Optimization of Autoencoders for Speckle Reduction in SAR Imagery Through Variance Analysis and Quantitative Evaluation

• DOI: 10.3390/math13030457
• Publication Date: 2025-01-30T09:01:39Z
• Authors: Ahmed Alejandro C...; Rubén Darío Vásqu...; Jean P. Diaz-Paz; Henry O. Sarmient...; Luis Gómez; Carlos M. Travies...
• Keywords: Variance reduction

Speckle reduction in Synthetic Aperture Radar (SAR) images is a crucial challenge for effective image analysis and interpretation remote sensing applications. This study proposes novel deep learning-based approach using autoencoder architectures SAR despeckling, incorporating of variance (ANOVA) hyperparameter optimization. The research addresses significant gaps existing methods, such as the lack rigorous model evaluation absence systematic optimization techniques learning models processing. methodology involves training 240 on real-world data, with performance metrics evaluated Mean Squared Error (MSE), Structural Similarity Index (SSIM), Peak Signal-to-Noise Ratio (PSNR), Equivalent Number Looks (ENL). By employing Pareto frontier optimization, identifies that effectively balance denoising preservation fidelity. results demonstrate substantial improvements speckle quality, validating effectiveness proposed approach. work advances application denoising, offering comprehensive framework