Deep Wiener Deconvolution Denoising Sparse Autoencoder Model for Pre-processing High-resolution Satellite Images

Abstract

The detection of geospatial objects in surveillance applications faces significant challenges due to the misclassification of object boundaries in noisy and blurry satellite images, which complicates the detection model's computational complexity, uncertainty, and bias. To address these issues and improve object detection accuracy, this paper introduces the Deep Wiener Deconvolution Denoising Sparse Autoencoder (DWDDSAE) model, a novel hybrid approach that integrates deep learning with Wiener deconvolution and Denoising Sparse Autoencoder (DSAE) techniques. The DWDDSAE model enhances image quality by extracting deep features and mitigating adversarial noise, ultimately leading to improved detection outcomes. Evaluations conducted on the NWPU VHR-10 and DOTA datasets demonstrate the effectiveness of the DWDDSAE model, achieving notable performance metrics: 96.32% accuracy, 86.88 edge similarity, 75.47 BRISQUE, 28.05 IQI, 38.08 PSNR (dB), 0.883 SSIM, 98.25 MSE, and 0.099 RMSE. The proposed model outperforms existing methods, offering superior noise and blur removal capabilities and contributing to Sustainable Development Goals (SDGs) such as SDG 9 (Industry, Innovation, and Infrastructure), SDG 11 (Sustainable Cities and Communities), and SDG 13 (Climate Action). This research highlights the model's potential for inclusive innovation in object detection applications, showcasing its contributions and novel approach to addressing existing limitations.