Simulated database construction method for SAR automatic target recognition

Abstract

This paper mainly deals with the method of building target database for implementing target automatic recognition technology in SAR images. To automatically identify a target, a database for the entire target of interest must be pre-built, consisting of SAR images of the targets for the full azimuth angles, multiple incident angles, polarizations, and the frequency bands of interest. It would be best if the measurement data were available when building the target database, but the database should be built by simulation because it cannot be measured if it is not a target of allies. There are three main types of technologies required to develop the target database construction technology: the first is the acquisition method of a three-dimensional CAD model, the second is the technology for generating the target's SAR image, and the third is the technology for verifying a series of simulation processes by comparing the measured images with the simulated images.

A method for estimating and producing three-dimensional shapes as close as possible to reality is proposed by using target indirect information such as two-dimensional drawing, photographic data from various angles, and plastic models for the production of three-dimensional precision models for the target when the direct access to the target is not possible. In order to verify how similar the three-dimensional shape of the target produced by the estimation to its original, it is to be compared with the ultra-precision model obtained by laser scanning the actual target by calculating the radar cross section (RCS, Radar Cross Section), two-dimensional Inverse Synthetic Aperture (ISAR), and three-dimensional scattering points for each model as well as the height differences between their silhouettes.

CAD models produced by any method is applied to electromagnetic numerical analysis techniques that can calculate the reflected waves generated when radio waves enter the target, to create target SAR images, which present the SBR (Shooting and Bouncing) techniques using the bistatic method to quickly generate images for targets with a large electrical length in the high frequency bands, satisfying a certain level of accuracy.

In order to use target SAR images generated by simulation as a database for target recognition, verification through comparative analysis with measured images is essential. Among the numerous evaluation methods proposed for Image Quality (IQ) assessment in optical images, the effectiveness of the structural similarity (SSIM) as an image similarity (IS) measure which is well applied to SAR images is shown by presenting the experimental results.