Research On Face Expression Recognition Based On Deep Learning

Facial expression recognition technology plays a crucial role in the field of humancomputer interaction,fundamentally changing the relationship between humans and computers,and providing better services for humans.The traditional manual process of extracting facial expression features is too complex,consumes a lot of manpower and resources,and the extraction of facial expression features is insufficient,resulting in poor recognition results.Deep learning models can perfectly extract feature information and effectively enhance the model’s generalization ability and robustness.However,the existing convolutional neural networks have deeper layers and larger model parameters,which are not conducive to practical applications.In order to solve the above problems,this article has conducted the following research.In order to achieve high accuracy in facial expression recognition under small parameter conditions,a facial expression recognition algorithm based on small convolutional kernels is proposed.The entire neural network uses a 3 × 3 small-scale convolutional kernel to extract facial expression features,and then completes feature classification through Soft Max.This algorithm fully utilizes the advantages of small convolutional kernels with low computational complexity and fewer parameters.The experimental results on the public dataset FER2013 and CK+datasets show that this algorithm can achieve accuracy of 64.57% and 96.43% with a small number of parameters.In order to further improve network performance and optimize network structure based on the facial expression recognition algorithm based on small convolutional kernels,a facial expression recognition algorithm based on optimized Res Net is proposed.The algorithm utilizes two small-scale convolution kernels and four improved residual modules to extract facial expression features.The improved residual module introduces deep separable convolutions to replace standard convolutions,reducing the number of parameters.The tail of the residual module also incorporates a channel attention mechanism to learn channel information and improve recognition accuracy.The experimental results on the FER2013 dataset and the CK+dataset show that the algorithm’s facial expression recognition accuracy is 70.58% and 99.28%,respectively.Compared to facial expression recognition algorithms based on small convolutional kernels,there is a significant improvement.In order to solve the problem of slow training Rate of convergence and improve the accuracy of the model,a facial expression recognition algorithm based on Transfer learning is proposed.The algorithm is divided into two parts: the pre training task and the facial expression recognition task.The pre training task is to use the MS-Celeb-1M dataset to complete the face recognition task in the facial expression recognition network based on optimized Res Net,and then use the knowledge learned from the task as the initial point of the facial expression recognition task,and train the network again to achieve better classification effect and Rate of convergence.The experimental results on the CK+dataset show that the method converged with only 15 epochs and achieved a high accuracy of99.59%.