Correction of amplitude scintillation effect in fully polarimetric SAR coherency matrix data

Abstract

In this paper, we investigate the correction of ionospheric amplitude scintillations appearing as image artifacts in low frequency synthetic aperture radar (SAR) data. A method has been proposed to separate amplitude and phase components of fully polarimetric SAR (POLSAR) coherency matrix [T], and then implement a combination of multi-level, two-dimensional Discrete Wavelet Transform and Fast Fourier Transform (DWT-FFT) to correct the amplitude scintillation-induced stripes in [T] matrix. In the methodology, [T] is decomposed into component matrices of amplitude and relative phases. To the amplitude [T] matrix, a combination of 2-D DWT is performed, using the Haar wavelet, followed by a non-linear 2-D FFT correction strategy to remove amplitude scintillation stripes. Using DWT, each amplitude [T] matrix element is decomposed into low- and high-frequency components known as approximate and detailed coefficients respectively. On applying a 2-D FFT on the detailed coefficients, amplitude scintillation-induced stripes are easily identified in the frequency spectrum and then removed using 2-D FFT correction strategy of averaging-post-thresholding. The proposed method is tested on the [T] matrix of several POLSAR datasets acquired from Advanced Land Observation Satellite/Phased Array type L-band synthetic aperture radar (ALOS/PALSAR) and Advanced Land Observation Satellite-2/Phased Array type L-band synthetic aperture radar-2 (ALOS-2/PALSAR-2) satellites. The performance of the correction technique is analysed both, qualitatively and quantitatively. While improvements in the post-correction datasets are visually observed, a supervised Wishart classification also performed on the [T] matrices of undisturbed (reference), disturbed and corrected data indicates significant increase in overall accuracy (OA) and kappa coefficient (k̂) parameters. The classification accuracies of the disturbed data (OA = 69.31%, k̂=0.54) improved to OA = 76.49%, k̂=0.68 after applying the proposed correction approach, which is comparable to the reference data values (OA = 81.57%, k̂=0.71). The proposed method performs significantly better than the existing technique of a simple 2-D FFT approach (OA = 73.49%, k̂=0.62). The contribution of dominant scattering powers, obtained from the seven-component scattering power decomposition of POLSAR coherency matrix (7SD) model, also demonstrates post-correction improvements. The dominant scattering power for targets increased by ~3–9% using the proposed approach, shows an improvement of ~3% over the existing 2-D FFT technique.