Human body image multi-attribute classification method based on priori prototype attention mechanism

Abstract

The invention discloses a human body image multi-attribute classification method based on a priori prototype attention mechanism, and belongs to the technical field of image processing. According to the scheme, the method comprises the steps that firstly, an attribute table and a corresponding human body image data set are constructed; and then constructing a multi-attribute classification neuralnetwork model: adding a priori prototype attention mechanism plug-in to the tail part of a conventional multi-classification neural network model, and changing the tail part of the multi-classification neural network into a multi-attribute classification network; training the constructed neural network model; and finally, performing multi-attribute classification recognition on the human body image based on the trained neural network model. According to the method, on one hand, the method of a traditional attention mechanism is reserved, point-by-point multiplication is carried out through thegenerated attention graph and the last convolution feature, and therefore the strong filtering performance of the traditional attention mechanism is reserved; and on the other hand, the centrality ofthe attention graphs is enhanced through a prior prototype attention graph linear combination mode. Therefore, the generalization ability of the model is greatly improved.

Description

Technical field[0001]The invention belongs to the technical field of human body image attribute classification, and specifically relates to a human body image multi-attribute classification method based on a priori prototype attention mechanism.Background technique[0002]The performance of multi-classification tasks on the imagenet dataset, a visualization database used for visual object recognition, is getting better and better. On this dataset, the classification ability of the network model has surpassed the classification ability of people. In contrast, the performance of the existing human image multi-attribute classification task is not very satisfactory. This task is different from the ordinary multi-class classification task. The difficulty is that the input is the entire picture of the person, but some analysis is required. Only the attributes of the local area are needed, and there is no local area information about the attributes on the existing data sets, which will cause...

AI Technical Summary

Problems solved by technology

Compared with the traditional attention mechanism, this method does not directly apply the generated attention map to the last convolution feature, resulting in a weaker ability to remove irrelevant features.

[0005] That is, the traditional attention mechanism, although the attention map can be directly used to filter out irrelevant features, but the attention map generated in this way is often not concentrated

Although the CAM-based attention mechanism can get a relatively concentrated CAM, this method does not have an attention map to help the network model filter out irrelevant features.

Images