Food Chem | in depth learning accurately predicts food categories and nutritional components based on ingredient statements

Introduce an article by Peihua Ma And other people in 2022 year 5 Published in June Food Chemistry Articles on . The author of this paper proposes a general technology based on deep learning to predict food types and nutrients .

Thesis link ：

https://www.sciencedirect.com/science/article/abs/pii/S0308814622012055

Abstract

Determine such as food classification 、 Creating attributes such as taxonomies and food nutrients can be a challenging and resource intensive task , Although it is important to better understand food . In this study , from USDA A new data set has been collected in the brand food database 134 k BFPD, And modified it , And three kinds of food classification and nutritional value were marked , And became an artificial intelligence (AI) Data sets , It covers the largest food type so far . Overall speaking , Multilayer perceptron (MLP)-TF-SE Method in use AI The highest learning efficiency was achieved in the food natural language processing task , The accuracy of food classification is as high as 99%, The estimation of calcium reaches 0.98 R2（0.93 ∼ 0.97 for calories 、 protein 、 sodium 、 Total carbohydrates 、 Total lipids, etc ）. Deep learning methods have great potential , It can be embedded into other food classification and regression tasks , And as an extension of other applications in the field of food and Nutrition .

chart 1. Food deep learning

（a） input data - Ingredient list and nutritional information from BFPD get .（b1） Use natural language processing （b2） Word parsing of component list by data coding , Combine ingredients 、 The category and nutrition information strings are converted into a matrix that the machine can learn .（c） Data sampling consists of two steps , Data rebalancing and splitting , And transform it into data tensor .（d） Learn more . The labeled tensor is used to train the deep learning model to achieve the goal .（e） The deep learning model can be applied in different fields , Such as food marketing analysis 、 Novel food design and personalized dietary suggestions .