SAR Image Despeckling Using a Convolutional
Synthetic Aperture Radar (SAR) images are often contaminated by a multiplicative noise known as speckle. Speckle makes the processing and interpretation of SAR images difficult. We propose a deep learning-based approach called, Image Despeckling Convolutional Neural Network (ID-CNN), for automatically removing speckle from the input noisy images. In particular, ID-CNN uses a set of convolutional layers along with batch normalization and rectified linear unit (ReLU) activation function and a component-wise division residual layer to estimate speckle and it is trained in an end-to-end fashion using a combination of Euclidean loss and Total Variation (TV) loss. Extensive experiments on synthetic and real SAR images show that the proposed method achieves significant improvements over the state-of-the-art speckle reduction methods.
We have proposed a new method of speckle reduction in SAR imagery based on CNNs. Compared to nonlocal filter- ing and BM3D image despeckling methods, our CNN-based method generates the despeckled version of a SAR image through a single feedforward process. Another uniqueness of our method is that the network is designed to recover the noise by convolutional layers and then the noisy input is divided by the estimated noise which results in the denoised output. This strategy, similar to residual learning, is inspired by the observation that a SAR image can be viewed as a product of a clean reference and noise. Results on synthetic and real SAR data show promising qualitative and quantitative results. This new process is also valuable for many SAR image understanding tasks such as road detection, railway detection, ship wake detection, texture segmentation for agricultural scenes and coastline detection.