Road Subsidence Early Warning Techniques Based On Ground Penetrating Radar

Road subsidence is becoming a long-term problem to be immediately solved in the national road construction engineering.Due to its good penetration,flexible ways of working and non-destructiveness of the sites,ground penetrating radar(GPR)techniques present great prospects in the application of road subsidence warning.The study on road subsidence early warning techniques based on GPR is present in this thesis.Firstly,due to the limitations of traditional PCA in clutter reduction,an improved PCA subspace method is proposed based on the 2D wavelet transform.Moreover,the combination of the improved subspace method and adaptive filtering ensures the signal fidelity and learning adaptability of adaptive filtering.Then,an adaptive clutter reduction algorithm based on wavelet transform and PCA,as well as adaptive filtering,is proposed.The experimental results suggest that the proposed method improves the signal to clutter ratio and target image definition.Secondly,DCT shows good performances in aspects of data compression and feature extraction.Combined with the initial target electromagnetic scattering waveform,DCT is utilized to extract features of GPR A-scan data.Moreover,the non-stationary characters of GPR data and short-term stability of speech signal is similar.Therefore,the MFCC of speech signal is introduced as a feature applying to GPR B-scan data.The simulation results based on GprMax2 D software demonstrate the effectiveness of the two methods in underground conventional materials recognition.Finally,differences between underground voids and pipelines are mainly about target materials and shapes.Then,a voids recognition method based on the DCT and polarization attributes is proposed.The Simulation model deals with three aspects,including the typical circular-end bow-tie,road layered structure and lossy and dispersive soil.The finite element method is employed to conduct numerical modeling of randomly generated model.The results of underground target classification based on support vector machine classifier indicate that the feature extraction method exhibits encouraging performance in terms of underground voids identification.