Research On Feature Extraction And Classification Of Typical Geological Structures In Elastic Wave Echo Signals

The subject is supported by National Natural Science Foundation of "The theoretical research on the fine detection of reflective channel wave in coal seam based on continuous source".The purpose of this dissertation is to complete the feature extraction and classification of elastic wave echoes,and to provide the basic theoretical support for the detection of elastic waves in the coal measure strata.In this paper,feature extraction algorithm based on wavelet packet decomposition and statistical analysis is used to extract features of geological echo signals.Firstly,the wavelet packet decomposition method is used to decompose the four scale of the geological echo signal,and then the statistical analysis method is used to calculate the singular values,standard deviations and maximum values of the 16 wavelet packet coefficients on the fourth scales after the wavelet packet decomposition.The statistical characteristics of the 48-dimensional wavelet packet coefficients are obtained and the feature space of the geological echo signals is thus formed.In order to improve the efficiency and accuracy of geological echo signal classification as much as possible,this paper uses feature selection and classification algorithm of BP neural network optimized by optimize to classify geological echo signals.The feature selection using genetic algorithm of geological echo signals reduces the characteristics of 48-dimensional geological echo signals to 26-dimensional.At the same time,genetic algorithm is used to optimize the weights and thresholds of the error back propagation neural network classifier,and the features of the geologic echo signals obtained by dimension reduction are input into the classifier for training and prediction.So four types of geological signals from the Geological Science Data Sharing Network of the Chinese Academy of Geological Sciences have been classified:fault echo signal,collapse column echo signal,water-rich area echo signal and mined out area echo signal.The accuracy of the test set reached 94.17%,and the average sensitivity,specificity,and positive predictive were 94.17%,98.28%,and 95.80%.Finally,the four kinds of echo signal are collected through the geological echo signal acquisition experiment system based on STM32F103RCT6,and they are used for validation of feature extraction and classification algorithm.The accuracy of classification reached 97%,and the average sensitivity,specificity and positive predictive were 97%,98.99%and 97.08%.The experimental results show that the feature extraction algorithm used in this paper can effectively extract the stable feature of the geological echo signal,and the high precision classification of four kinds of geological echo signals is realized by the neural network classifier optimized by genetic algorithm.Therefore,BP neural network feature classifier optimized by genetic algorithm can be used to identify geological features effectively,which is of great significance for coalfield exploration and formation detection.