GPR B-Scan Image Denoising via Multi-Scale Convolutional Autoencoder with Data Augmentation

Ground-penetrating radar (GPR), as a non-invasive instrument, has been widely used in civil engineering. In GPR B-scan images, there may exist random noise due to the influence of environment and equipment hardware, which complicates interpretability useful information. Many methods have proposed eliminate or suppress noise. However, existing an unsatisfactory denoising effect when image is severely contaminated by This paper proposes multi-scale convolutional autoencoder (MCAE) denoise data. At same time, solve problem training dataset insufficiency, we designed data augmentation strategy, Wasserstein generative adversarial network (WGAN), increase MCAE. Experimental results conducted on both simulated, generated, field datasets demonstrated that scheme promising performance for denoising. terms three indexes: peak signal-to-noise ratio (PSNR), time cost, structural similarity index (SSIM), can achieve better suppression compared with state-of-the-art competing (e.g., CAE, BM3D, WNNM).