| With the development of human civilization and the process of science and technology, aswell as the gradual deepening of human understanding of nature, discovery demand of theunderground world for people becomes more and more urgent and thorough. GPR(GroundPenetrating Radar) is one of the most important promising shallow remote sensing technology inrecent years, with rapid, nondestructive, high resolution and other prominent advantages. GPRrecords contain a wealth of geological information, which makes the ground penetrating radarprocessing and interpretation has becomes a very important work in mineral prospecting,geological exploration and other fields. In order to get more detailed stratigraphic informationeffectively, finding and producing more resources, we need to find a fast and effective method toextract the effective information in GPR signal and excavate its representative practicalsignificance.As an important branch of data mining, time series data mining has been widely studied inrecent years, often used in finance, science, engineering, medical and other fields. Different fromthe traditional static data, time series data is a complex data object that describes the changeprocess of things, and time series data is ubiquitous. In the field of the geoscientific research,ground penetrating radar data can be viewed as a set of data related to time. Time series miningmethod for ground penetrating radar data analysis, can achieve multi-mode research on GPRtime series data, tracking the significant wave effectively and enhancing the identification effectof weak signals, thus get better geological structure interpretation.As a collection of time-series signal, GPR data has a complex form, often accompanied by avariety of deformation, and it is difficult to predict its time and degree of the deformation inadvance. The traditional method of data processing is the time-frequency transform, andhandling the attenuation by automatic gain; but the parameters selection of these methods greatlyinfluenced by subjective factors, affecting the efficiency and reliability of data interpretation.Meanwhile, due to the uncertainty of the target detection, the number of GPR data classificationalso has great blindness, affecting the visualization rendering of the final clustering results. Tosolve these problems, this paper applies the theory of time series mining on ground penetratingradar data, and put forward a GPR data clustering algorithm based on time series mining.The GPR data clustering method based on time series mining, can effectively track theattenuated signal, further tap the potential information and improve the resolution of clusteringresults. The first step of the algorithm is to analyze the time characteristics of GPR signal andextract its time sequence features; then, on the basis of these features, choose a reasonable and effective similarity measure to cluster the time series GPR data, and eventually form a sectionalview of the detection structure. According to the GPR data classification problem, propose anoptimization algorithm based on threshold for GPR time series data clustering; this algorithmimplements secondary classification on feature datasets and top-down adaptive optimization ofthe classification number, overcomes the subjectivity of the definition of the initial clusteringcenter and achieves top-down, and is able to reflect more data information under lessclassification number. |
