海杂波背景下微弱信号特征提取及检测方法研究

The weak signal detection under sea clutter background is the research emphasis in the field of radar signal processing. Traditional detection methods have the problems such as low precision, poor generalization and worse real-time. This project combines with the empirical mode decomposition, fractal and stochastic resonance, which studies the weak signal detection method and the intrinsic physical characteristics of sea clutter under the deep learning framework. In addition, the weak signal detection model for complex sea condition is set up, which can improve performance of weak signal detection under different sea condition. The main research contents include:①the noise problem of mixed chaotic signal(contain noise and weak signal) are analysed, the denoising encoder of deep learning is improved, and the pretreatment method in different condition of sea clutter is researched;②the energy migration of the limited boltzmann machine in stochastic resonance system is researched, a new model of deep learning energy is proposed, and the detection effects of different dimension oscillator are compared, which improves the training efficiency;③the computational complexity of deep learning is reduced combined with the fractal theory, the relations of fractal characteristics and different sea condition signals in different scales, time and frequency are studied, which is the discriminant standard to improve the operation efficiency of weak signal detection;④the proposed method is validated and optimized by the observed sea clutter data. Through these researches, the small target positioning of the sea surface and hidden object detection are achieved, which has important theoretical significance and practical application value.

海杂波背景下微弱信号检测是雷达信号处理领域的研究热点，传统检测方法存在精度低、泛化性差和实时性欠佳的问题。本项目结合经验模态分解、随机共振和分形等理论研究海杂波的内在物理特性，并统一在深度学习框架内，针对复杂海情，建立海杂波背景下微弱信号检测模型，提高不同海情下的微弱信号检测性能。主要研究内容包括：①研究混沌混合信号（包含噪声和微弱信号）的噪声干扰问题，分析微弱信号经验模态分解分量与背景噪声幅度的对应关系，提出针对不同海情信号的自适应预处理方法；②研究复杂海情条件下海杂波在随机共振系统中的能量迁移情况，分析不同维数振子的检测性能，建立符合海杂波特性的能量模型，提高检测效率；③研究不同海情、高低尺度和时频域条件下分形特征参量的变化特性，对海情进行分类，提高微弱信号检测精度；④采用实测海杂波数据，验证和优化所建立的信号检测模型。本项目研究对海面小目标的检测具有重要的理论意义和实际应用价值。

海杂波背景下微弱信号检测是雷达信号处理领域的研究热点，传统检测方法存在精度低、泛化性差和实时性欠佳的问题。依据项目计划书要求，我们结合经验模态分解、随机共振和多尺度分形理论研究海杂波的内在物理特性，针对复杂海情，利用优化的浅层学习和深度学习网络建立了海杂波背景下微弱信号检测模型，提高了不同海情下的检测性能。研究混沌混合信号（包含噪声和微弱信号）的噪声干扰问题，对微弱信号进行经验模态分解，分析分解后的固有模态分量与背景噪声幅度的对应关系，提出了针对不同海情信号的自适应预处理方法；研究复杂海情条件下海杂波在随机共振系统中的能量迁移情况，分析系统参数对待测信号和噪声的影响，比较一维朗之万方程与二维Duffing混沌振子的检测性能，建立了符合海杂波特性的能量模型；以分形特征量Hurst指数为研究对象，从时间和空间的角度，研究了不同海情、高低尺度和时频域条件下海杂波分形特征参量的变化特征，对海情进行分类。用优化的支持向量机、回声状态网络、核极限学习机、深度置信网络，以及长短期记忆网络等模型学习海杂波内部的特征，建立预测模型，从预测误差中检测出微弱目标信号，采用实测海杂波数据，验证和完善所建立的信号检测模型，完成了混沌海杂波背景下微弱信号检测的任务。构建了海杂波信号预处理-预测模型构建-微弱目标检测的研究体系。随着研究不断深入，在信噪比为-104.2473dB时，预测信号的均方根误差低至0.0001463，检测时间减少了47%。这有效地提高了检测精度，满足实时性检测的需求，对海面小目标的识别和海面安全监测，具有重要的理论和实用价值。