Research On Facial Expression Recognition Algorithm Based On Deep Learning

With the increasing artificial intelligence interaction in various fields,the demand for artificial intelligence has drawn increasing attention.In daily communication and interaction,reading and recognizing facial expressions plays a key role,which is also the basis of artificial intelligence interaction.In this developing era,the application of human-computer interaction is very extensive,such as the monitoring of human fatigue driving in the process of driving,or special psychological counseling.In the past studies,people used deep learning network,cascade network and other methods to build networks to predict emotions,and achieved a series of ideal results.Based on the idea of deep learning,this paper further explores and improves the shortcomings of existing facial expression recognition algorithms.The main contents and research results of this paper are as follows:Aiming at the problem that the convolutional neural network training process needs a large amount of data,the scale-invariant feature transformation algorithm is introduced.The unique feature of this algorithm is that it does not need a large number of data sets to extract important features and obtain ideal results,so a small sample expression recognition algorithm integrating different features under deep learning is proposed.Thought is the convolution of the proposed algorithm neural network model and scale invariant feature transform model,first of all to a single convolution neural networks are compared,and explore the selection of appropriate convolutional neural network model,and then by using the scale invariant feature transform model to extract features,finally to deep learning model and scale invariant feature transform the algorithm fully takes advantage of the scale invariant feature transform of local characteristics,at the same time combines the automatic features of deep learning,increase the input characteristics of diversification.In order to further improve the accuracy in facial expression recognition,thispaper proposes a method,by combining automatic characteristics of convolution neural network learning and manual characteristics of visual word bag model,so as to improve the recognition rate of facial expression recognition first,adopt VGG13 convolutional neural network model,obtain the convolutional neural network automatically characteristics and visual word bag model was used to extract manually after the fusion of these two types of features,due to the introduction of local learning method,to the requirement of feature selection is higher,therefore,adopt the method of cosine measure,finally using support vector machine classifier method to predict the class label of each image.For facial expression recognition,the parameters of the model and the problem of large amount of calculation by introducing depth separable convolution,puts forward the model of facial expression recognition based on depth of separable convolution first build depth separable convolution model,compared with the traditional convolution can largely reduce the parameters and the amount of calculation,the second will be the whole connection layer,switch to the global average pooling layer instead of,fully solved due to the connection layer parameters increase abruptly,lead to the parameters of the fitting problem of too much of the algorithm than the algorithm of similar parameters,such as reduced nearly twenty times.