP nerve network recognition mode was adopted. RESULTS： The microstructures in the micro images of the samples were kept integrated. The measured data of 27 characteristics were acquired in each group of sample. Through principal component analysis, the parameters of 11 main components were selected to establish BP never network. The average recognition rate of BP nerve network was 100% between 2 medicinal material relatives （n=50）. CONCLUSIONS： The method can be used for micro-image auto Stitching of Chinese medicinal materials and image recognition ofA. bidentata and C. officinalis.

关 键 词：牛膝 川牛膝 图像拼接 模式识别 MATLAB软件 

分 类 号：R282.5[医药卫生—中药学]