SAR target identification method based on feature dimension reduction

Abstract

The invention discloses an SAR target identification method based on feature dimension reduction, and relates to research on a target identification method in SAR image automatic target identification. The method includes the steps that firstly, an SAR training image set is preprocessed, and a training sample set is established; secondly, a linear model is established for the training sample set, an optimal dimension reduction matrix is acquired, dimension reduction is conducted on the training sample set, a dimension-reduced training sample set is obtained, and a probability density function of the dimension-reduced training sample set is acquired; thirdly, SAR test images are preprocessed, and test samples are established; fourthly, dimension reduction is conducted on the test samples through the optimal dimension reduction matrix, dimension-reduced test samples are obtained, the likelihood probability of the dimension-reduced test samples is acquired, and the dimension-reduced test samples are identified through a Bayes classifier. The method mainly achieves the purpose of SAR target identification under the small sample condition, compared with the prior art, the identification performance of the method is obviously improved, and the method can be used for SAR target identification.

Description

Technical field [0001] The invention belongs to the field of radar automatic target recognition, and relates to the study of a target discrimination method in SAR image automatic target recognition, in particular to a SAR target discrimination method based on feature reduction. Background technique [0002] Synthetic Aperture Radar (SAR) is an active sensor that uses microwaves to perceive, and can perform all-weather, all-weather reconnaissance on targets of interest. In recent years, SAR has become an indispensable means of military reconnaissance, and SAR image automatic target recognition technology has also become a hot topic in domestic and foreign research. [0003] The Lincoln Laboratory of the United States proposed a three-level processing flowchart for automatic target recognition of SAR images and is widely used. The process consists of three basic stages: detection, identification, and classification. First, detect the potential target pixels in the entire SAR image ...

AI Technical Summary

Problems solved by technology

Theoretically, the proposed features can represent the scattering difference between the target and the clutter, and can be used for identification. However, in practice, it is found that if all the features are used for target identification, there will be some problems as follows: (1) individual features Because the influence of noise and other factors is invalid for identification; (2) Too many features can easily cause redundancy of mutual information; (3) Too many features will increase the amount of calculation and even lead to the disaster of dimensionality

In practice, especially when the training set is limited, it is inaccurate to estimate these statistics directly in the high-dimensional space, and the final dimensionality reduction matrix is ​​not necessarily optimal, which will affect the final identification performance.

Images