计算机应用

• 人工智能与仿真 •    下一篇

基于深度残差网络的轻量级人脸识别方法

贺怀清,闫建青,惠康华   

  1. 中国民航大学 计算机科学与技术学院,天津 300300
  • 收稿日期:2021-05-27 修回日期:2021-09-03 发布日期:2021-09-03 出版日期:2021-09-15
  • 通讯作者: 闫建青

Lightweight face recognition method based on deep residual network

  • Received:2021-05-27 Revised:2021-09-03 Online:2021-09-03 Published:2021-09-15

摘要: 针对深度残差网络在小型移动设备的人脸识别应用中存在网络结构复杂、时间开销大等问题,提出一种基于深度残差网络的轻量级模型。首先通过对深度残差网络的结构进行精简优化,结合知识转移方法,从深度残差网络(教师网络)中重构出轻量级残差网络(学生网络),在保证精度的同时,降低网络结构复杂度;然后在学生网络中通过分解标准卷积减少模型的参数,降低特征提取网络的时间复杂度。实验结果表明,在LFW、VGG-Face、AgeDB和CFP-FP等四个不同数据集上,提出模型的识别精度接近主流人脸识别方法的同时,单张推理速度达到16ms,速度提升了10%-20%。所提模型在推理速度能够得到有效的提升的同时识别精度基本不下降。

关键词: 深度残差网络, 人脸识别, 轻量级, 知识蒸馏, 深度可分离卷积

Abstract: Aiming at the problems that deep residual network had complex network structure and high time cost of face recognition applications of small mobile devices, a lightweight model based on deep residual network was proposed. Firstly, by simplifying the structure of the deep residual network and combining the knowledge transfer method, a lightweight residual network (student network) was reconstructed from the deep residual network (teacher network), which reduced the network structural complexity while ensuring accuracy. Then, in the student network, the parameters of the model were reduced by decomposing standard convolution, which reduced the time complexity of the feature extraction network. The experimental results show that on four different datasets such as LFW(Labeled Faces in the Wild), VGG-Face(Visual Geometry Group Face), AgeDB(Age Database)and CFP-FP(Celebrities in Frontal Profile with Frontal-Profile), the recognition accuracy of the proposed model is close to mainstream face recognition methods, while the speed of reasoning reaches 16ms every image, and the speed is increased by 10%-20%. The speed of reasoning with the proposed model can be effectively improved while the recognition accuracy is basically not reduced.

Key words: deep residual network, face recognition, lightweight , knowledge distillation, depthwise separable convolution

中图分类号: