Batch despeckling of SAR images by a convolutional neural network-based method
H Elhajj Imad, Yassine Tounsi, Benjelloun Mohammed, Nassim Abdelkrim · 2020
Synthetic aperture radar (SAR) is a very high and accurate technology for remote sensing. SAR images are very exploited for change detection and monitoring because of their high resolution. Because of the coherent retro diffusion of the backscattered radar waves with the rigorous earth surface, SAR images are characterized by a multiplicative noise called speckle that influences their analysis. For this reason, several methods and algorithms have been introduced for speckle noise reduction from these images. This paper proposes and exploits a convolution neural network (CNN) based technique for SAR image denoising. The performance of the introduced technique is studied firstly by using numerical simulation where we added speckle noise with a different variance; then, a quantitative appraisal is presented by using three metrics: Peak to signal noise ratio (PSNR), Edge preservation index (EPI) and image quality index (Q). Finally, we apply the proposed method on a real SAR image obtained by the Sentinel-l satellite and ERSSAR sensor.