ABSTRACT
Hyper-wavelet transforms such as the non-subsampled shearlet transform (NSST) are popular algorithms to remove random noise and clutter in ground penetrating radar (GPR) images. However, due to the effect of time-varying gain, random noise and clutter in GPR images are non-stationary. Moreover, random noise which is poorly correlated among pixels and clutter which shows a certain correlation among adjacent pixels are usually mixed together. Therefore, it is difficult to determine the threshold function of hyper-wavelet transforms, resulting in a decrease of noise and clutter removal performance. To address this issue, a novel two-step algorithm is proposed to remove non-stationary random noise and clutter in the GPR image separately. In the first step, Self2Self, a self-supervised denoising algorithm, is employed to remove the non-stationary random noise. In the second step, a time-varying threshold function based on NSST and an edge area protection method based on the Canny algorithm are proposed to effectively remove non-stationary clutter in the GPR image. Experimental results show that the proposed method has an excellent performance in removing non-stationary random noise and clutter while effectively protecting the edge information of the GPR image.
Data Availability Statement
The data that support the findings of this study are available from the corresponding author upon reasonable request.
Disclosure statement
No potential conflict of interest was reported by the author(s).