Study Of Signal Classification Method Based On Waveform Feature

Modern signal processing technology has penetrated into every aspect of our lives.Signal frequency, amplitude and phase change can record and characterization a widevariety of waveform information. For seismic signal, the reflection signal of differentmedium has different waveform feature. So by appropriate methods to extract thewaveform feature and classifying them, we can distinguish different undergroundmedium, the signal classification techniques based on waveform feature have importantapplications, for example used in the search for underground oil and gas reservoir.This thesis reviewed the background of signal classification based on waveformfeature and the present situation of research, the technology process are introduced indetail which include the waveform de-noising preprocessing, feature extraction andfeature selection, classification algorithm steps. At the same time we studied severalpractical problems such as feature extraction and classification algorithm, and formed aset of suitable method for the actual signal classification based on the waveform feature.Specific innovation works are as follows:1. In post-stack seismic signal waveform classification technique, traditionalmethods are using unsupervised classification algorithm, and ignored thepriori-information of well logging data, and thus the classification results are datadriven, the contact of classification results and the actual distribution of oil and gasunderground are not close enough. A semi-supervised EM algorithm is proposed in thisthesis and the priori-information with label was combined with a large number ofsamples without label to classify algorithm. Rationality and accuracy of the results hasimproved. The proposed algorithm were used for the classification of F3data in Dutchnorth sea oil field, Yuanba data of gas field in Sichuan province, Hebaochang data inSichuan province, all these data are post-stack seismic signal, the classification resultscompared with the traditional method has larger ascension.2. Different incident Angle of seismic reflection are changed, The change rule ofdifferent seismic reflection of underground medium are different, and post-stack seismicsignals will obtained by stack different Angle of incidence of pre-stack seismic signal, it can’t expression the change of incidence Angle seismic reflection. It means post-stackseismic signal waveform classification can’t classification seismic signal according tothe change rule of seismic reflection with different incident Angle. According to thisproblem, a pre-stack seismic signal waveform classification algorithm is proposed inthis thesis, and compared with the results of post-stack seismic signal classification, thecontinuity and detail has ascension; At the same time, with chebyshev polynomialfitting for feature extraction and variance method for feature selection, while reduce thepre-stack seismic signal redundancy which improves the classifier performance. Theproposed algorithm was applied in Sulige gas field which located at Inner Mongolia,and compared with the same area of post-stack waveform classification results, showsthat the proposed method is feasible.