Expression label correction and identification method based on separable residual attention network

Abstract

The invention discloses an expression label correction and recognition method based on a separable residual attention network. The method comprises the following steps: 1, collecting facial expression data and preprocessing an MERC method; 2, establishing a separable residual attention-based facial expression feature extraction network, wherein the separable residual attention-based facial expression feature extraction network sequentially comprises a shallow network module, a separable residual attention module DSA and a weight output module; 3, using a label correction module LA to correct uncertainty expression sample labels with low weights; and 4, performing iterative training on the network by combining the self-attention weight cross entropy loss LSCE, the sorting regularization loss LRR and the category weight cross entropy loss LCCE. According to the method, interference removal can be carried out on facial expression data samples, label correction can be carried out on uncertain samples, the problem of class imbalance is solved, and finally, the deep separable residual attention module is used, so that the recognition precision of facial expressions can be improved while network parameters are reduced.

Description

technical field [0001] The invention relates to the classification and discrimination of convolutional neural network, depth separable network, residual network, attention module and final facial emotion calculation, belongs to the field of computer vision, and is specifically a kind of Lu Great way to recognize facial expressions. Background technique [0002] According to the research of psychologist A. Mehrabia, in the daily communication of human beings, the information transmitted through language only accounts for 7% of the total information, while the information transmitted through facial expressions reaches 55% of the total information, so We show our expressions to the outside world every day, and we are also receiving expressions from others. With the rapid development of society, more and more mental diseases appear, such as insomnia, anxiety, depression and so on. Non-contact facial expression analysis is playing an increasingly important role in daily life, h...

AI Technical Summary

Problems solved by technology

[0006] (3) Discrimination of unknown emotions

During network training, related research rarely considers the imbalance of different categories of samples and the problem of uncertain labels.

The above problems will seriously affect the learning effect of the network and the discrimination of unknown samples.

Images