Target feature extraction method used for synthetic aperture radar

Abstract

The invention belongs to the radar technology field and particularly relates to a target feature extraction method used for a synthetic aperture radar. The target feature extraction method combines peak value area segmentation of an image domain and sparse Bayes learning of a signal domain to perform feature extraction, segments a strong scattering area through an PRS algorithm in order to avoid an effect on an SBL algorithm by strong scattering points, uses the SBL algorithm to autonomously study to extract weak scattering featurepoints to compensate insufficient segmentation on an weak target by the PRS algorithm based on the image domain. The target feature extraction method uses the advantages of the PRS algorithm and the SBL algorithm to show a good effect in weak object feature extraction and is less in adjustable parameters and strong in robustness. After the PRS algorithm segments the strong scattering area, the number of dimensions of a measurement matrix is reduced, which reduces computation burden and improves calculation efficiency. As a result, the target feature extraction method can be widely applied to the SAR object feature extraction field and provides a basis for a follow-up automatic target recognition.

Description

technical field [0001] The invention belongs to the technical field of radar, and relates to a target feature extraction method for synthetic aperture radar (SAR). Background technique [0002] Synthetic Aperture Radar (SAR) target feature extraction is a very important research field in SAR signal processing, and its accuracy directly affects the accuracy of target recognition, so it is of great significance to accurately extract target features. The essence of feature extraction is to map sampled data from high-dimensional space to low-dimensional feature space. The existing SAR feature extraction methods can be divided into two categories: the first category is based on the image domain, through the Nyquist sampling theorem and matched filtering Theory for feature extraction, such as watershed algorithm. This type of algorithm only uses amplitude information but not phase information, so it is easily affected by clutter and noise, which reduces the accuracy of feature ex...

AI Technical Summary

Problems solved by technology

For convex relaxation algorithms, the main idea is to The norm non-convex problem is equivalent to Norm convex problem to solve, such as basis pursuit algorithm (Basis Pursuit, BP), interior point method, etc., such algorithms are computationally complex, and the calculation efficiency is low when used for big data, which cannot meet the real-time requirements; for greedy algorithms, other The main idea is to gradually approach the original signal by obtaining a local optimal solution each iteration, such as Matching Pursuit (Matching Pursuit, MP) and Orthogonal Matching Pursuit (OMP), which are fast in calculation speed. It is easy to implement, but it often needs to know the sparsity of the target scene. The result of feature extraction depends greatly on the setting of sparsity, and it does not consider the spatial structure information of the scene, and is easily affected by noise; while Bayesian algorithms, such as sparse Bayesian The Sparse Bayesian Learning algorithm (Sparse Bayesian Learning, SBL), its main idea is based on the Gaussian prior assumption, can learn and determine all the hyperparameters in the algorithm independently, and obtain a more sparse solution. The algorithm introduces the spatial structure information of the target, It has a certain anti-noise ability and has a smaller computational complexity than the convex optimization algorithm

However, in practice, the CS algorithm is easily disturbed by strong points, which makes the extraction performance of weak scattering features unsatisfactory. At the same time, when the target scene is too large, the dimension of the constructed measurement matrix is ​​too large, which makes the actual processing The amount of calculation is large, the efficiency is low, and it is not convenient for effective application

Images