A Lightweight Network Facial Expression Recognition Method with Equalization Loss

Abstract

The invention claims to protect a light-weight network facial expression recognition method integrating balanced loss, belonging to the technical field of pattern recognition. It includes the following steps: first, the sample category loss function is proposed, and the weight is set by class_weight, and the sample category loss is integrated into the network training; secondly, the sample quality loss function is proposed, and the method of locating key points in the expression area is used to filter out the expression quality The good and bad image samples are integrated into the loss function through the influence of weights; then, using the network attention mechanism, a multi-dimensional attention loss function is designed, and the features formed by the two network attention mechanisms are used as discriminant labels and predicted values. The metric index, thereby improving the classification accuracy of the network model; finally, the three losses mentioned above are cascaded and fused in the Keras framework-based network model to form EQ‑loss, and added to the lightweight network framework, Implement end-to-end facial expression recognition.

Description

technical field [0001] The invention belongs to the technical field of computer pattern recognition, in particular, to a facial expression recognition method. Background technique [0002] As an important branch of face recognition, facial expression recognition technology takes into account a variety of subject knowledge, and has become a relatively new and promising research direction. Facial expression recognition technology has the characteristics of non-interference, low operating cost and strong interactivity, making it applicable to the fields of human-computer interaction, treatment of mentally ill patients, affective computing, and distance education. The previous research on facial expression recognition mainly focused on the static frontal face, but with the development of technology and the popularization of applications, the demand for facial expression recognition that changes under uncontrollable conditions is increasing. Due to the change of camera angle and...

AI Technical Summary

Problems solved by technology

At present, the convolutional neural network used for facial expression recognition tasks mainly includes the expression peak supervised network PPDN, the IL-CNN that expands the difference between classes through the Island loss function, the network FaceNet2ExpNet that passes the two stages of face authentication to expression recognition, and the use of human DAM-CNN, etc., which improve task accuracy by focusing on local areas of the face. These networks have achieved better accuracy by designing deep network structures and optimizing loss functions. However, due to the number of network layers, model complexity, and parameter quantities Constantly increasing, so that the calculation conditions are limited and cannot be applied to portable devices

Images