以具有未知循環位移之高解析度距離輪廓做自動目標物識別

Automatic Target Recognition Based on High Resolution Range Profiles with Unknown Circular Range Shift

林昆佑(Kun-Yo Lin)

指導教授 ： 黃正光

元智大學/工程學院/通訊工程學系/碩士(2006年)

https://doi.org/10.6838/YZU.2006.00256

摘要

在本篇論文中，我們提出一個基於高解析度距離輪廓(high resolution range profile, HRRP)之自動飛行目標識別(automatic aircraft target recognition, ATR)方法的架構。此架構可以拆解成下列兩個主要部份。第一部份我們探討如何去產生一個高解析度距離輪廓，其中包含了雷達截面積(radar cross section, RCS)的塑模、模擬、步階頻率波型(step frequency waveform, SFW)的設計、以及如何利用反傅利葉轉換來處理合成的高解析度距離輪廓，並用此藉以建立目標物高解析度距離輪廓的資料庫，以供後續自動目標物識別演算法使用。此外我們假設已知目標物飛行的軌跡、水平角和仰角，並且憑藉著統計分類的技巧來進一步發展自動飛行目標的識別方法。我們提出兩種不同的識別演算法，包含最大事後機率(maximum a posteriori, MAP)，以及平方分類器(quadratic classifier)基於統計特徵的分類器 (feature-based statistical classifier)。我們會先就基礎的平方分類器切入鑑別演算法。而在最大事後機率決策法則中我們是採用訊號空間的觀念。利用格蘭樞密特正交化(Gram-Schmidt orthogonalization ,GSO)步驟去建構出訊號空間，然後將接收到的高解析度距離輪廓投影到訊號空間上。最後，我們可以依據觀測向量與各信號向量間之最小距離做為目標鑑別準則。本文中已包含的模擬結果，可以證明這些近似法的可行性。

關鍵字

自動目標物識別 ； 高解析度距離輪廓 ； 步階頻率波型 ； 訊號空間法則 ； 統計分類

並列摘要

In this thesis, an automatic aircraft target recognition (ATR) framework is presented, which is based on the high resolution range profiles (HRRP) of aircraft targets. This work is divided into two major parts. First, we consider the generation of the HRRP, which includes the modeling and simulation of radar cross section (RCS), the design of step frequency waveform (SFW), and IFFT processing for HRRP synthesis. In practice, a possible circular shift of the received HRRP relative to the template HRRPs in target library may exist. In such a situation, we resort to the statistical classification technique to develop ATR algorithms. We propose two kind of recognition algorithms including maximum a posteriori (MAP) decision rule, and quadratic classifier. In MAP criterion, we use the concept of signal space. we adopt the Gram-Schmidt orthogonalization (GSO) procedure to construct a signal space, and then project the received HRRP onto the signal space. Finally, the target classification can be done in terms of MAP decision rule. Generally speaking, the standard approaches for ATR use the entire range profile as the feature vector. Training the classifier is simply a statistical parameter problem. The estimation of the parameters is based on the observation of the target over a small range of viewing aspects. And the quadratic classifier is the most popular choice. Simulation results are also included to demonstrate the feasibility of these approaches.

並列關鍵字

automatic target recognition ； high resolution range profile ； step frequency waveforms ； statistical classification ； signal space method

參考文獻

[1] D. R. Wehner, High Resolution Radar, second edition,

Pattern Recognition, 2003.

radar models for joint tracking and recognition,”

recognition using sequences of high resolution radar

range-profiles,’’ IEEE Transactions on Aerospace and