Polarization Euler feature fusion deep learning-based marine target identification method

Abstract

The invention discloses a polarization Euler feature fusion deep learning-based marine target identification method. The method comprises the following steps of S1, analyzing a polarization scatteringmechanism of a marine artificial target; S2, based on the analysis result of the step S1, constructing a corresponding offshore geometry polarization distance matrix data set, and performing multi-polarization feature extraction on the samples of the polarization distance matrix data set; S3, based on the step S2, training a multi-polarization feature fusion deep learning model; and S4, carryingout identification, test and verification on the multi-polarization feature fusion deep learning model in the step S3 by adopting the offshore artificial target. The method has the advantages that themethod starts from the basic polarization scattering mechanism of the marine target, utilizes HRRP and polarization information to realize the target identification based on the deep learning fusion,eliminates the adverse effects caused by orientation sensitivity, improves the radar target identification precision, and has the advantages of high detection rate, low false alarm rate, flexible expansion application and full-automatic processing process.

Description

technical field [0001] The invention relates to the technical field of radar target recognition, in particular to a sea target recognition method based on polarization-Eulerian feature fusion deep learning. Background technique [0002] The polarization scattering feature can effectively characterize the symmetry, surface roughness, and orientation of the target. It is another important feature besides the amplitude and phase information of the radar echo and the Doppler frequency shift. This feature has great application potential in anti-clutter interference, target signal filtering and enhancement, target detection and recognition, etc. Radar High Resolution Range Profile (HRRP) is an important eigenvector of the scattering source distribution of the target in the direction of the radar line of sight obtained by using broadband radar signals. It is widely used because of its advantages of easy acquisition and strong real-time performance. for radar target recognition. H...

AI Technical Summary

Problems solved by technology

This method has the advantages of retaining relatively complete information and learning features layer by layer; however, the DBN model is directly trained based on complex scene SAR data, without starting from the basic scattering mechanism of the research object, so the extended applicability of the recognition model is limited

[0005] In terms of papers, Baird et al. from the University of Massachusetts Lowell analyzed and discussed in detail the advantages of the characteristics of scattering images in the Euler parameter space in radar target recognition in their papers, and in their paper published in 2007 "Development and assessment of a complete ATR algorithm based on ISAR Eulerimagery" proposes a fully automatic radar target algorithm, by comprehensively utilizing the Euler parameters optimized by the azimuth persistence of inverse synthetic aperture radar (ISAR) images, compared with traditional polarization parameter greatly improves the performance of radar target recognition; however, this method only uses the characteristic information in the Euler parameter space, ignoring the beneficial contribution of the traditional polarization amplitude and phase characteristic parameters to target recognition, and the target recognition still uses the It is a traditional template matching recognition method, which needs to establish a complex target scattering image feature library

[0006] In their 2018 paper "General Purpose PolSAR Classifier with Convolutional Neural Network", Suo Li et al. of Fudan University proposed a convolutional neural network (CNN)-based polarization SAR object classification algorithm, by directly extracting the polarization coherence matrix parameters, and normalized to construct a 6-channel polarization feature vector for CNN model training and ground object classification and recognition; however, the scattering mechanism of ground objects is significantly different from that of artificial targets in the sea environment, and it is difficult to apply to Maritime target recognition; and the CNN model is directly trained based on the SAR data measured in complex scenes, without starting from the basic scattering mechanism of the research object, so the extended applicability of the recognition model is limited

Images