摘要
目的基于移动终端拍摄的食物图像对慢性病患者的日常饮食进行智能营养评估。方法构建基于人工智能的膳食评估系统,利用深度学习技术与图像处理方法,实现食物图像的智能分割、识别与营养素估算,使慢性病患者仅依据智能手机拍摄的食品图像即可得到食物的营养素信息。该系统同时支持172类中餐食谱与353种食材的细粒度识别,并在Vireo Food^(-1)72食谱数据集上得到了验证。结果基于卷积神经网络模型的食谱预测准确率为89.72%,食材评估指标微平均(micro-averaging,Micro-F1)提升至79.06%,宏平均(macro-averaging,Macro-F1)提升至64.28%,在Vireo Food^(-1)72食谱数据集上取得了目前食材分类的最佳性能;基于食谱与食材识别结果对食物营养素进行估计,估计值与参考值误差均处于合理的范围内。结论本系统可实现针对慢性病人群的智能膳食评估,便于患者进行每日饮食的自我监督,且有助于辅助营养师完成患者的日常饮食记录与评估,具有实用价值与研究意义。
Objective To built a diet evaluation system based on artificial intelligence to evaluate the daily diet of patients with chronic diseases.Methods Deep learning technology and traditional image processing method were used to realize intelligent segmentation,recognition and nutrient estimation of food image,so that patients with chronic diseases could obtain the nutrient information of food directly with only the food images taken by smart phones.The system also supported the fine-grained recognition of 172 Chinese food recipes and 353 food ingredients which had been verified in the food dataset Vireo Food^(-1)72.Results The predictive accuracy of the recipes based on the convolutional neural network model was 89.72%,the micro-averaging(Micro-F1)and macro-averaging(Macro-F1)of ingredients improved to 79.06%and 64.28%respectively.The state-of-the-art performance of ingredients recognition was achieved on food dataset Vireo Food^(-1)72;The food nutrients were estimated based on the results of recipe classification and ingredients recognition,and the error between the estimated values and the reference values was within a reasonable range.Conclusion The system could realize intelligent dietary assessment for patients with chronic diseases,which could facilitate patients to self-supervise their daily diet and assist nutritionists to complete the daily diet record and assessment.It had practical value and research significance.
作者
马兰芳
薛怡蓉
Ma Lanfang;Xue Yirong(Dean s office,Beijing University of Posts and Telecommunications Hospital,Beijing 100876,China;School of Artificial Intelligence,Beijing University of Posts and Telecommunications,Beijing 100876,China)
出处
《首都医科大学学报》
CAS
北大核心
2021年第2期262-268,共7页
Journal of Capital Medical University
基金
中央高校基本科研业务费专项资金资助(2020XD-A02-1)。
关键词
慢性病
食物图像
深度学习
超像素分割
图像分类
营养估计
chronic diseases
food image
deep learning
super-pixel segmentation
image classification
nutrition estimation
