Speckle Noise Reduction in Satellite Images
Irene Mary Mathew, D Akhilaraj, Joseph Zacharias · 2023
Satellite images play a crucial role in various applications due to their unique characteristics, large coverage area, and availability over time. These are captured by remote sensing instruments and sent over long distances, which can introduce various types of noise into them such as speckle noise, atmospheric noise, and more. Satellite image denoising is relevant because it enhances the quality and utility of satellite data for various image processing tasks. This paper introduces a deep learning model based on the UNet architecture, to remove speckle noise from images. The performance of the model has been evaluated on satellite images as well as standard denoising datasets like Kodak24, CBSD68, and McMaster. When compared with the RDUNet model, the proposed UNet with single and multi-head attention module gives better PSNR and SSIM values.