Research On Key Problems Of Face Recognition Parallel Algorithm Based On Multi-dimensional Genetic Neural Network

With the development of computer technology,parallel technology is a new research direction,and it is a hot topic in the field of face recognition in recent years.The neural network classifier in face recognition is the base of face recognition.In the paper,based on the parallel application of neural network in face recognition,parallel PCA algorithm,parallel genetic algorithm,multi BP neural network,multi Island genetic algorithm to optimize the core algorithm of the neural network and multi universe quantum genetic neural network were studied.First,the algorithm of facial feature extraction is studied.The feature extraction of face recognition is usually a necessary step in face recognition process,and compared with other algorithms,PCA algorithm has a higher robustness.But PCA algorithm is limited by the hardware,and compared with the other algorithms it does not have the advantage in hardware processing speed.To solve these problems,this paper proposes a thread level parallel optimization algorithm based on PC A algorithm.The PCA algorithm is optimized by using OpenMP and SSE at the same time.In the process of feature extraction,the experiment shows that the face data can be detected by the system database,and the parallel PCA algorithm can achieve a processing speed 1.67 times faster than the original algorithm,which means the system efficiency is greatly improved.Second,in the process of face recognition,neural network classifier weight genetic algorithms is studied.Genetic algorithm is one of the most commonly used algorithm for neural network optimization,but with the continued process of genetic evolution,the trend of optimal solutions often occur in the same gene and a repeat of this individual,lead to the reduction of the overall algorithm efficiency.In order to solve this problem,a memory genetic algorithm is proposed,and the concept of the gene pool,the distribution of their individual fitness value are introduced.By comparing the fitness value,the repetition of the individual is avoided,and the efficiency of the algorithm is improved.Based on the proposed memory genetic algorithm,this paper further optimizes the memory genetic algorithm in multi thread and parallel optimization.In solving the practical problems of TSP,the experimental results show that,compared with the classical genetic algorithm,the multi thread parallel genetic algorithm has faster convergence speed and higher system efficiency.Third,the neural network algorithm used in face recognition is studied,and a multi-dimension parallel BP neural network algorithm is proposed.BP neural network algorithm is a commonly used face recognition classifiers,but the limitation of their algorithm in data convergence and speed has been an important factor to restrict its applications.To solve these problems,a multi-dimensional parallel algorithm of BP neural network is proposed.BP neural network is divided into several "mirror network",the data is divided into the same number of data sets.Once a train is completed,a data exchange is done in the processing center.The experimental results show that,compared with the original BP neural network,the multi-dimensional parallel BP neural network has faster convergence rate,which is 10 times faster than that of the original one.Fourth,the optimization of neural network weights based on parallel genetic algorithm is studied,and a parallel genetic neural network algorithm is proposed.In the optimization problem of neural network,the method is generally divided into topology optimization and weight optimization.This paper presents a new adaptive genetic algorithm which can quickly search for the optimal solution.By dividing the initial population,parallel optimization is done in the new adaptive genetic algorithm,and the weights are optimized.The experimental results show that,compared with the original BP neural network,the new parallel genetic neural network algorithm is superior in terms of convergence speed,and can achieve a faster running speed on the multi core PC platform.Finally,this paper further studies the problem of parallel quantum genetic algorithm neural network weight optimization.In the traditional genetic algorithm,the crossover and mutation process often occupy the main system resources.In order to avoid the cumbersome process of crossover and mutation,the concept of quantum computation is introduced in the paper.In the process of quantum computation,the quantum rotation gate is used to realize the updating operation of the chromosome.The quantum crossover overcomes the premature convergence phenomenon,which avoids the complicated process of crossover and mutation in genetic algorithm.The expperimental results show that,compared with the original genetic algorithm,quantum genetic algorithm has a significant improvement in both the robustness and the experimental speed.In the multi core PC machine,the operation speed of large scale data could attain 14 times of the speed of traditional operation.