NONSTATIONARY FEATURE EXTRACTION TECHNIQUES FOR AUTOMATIC CLASSIFICATION OF IMPACT ACOUSTIC SIGNALS

摘要: Condition monitoring of wooden railway sleepers applications are generally carried out by visual inspection and if necessary some impact acoustic examination is intuitively skilled personnel. In this work, a pattern recognition solution has been proposed to automate the process for achievement robust results. The study presents comparison several techniques together with various nonstationary feature extraction classification emissions. Pattern classifiers such as multilayer perceptron, learning cector quantization gaussian mixture models, combined Short Time Fourier Transform, Continuous Wavelet Discrete Transform Wigner-Ville Distribution. Due presence different technqies, data fusion investigated. Data in current case mainly investigated on two levels, level classifier respectively. Fusion at demonstrated best results an overall accuracy 82% when compared human operator. _______________________________________________________________________________ Hogskolan Dalarna Tel: +46-23-778800 Roda Vagen 3, 781 88 Fax: +46-23-778050 Borlange, Sweden Http://www.du.se 3 Yasemin Bekiroglu Degree projectJune 2008