北京生物医学工程

基于AdaBoost级联框架的舌色分类

Tongue color classification based on AdaBoost cascade framework

作者：王奕然张新峰

单位：北京工业大学信息学部（北京 100124）

分类号：R318

出版年·卷·期（页码）：2020·39·1（8-14）

摘要：

目的基于图像处理的舌质颜色分析是中医舌诊现代化的重要内容，提高舌色的正确识别率是其中的关键问题。本文使用集成学习的分类方法来探讨舌色分类问题，以达到客观，准确地识别中医舌色的目的。方法首先通过AdaBoost算法对舌图像进行初步分类，再将该算法与级联框架进行结合，然后通过“一对其余”的方法将AdaBoost从二分类扩展到多类来完成舌质颜色的分析。最后通过实验进行了验证，并与其他方法所得出的结果进行对比。结果针对各类舌质颜色分类问题，使用随机森林与传统的AdaBoost分类器进行分类的正确率分别在78.0%-90.2%与89.4%-95.5%区间，而基于AdaBoost级联框架的分类器的各类舌质分类正确率在93.0%-98.7%之间。结论基于AdaBoost级联框架的舌质颜色分类方法与其他经典方法相比，具有较高的正确分类率，为基于图像处理的中医舌诊辅助诊断奠定了一定的基础。

Objective The color analysis of tongue based on image processing is an important part of the automatic analysis of tongue images in Chinese medicine. It is the key to improve the correct recognition rate of tongue color. This paper uses the classification method of integrated learning to explore the problem of tongue color classification [no, nuclear, and corresponding English] to achieve objective and accurate identification of TCM tongue color.Methods Firstly, the tongue image is preliminarily classified by AdaBoost algorithm, and then the algorithm is combined with the cascade framework. AdaBoost can be extended from two categories to multiple classes by a "one pair of rest" approach. In this way, the analysis of the color of the tongue can be achieved.Finally, the experiment is carried out and the results obtained by other methods were compared. ResultsFor various tongue color classification problems, the accuracy of classification using random forest and traditional AdaBoost classifiers is in the range of 78.0% -90.2% and 89.4% -95.5%, respectively, and the types of classifiers based on the AdaBoost cascade framework The accuracy of tongue classification is between 93.0% -98.7%.Conclusions Compared with other classical methods, the tongue color classification method based on AdaBoost cascading framework has a higher correct classification rate, which lays a foundation for the diagnosis of TCM tongue diagnosis based on image processing.

参考文献：

[1]李经纬．中医大辞典［M］．2版.北京:人民卫生出版社，2005．

[2]杨新宇,梁嵘,王召平,等. 基于色度学的舌色分类研究现状与分析[J]. 北京中医药大学学报,2012,35(8):539-542+577.

[3]沈兰荪. 用于舌诊客观化的图像分析技术[C]//中国中西医结合学会.第二次世界中西医结合大会论文摘要集.北京:中国中西医结合学会,2002:1.

Shen LX. Image analysis technology for objectification of tongue diagnosis[C]// Chinese Integrative Medicine Association. Second World Congress of Integrated Traditional Chinese and Western Medicine Abstracts.Beijing: China Association of Integrative Medicine, 2002 :1.

[4]黄勃. 舌数字图像颜色计算机分析与分类[D].长沙：湖南中医药大学,2013.

Huang B. Computerized tongue color analysis and classification[D].Changsha：Hunan University of Chinese Medcine,2013.

[5]樊艳,梁嵘,王召平,等.数字舌图的色彩校正方法研究[J].世界科学技术,2005(4):34-36，85.

Fan Y, Liang W, Wang ZP, et al. Research on color correction method of digital tongue map[J]. World Science and Technology, 2005(4): 34-36，85.

[6]李博聪,黄庆梅,陈松鹤,等.基于CIELAB空间的中医舌色分析方法[J].世界科学技术-中医药现代化,2007(3):28-32，54.

Li BC, Huang QM, Chen SH, et al. TCM color analysis method based on CIELAB space[J]. World Science and Technology-Modernization of Traditional Chinese Medicine, 2007(3): 28-32，54.

[7]李敏,徐学军.基于数字化技术对糖尿病两虚证舌色的研究[J].电子制作,2016(11):80-81，83.

[8]张铮 ,徐超，任淑霞，等.数字图像处理与机器视觉:Visual C++与Matlab实现[M].北京：人民邮电出版社，2015.

[9]Viola P, Jones M. Rapid object detection using a boosted cascade of simple features[C]// Proc of IEEE Conference Computer Vision and Pattern Recognition.Cambridge:IEEE Computer Society,2001:905-910.

[10]Zhou ZH. Large margin distribution learning .artificial neural networks in pattern recognition[M]. Berlin:Springer International Publishing, 2014:1-11.

[11]陈伟民，张凌，宋冬梅，等.基于 AdaBoost改进随机森林的高光谱图像地物分类方法研究［J］．遥感技术与应用，2018，33(4)：612-620．

Chen WM,Zhang L,Song DM,et al.Classification method based on AdaBoost improved random frrest［J］．Remote Sensing Technology and Application,2018,33(4):612-620.

服务与反馈：

【文章下载】【加入收藏】

提示：您还未登录，请登录！点此登录