基于优化后向传播神经网络的基础心音分类

针对后向传播（BP）神经网络高度依赖初始权值、收敛慢且易陷入局部极值，标准人工蜂群算法开发能力弱、局部搜索能力差等问题，提出一种基于改进人工蜂群算法优化BP神经网络的分类方法。引入自适应和全局最优策略改进人工蜂群算法中跟随蜂蜜源全局搜索、概率选择算法，使用当前迭代的最优解来提高其开发能力。此外，利用混沌系统产生初始种群，以增强人工蜂群算法全局收敛性。最后，将本文算法应用到基础心音分类。结果表明本文算法较经典分类算法分类准确率有较大的提升。梅尔频率倒谱特征参数下，本文算法的分类准确率达到94%以上。

Fundamental heart sound classification based on optimized back-propagation neural network

For solving the problems of back-propagation (BP) neural network such as highly relying on initial weights, slowconvergence and easily falling into local extremum, and the weak development capability and poor local search ability ofstandard artificial bee colony (ABC) algorithm, a classification method based on improved ABC algorithm is proposed tooptimize BP neural network. The adaptive and global optimal strategies are introduced to improve the global search andprobability selection algorithm of honey sources in ABC algorithm, and the optimal solution of the current iteration is used toimprove the development capability. In addition, chaotic systems are used to generate initial populations, thus enhancing theglobal convergence of ABC algorithm. Finally, the proposed algorithm is applied in fundamental heart sound recognition. Theexperimental results show that the classification accuracy of the proposed algorithm is superior to that of the classicalclassification algorithms. Based on Mel-scale frequency cepstral coefficients, the proposed algorithm can achieve aclassification accuracy rate above 94%.