W-Band Multi-Aspect High Resolution Range Profile Radar Target Classification Using Support Vector Machines.

Tomasz Jasinski,Irina Antipov,Graham Brooker

doi:10.3390/s21072385

Abstract

Millimeter-wave (W-band) radar measurements were taken for two maritime targets instrumented with attitude and heading reference systems (AHRSs) in a littoral environment with the aim of developing a multiaspect classifier. The focus was on resource-limited implementations such as short-range, tactical, unmanned aircraft systems (UASs) and dealing with limited and imbalanced datasets. Radar imaging and preprocessing consisted of recording high-resolution range profiles (HRRPs) and performing range alignment using peak detection and fast Fourier transforms (FFTs). HRRPs were used because of their simplicity, reliability, and speed. The features used were fixed-length, frequency domain range profiles. Two linear support vector machine (SVM)-based classifiers were developed which both yielded excellent results in their general forms and were simple to implement. The first approach utilized the positive predictive value (PPV) and negative predictive value (NPV) statistics of the SVM directly to generate target probabilities and consequently determine the optimal aspect transitions for classification. The second approach used the Kolmogorov–Smirnov test for dimensionality reduction, followed by concatenating feature vectors across several aspects. The latter approach is particularly well-suited to resource-constrained scenarios, potentially allowing for retraining and updating in the field.

Highlights

It is well-known that radar target imaging is heavily aspect dependent
There are many options for performing feature selection for support vector machine (SVM) classification such as F-score and random forests [44], though we have used the Kolmogorov–Smirnov (K–S) test; this is a nonparametric approach that is consistent with the data-driven methodology employed in this paper
The data consisted of high-resolution range profiles (HRRPs) and were preprocessedCbaylcpuleartfeodrmistirnibgutriaonngfeunacltiigonnmfoernetacahndtartgaekti:nFgn(xF)FTs, with all classifications carried out in the fCreaqlcuuelantecyK–dSosmtaatiisnti.cW: Dhn,mile this data were sufficient for classification, the datasets later provenddtofobr e highly imbalanced and too small for an ideal investigation of claDsestiefircmaitnieonuampbperrooafcfheeatsu. rHesotwo uesveer, these shortcomings turned out to be fortuitous in that tCheonycfaotecnuasteedfeaantudressteinetroefdearteusreeavrecchtotro: wx ards particular solutions

Summary

Introduction

It is well-known that radar target imaging is heavily aspect dependent. Target classification performance can be vastly improved by selecting and combining information from different aspects [1,2]. While multiaspect radar classification have been extensively researched, it is the application of existing multiaspect radar imaging techniques to millimeter-wave radar techniques that has the potential to revolutionize how radar systems are used. Of particular relevance are applications involving multiple unmanned aircraft systems (UASs). Millimeter-wave radar is well-suited to short-range (i.e., 5–40 km) imaging in clear weather only, primarily due to relatively high atmospheric absorption. Millimeter-wave radar allows small, cheap and high-resolution (in terms of angle, range and Doppler) radar systems to be developed for niche applications using conventional radar imaging techniques

Objectives

Findings

Discussion

Conclusion