文献类型：期刊文献

中文题名：基于二级分类器的齿痕舌分类模型研究

英文题名：Research on classification model of teeth-printed tongue based on two level classifier

作者：颜建军[1];李东旭[1];郭睿[2];燕海霞[2];王忆勤[2]

机构：[1]华东理工大学,机械与动力工程学院智能感知与诊断研究所,上海200237;[2]上海中医药大学四诊信息综合实验室,上海201203

年份：2022

卷号：37

期号：4

起止页码：2181

中文期刊名：中华中医药杂志

外文期刊名：China Journal of Traditional Chinese Medicine and Pharmacy

收录：CSTPCD;;北大核心:【北大核心2020】；CSCD:【CSCD2021_2022】；

基金：国家自然科学基金面上项目(No.81673880)。

语种：中文

中文关键词：二级分类器;舌诊客观化;齿痕;机器学习;深度学习;凸包算法;随机森林

外文关键词：Two-level classifier;Objectification of tongue diagnosis;Teeth-print;Machine learning;Deep learning;Convex hull algorithm;Random forest

摘要：目的:解决目前齿痕舌的识别方法上存在效率低、准确率不高的问题。方法:本研究提出了一种基于二级分类器的齿痕舌分类模型。首先准备齿痕舌和非齿痕舌两类舌图像样本,利用深度学习分割模型对舌体进行分割;再利用凸包算法提取齿痕舌图像中的每个齿痕的几何形状特征,训练齿痕识别模型;然后利用该模型预测舌体凸包凹缺陷的齿痕候选区域,构建齿痕舌识别的特征向量,基于随机森林建立齿痕舌分类模型;最后进行模型分类测试和结果评估。结果:采用该方法得到的总体分类准确率达到93%。结论:该方法取得了较好的齿痕舌分类效果,为齿痕舌的识别研究提供了一种新的思路,对舌诊客观化和现代化具有一定借鉴意义和实用价值。
Objective:To solve the problem of the current method for identifying teeth-printed tongue has low efficiency and low accuracy.Methods:Model was established by a classification of teeth-printed tongue based on two-level classifier.First,prepare two kinds of tongue image samples of teeth-printed tongue and non-dented tongue,and use the deep learning segmentation model to segment the tongue;use the convex hull algorithm to extract the geometric characteristics of each teeth-print in the teethprinted tongue image.Train the teeth-print recognition model;then use the model to predict the teeth-print candidate area of the tongue convex hull defect,construct the feature vector of the teeth-printed tongue recognition,build the teeth-printed tongue classification model based on the random forest;finally carry out the model classification test and result evaluation.Results:The overall classification accuracy rate obtained by this method reached 93%.Conclusion:This method has achieved a good classification effect of indented tongue,provides a new idea for the identification of indented tongue,and has certain reference significance and practical value for the objectification and modernization of tongue diagnosis.