Automatic target recognition scheme for a high-resolution and large-scale synthetic aperture radar image

Abstract

Existing automatic target recognition of synthetic aperture radar (SAR ATR) schemes mainly focus on target chips, but there is very little research for a large-scale and highresolution SAR image that is more practical for SAR image interpretation. How to recognize targets efficiently and accurately from a large-scale and high-resolution SAR image is still a challenge. We present a scheme based on the combination of a salient detection approach, an active contour model (ACM), an affine-invariant shape descriptor, and the corresponding shape context. During the detection stage, the spectral residual approach is utilized to efficiently preselect salient regions. The proposed convex ACM, based on a ratio distance and distribution metric which makes it more robust to multiplicative speckled noise, is then adopted to get accurate candidate target chips. For the discrimination stage, a cumulative sum of multiscale lacunarity feature is proposed to select vehicle chips from clutter chips. Finally, affine-invariant shape features, obtained from the contours by our proposed ACM, are combined with a corresponding shape context to make the classification more accurate. Experimental results demonstrate that our SAR ATR system, integrating all the proposed methods, is feasible in ATR from a high-resolution and large-scale SAR image.

Fast and Accurate Target Detection Based on Multiscale Saliency and Active Contour Model for High-Resolution SAR Images

“…As for the two-parameter CFAR detector, its time complexity is O(M 2 ), which has been analyzed in our early research [35]. Then, the detection result by CFAR is treated with some postprocessing methods [33].…”

“…Of course, a more complicated discrimination method, which was used in our early research [35], is also tried to select the desired target chips in Section IV-A4.…”

“…Furthermore, P (φ) is the level set regularization term [8] that can regularize the level set function by penalizing the deviation of φ from a signed distance function, and μ > 0 and v > 0 are constants that denote the weights of the length and level set regularization terms, respectively. The weight ω used to combine the modified CV and RSF models should be sufficiently small to handle SAR images corrupted by intensity inhomogeneity, and a proper ω makes it possible that the influence of the two ACMs on the curve evolution is complementary (see a more detailed discussion of the weight settings in Section 2.1.3 of our early research [35]). The energy functional E modified LGIF can be minimized with respect to φ by solving the following gradient descent flow equation: Because F 1 (x) and F 2 (x) are obtained from the RSF and modified CV models, they are called the LIF and GIF forces, respectively.…”

“…Since the salient regions are only a part of the original SAR image, the maximum time complexity of this step is O(M ); if we adopt another method (see Section IV-A4), i.e., the cumulative sum of a multiscale lacunarity feature [35] is utilized to discriminate the target chips from the clutter chips, its time complexity is also O(M ), which has been analyzed in [35]. Therefore, based on the aforementioned analysis, either way we select out the target chips, the time complexity of this step is O(M ).…”

“…Now, the following task is how to select the desired salient regions from the aforementioned obtained salient regions, and we think that there are two common solutions: one is using some discrimination methods, and the other is simply considering the salient regions with strong scattering centers to be the target regions if the image scene is not very complex. As for the former methods, the goal of discrimination is to reject the regions containing natural-clutter false alarms while accepting the regions containing real targets, but these methods often need a lot of samples to be trained for target discrimination [35]; this is another important and complicated task in a SAR ATR system.…”