A lightweight face recognition method based on depthwise separable convolution and triplet loss

At present, there are two main challenges for large-scale face recognition based on deep learning. One is to design an appropriate loss function to enhance the discrimination ability. The other is that the deployment environment of face recognition often has problems such as low performance and high real-time requirements. In order to solve the above two problems, this paper designs a method combine L2 loss and triplet loss to form a loss function, and uses the Depthwise Separable Convolutions to improve the face residual network so that reduce the amount of network parameters. These methods can effectively enhance the robustness of the system, and reduce the amount of network parameters to improve real-time performance and reduce performance requirements. The experimental results show that the method used in this experiment has a recognition rate of 99.36% on the LFW standard test set, and the amount of parameters of the improved network model is about 45% less than that of the face residual network.