Scattering Center Extraction and Recognition Based on ESPRIT Algorithm

Abstract

Inverse Synthetic Aperture Radar (ISAR) generates high quality radar images even in low visibility. And it provides important physical features for space target recognition and location. This thesis focuses on ISAR rapid imaging, scattering center information extraction, and target classification. Based on the principle of Fourier imaging, the backscattering field of radar target is obtained by physical optics (PO) algorithm, and the relation between scattering field and objective function is deduced. According to the resolution formula, the incident parameters of electromagnetic wave are set reasonably. The interpolation method is used to realize three-dimensional (3D) simulation of aircraft target, and the results are compared with direct imaging results. CLEAN algorithm extracts scattering center information effectively. But due to the limitation of resolution parameters, traditional imaging can’t meet the actual demand. Therefore, the super-resolution Estimation of Signal Parameters via Rotational Invariance Techniques (ESPRIT) algorithm is used to obtain spatial target location information. The signal subspace and noise subspace are orthogonal to each other. By combining spatial smoothing method with ESPRIT algorithm, the physical characteristics of geometric target scattering center are obtained accurately. In particular, the proposed method is validated on complex 3D aircraft targets and it proves that this method is applied to most scattering mechanisms. The distribution of scattering centers reflects the geometric information of the target. Therefore, the electromagnetic image to be recognized and ESPRIT image are matched by the domain matching method. And the classification results under different radii are obtained. In addition, because the neural network can extract rich image features, the improved ALEX network is used to classify and recognize target data processed by ESPRIT. It proves that ESPRIT algorithm can be used to expand the existing datasets and prepare for future identification of targets in real environments. Final a visual classification system is constructed to visually display the results.