Method for training food image classification model and image classification method

Abstract

The invention provides a method for training a food image classification model and an image classification method, and the method comprises the steps: respectively constructing a multi-scale food rawmaterial neural network and a multi-scale food category neural network, and carrying out the multi-scale division of a target image; respectively carrying out multi-scale fusion on each type of features of the target image, and fusing the three types of features after fusion; and inputting the fused features into a classifier based on the fused features for classification. Complementary fusion ofadvanced food semantic distribution and depth visual features is innovatively proposed, raw material attribute information is further fused with the advanced food semantic distribution and depth visual features, the problems of non-rigid structures and geometric deformation of food images are solved, and recognition of the food images is better facilitated. Moreover, through a multi-scale fusion mode, the defect that the food image does not have the spatial arrangement characteristic is overcome, and the classification performance is improved to the maximum extent.

Description

technical field [0001] The invention relates to the field of image recognition, in particular to food image classification. Background technique [0002] Food is the material basis of people's life, and good eating habits can prevent various chronic diseases (such as obesity, diabetes, etc.). Food image classification has a wide range of practical applications, such as smart bracelets to analyze your diet and nutrition, smart restaurant self-checkout, etc. [0003] However, there are also some difficulties in food image classification: (1) in real life, food images contain background information that has nothing to do with food; (2) food images in the same category may have obvious differences, but they are different from different categories (3) The food image does not have any unique spatial shape, nor does it have a unique appearance that changes with cooking methods, thus lacking a rigid structure. [0004] In order to solve the above problems, some work is based on Fa...

AI Technical Summary

Problems solved by technology

[0006] The purpose of the present invention is to solve the problem that the above prior art does not study the non-rigid structure of food images and does not consider its geometric deformation. Therefore, a food image classification model training method based on fusion of multi-scale and multi-view features and image Classification

Images