计算机工程 ›› 2019, Vol. 45 ›› Issue (5): 143-148,154.doi: 10.19678/j.issn.1000-3428.0050465

• 人工智能及识别技术 • 上一篇    下一篇

基于深度自编码多维特征融合的慢动目标检测

张文涛a,许治国a,郑霖b,杨超b   

  1. 桂林电子科技大学 a.电子工程与自动化学院; b.广西无线宽带通信与信号处理重点实验室,广西 桂林 541004
  • 收稿日期:2018-02-09 出版日期:2019-05-15 发布日期:2019-05-15
  • 作者简介:张文涛(1976—),男,教授、博士、博士生导师,主研方向为雷达信号处理、光电检测技术;许治国,硕士研究生;郑霖,教授、博士;杨超,博士研究生。
  • 基金项目:

    国家自然科学基金(61371107,61565004,61571143);国家科技重大专项“高精度位置测量系统在线标定技术研究”(2017ZX02101007-003)。

Slow Moving Target Detection Based on Multi-dimensional Feature Fusion Using Deep Autoencoder

ZHANG Wentaoa,XU Zhiguoa,ZHENG Linb,YANG Chaob   

  1. a.School of Electronic Engineering and Automation; b.Guangxi Key Laboratory of Wireless Wideband Communication and Signal Processing,Guilin University of Electronic Technology,Guilin,Guangxi 541004,China
  • Received:2018-02-09 Online:2019-05-15 Published:2019-05-15

摘要:

针对强杂波环境下慢动目标检测存在的多普勒频移低、杂波干扰强、特征提取困难等问题,提出一种多维特征融合的检测算法。利用时频变换和脉冲压缩解析回波信息,提取目标回波时频域和距离像的特征,将特征串联输入到深度自编码网络中进行融合。深度自编码网络通过自主学习提取目标不同维度的特征,增强多维特征联合检测性能。仿真结果表明,与直接利用单域特征的深度自编码以及利用SVM进行目标检测的算法相比,该算法能有效融合时频域与距离像特征,实现特征互补,提高目标检测的鲁棒性与识别精度。

关键词: 目标检测, 深度自编码, 特征提取, 多维特征融合, 时频变换, 脉冲压缩

Abstract:

To address the problems of low Doppler frequency shift,strong clutter interference,difficult feature extraction in slow moving target detection under strong clutter environment,a target detection method based on multi-dimensional feature fusion is proposed.Firstly,the echo information is analyzed by using time-frequency transform and pulse compression,and the features of the target echo in the time-frequency domain and range image are extracted.Then,the features are connected in series into deep autoencoder network for fusion.Finally,the deep autoencoder network is used to extract different dimensions of target characteristics,and enhance the joint detection performance of multi-dimensional features based on autonomous learning.Simulation results show that compared with the target detection based on single domain feature using deep autoencoder and the target detection using SVM,the proposed method can fuse the features of the time-frequency domain and range image effectively,having advantages of both,and improve the robustness and recognition precision of target detection.

Key words: target detection, deep autoencoder, feature extraction, multi-dimensional feature fusion, time-frequency transform, pulse compression

中图分类号: