A Novel Approach for SAR Target Detection Based on Unsupervised
Complex-Valued Extreme Learning Machine
Strong clutter seriously affects target-of-interest detection in
synthetic aperture radar (SAR) images. This letter proposes an
unsupervised target detection method (U-TDM) based on a complex-valued
extreme learning machine (CV-ELM), the essence of which is to transform
the problem of target detection into a pixel binary classification
problem. The SAR image is first divided into several unlabeled patches,
and fuzzy c-means (FCM) is used to construct the reference target patch
set and the clutter patch set. Based on these two patch sets, CV-ELM is
used to classify the neighboring patch of the pixel to be detected.
Since the pixel intensity and distribution of target-of-interest and
clutter are different, unsupervised pixel classification could be
realized without ground-truth through U-TDM. Experimental results on
GF-3 data and Sentinel-1 data show the efficiency of the proposed method
in target detection with a heterogeneous clutter environment.