| 引用本文: |
李秋华, 史国峰, 李玥博, 任路.基于卷积神经网络的“舌边白涎”舌象识别研究[J].湖南中医药大学学报,2024,44(7):1254-1260[点击复制] |
|
| |
|
|
| 本文已被：浏览 418次 下载 228次 |
| 基于卷积神经网络的“舌边白涎”舌象识别研究 |
| 李秋华,史国峰,李玥博,任路 |
| (辽宁中医药大学, 辽宁 沈阳 110847;辽宁中医药大学附属第二医院, 辽宁 沈阳 110034) |
| 摘要: |
| 目的 通过机器学习分析“舌边白涎”舌象特性，对舌象进行局部特征识别研究，探讨卷积神经网络算法在舌象识别应用中的性能。方法 使用Python进行图像预处理，搭建用于舌象识别的视觉几何组16层（visual geometry group 16, VGG16）卷积神经网络模型，分析其对“舌边白涎”舌象鉴别分析的效果，并结合热力图分析“舌边白涎”典型舌象表现。结果 基于PyTorch框架，进行卷积神经网络的舌象鉴别研究，VGG16及残差网络50层（residual network 50, ResNet50）模型验证准确率均较高，达到80%以上，且ResNet50模型优于VGG16模型，可为舌象识别提供一定参考。基于加权梯度类激活映射（gradient-weighted class activation mapping, Grad-CAM）技术，通过舌苔舌色差异分布的网络可视化，有助于直观进行模型评估分析。结论 基于卷积神经网络模型对舌象数据库进行分析，实现“舌边白涎”舌象识别，有助于临床诊疗的客观化辅助分析，为舌诊智能化发展提供一定借鉴。 |
| 关键词: 卷积神经网络 视觉几何组 Python 人工智能 舌边白涎 |
| DOI：10.3969/j.issn.1674-070X.2024.07.016 |
| 投稿时间：2023-12-05 |
| 基金项目:中国博士后科学基金项目（2021MD703842）；国家中医药管理局中医药国际合作专项项目（0610-2140NF020629）；辽宁省自然科学基金计划重点项目（20180540043）。 |
|
| Tongue image recognition of “white saliva on the tongue side” based on convolutional neural network |
| LI Qiuhua, SHI Guofeng, LI Yuebo, REN Lu |
| (Liaoning University of Chinese Medicine, Shenyang, Liaoning 110847, China;The Second Hospital of Liaoning University of Chinese Medicine, Shenyang, Liaoning 110034, China) |
| Abstract: |
| Objective To study the local feature recognition of tongue image through machine-learning analysis on the characteristics of "white saliva on the tongue side", and to explore the performance of convolutional neural network algorithms in identifying tongue images application. Methods Python was used for image preprocessing, and visual geometry group 16 (VGG16) convolutional neural network model was built for tongue image recognition. The effect of the model on tongue image recognition of "white saliva on the tongue side" was identified and analyzed, and the typical tongue image performance of "white saliva on the tongue side" was analyzed combined with heat map. Results Based on PyTorch framework, tongue image identification research of convolutional neural network was carried out. The verification accuracy of VGG16 and residual network 50 (ResNet50) models were high, reaching over 80%, and the ResNet50 model outperformed the VGG16 model, providing a certain reference for tongue image recognition. Based on gradient-weighted class activation mapping (Grad-CAM) technology, the network visualization of the difference distribution of tongue coating and tongue color was helpful for intuitive model evaluation and analysis. Conclusion The analysis of tongue image database based on convolutional neural network model can realize the tongue image recognition of "white saliva on the tongue side", which is helpful for objectified auxiliary analysis in clinical diagnosis and treatment, and provides some references for the intelligent development of tongue diagnosis. |
| Key words: convolutional neural network visual geometry group Python artificial intelligence white saliva on the tongue side |
|
二维码（扫一下试试看！） |
|
|
|
|
