Food Recognition Model Based on Deep Learning and Attention Mechanism

Since food culture and the Internet technology has developed, it is popular to share food photos through the Internet. How to mine the useful information contained in these food images has posed a challenge to us. Image-based food recognition technology has a broad application prospect. It can not only quickly identify food category, ingredients and cooking methods, providing people with relevant recipe information, but also predict food nutrition information, which can be used in nutritional analysis, scientific dietary matching and medical health management. Considering the above problems, in this paper we conduct research and analysis from two aspects: dataset construction and recognition model design. The main contributions of this paper are as follows: (1) Since there is an absence of public datasets which contain both food cooking methods and calorie information, we construct a food dataset with rich food attributes. (2) Existing food calorie prediction methods usually need to go through multiple calculation steps while ignoring the influence of cooking methods. In addition, the mutual occlusion of ingredients, the changes in shape, color and texture of ingredients after different cooking methods, and the similarity of different types of food in terms of shape and color, all make the food image recognition tasks hard to solve.To solve these problems, a food recognition model based on multi-task attention network is proposed.