Fast and Robust RBF Neural Network Based on Global K-Means Clustering with Adaptive Selection Radius for Sound Source Angle Estimation

The sound source localization technique is widely applied to target detection and localization. However, the application of conventional sound source localization methods is limited in actual environment because of estimation accuracy, computational complexity, and flexibility of the environment. In order to improve the sound source localization performance in actual environment, a fast and robust radial basis function (RBF) neural network based on global K-means clustering with adaptive selection radius is proposed in this paper. In the proposed method, an adaptive selection radius is calculated according to the population density sampling method to remove unnecessary points around cluster centers during the global K-means clustering; thus, compared with the conventional neural network, a fast optimization of hidden layer neuron parameters can be achieved. Afterward, the RBF neural network is trained to locate the sound source by solving nonlinear equations of the time difference of arrival and sound source location. Because of the adoption of adaptive selection radius in global K-means clustering, the proposed method can provide desirable performance with low computational complexity. Based on the simulated and actual experimental data, the proposed method is verified and an improved performance is achieved compared with that of conventional neural network sound source localization methods.