Improvement of computation time and accuracy of 3-D scattering center extraction algorithm using the shooting and bouncing ray technique

Abstract

3-D inverse synthetic aperture radar (ISAR) image and 3-D scattering center model successfully char-acterizes a complex radar target signature. The model-based automatic target recognition system is one of the prominent applications. An unknown target can be identified through matching the radar measurement against candidate radar target signatures extracted from computer-aided design (CAD) models. Therefore, improvement in the computation time and accuracy of 3-D scattering center extraction algorithm is essential to achieve precise target identification in near real-time. Modified 3-D scattering center extraction algorithm from target CAD model using the SBR technique is proposed to enhance computation time and accuracy. The proposed algorithm generates a 3-D ISAR image from which a 3-D scattering center is then extracted using the CLEAN algorithm. In the first stage, a 3-D ISAR image is generated using the SBR technique. In general, the shooting and bouncing ray (SBR) technique is used to calculate the radar cross-section (RCS) of an electrically large and complex-shaped target. Incident rays, which are shot into the target, are traced based on the laws of geometrical optics (GO) in the SBR tech-nique. Since the introduction of a fast Fourier transform (FFT)-based convolution scheme between impulse trains, which include the last hit point information from the SBR technique, and the known ray spread func-tion, remarkable progress has been made in reducing the computation time. However, interpolation, which should antecede the FFT operation, causes the distortion of the image. The present work is an attempt to gen-erate an accurate image without loss of computation time and memory efficiency. By removing dominant interpolation error contribution from the previous ray spread function, the distortion is prevented without the use of the high oversampling, which leads to increases in computation time and memory requirements. In ad-dition, interference from the side lobe level is alleviated using the Chebyshev window-based ray spread func-tion at the expense of the main lobe width. In the second stage, 3-D scattering center model is extracted from the generated 3-D ISAR image by using the modified CLEAN algorithm. The conventional CLEAN algorithm extracts the strongest peak itera-tively based on the assumption that the scattering centers are isolated. In a realistic target, however, both inter-ferences from the closely spaced point targets and additive noise distort the extraction process. Specifically, due to the contiguous target problem, actual scattering centers are buried under numerous main lobes and side lobes of the other point targets and additive noise. Therefore, the matched filtering of which impulse response is the known point spread function was utilized. In the proposed CLEAN, a point scatterer most correlated with the point spread function is extracted to improve the extraction accuracy. The proposed CLEAN algo-rithm optimally enhances the extraction accuracy in the presence of the massive distortions, which can be modeled white Gaussian noise. In addition, interference from the side lobe level is alleviated using the Cheby-shev window-based point spread function at the expense of the main lobe width. The previous comprehensive 3-D extraction algorithm (including image formation and scattering cen-ter extraction) circumvents the direct extraction by separating the dimension into three 1-D cases. The key reason is that the 3-D image formation using the previous FFT-based convolution is unable to be practically implemented given the constraints of memory. Bhalla proposed separating the 3-D image formation into three 1-D range profile generations. Because the 3-D image is calculated only at candidates (i.e., intersections of the 1-D results), the memory requirement is considerably mitigated. However, computation time increases signifi-cantly due to the ray-sum process in the 3-D image formation, which involves millions of launched rays. Fur-thermore, the original CLEAN is used at three 1-D scattering center extractions and one 3-D scattering center extraction. The extraction accuracy is thus deteriorated. To alleviate the computation time and accuracy problem, direct and simple implementation based on the proposed 3-D image formation scheme and the pro-posed CLEAN algorithm is proposed. The extraction algorithm is accelerated by replacing the ray-sum calcu-lation with the FFT-based 3-D convolution. The extraction accuracy is also improved by using the matched filter-based CLEAN algorithm. Numerical simulations of realistic targets are presented to demonstrate the per-formance of the proposed 3-D scattering center extraction algorithm.