针对混沌海杂波背景下的微弱信号检测问题,结合互补集成经验模态分解理论,提出了一种优化的核极限学习机微弱信号检测方法。采用互补集成经验模态分解法将混沌信号分解为一系列固有模态函数,通过核极限学习机对经相空间重构后的各模态函数分别建立预测模型,利用人工蜂群算法对核极限学习机的正则化系数和核参数进行优化,重构预测信号,从预测误差中检测出混沌海杂波背景中的微弱信号(瞬态信号与周期信号)。分别以Lorenz 系统和IPIX 雷达海杂波数据为例进行了仿真,并研究了不同强度的噪声对微弱信号检测的影响。结果表明:该方法可以有效地从混沌背景中检测出微弱目标信号,当系统不存在噪声时,Lorenz 系统得到的均方根误差0. 000 000 12 (-118. 959 1 dB)比传统极限学习机方法的均方根误差0. 001 345 08(-80. 154 7 dB)降低了4 个数量级;若SNR逸0 dB,噪声对微弱目标信号检测的影响可以忽略;但当SNR<-7 dB,则无法检测出微弱信号。

According to the weak signal detection in chaotic sea clutter, a new detection method of optimized kernel extreme learning machine(KELM) based on complete ensemble empirical mode decomposition(CEEMD) theory is proposed. By CEEMD, chaotic signals containing the weak target can be decomposed into a series of intrinsic mode functions(IMFs), and then the prediction models of each IMF reconstructed by phase space are established by KELM. Using artificial bee colony algorithm to continuously optimize the regularization coefficient and the kernel function parameter of KELM, after reconfiguration of predictive signal, the weak signals (transient signal and periodic signal) submerged in chaotic sea clutter background can be detected from the prediction error. Lorenz attractor and the data from the IPIX radar sea clutter database are used in the simulation, and the influence of noise at different intensities on weak signal detection is investigated in depth. The results show that the proposed method can effectively detect the weak target from chaotic signal background. When there is no noise in the system, by using the proposed method, the root mean square error can be reduced by four orders of magnitude, reaching 0. 000 000 12(SCR = -118. 959 1 dB), while the conventional ELM can only reach 0. 001 345 08 under the condition of SCR =-80. 1547 dB. In addition, the noise influence on the target detection performance can be ignored if SNR≤0 dB.

TN957.52

国家自然科学基金资助项目(61671248);国家重点研发计划资助项目(2018YFC1506102);江苏省重点研发计划资助项目(BE2018719);江苏省研究生科研创新计划资助项目(KYCX18_1038);江苏省“信息与通信工程冶优势学科计划资助课题