An Intelligent pattern recognition for nonstationary signals based on the time-frequency entropies

ABSTRACT PhD Thesis AN INTELLIGENT PATTERN RECOGNITION FOR NONSTATIONARY SIGNALS BASED ON THE TIME - FREQUENCY ENTROPIES Ibrahim TÜRKO?LU Firat University Graduate School of Natural and Applied Sciences Department of Electrical - Electronics Engineering 2002, Page: 112 Intelligent recognition systems have gained importance in various areas with the technological developments, and pattern recognition constitutes the bases of this systems. Pattern recognition is to recognize unknown patterns by assigning them into a known class or known patterns belonging to a known class. Pattern recognition includes two steps: Feature extraction and classification. The most important indication is to select the true features. In another word extracting more precise feature is much more important than designing more complex classifier in advanced pattern recognition techniques. In this thesis, a powerful feature extraction was aimed for non-stationary signal patterns. With this aim integrated feature extraction methodologies were developed which depend on wavelet transform, entropy, parametric and non- parametric time- frequency spectral analysis methods for non-stationary signals. By placing the feature extraction methods developed into the artificial neural network pattern classifier, adaptive pattern recognition approach was presented by the new wavelet neural network models. The efficiency and reliability of these methods were examined by using Doppler heart signals from the cardiology clinic of Firat Medicine Center. The success of the system was compared with the doctor-based diagnosis and correct diagnosis about of %96 was achieved. This method will increase the reliability of automatic intelligent diagnosis systems by helping the diagnosis by the practitioner or by directly diagnosing in various fields such as biomedical and telecommunications. Keywords : Pattern recognition, feature extraction, adaptive feature extraction, intelligent diagnosis system, parametric models, nonparametric models, nonstationary signal, Doppler heart signals, artificial neural networks, entropy, wavelet networks. xm