摘要
为了在复杂混沌噪声背景中快速准确提取有用信号,提出基于复杂非线性系统相空间重构理论,采用改进极限学习机(ELM)预测单步误差检测微弱信号的方法。采用改进K均值聚类算法选择最优族作训练集,改进极限学习机选择权值和偏置的方法进一步提高检测的精度和速度,采用Lorenz系统建立了混沌噪声序列的一步预测模型,从预测误差中检测湮没在混沌噪声中的微弱目标信号(包括周期信号和瞬态信号),然后使用加拿大Mc Master大学IPIX雷达数据,在海杂波噪声中提取漂浮物信号作为实验研究。结果表明该方法能够有效检测混沌背景噪声中极微弱信号,同时抑制噪声对混沌背景信号的影响,与径向基函数(RBF)神经网络等传统算法相比,预测精度提升了25%,检测门限提高了-5 d B,同时训练用时减少77.1 s,在实际应用中具有更明显优势。
Aiming at the problem of extracting the useful signal in the complex background of chaotic noise rapidly and accurately, the phase space reconstruction theory based on complex nonlinear system was proposed, and the method of improved Extreme Learning Machine （ELM） was used to predict the single step error and detect the weak signal. The improved K-means clustering algorithm was used to select the optimal family as training set, the improved extreme learning machine chose the weight value and the offset to improve the detection accuracy and speed. The one step prediction model of chaotic noise sequence with Lorenz system was established, and the weak target signals （ including periodic signal and transient signal） that lost in the chaotic noise were detected, then the IPIX radar data of Canada Mc Master University were used, and the floater signal in sea clutter noise was extracted to do the experimental research. The results show that the method can effectively detect the very weak signal in chaos background noise, at the same time, the influence of noise was restrained to the chaotic background signal, compared with the traditional algorithms such as Radial Basis Function （RBF）, the prediction accuracy is increased by 25%, the detection threshold is increased by -5 dB, the training time is reduced by 77.1 s, it has more obvious advantages in practical application.
出处
《计算机应用》
CSCD
北大核心
2015年第3期896-900,共5页
journal of Computer Applications
基金
国家自然科学基金地区科学基金资助项目(61263004)
关键词
混沌噪声
极限学习机
海杂波
微弱信号检测
耗时运算
chaotic noise
Extreme Learning Machine （ELM）
sea clutter
weak signal detection
time-consumingoperation
