Automatic micro-expression identification method of macro-to-micro conversion model based on depth learning

Abstract

The invention provides an automatic micro-expression identification method of a macro-to-micro conversion model based on depth learning. The method comprises steps that A, micro-expression sample processing, 1), a micro-expression data set sample and a macro-expression data set sample are pre-processed; 2), sample pairs of a cross-modal tuple loss function are constructed; B, cross-modal macro-to-micro conversion model training, 3), the AU detection network is trained, an AU detection network parameter is initialized, and a flexible maximum value loss function is trained; 4), the AU detection network parameter is fixed, a cross-modal macro-to-micro conversion model parameter is initialized, and the cross-modal macro-to-micro conversion model is trained; and C, micro-expression identification, according to a trained convolutional neural network model, a test parameter is initialized, a test sample is inputted to the trained convolutional neural network model, and an identification rate is outputted after forward network propagation. The method is advantaged in that robustness is realized compared with methods in the prior art.

Description

technical field [0001] The invention relates to a micro-expression automatic recognition method based on a 'macro-to-micro conversion model' based on deep learning, and belongs to the technical field of deep learning and pattern recognition. Background technique [0002] Micro-expressions express the true emotions that people try to cover up and hide. They are a set of time-continuous image sequences, and the duration is generally between 250ms and 500ms. Research on micro-expressions can help reveal people's psychological changes in characteristic scenes, for example, Expose prisoners' lies, evaluate people's inner emotional state, and then promote the development of criminology and psychology. Compared with facial expressions, micro-expression recognition is more challenging. First, unlike expressions, micro-expressions use a sequence of images to jointly represent an emotional label, but the duration is short (usually less than 500ms). If a 60-frame camera is used to rec...

AI Technical Summary

Problems solved by technology

In order to solve the problem of micro-expression recognition, convolutional neural network seems to be a good tool, however, convolutional neural network needs to be trained with a higher number of data sets in order to obtain representative features, otherwise it will cause excessive Fitting and other issues, and the micro-expression data set has less than about 1000 samples, which is not enough for convolutional neural network training

Images