With the continuous development of deep learning, convolutional neural networks have achieved good results in the field of face recognition. However, deep convolutional neural networks have difficulty in convergence and optimization during the training process. The emergence of residual networks alleviates these problems. In addition, the channel attention mechanisms can help networks to selectively learn features that contain useful information, which can enhance the expressive ability of the network. In this paper, we first use the mish function to improve the original residual network to obtain the improved residual network named RESNET_IR, and then the CAM which is a kind of the channel attention mechanisms is introduced into the RESNET_IR to obtain the final network model named CAMRESNET_IR, making the extracted facial features more discriminative. The experimental results on LFW, CFP-FP, and AgeDB-30 show that our model can improve the performance of face recognition and maintain better results when the illumination, posture, and age change.
展开▼