GPR noise suppression algorithm based on compressed parallel non-local mean filtering method

The quality of ground penetrating radar (GPR) data plays an important role in data processing and subsequent geological interpretation. When GPR data is seriously polluted by random noise, the effective signal will be largely masked. Conventional non-local mean (NLM) filtering can suppress random noise effectively, but it is limited in the application of large-scale data processing because of its high computational complexity. In this paper, a GPR noise suppression algorithm based on compressed parallel NLM filtering method is presented. The algorithm is based on data integration and uses data center symmetry to form a data compression algorithm, which reduces the computational complexity. By calculating the standard deviation of similarity, an adaptive filter parameter estimation algorithm is implemented to further improve the denoising effect. By introducing parallel algorithm, the computational efficiency is further improved. In the test of synthetic data, the denoising effect of the proposed method is better than other methods by comparing the mean root square error and the mean signal-to-noise ratio. In the test of real data in Dauphin Island, Alabama, the calculation time of the proposed method is 9.7 seconds, which is significantly better than the conventional NLM filtering algorithm, which verifies the feasibility and effectiveness of the proposed method.