一种改进的LeNet-5嵌入式人脸识别方法

给出了一种基于LeNet-5改进的人脸识别方法,以其能适用于资源及计算能力有限的嵌入式系统.把典型卷积神经网络LeNet-5的结构,设计为由两个卷积采样层、一个全连接隐藏层和一个分类输出层,降低了网络结构复杂度.而且减少了卷积核的个数、改进了池化方式以及分类输出方式,降低了计算复杂度.实验证明,在保证训练和测试精度的同时,该方法提高了在嵌入式平台上进行单人脸识别的速度.

An Improved Embedded Face Recognition Method based on LeNet-5

A face recognition method based on LeNet-5 is presented, which can be applied to embedded systems with limited resources and computing power. We designed the structure of the typical convolutional neural network LeNet-5 to consist of two convolution sampling layers, a fully connected hidden layer, and a classified output layer, which reduces the complexity of the network structure. Moreover, we have reduced the number of convolution kernels, improved the pooling method, and classified output methods, which reduces the computational complexity.Experiments show that while ensuring the accuracy of training and testing, this method improves the speed of single face recognition on embedded platforms.