Speckle noise reduction in sar images using improved filtering and supervised classification

Synthetic aperture radar (SAR) is a remote sensing device that extracts the earth's surface's geo and biophysical characteristics. Classification performance is a major phase in SAR processing. Speckle noise occurs in SAR due to the coherent combination of backscatter signals from different sources. One of the approaches for suppressing the noise from SAR is to utilize local statistics. The proposed architecture evaluates the robustness of several improved filters like Improved Lopez, Improved Boxcar, Improved Guided filter and improved Lee-sigma and verifies their effects on classification accuracy. These filters were designed to overcome the suppression of target points and the blurring of edges. The supervised Wishart classifier with an improved Sparrow Search Algorithm (WC-ISSA) is utilized in the classification. SSA is used to optimize WC parameters and improve classification performance. One of the essential parameters in speckle noise filtering is the size of the sliding window. The window size varies, and the improved filters' performance is evaluated. Further, a growing self-organizing map (GSOM) is used to improve blurring performance. The proposed model is used for deblurring and enhancing the performance of smoothing images. The overall evaluation is carried out on the Matlab platform. The performance of the improved filters is compared to the standard filters, and the performances are compared on the virtual SAR dataset. The implemented results proved that the Extended Lee-sigma performed better than other filters. The PSNR and SSIM obtained by the proposed model were found to be 65.72 and 99.92%, respectively, which is considered to be more effective than other models already in use.