Facial expression recognition based on deep convolutional neural network

In interpersonal communication, facial expressions serve as an important way for people to communicate with each other. Through small changes of the face, people can express a variety of emotions. However, the existing facial expression recognition technology has the disadvantages of low recognition rate, slow speed and poor generalization. Based on these problems, we propose a new facial expression recognition method which uses the convolutional network based on convolution block to realize the recognition of facial expression. Firstly, we expand the existing expression dataset to effectively improve the generalization performance of training samples and solver other issues such as single background; Secondly, for the convolution block in convolutional neural network, multi-layer small convolution kernels are mostly used instead of large convolution kernel. This not only reduces the parameters and improves the practical application of convenience, but also makes the network more sensitive to image details and more significant recognition effect. In this paper, we use three different methods to experiment on 12 expressions. The above method has the obvious advantage that the error rate of expression recognition is reduced to 13.7%. The experimental results show that the proposed method has a good recognition rate and training speed, which has a certain promotion effect and reference value for more accurate facial expression recognition in the future.