音響信号を利用したカーネル判別器による疲労紙幣検出:Divergence-basedカーネルを用いた時変スペクトル判別

概要

論文の詳細を見る
In Japan, the quality of paper banknotes in circulation is maintained at a high level. The main reasons for this are as follows:(1) to facilitate the detection of counterfeit banknotes, and (2) to prevent paper jam problems in automated teller machines. Old, stained, or badly wrinkled banknotes frequently cause these problems. Such banknotes are removed from circulation and discarded by the Bank of Japan. In this paper, we introduce the time-varying spectrum classification method based on the SVM and RVM with the divergence-based kernel and apply this method to the detection of worn-out banknotes. The divergence gives a metric between two probability distributions, and the divergence-based kernel was proposed as a measure between two probability distributions in the kernel method. Here, two data sets are used for verification of classification performance, and the sound of flicking of banknotes is considered as the analyzed object. The obtained result shows a higher classification performance than some common methods.
公益社団法人計測自動制御学会の論文