Transformer-based SAR Image Despeckling
This addresses the difficulty in processing and interpreting SAR images for remote sensing applications, but appears incremental as it applies a known transformer architecture to a specific domain.
The paper tackles the problem of speckle noise in Synthetic Aperture Radar (SAR) images by introducing a transformer-based network for despeckling, achieving significant improvements over traditional and convolutional neural network-based methods on both synthetic and real images.
Synthetic Aperture Radar (SAR) images are usually degraded by a multiplicative noise known as speckle which makes processing and interpretation of SAR images difficult. In this paper, we introduce a transformer-based network for SAR image despeckling. The proposed despeckling network comprises of a transformer-based encoder which allows the network to learn global dependencies between different image regions - aiding in better despeckling. The network is trained end-to-end with synthetically generated speckled images using a composite loss function. Experiments show that the proposed method achieves significant improvements over traditional and convolutional neural network-based despeckling methods on both synthetic and real SAR images.