HSIFoodIngr-64: A Dataset for Hyperspectral Food-Related Studies and a Benchmark Method on Food Ingredient Retrieval

Food-related issues have attracted increasing attention recently due to its various applications in our daily life, for example, restaurant service and dietary assessment, which are essential human health management understanding food characteristics. Many existing datasets proposed food-relevant tasks, e.g., detection segmentation. However, most of the labelled with only dish-level annotations lack detailed information corresponding ingredients. Hyperspectral imaging (HSI), can explore emissive characteristics different objects a long region spectral bands, is great potential ingredient analysis. In present work, new image dataset HSIFoodIngr-64 containing 3,389 pairs HSI RGB images 21 dish classes 64 categories was established. It first HSI-based dataset, all ingredient-level pixel-wise masks. Therefore, be applied food-centric tasks. Furthermore, this research benchmark method on retrieval task, consists two sub-networks called IPSN prediction sub-network ERSN edge refinement sub-network. Extensive experiments were conducted effectiveness verified. expected provide perspective analysis inspire more efforts food-related issues. We made public at: https://doi.org/10.7910/DVN/E7WDNQ.