CAJ | 학술논문

反向传播( BP)神经网络算法在手势识别中得到了广泛的应用。为了对算法进行改进以提高BP神经网络的学习效率,提出一种基于量子粒子群优化BP神经网络的手势识别训练算法。在手势识别过程中,首先采用量子粒子群算法( QPSO)训练BP神经网络,获得优化的BP神经网络权值和阈值；合理地定义并提取BP神经网络的手势识别样本；最后采用训练过的BP神经网络对动态手势进行识别。该算法简单,不依赖初始值,并且收敛速度快,尤其对于高维复杂问题,能保证收敛到最优解。实验结果表明,该算法平均训练时间达到5．15 s,识别正确率达到95．1%,效果明显优于一般的BP神经网络算法。
반향전파( BP)신경망락산법재수세식별중득도료엄범적응용。위료대산법진행개진이제고BP신경망락적학습효솔,제출일충기우양자입자군우화BP신경망락적수세식별훈련산법。재수세식별과정중,수선채용양자입자군산법( QPSO)훈련BP신경망락,획득우화적BP신경망락권치화역치；합리지정의병제취BP신경망락적수세식별양본；최후채용훈련과적BP신경망락대동태수세진행식별。해산법간단,불의뢰초시치,병차수렴속도쾌,우기대우고유복잡문제,능보증수렴도최우해。실험결과표명,해산법평균훈련시간체도5．15 s,식별정학솔체도95．1%,효과명현우우일반적BP신경망락산법。
Back Propagation ( BP ) neural network algorithm is widely used in hand gesture recognition. In order to improve learning efficiency of the BP neural network, the authors proposed a hand gesture recognition algorithm based on Quantum-behaved Particle Swarm Optimization ( QPSO) of BP neural network. In the process of gesture recognition, first, the QPSO algorithm was used to train the BP neural network and get the weights and thresholds of the optimized BP neural network. Experiment program defined and extracted gesture recognition samples reasonably for the BP neural network. Finally, the dynamic gestures were recognized by the trained BP neural network. The proposed algorithm is simple, does not depend on the initial value, and has a fast convergence speed, especially for high dimensional complex problems, it can guarantee the convergence to the optimal solution. The experimental results indicate that the average training time of the new algorithm can reach 5. 15 seconds, the correct recognition rate of the new algorithm can reach 95. 1%. The new algorithm has better effects than the general BP neural network algorithm.