【摘要】 利用ResNet残差卷积神经网络在图像分类上的研究成果,创新性地调整了网络结构及参数初始化方法,使得残差神经网络在不同的人脸库上都得到了较高的准确率.考虑到人脸识别的即时性和准确性,提出了一种相对较浅层次的网络结构,使得其可以有很好的应用.同时,为整个网络搭建了一个仿真平台,便于直观地观察识别结果.更多还原

【Abstract】 Based on the results of ResNet convolution neural network researches in image classification,the network structure and parameter initialization method are innovatively adjusted,which makes the residual neural network get higher accuracy in different face databases.Considering the instantaneity and accuracy of face recognition,this paper proposes a relatively shallow network structure,which makes it possible to have a good application.At the same time,a simulation platform is built for the whole network to observe the recognition results directly.更多还原