Scholarlink

<p>With advancements in artificial intelligence, automatic food nutrition estimation methods have emerged. However, more effective food nutrition estimation methods must be developed because of their limited accuracy. Thus, in this study, we aimed to address the issue of low accuracy in nutrition assessment by developing a novel Depth Image Multi-scale Fusion Network (DIMF-Net). To validate its effectiveness, DIMF-Net was compared with the existing methods.&#xa0;DIMF-Net, integrated with dual attention mechanisms and a Multi-scale Feature Fusion Module, was used to estimate calories, mass, fat, carbohydrates (carb), and protein from food images. An ablation study was conducted to identify the significance of the input elements on the output. The mean values, mean absolute error (MAE), and percentage of mean absolute error (PMAE) were used as metrics to compare DIMF-Net with other existing methods to ensure a comprehensive assessment.&#xa0;The mean values of calories, mass, fat, carb, and proteins were closest to the ground truth, with no statistically significant differences observed. DIMF-Net outperformed existing methods by achieving the lowest MAE and PMAE.&#xa0;DIMF-Net was developed as an effective solution for food nutrition estimation. The results demonstrated its accuracy and reliability, outperforming the existing methods for dietary assessment.</p>

错误：搜索内容不能为空，请输入英文关键词
错误：关键词超出字数限制，请精简
高级检索

Depth image multi-scale fusion network: a novel approach for food nutrition estimation

  • 期刊论文

  • 发表日期：

    2025年11月28日

查看全文
  • Xiao Shan,
  • Masato Tagi,
  • Ruiqing Liu,
  • Takeshi Konishi,
  • Jun Hirose

摘要

With advancements in artificial intelligence, automatic food nutrition estimation methods have emerged. However, more effective food nutrition estimation methods must be developed because of their limited accuracy. Thus, in this study, we aimed to address the issue of low accuracy in nutrition assessment by developing a novel Depth Image Multi-scale Fusion Network (DIMF-Net). To validate its effectiveness, DIMF-Net was compared with the existing methods. DIMF-Net, integrated with dual attention mechanisms and a Multi-scale Feature Fusion Module, was used to estimate calories, mass, fat, carbohydrates (carb), and protein from food images. An ablation study was conducted to identify the significance of the input elements on the output. The mean values, mean absolute error (MAE), and percentage of mean absolute error (PMAE) were used as metrics to compare DIMF-Net with other existing methods to ensure a comprehensive assessment. The mean values of calories, mass, fat, carb, and proteins were closest to the ground truth, with no statistically significant differences observed. DIMF-Net outperformed existing methods by achieving the lowest MAE and PMAE. DIMF-Net was developed as an effective solution for food nutrition estimation. The results demonstrated its accuracy and reliability, outperforming the existing methods for dietary assessment.