基于深度神经网络与MPI并行计算的人脸识别算法研究

Research on Face Recognition Algorithm Based on Deep Neural Networks and MPI Parallel Computing

针对实际环境中干扰因素多和计算量大,导致人脸识别准确度下降和系统算力不足的问题,提出了一种基于深度神经网络与MPI并行计算的人脸识别算法.首先,分析深度神经网络模型,设计关键训练步骤,同时收集各类人脸图像,建立训练样本库.然后,结合深度神经网络模型,对样本库数据进行训练,生成识别框架,并借助TensorFlow开源模型与Python来实现算法,进而达到识别人脸的目的.最后,基于MPI并行计算技术,搭建高性能并行计算平台,对所提算法进行分段优化与集成,实现识别系统的高速计算效率.实验测试结果显示:与已有的相关识别技术相比,所提算法具有更高的人脸识别准确度与抗干扰能力,从而可为高端智能监控系统提供技术基础.

In view of the problems of many interference factors and large amount of calculation in the actual environment,which lead to the decrease of the accuracy of face recognition and the insufficiency of the calculation power of the system,a face recognition algorithm based on deep neural networks and MPI parallel computing is proposed in this paper.First,the deep neural network model is analyzed,and the key training steps are designed.At the same time,all kinds of face images are collected,and the training sample database is established.Then,combined with the deep neural network model,we train the sample database data,generate the recognition framework,and implement the algorithm with the help of tensorflow open source model and python,so as to achieve the purpose of face recognition.Finally,based on MPI parallel computing technology,a high-performance parallel computing platform is built to optimize and integrate the proposed algorithm in sections to achieve high-speed computing efficiency of the recognition system.The experimental results show that the proposed algorithm has higher accuracy and anti-interference ability than the existing related recognition technology,and provides technical basis for high-end intelligent monitoring system.