Abstract
Discrimination of aircraft type by its noise is performed using a cooperative learning model applied to neural networks. There are three independent lower networks and an upper network, which unifies the outputs of the lower networks, in this model. Such unified networks are implemented to recognize takeoff and approach noise respectively.
Front and rear noise of aircrafts are transformed individually to spectram data with help of a fast Fourier transformation. Then, the data are added synchronously and normalized for emphasis on the characteristics of aircraft noise. The networks are trained on these data using a backpropagation training method. Giving these discrimination data to the neural networks, the discrimination rates were 94.5% in approach and 95.7% in takeoff. These rates are somewhat higher than those obtained using the discriminant analysis method. These discrimination rates are increased because of two reasons : 1) the normalization of data leads to lowering errors, and 2) in the cooperative learning model each network works to improve the faults on others.
