Evaluation of multiple SAR speckling filter techniques performance in irrigated rice areas

Abstract

Monitoring irrigated rice crops is essential for efficient agricultural management, and Synthetic Aperture Radar (SAR) images are beneficial due to their capability to function in all weather conditions. However, speckle noise in SAR images complicates the classification of land cover. This study evaluates speckle filtering techniques to enhance the image quality of SAR for rice field monitoring. We extend previous work by comparing Bayesian filters in the spatial domain with advanced transform domain methods, including block matching 3D (BM3D) and Discrete Fourier Transform-extracted Edge (DFT Edge) techniques, across different rice growth stages. Twenty-two Sentinel-1B SAR images from the municipality of Turvo, Santa Catarina, Brazil were analyzed. Filters were assessed using metrics for speckle suppression and edge preservation. Our experiments reveal that BM3D with a sigma parameter of 10 (BM3Dsigma10 ) provided superior results, balancing effective speckle suppression and edge preservation. Specifically, BM3Dsigma10 was superior in preserving edge details compared to other techniques. However, as the σ value increased, a loss in resolution was observed, even for the DFT Edge method, which, while efficient in edge preservation, often resulted in reduced detail resolution. These findings highlight the importance of selecting appropriate filtering techniques for accurate agricultural monitoring using SAR data.