基于BP神经网络的声源定位研究

Sound Source Localization Based on BP Neural Network

传统声源定位算法计算量较大,不适合复杂多变的场景,因此提出基于BP神经网络的声源定位算法。该算法以麦克风阵列接收声音的时间差作为输入,以声源位置坐标作为输出。为了得到较理想的BP神经网络模型,借助MATLAB仿真实验,分别讨论了神经网络中神经元个数、隐藏层层数以及反向传播函数对算法精确性的影响。其中,结合不同反向传播函数的特点和局限性,选择5种反向传播函数分别在不同数据量下进行对比实验。实验结果表明,采用7个神经元、3层隐藏层与LM反向传播函数的BP神经网络模型较适合该研究场景,准确度可达到较高水平,R值可达到1,MSE值在e-11数量级下,位置误差在毫米范围内。

Traditional sound source localization algorithms require heavy computation and are not suitable for complex and changeable scenes.A sound source localization algorithm based on BP neural network is proposed.The algorithm takes the time difference of the sound received by the microphone array as input and the position coordinates of the sound source as output.In order to obtain an ideal BP neural network model,discusses the influence of the number of neurons,the number of hidden layers and the back propagation function on the accuracy of the algorithm through simulation experiments under the environment of MATLAB.As for the third factor,considering the characteristics and limitations of different back propagation functions,selects five kinds of back propagation functions for comparative experiments under different data volumes.The results show that the BP neural network model with 7 neurons,3 hidden layers and LM back propagation function is suitable for this research scenario.The accuracy reaches a high level,the R value reaches 1,the MSE value is in the order of e-11,and the position error is within the millimeter range.