Comparison of Feature Extraction Methods for Automated Target Recognition by Reducing Speckle Noise in SAR Data

Automatic Target Recognition (ATR) utilization holds significant importance in the defense domain; it serves as a fundamental step in augmenting intelligence and facilitating the self-sufficient functioning of defense platforms. Synthetic Aperture Radar (SAR) is an attractive option for ATR since it can produce high-resolution images even in adverse conditions, such as through clouds and in darkness, by penetrating the environment. Despite its advantages, automatic target recognition for SAR images is still challenging due to factors such as the variability in target appearance, complex backgrounds, and fluctuations in imaging circumstances. In this study, we evaluated different feature extraction methods using SAR images obtained from the essential MSTAR database. The goal of this study was to determine the effectiveness of these techniques in producing target images with enough resolution for recognition. To ascertain the effectiveness of different techniques in producing high-resolution target images for recognition, we conducted a comparison of linear and non-linear Support Vector Machine (SVM) and Random Forest (RF) methods on the task of target classification. Later, we evaluated the impact of speckle noise reduction on the accuracy of the classifiers. The insights gained from our findings provide valuable guidance for selecting appropriate feature extraction methods and classifiers for the purpose of automatic target recognition through the use of SAR imagery.