中医舌象的高光谱特征提取与参数分类研究  

Study on hyperspectral feature extraction and parameter classification of traditional Chinese medicine tongue images

作　　者：张冬[1,2] 庞稳泰 王可仪 杨丰文 张俊华[1,2] ZHANG Dong;PANG Wentai;WANG Keyi;YANG Fengwen;ZHANG Junhua(Institute of Traditional Chinese Medicine,Tianjin University of Traditional Chinese Medicine,Tianjin 300193,China;Xin-Huangpu Joint Innovation Institute of Chinese Medicine,Guangzhou 510000,China)

机构地区：[1]天津中医药大学中医药研究院,天津300193 [2]广东省新黄埔中医药联合创新研究院,广州510000

出　　处：《中华中医药杂志》2025年第1期189-194,共6页China Journal of Traditional Chinese Medicine and Pharmacy

基　　金：广东省新黄埔中医药联合创新项目(No.2022IR024);国家自然科学基金青年科学基金项目(No.82205316)。

摘　　要：目的:采用高光谱图像对中医舌象进行特征提取与分类研究,扩展中医舌象采集形式与表达方式。方法:制定纳入排除标准,完成对中医舌象的高光谱图像采集与获取并对其进行数据清洗;提取舌质与舌苔区域的高光谱曲线特征进行可视化,对比其光谱特征差异;最后基于机器学习模型进行舌象参数的分类。结果:通过对228例中医舌象高光谱图像数据采集与特征提取,不同舌色、苔色、老嫩舌、润燥苔、腻苔间在高光谱特征上存在一定的差异。基于高光谱特征进行上述5种不同舌象参数的分类准确性分别达到了68.49%、79.37%、82.73%、83.46%、86.76%。结论:高光谱图像能够在连续波段范围内展示舌象的光谱信息,基于高光谱图像进行中医舌象参数分类具有较好的准确率,对中医舌诊客观化研究具有一定的意义。Objective:Using hyperspectral images for feature extraction and classification of traditional Chinese medicine tongue images,expanding the collection and expression methods of traditional Chinese medicine tongue images.Methods:Develop inclusion and exclusion criteria,complete the collection and acquisition of hyperspectral images of traditional Chinese medicine tongue images,and perform data cleaning on them;Extract hyperspectral curve features of tongue texture and tongue coating area for visualization,and compare their spectral feature differences;Finally,based on the machine learning model,classify tongue image parameters.Results:Through the collection and feature extraction of hyperspectral image data from 228cases of traditional Chinese medicine tongue images,there were certain differences in hyperspectral features among different tongue colors,fur colors,old and tender tongues,moist and dry fur,and greasy fur.The classification accuracy of the five different tongue image parameters based on hyperspectral features reached 68.49%,79.37%,82.73%,83.46%,86.76%,respectively.Conclusion:Hyperspectral images can display spectral information of tongue images within a continuous band range.Based on hyperspectral images,traditional Chinese medicine tongue parameter classification has good accuracy,and has certain significance for the objective research of traditional Chinese medicine tongue diagnosis.

关 键 词：中医舌诊 高光谱图像 特征提取 参数分类 机器学习 

分 类 号：R241.25[医药卫生—中医诊断学] TP391.41[医药卫生—中医临床基础] TP181[医药卫生—中医学]