Scholarlink

<p>Accurate calorie estimation from food images is a critical yet challenging task in automated dietary monitoring systems. This paper presents a synergistic hybrid approach for calorie estimation using multi-view food images, integrating enhanced segmentation, classification, and portion size estimation through deep learning techniques. The proposed framework combines the Enhanced Squirrel Search Algorithm (ESSA) for optimized segmentation and a Convolutional Neural Network (CNN) for food classification and feature extraction. To improve portion size accuracy, a grid-based superimposition and perspective transformation technique is applied on both top and side-view images, enabling scale factor computation and volumetric estimation. The system is evaluated on benchmark datasets UEC-Food101 and Nutrition5k, achieving a mean relative error of 3.95% in calorie prediction. This work addresses critical limitations of single-view and classification-only models by incorporating multi-angle geometry and feature synergy. The results demonstrate the potential of the proposed system for real-world dietary assessment applications, offering improved accuracy, scalability, and robustness across diverse food types.</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

A synergistic hybrid approach for calorie estimation using multi-view food images and deep learning-based segmentation and classification

  • 期刊论文

  • 发表日期：

    2026年2月6日

查看全文
  • Megha Chopra,
  • Archana Purwar

摘要

Accurate calorie estimation from food images is a critical yet challenging task in automated dietary monitoring systems. This paper presents a synergistic hybrid approach for calorie estimation using multi-view food images, integrating enhanced segmentation, classification, and portion size estimation through deep learning techniques. The proposed framework combines the Enhanced Squirrel Search Algorithm (ESSA) for optimized segmentation and a Convolutional Neural Network (CNN) for food classification and feature extraction. To improve portion size accuracy, a grid-based superimposition and perspective transformation technique is applied on both top and side-view images, enabling scale factor computation and volumetric estimation. The system is evaluated on benchmark datasets UEC-Food101 and Nutrition5k, achieving a mean relative error of 3.95% in calorie prediction. This work addresses critical limitations of single-view and classification-only models by incorporating multi-angle geometry and feature synergy. The results demonstrate the potential of the proposed system for real-world dietary assessment applications, offering improved accuracy, scalability, and robustness across diverse food types.