Scholarlink

<p>The intelligent management and reasonable nutrition of food depend on the continuous improvement of cultivar discrimination technology. A model for food classification and detection based on image deep learning is proposed, which is used to discriminate hundreds of commonly used foods such as fruits, vegetables, and meat. We use smartphones to collect food images and form a dataset under uncontrolled lighting and camera parameters such as focal length and camera stability. It also discriminates partially occluded objects with similar features and achieves good results. Testing on a detection dataset yielded a mAP value of 94.35% and an inference time of 27.53&#xa0;ms, with a slight increase in parameter scale. Further enhancements in feature fusion and loss function testing showed that IDNet, integrating both BiFPN and WIOU, achieved the best overall performance with a mAP value of 96.24% and an inference time of only 29.86&#xa0;ms. Experiments are conducted using diverse foods placed in the refrigerator, confirming its superiority.</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

Image-Based Deep Learning Model for Food Classification and Detection

  • 期刊论文

  • 发表日期：

    2026年2月20日

查看全文
  • Yi Huang,
  • Aizaz Ali Shah,
  • Shanglong Xu,
  • Chao Wu,
  • Sheng Yan

摘要

The intelligent management and reasonable nutrition of food depend on the continuous improvement of cultivar discrimination technology. A model for food classification and detection based on image deep learning is proposed, which is used to discriminate hundreds of commonly used foods such as fruits, vegetables, and meat. We use smartphones to collect food images and form a dataset under uncontrolled lighting and camera parameters such as focal length and camera stability. It also discriminates partially occluded objects with similar features and achieves good results. Testing on a detection dataset yielded a mAP value of 94.35% and an inference time of 27.53 ms, with a slight increase in parameter scale. Further enhancements in feature fusion and loss function testing showed that IDNet, integrating both BiFPN and WIOU, achieved the best overall performance with a mAP value of 96.24% and an inference time of only 29.86 ms. Experiments are conducted using diverse foods placed in the refrigerator, confirming its superiority.