A method and system for fine-grained image classification

Abstract

The invention discloses a fine-grained image classification method and system. The method includes the following steps: Step S1, performing feature extraction on an input picture to obtain a feature map. Step S2, build a knowledge map by counting the correlation between category labels and attributes in the data set; Step S3, use the GGNN network to express the features of the constructed knowledge map, and iteratively update the knowledge map to obtain the feature representation of the knowledge map; Step S4, the The feature map extracted in step S1 is fused with the advanced knowledge obtained through the GGNN network in step S3, and the network classification is guided through the combination of advanced knowledge and feature map. The present invention performs fine-grained classification through knowledge guidance and embedding, making the network pay attention to the picture In the more discriminative regions, stronger classification features are learned, thereby improving the classification accuracy of the network.

Description

technical field [0001] The invention relates to the technical fields of CNN image classification, computer vision, etc., and in particular to a method and system for fine-grained image classification based on a knowledge-embedded feature learning network to process fine-grained image classification. Background technique [0002] The task of image classification occurs frequently in everyday life. It distinguishes different types of images according to the semantic information of the image, which is an important basic problem in computer vision and the basis of other high-level visual tasks such as image detection, image segmentation, object tracking, and behavior analysis. [0003] Convolutional neural networks in deep learning models have achieved very good results in the image domain in recent years. Because it directly takes image pixel information as input, it retains all the information of the input image to a great extent, extracts features and performs high-level abs...

AI Technical Summary

Problems solved by technology

However, the above conventional methods have the following disadvantages: the local dependence model requires a large number of annotations, which limits it to large data sets; the introduction of the visual attention network can only roughly locate the position that needs to be distinguished if it lacks supervision information

Images