Target Recognition by Means of Polarimetric ISAR Images

Target Recognition With Radar Images via Parameterized Dictionary Sets

“…Recently, the local features involved in radar images have received more active attention [9][10][11][12][13]. In [9], the local non-negative matrix factorization from the feature space is considered to recognize the ISAR image.…”

“…In [10], some information embedded in the statistical model is presented to recognize interest targets on the synthetic aperture radar (SAR) image. In [11][12][13], the scatterers on the SAR or ISAR image are also attempted to realize the recognition task. In general, the local features in the above methods economically represent the radar images of interest targets.…”

“…The second one is that the scatterer extraction can be a time-consuming procedure due to the prohibitively large dimension of an ISAR image. Large computation burden is unavoidable embedded in the recognition methods in [11][12][13] for extracting the scatterers and matching them to the stored templates.…”

“…Unlike the existing features [4][5][6][7], the 2-D location parameters of main high-energy scatterers on an ISAR image provide the image with very sparse representations and illuminate the scattering model of a target. Furthermore, the scattering model is often proportionately changed with the target scale and remains more stable over the wide angular region than the scatterer's amplitude [11][12][13][14]. Therefore, 2-D location parameters of main highenergy scatterers can be less sensitive to the variations of the ISAR image and more robust for target recognition.…”

“…Therefore, 2-D location parameters of main highenergy scatterers can be less sensitive to the variations of the ISAR image and more robust for target recognition. Main contribution of the proposed method in this paper is that target recognition is carried out with the parameterized dictionary set of every target and avoids the online scatterer extraction as the methods in [11][12][13]. Furthermore, the decomposition procedures involved in the proposed method only use the local image at the 2-D location parameters of main high-energy scatterers, and the image outside the scatterer locations is filtered out, which results in good generalization performance against noise or clutter.…”