Semi-supervised face emotion recognition method

Abstract

The invention discloses a semi-supervised face emotion recognition method. The method comprises the following steps: acquiring a basic image data sample; predicting the label data, and calculating the distance between a predicted value and a label value as supervision loss; generating a smooth pseudo label from the label-free data, and taking the distance between the smooth pseudo label and the predicted value as semi-supervised loss; constructing triple loss based on similarity learning, including clustering depth features of image data, constructing triads, calculating similarity triple loss, constructing a complete loss function, and updating network parameters according to gradient descent; and obtaining a semi-supervised face emotion recognition model, and performing emotion analysis on the current face image data. According to the method, the recognition in the continuous domain data set is accurate and efficient, and meanwhile, the excellent recognition rate in the discrete domain data set is obtained by adjusting the tail end; according to the invention, through the triple loss function based on similarity learning, the similarity between face images can be learned, and the recognition rate of emotion changes is improved.

Description

technical field [0001] The invention belongs to the technical field of computers, and in particular relates to a semi-supervised facial emotion recognition method. Background technique [0002] Accurate recognition of human emotions can help artificial intelligence achieve human-computer interaction. At present, the mainstream methods of emotion recognition include feature-based methods and deep learning-based methods. The method of feature extraction is mainly based on the oriented gradient histogram and the local binary pattern, but these features are not suitable for the nonlinear structure of the data, and the accuracy is not high. Using deep learning techniques, emotion recognition based on facial expressions has been significantly improved. Deep learning technology requires a large amount of labeled data to train neural networks, which will result in a lot of manpower and material resources during the training process. At present, common datasets include discrete do...

AI Technical Summary

Problems solved by technology

Deep learning technology requires a large amount of labeled data to train the neural network, which will consume a lot of manpower and material resources in the training process

Common datasets currently include discrete domain datasets and continuous domain datasets. Among them, discrete domain datasets contain limited data and are difficult to express rich human emotions, while continuous domain dataset labels are expensive and difficult to train.

Images