An Evaluation of Denoising Methods for Satellite Imagery: A Comparative Analysis
Malavika Jayakumar, Rahul Gudivada, Harsh Nair, S. Jenicka · 2023
Satellite images often suffer from noise, which hampers data reliability and analysis accuracy. This study presents a comparative evaluation of denoising algorithms for satellite imagery, with a specific focus on a deep learning method called Deep Convolutional Neural Network (DnCNN). This study includes a comparison of spatial filters such as NLM, and weighted mean filters, frequency domain filters such as gaussian filters, wavelet transform-based filters, and the DnCNN model with an attention module. To assess the effectiveness of these techniques in denoising images, various evaluation metrics such as peak signal-to-noise ratio (PSNR), structural similarity index (SSIM), and mean squared error (MSE) are utilized. Moreover, a novel approach based on the DnCNN network is presented, leveraging multiple layers and attention modules to selectively enhance important image features while suppressing noise. The proposed methodology highlights the DnCNN network's ability to capture intricate details and improve denoising performance. Results demonstrate the superiority of deep learning-based algorithms over alternative methods. The integration of attention modules within the DnCNN model further enhances the denoising process by leveraging image correlations and extracting pertinent information. This study offers valuable findings on effective methods for reducing noise in satellite imagery, highlighting the potential of deep learning and attention mechanisms in denoising satellite images.