GPR Data Feature Extraction And Pattern Classification Based On Time Series Analysis

Ground penetrating radar(GPR) has been widely used in various fields such as environmental engineering as a kind of shallow physical detection technology with intuitive results, and it has characteristics of high resolution and efficiency, nondestructive. Due to the complicated geological environment, the underground propagation rule of electromagnetic wave is changeable and attenuated, the received echo signal data is inevitably mixed with noise and various clutter interference, which leads to the effectiveness in the field of application has been restricted, although the theory involved in GPR technology is widely used. How to identify all kinds of noise and clutter correctly, extracting useful information has become an important part of the interpretation of GPR, the key to solving this problem lies in the GPR data processing and analysis.According to the existing GPR data processing methods and techniques, this paper found that mainly focus on single data processing technology, the purpose is to improve the signal-to-noise ratio, but also have different limitations. Then in the process of GPR data analysis after processing, professionals often rely on prior knowledge. It may leading to the detection results have mistakes or multiple solutions, especially in the detection of complex background and with a large amount of data, the processing results of these methods and techniques will be the key to achieve the desired effect. According to the detection principle of GPR technology, GPR data is collected from the real-time sampling processing with equal interval sampling pulse, which has the characteristic of time series. While in the field of time series analysis, time series data mining can effectively mine the useful information and knowledge of data. It has become an increasingly important research topic in recent years. Therefore, in the research processing of GPR data, first it study in deth several the existing data processing methods of GPR data, then it combined with the time series data mining method according to the target feature sequence extracted from the GPR data, finally according to the extraction of characteristics of the GPR data for pattern classification, and the processing and analysis results through the visual effect is presented, which provides a visual and effective basis for the professional staff to reducing the possibility of erroneous and multiple solutions.The main work of this paper includes:Firstly, in view of the limitations of the existing single method for GPR data processing, this paper proposed an adaptive GPR data processing method based on wavelet transform and EMD(empirical mode decomposition). Using wavelet transform to extract the high frequency signal and EMD to extract the low frequency signal, the GPR data denoises by two kinds of signals be input to adaptive filter as the noise. This method improves the signal-to-noise ratio of GPR data, lay a solid foundation for the next step to better extract effective features.Secondly, aiming at the lack of priori knowledge of the GPR data, combined with the time series characteristics of GPR data, the method constructing the local feature descriptors of GPR data based on time series shapelet is proposed, by using time series shapelet to extract the characteristic sequence of GPR channel data, effectively extract the different characteristics of the media information contained in GPR data, which laid a solid foundation for the latter pattern classification.Thirdly, aiming at the characteristics of GPR data with noise or mutation and GPR channel data may exist in a variety of media, this paper presents nearest neighbor pattern classification method based on the subsequence threshold, in the process of pattern classification set up subsequence threshold and use DTW metric in recent neighbor classification, each subsequence of GPR channel data can be classified.Finally, this paper presents the results of different media classification of GPR data by visual method, which provides an intuitive explanation for the professional.