A food image recognition method based on deep learning

Abstract

The invention provides a deep learning-based food image identifying method. The method comprises the following steps that a food image database, a deep learning network and a classifier are adopted; images of the food image database are input to the deep learning network; after calculation layer by layer, representative characteristics, including edge characteristic combinations of the images, the basic shape characteristic combinations of the images and the color characteristic combinations of the images, are output; the classifier classifies the images by using the characteristic combinations. The method fully takes the self-studying advantage of deep learning; label-free image data can be used to perform non-supervision study; when an image is input, the characteristics can be quickly and accurately extracted, abstract is performed layer by layer until the concept of some food is formed, and foods are classified by the classifier; for the acquired images shot from any angle, obtained local characteristics are the same substantially, and the problem on shooting angle limitation is solved.

Description

technical field [0001] The invention relates to the field of artificial intelligence, in particular to a food image recognition method based on deep learning. Background technique [0002] S.Ysng et al. proposed a system for identifying fast food, which can identify fast food in KFC and McDonald's. However, due to the diversity of fast food, the accuracy of the system is not high. In addition, there are also specific requirements for the camera device and image position, and the operation is relatively cumbersome. [0003] Face recognition based on deep learning: Face recognition using deep learning can be applied to access control systems, security inspection systems, emotion recognition, etc., but due to the particularity of the application, it is difficult to promote it in daily life. [0004] Moreover, the above-mentioned traditional image recognition methods only extract some representative features of the image, such as SIFT and SURF, which have certain limitations, a...

AI Technical Summary

Problems solved by technology

[0005] In order to solve the shortcomings and deficiencies in the prior art, the present invention proposes a food image recognition method based on deep learning, which combines deep learning with food image recognition, adopts a layer-by-layer initialization training method, and gives full play to the self-learning ability of deep learning. Advantages, can effectively solve the problems of slow training, difficult training, and difficult adjustment of parameters

Images