Neuro-fuzzy classification of the new and used bills using acoustic data

The proposed technique is based on an extension concept of an adaptive digital filter (ADF), a neural network (NN) with error back-propagation (BP), and fuzzy inference. Two-stage ADF is used in order to extract the desired bill sound from observation data in which the noise is included. The output signal of two-stage ADFs is transformed into spectral data by the fast Fourier transform (FFT), and it becomes an input pattern of the NN. Then, the discrimination result of the NN is finally judged by the fuzzy inference in a new bill or an exhausted bill. It is shown that the proposed technique is effective for the new and used discrimination of bill money for the experimental results presented.