电子测量与仪器学报

毕雪雯,行鸿彦.结合优化特征模态分解与谱熵特征的海面小目标检测[J].电子测量与仪器学报,2025,39(12):115-128

结合优化特征模态分解与谱熵特征的海面小目标检测

Sea-surface small target detection combining optimized feature modedecomposition and spectral entropy feature

DOI：

英文关键词:sea clutter target detection feature mode decomposition multiscale envelope spectrum entropy deep extreme learning machine

基金项目:国家自然科学基金（62171228）项目资助

作者	单位
毕雪雯	南京信息工程大学电子与信息工程学院南京210044
行鸿彦	1.南京信息工程大学电子与信息工程学院南京210044；2.南通理工学院电气与能源工程学院南通226001

Author	Institution
Bi Xuewen	School of Electronics and Information Engineering, Nanjing University of Information Science & Technology, Nanjing 210044, China
Xing Hongyan	1.School of Electronics and Information Engineering, Nanjing University of Information Science & Technology, Nanjing 210044, China;2.School of Electrical and Energy Engineering, Nantong Institute of Technology, Nantong 226001, China

摘要点击次数: 500

全文下载次数: 466

中文摘要:

针对海杂波背景下海面小目标检测中特征提取复杂、检测率低等问题,分析了海杂波与目标回波的数据特征,研究了特征模态分解（feature mode decomposition，FMD）在海杂波信号处理中的适用性,提出了一种结合优化特征模态分解与谱熵特征的海面小目标检测方法。使用结合互生生物搜索（symbiotic organism search，SOS）与粒子群优化（particle swarm optimization，PSO）的混合智能算法进行参数寻优,利用多尺度包络谱熵（multi-scale envelope spectrum entropy，MSESEn）提取信号特征,构建了虚警可控的深度极限学习机分类器模型（deep extreme learning machine，DELM）。将归一化之后的特征数据输入模型中,通过对比预测值与决策阈值的大小实时更新判决门限,实现了控制模型的虚警率,提高了算法的可靠性与检测效率。采用IPIX数据集进行验证,在HV极化方式下检测率平均提高了18%,说明了所提方法性能优于傅里叶变换与三特征检测方法。

英文摘要:

Aiming at the problems of complex feature extraction and low detection rate in sea-surface small target detection under the background of sea clutter, the data characteristics of sea clutter and target echoes are analyzed, and the applicability of feature mode decomposition (FMD) in sea clutter signal processing is studied. Based on this, a sea-surface small target detection method combining optimized feature mode decomposition and spectral entropy features is proposed. A hybrid intelligent algorithm combining symbiotic organism search (SOS) and particle swarm optimization (PSO) was used for parameter optimization, and multi-scale envelope spectrum entropy (MSESEn) was used to extract signal features. A deep extreme learning machine (DELM) classifier model with controllable false alarm is constructed. The normalized feature data is input into the model, and the decision threshold is updated in real time by comparing the predicted value and the decision threshold. The false alarm rate of the control model is realized, and the reliability and detection efficiency of the algorithm are improved. The IPIX data set is used for verification, and the detection rate is improved by 18% on average under HV polarization mode, which shows that the performance of the proposed method is better than that of Fourier Transform and three-feature detection method.

查看全文查看/发表评论下载PDF阅读器