Convolutional neural network based on multi-channel feature fusion and dynamic sample weights
Xiaoyue Cheng, Longzhang Zhao, Zhichao He, Jiapeng Shi · 2019
In order to solve the problem of insufficient feature extraction and slow convergence in face recognition, a convolutional neural network face recognition algorithm based on multi-channel feature fusion and dynamic sample weights is designed. Firstly, the theoretical analysis of the shortcomings of traditional CNN in face recognition is carried out. Then multi-channel feature fusion and dynamic sample weight theory are designed. The network structure of multi-channel feature fusion and dynamic sample weight CNN is determined through multiple experiments. In addition, the above algorithm is used to identify the partial occlusion and rotation test set. In the end, the method designed in this paper is compared with other methods for face recognition rate and convergence speed. It is finally proved that compared with other face recognition methods, the method designed in this paper not only has higher recognition rate and faster. Convergence speed, and strong robustness in image occlusion and rotation.