基于深度学习的佩戴口罩下的人脸识别
Masked Face Recognition Based on Deep Learning

深度学习卷积神经网络在图像处理中的应用引起了国内外许多学者的广泛关注。识别和验证有遮挡物下的人脸将是深度学习领域里持续受到关注的课题，我们需要更有效的方法来实现实时佩戴口罩检测和面部识别。从传统的机器学习算法到现在的深度学习卷积神经网络，图像识别效率、图像识别精度和网络训练速度的优化始终都是第一要义。为解决传统神经网络的梯度消失和网络退化问题，本文提到了一种基于改进型激活函数LeakyReLU的ResNet18残差神经网络的口罩遮挡下的人脸识别方法。利用Python语言构建PyTorch框架下的ResNet18残差神经网络模型，训练结果显示，改进型激活函数LeakyReLU在两轮训练后产生的结果比同等训练条件下ReLU函数的识别精确度高，因此，ResNet18卷积神经网络模型较其他人脸遮挡识别方法在识别准确度上有所提升。
The application of deep learning convolutional neural network in image processing has attracted wide attention of many scholars at home and abroad. Recognizing and verifying faces under occlusions will continue to be a hot topic in the field of deep learning. We need more effective methods for real-time mask wearing detection and face recognition. From the traditional machine learning algorithm to the current deep learning convolutional neural network, the optimization of image recognition efficiency, image recognition accuracy and network training speed is always the first essential. In order to solve the problem of gradient disappearance and network degradation of traditional neural network, this paper mentions a ResNet18 residual neural network based on improved activation function LeakyReLU for face recognition under mask occlusion. Python language is used to build a ResNet18 residual neural network model under the PyTorch framework. The training results show that the improved activation function LeakyReLU produces higher recognition accuracy than the ReLU function under the same training conditions after two rounds of training. The ResNet18 convolutional neural network model has improved the recognition accuracy compared with other face occlusion recognition methods.