基于卷积神经网络的中医面色提取识别研究  

Research on TCM color extraction and recognition based on the convolutional neural network

作　　者：孙康宁 孙琦 李新霞[1] 戴彩艳 SUN Kang-ning;SUN Qi;LI Xin-xia;DAI Cai-yan(School of Artificial Intelligence and Information Technology,Nanjing University of Chinese Medicine,Nanjing 210023,China;School of Computer Science and Engineering,Nanjing University of Science and Technology,Nanjing 210094,China)

机构地区：[1]南京中医药大学人工智能与信息技术学院,南京210023 [2]南京理工大学计算机科学与工程学院,南京210094

出　　处：《中华中医药杂志》2021年第7期4286-4290,共5页China Journal of Traditional Chinese Medicine and Pharmacy

基　　金：江苏省青年基金项目(NO.BK20180822);江苏省高等学校自然科学研究面上项目(No.18KJB52004)。

摘　　要：目的:研究对比3种机器学习与深度学习算法在中医面色提取识别中的应用效果。方法:将人脸的面色分类分为整体面色分类与局部面色分类,基于人脸检测、人脸68个特征点定位技术,提出人脸的8个局部感兴趣区域的提取方法。采用两批训练集对卷积神经网络、支持向量机、k-means算法进行模型构建并分析3种模型识别面色的准确率。结果:在小样本的情况下卷积神经网络与支持向量机对面色识别的效果较好,当训练集的数量增加到1230张时,卷积神经网络的准确率提高为95.107%,明显优于其他两种算法。结论:将深度学习算法结合中医面诊理论对中医临床诊断的标准化、定量化研究具有推动作用。Objective: To compare the application effects of three machine learning and deep learning algorithms on TCM color extraction and recognition. Methods: The facial color classification of the face was divided into the overall facial color classification and the local facial color classification. Based on the face detection and the 68 feature points localization techniques of the face, the extraction methods of the eight local interest regions of the human face were proposed. The two sets of training sets were used to construct the convolutional neural network, support vector machine and k-means algorithm, and analyze the accuracy of the three models to identify the color. Results: In the case of small samples, convolutional neural network and support vector machine had better effect on facial color recognition. When the number of training sets increases to 1 230, the accuracy of convolutional neural network was improved to 95.107%, which was obviously better than the other two. Conclusion: The combination of deep learning algorithm and TCM face diagnosis theory has promoted the standardization and quantitative research of TCM clinical diagnosis.

关 键 词：中医面诊 机器学习 面色分类 深度学习 

分 类 号：TP391.41[自动化与计算机技术—计算机应用技术] TP183[自动化与计算机技术—计算机科学与技术] R241[医药卫生—中医诊断学]