【作者】 陈磊；

【导师】 安良；

【作者基本信息】 东南大学 ， 信号与信息处理， 2023， 硕士

【摘要】 被动声呐具有探测距离远、隐蔽性好等优点,是水声目标探测的主要手段之一。传统的被动声呐是由声呐员在预成多波束宽带能量图上选取峰值位置对应的波束作为潜在目标波束,然后由计算机对潜在目标波束所在方位进行跟踪,得到目标跟踪波束信号,继而进行目标特征提取和判别。然而,在强干扰背景下,宽带能量较弱的目标(以下简称“弱目标”)在多波束宽带能量图上往往得不到凸显,对弱目标的跟踪和判别变得愈发困难。本文针对强干扰背景下弱目标检测需求,利用目标辐射噪声特征的稳定性和可辨识性,开展了基于辐射噪声高维特征联合的检测方法研究,对现有被动声呐目标检测框架进行改进,提升了对弱目标的检测能力。论文的主要工作和创新如下:首先,对现有被动声呐目标探测面临的问题进行深入分析,讨论了特征提取前置的被动声呐探测方法的可行性。针对特征提取前置后,传统目标特征提取依赖的跟踪波束信号缺失问题,提出一种基于高维特征联合的全空域目标检测框架。该框架直接对被动声呐阵列信号处理得到的预成多波束信号进行多种特征提取,分别利用各特征进行目标检测,再进行多特征联合检测判决,输出目标检测结果。其次,针对基于高维特征联合的全空域目标检测框架中的单特征检测问题,设计了基于后验概率预测的检测器。该检测器利用最大后验概率准则,对各特征的后验概率分布进行建模,利用特征样本对特征空间的后验概率分布进行估计。基于估计的后验概率分布来计算新样本的预测后验概率,根据设定的门限实现对目标的检测。通过对预成多波束信号提取得到的辐射噪声线谱、梅尔频率倒谱系数和功率谱熵三种辐射噪声信号特征,对检测器进行了仿真验证。再次,针对基于高维特征联合的全空域目标检测框架中的多种高维特征联合检测问题,构建多检测器输出结果联合判决的高维特征联合检测模型。基于证据理论,提出一种信任度映射的Dempster合成规则,该规则依据先验知识对不同证据分配不同的信任度映射函数,再对映射结果进行Dempster合成,实现对多种特征检测结果的联合,有效提升了目标检测率,同时降低干扰带来的虚警。最后,利用仿真和海试数据对本文提出的检测方法进行验证。结果表明,本文提出的联合检测方法能够在强干扰背景下,对预成多波束能量检测失效的弱目标实现成功检测;相对单一特征检测,多种高维特征的联合能够有效提升检测的准确性。更多还原

【Abstract】 Passive sonar has the advantages of long detection range and good concealment,making it one of the main means of underwater target detection.In the traditional procedure of passive sonar system,the sonar operator firstly selects the beam corresponding to the energy peak on the BTR(Bearing-Time Records)as the potential target beam,and then tracks the direction of the potential target beam to obtain the target signal,lastly extract features for identification.However,in the presence of strong interference,targets with weak broadband energy(hereinafter referred to as ”weak targets”)are often not highlighted on the BTR,making it increasingly difficult to track and identify weak targets.In this paper,the demand for weak target detection in the strong interference background is focused.By utilizing the stability and distinguishability of target radiated noise characteristics,a detection method based on high-dimensional feature fusion of radiated noise is proposed to improve the detection performance of weak targets within the existing passive sonar detection framework.The main work and innovation of this paper are as follows:Firstly,an in-depth analysis of the problems faced by existing passive sonar system is conducted,and the feasibility of extracting features before energy detection is discussed.Aiming at the problem of missing tracking beam signals that are relied upon for feature extraction in traditional detection procedure,a full-space target detection framework based on high-dimensional feature fusion is proposed.This framework directly extracts multiple features from the preformed multi-beams obtained through the passive sonar array signals,and uses each feature for target detection.Then,the fusion of multi-feature detection results is made to output the target detection results.Secondly,for the problem of single feature detection in the full-space detection framework based on high-dimensional feature fusion,a detector based on posterior probability prediction is designed.The detector estimates the posterior distribution probability of each feature space using feature samples by MAP(Maximum A Posteriori)criterion.Then the predicted distribution of new samples based on the estimated posterior distribution is calculated,and the target is detected by a given threshold.The detector is simulated and validated by extracting three types of radiated noise signal features,namely,radiated noise line spectrum,MFCC(Mel Frequency Cepstral Coefficients),and power spectral entropy,from pre-computed multi-beams.Thirdly,for the problem of multi-features fusion detection in the full-space detection framework,a high-dimensional feature fusion model with multiple detector output results is constructed.Based on D-S theory(Dempster-Shafer Theory),a Dempster rule with trust degree weighted mapping is proposed.This rule assigns different trust degree mapping functions to different evidence based on prior knowledge,and then performs Dempster rule on the mapping results to achieve joint detection of multiple features,effectively improving the target detection rate while reducing false alarms caused by interference.Finally,the detection method proposed in this paper is verified by simulation and sea trial data.The results show that in the context of existing strong interference,the features fusion method proposed in this paper can successfully detect weak targets that fail to be detected by energy detection.Compared with single feature detection,the multiple high-dimensional features fusion can effectively improve the accuracy of detection.更多还原