Adaptive Interference Detection and Removal in SAR Images Based on the Order-2 Tracy-Widom Distribution

This paper proposes a novel eigenvalue-based detector for adaptive detection of radio interference in synthetic aperture radar (SAR) images. The proposed method leverages the increased eigenvalues caused by interference in single-look-complex SAR image blocks, where the interference is expected to have a small set of eigenvalues, particularly with a dominating one. Specifically, the method segments the image into multiple blocks, computes the eigenvalues of each block’s covariance matrix, and compares the largest eigenvalue with a threshold to determine the presence of interference under the criteria of constant false alarm rate. The threshold is determined by the order-2 Tracy-Widom distribution of the largest eigenvalue for each image block under the assumption of image homogeneity. The proposed detector is integrated with our previously proposed block subspace filter for adaptive interference detection and removal. Experimental results on Sentinel-1 double-polarization and Gaofen-3 HV-polarization images demonstrate the effectiveness of the proposed method on adaptive detection and suppression of interference artefacts in SAR images.