摘要
目的传统中药材种类繁多,面对日益增长的市场需求,亟需引入智能化的人工智能识别技术,以提升识别准确性和效率。本研究旨在基于不同YOLO神经网络,开发一款能够自动识别163种中药材的人工智能模型,并构建配套的手机端应用程序。方法2020年1月—2024年10月期间,收集了包含163种中药材图像的两个数据集,用于不同架构和规模的YOLO模型迁移学习与微调训练。通过在验证集和测试集上评估模型的准确率、灵敏度、特异度、精确度、ROC曲线下面积、F1分数等性能指标,筛选出表现最佳的模型。为提升模型的透明性与可解释性,应用梯度加权分类激活映射技术。最终,基于Streamlit框架将该模型开发为一款便捷的手机应用程序。结果本研究共纳入了276767张图像,开发了6种YOLO神经网络模型,包括v8n、v8s、v8m、v11n、v11s、v11m。经比较,YOLOv11s表现最佳,在内部验证集中准确率、灵敏度、特异度分别为98.91%、98.95%、99.99%。在外部测试集中,该模型准确率、灵敏度、特异度分别为98.68%、98.68%、99.99%,展现出良好的性能。所开发的智能手机应用可快速实时识别163种中药材,并直观显示预测结果及置信度排名。结论基于最新提出的YOLOv11s神经网络开发的人工智能模型及手机应用,能够快速而准确地识别163种中药材,为医师在中药材鉴别工作中提供有力支持,具有较好的应用前景。
Objective The variety of traditional Chinese medicinal ingredients is vast,and in response to the ever-increasing market demand,there is an urgent necessity to integrate intelligent artificial intelligence identification technology to enhance both accuracy and efficiency.Methods Between January 2020 and October 2024,two datasets containing images of 163 types of Chinese medicinal materials were collected.These datasets were used for transfer learning and fine-tuning with YOLO neural network models of varying architectures and sizes.The models’performance was evaluated on validation and test sets using metrics such as accuracy,sensitivity,specificity,precision,area under the ROC curve,and F1 scores.The bestperforming model was selected.To improve the model’s transparency and interpretability,the gradientweighted class activation mapping technique was employed.Finally,the model was integrated into a userfriendly mobile application using the Streamlit framework.Results A total of 276767 images were included in this study,and six YOLO neural network models were developed,namely YOLOv8n,v8s,v8m,v11n,v11s,and v11m.Among them,YOLOv11s performed the best,achieving an accuracy of 98.91%,sensitivity of 98.95%,and specificity of 99.99%on the internal validation set.On the external test set,the model achieved an accuracy of 98.68%,sensitivity of 98.68%,and specificity of 99.99%,demonstrating excellent performance.The developed smartphone application enabled rapid real-time recognition of 163 types of Chinese medicinal herbs,intuitively displaying prediction results and confidence rankings.Conclusions The AI model and mobile application developed using the latest YOLOv11s neural network enable fast and accurate identification of 163 types of Chinese medicinal materials.This provides strong support for physicians in Chinese medicinal material identification tasks and holds promising application prospects.
作者
戴一奇
张子豪
奚美娟
夏开建
王甘红
陈健
DAI Yiqi;ZHANG Zihao;XI Meijuan;XIA Kaijian;WANG Ganhong;CHEN Jian(Department of Gastroenterology,Changshu No.1 People’s Hospital,Suzhou,Jiangsu Province 215500;Shanghai Hao Brothers Educational Technology Co.,Ltd.,Shanghai 200434;Department of Gastroenterology,Changshu Traditional Chinese Medicine Hospital,Suzhou,Jiangsu Province215500;Center of Intelligent Medical Technology Research,Changshu No.1 People’s Hospital,Suzhou,Jiangsu Province 215500)
出处
《北京生物医学工程》
2026年第2期153-162,共10页
Beijing Biomedical Engineering
基金
苏州市第二十三批科技发展计划(临床试验机构能力提升)项目(SLT2023006)
苏州市科技攻关计划(医疗卫生创新)项目(SYW2025034)资助。
