抄録
This paper is concerned with bill money classification between new and used bills based on acoustic data by neural networks. The acoustic data is measured by a microphone located inside a banking machine. We propose a method to extract the spectral density of bill money acoustic data from the measured data which contains both bill sound and motor sound by the banking machine. The extraction method is based on the relation of correlation functions. Applying the FFT to the correlation functions, we can extract the spectral data of only bill money. From the spectral data, we classify the bill into new or used one. The proposed method by neural networks presents better classification performance than a template matching classification method.
