An Improved Lightweight Method of Convolutional Neural Network for Face Recognition
Yankun Wang, Jingyan Huang, Wei Xiong, Peiqin Li · 2022 3rd International Conference on Computer Vision, Image and Deep Learning & International Conference on Computer Engineering and Applications (CVIDL & ICCEA) · 2022
With the continuous development of convolutional neural networks, more hardware resources are required. Lightweight convolutional neural networks show their own advantages in this aspect. In this paper, three kinds of lightweight convolutional neural networks with different layers are designed based on DenseNet’s fully connected network structure, and training tests are carried out on self-built face database. Test results show that, within limits to increase the depth of the network can strengthen the training speed of network convergence, 7 layers network parameters increased by 88% compared with 4 layers, but the training convergence rate is increased by 3 times, also proved that the whole connections of network structure on the lightweight convolution neural network is feasible, and provide the method to the lightweight convolution neural network performance improvement.