基于视觉的腰背部中医通络机器人穴位定位方法  

A Vision Based Acupoint Localization Method for Traditional Chinese Medicine Tongluo Robot in the Lumbar and Back

作　　者：费红琳 黄理杰 陆东海 郭程湘 杨仁轩[3] FEI Honglin;HUANG Lijie;LU Donghai;GUO Chengxiang;YANG Renxuan(Guangdong Guoshi Medical and Health Research Institute,Guangzhou 510530,China;Guangdong New Huangpu Traditional Chinese Medicine Joint Innovation Research Institute,Guangzhou 510530,China;Guangdong Provincial Hospital of Traditional Chinese Medicine,Guangzhou 510120,China)

机构地区：[1]广东省郭氏医学保健研究所,广东广州510530 [2]广东省新黄埔中医药联合创新研究院,广东广州510530 [3]广东省中医院,广东广州510120

出　　处：《现代中医药》2023年第5期24-30,共7页Modern Chinese Medicine

基　　金：广东省新黄埔中医药联合创新研究院联合创新研究项目(2021IR003);中医畅气通络“智能佛手”机器人-广州高新技术产业开发区。

摘　　要：目的加快中医药守正创新、传承发展,打破传统中医药传承方式方法的局限性。方法以全国名老中医药专家的中医畅气通络特色疗法技术为切入,将该中医成果转化为中医通络机器人,其中包含了机器人力控技术、3D视觉识别技术、人工智能辅助定位技术、语音交互技术等,循经取穴是中医成果转化中非常核心的技术,本项目使用视觉相机作为工具,进行训练样本采样、深度学习模型训练得到智能中医穴位定位模型。并根据训练完成的模型进行样本的预测,实现对未知的人体数据进行穴位识别。其间对多种人工智能模型精度进行了系统的评估,包括基于位置回归的RESNET、基于热图回归的UNET以及基于注意力机制的DMNET网络。结果基于位置回归的RESNET方案鲁棒性很高,能够保留不同穴位之间的结构相关性。最终使用基于RESNET的网络结构进行最终的模型部署实验,评估结果在不同的穴位上存在一定差异。结论应用该方法的中医畅气通络机器人可实现名老中医的部分特色手法。Objective To accelerate the innovation,inheritance and development of traditional Chinese medicine,and break the limitations of traditional Chinese medicine inheritance methods.Methods The method of this project is derived from the Changqi Tongluo characteristic therapy technology of traditional Chinese medicine,which is developed by renowned experts in traditional Chinese medicine in China.The traditional Chinese medicine achievement is transformed into a traditional Chinese medicine Tongluo robot,which includes robot force control technology,3D visual recognition technology,artificial intelligence assisted positioning technology,voice interaction technology,etc.Acupoint selection along the meridian is a very core technology in the transformation of traditional Chinese medicine achievements.This project uses a visual camera as a tool to conduct training sampling,deep learning model is trained to obtain an intelligent traditional Chinese medicine acupoint localization model.And predict samples based on the trained model to achieve acupoint recognition of unknown human data.During this period,a systematic evaluation was conducted on the accuracy of various artificial intelligence models,including RESNET based on position regression,UNET based on heat map regression,and DMNET network based on attention mechanism.Results The RESNET scheme based on position regression has high robustness and can preserve the structural correlation between different acupoints.The final model deployment experiment was conducted using a RESNET based network structure,and the evaluation results showed certain differences in different acupoints.Conclusion The application of this method in the traditional Chinese medicine Changqi Tongluo robot can achieve some unique techniques of renowned traditional Chinese medicine.

关 键 词：畅气通络疗法 通络机器人 穴位定位 中医按摩机器人 中医特色疗法 

分 类 号：R245-33[医药卫生—针灸推拿学]