Fine-grained image recognition method and device for high-order association discovery based on graph structure representation

Abstract

Embodiments of the present invention provide a high-order association discovery fine-grained image recognition method and device for graph structure representation, wherein the method includes: inputting the image to be classified into a convolutional neural network feature extractor with multiple stages, and extracting the final stage Two-layer network feature map, construct a mixed high-order attention module according to the network feature map, and form a high-order feature vector pool according to the mixed high-order attention module, use each vector in the vector pool as a node, and use high-order features Semantic similarity among groups, forming representative vector nodes by grouping, performing global pooling on representative vector nodes to obtain classification vectors, and based on classification vectors, fine-grained classification results can be obtained through fully connected layers and classifiers without relying on additional components The feature extraction network realizes the rapid and accurate extraction of image features, uses the information of the image itself to construct the correlation relationship between image features, obtains the fine representation of fine-grained features, and improves the image recognition effect.

Description

technical field [0001] The invention relates to the technical field of image processing, and in particular, to a method and device for fine-grained image recognition of high-order correlation discovery of graph structure characterization. Background technique [0002] Image fine-grained recognition is a technique for inductively classifying input images according to a large number of fine-grained classification categories and algorithmic recognition. This technology can be widely used in various fields such as Internet analysis, face verification, pedestrian recognition, and smart cities. [0003] At present, most fine-grained recognition of images tends to use regional features, or component-level representations, to enhance the recognition effect of images. However, such methods need to rely on additional component feature extraction networks, and the features extracted from components are prone to unstable representations, which affect the image recognition effect. SUM...

AI Technical Summary

Problems solved by technology

However, this type of method needs to rely on an additional component feature extraction network, and the features extracted by the component are prone to unstable representations, which affect the image recognition effect

Images