【作者】 王成；

【导师】 殷敬伟； 朱广平；

【作者基本信息】 哈尔滨工程大学 ， 水声工程， 2024， 博士

【摘要】 对水下目标的正确识别是建立在有效的特征提取之上,目标回波亮点是目前研究最多的水下目标特征之一,因此亮点及亮点的增强和提取成为了主动声呐识别的关键。本文以主动声呐获得的回波为研究对象,以从回波中有效增强和提取亮点为目标,重点围绕提升亮点的分辨力、混响抑制等关键问题和技术难点展开研究,为主动声呐识别提供关键技术基础。本文的主要贡献和创新性研究成果可总结为如下内容:首先,研究了基于分数阶反卷积的高分辨力回波亮点增强技术。目标回波亮点参数反映了目标特征参数,因此对目标亮点参数估计的越准确,对目标识别越有利。为了有效增强亮点的分辨力,从时频分析的基本定义入手,理论分析了影响时频分辨力的因素。并采用短时分数阶傅里叶变化替代短时傅里叶变换,以去除信号调频率对分辨力的影响。然后针对窗函数引起的时频域模糊,推导了目标回波与窗函数在分数阶域的卷积公式,并引入反卷积理论去除了窗函数的限制,在短时分数阶傅里叶变换的基础上进一步提升了时频分辨力。数值仿真和外场试验结果验证了基于分数阶反卷积的目标亮点增强方法可以有效提高亮点的分辨力,且其对背景干扰具有一定抑制效果。其次,研究了时频域抗混响目标回波亮点增强技术。主动声呐目标亮点信号会受到严重的混响干扰,针对抑制混响增强目标亮点强度的问题,本文对相应的抗混响技术进行了研究。一方面,从理论分析了混响与目标回波在时频域上的能量分布差异,得到了混响与目标回波在时频域上分别满足低秩与稀疏性这一结论,并据此使用基于低秩稀疏分解的时频滤波方法实现了混响的抑制。通过数值仿真实验和试验数据首先验证了混响和目标回波信号在时频域的分布特性,并且验证了时频滤波方法可以有效地实现混响与目标回波在时频域上的分离,提升信混比。另一方面,对声学参量阵窄波束发射实现空域滤波的抗混响技术进行了研究。参量阵的发射信号预调制是参量阵应用的关键技术,对此依据“Berktay远场解”研究了Kalman递推滤波调制算法,从信号处理的层面上提升了参量阵的性能。数值仿真结果和实验数据结果验证了,该算法可以降低参量信号的失真度,从而有效提升声学参量阵目标亮点增强的性能。再次,研究了混响背景下波束域目标回波亮点增强技术。理论和试验都证明了水下复杂目标的亮点特征会随角度的变化而变化,仅仅通过单一角度的目标回波亮点很多情况下并不能成为正确目标识别的依据。针对该问题,为获取目标亮点的时空分布,本文开展了波束域的目标亮点强度增强研究。波束域的混响会严重干扰亮点,甚至会形成虚假亮点,对此研究了多帧波束域数据联合处理的方法,依据多帧数据中混响潜在的低秩结构,将混响背景下的波束域(图像)目标亮点增强问题转化成了低秩矩阵恢复问题。基于此提出了基于迭代加权最小二乘法的波束域抑制混响的二维回波亮点增强技术,试验数据验证了该方法可以实现目标亮点与混响的有效分离,提升信混比,但是该方法的收敛性不稳定。因此提出了基于交替乘子法的亮点增强方法,该方法从凸优化的角度实现了低秩矩阵的恢复,实验数据验证了该方法的收敛性和良好的性能。在贝叶斯框架下提出了基于变分稀疏贝叶斯鲁棒主成分分析的亮点增强方法,解决了参数优化难题。该方法的参数都被视为随机量,并通过先验分布和贝叶斯推理的组合来进行处理,这使得该方法不需要大量的参数调整,可以通过闭式方程自动推断模型中涉及的参数,实验数据验证了该方法的有效性。针对帧间混响非相关部分的抑制问题,在已有的贝叶斯框架下通过对多帧数据中的非低秩部分进行混合高斯建模,去除了帧间非相关混响的干扰,进一步提升了抑制混响增强波束域二维目标亮点的性能。并通过外场实验数据验证了所提的方法的有效性。最后,研究了空间投影域目标回波亮点提取技术。本文受合成孔径声呐启发,提出了基于后向投影算法的空间投影域目标回波亮点提取技术。通过后向投影算法将多站位接收的目标回波映射到投影域中,实现多站位目标亮点在同一平面的映射。针对回波包络杂乱不平滑,后向投影图像尖锐部分影响回波亮点这一问题,提出了基于一维Kalman滤波的滤波器的平滑方法。数值仿真和试验数据验证了本文提出的方法可以获取到相较于单一角度更多的目标散射信息,实现了目标亮点的空间信息增强。更多还原

【Abstract】 Efficient augmentation of features is the foundation for accurate underwater target recognition.Presently,one of the most researched underwater target features is target echo highlights.Therefore,the enhancement of highlights and highlights has become the key to active sonar recognition.This thesis takes the echoes obtained from active sonar as the research object,with the goal of effectively enhancing highlights from the echoes.Research will focus on key issues and technical difficulties such as improving the resolution of highlight enhancement and reverberation suppression,providing a key technical foundation for active sonar recognition.In particular,the following succinctly describes the primary contributions and novel research findings of this thesis:Firstly,a fractional deconvolution-based high resolution target highlight enhancement method was examined.The more accurately the target highlight parameters are estimated,the more beneficial it is for target recognition,since the target parameters are reflected in the target echo.This thesis begins with the basic statement of time-frequency analysis and theoretically analyzes the factors that affect time-frequency resolution.And to eliminate the impact of signal frequency modulation rate on resolution,employ short-time fractional Fourier transform rather than short-time Fourier transform.Next,this thesis constructs the convolution formula between the window function and the target echo in the fractional order domain,which addresses the ambiguity in the time-frequency domain produced by the window function.Furthermore,the window function restriction is eliminated with the inclusion of deconvolution theory,further enhancing time-frequency resolution with the use of short-time fractional Fourier transform.The efficacy of the target highlight enhancement method based on fractional deconvolution in improving highlight resolution has been confirmed by the outcomes of numerical simulation studies and field investigations.Additionally,it inhibits background interference to some extent.Secondly,the technique for enhancing highlights of target echoes in the time frequency domain during suppress reverberation was studied.Not only does resolution restrict target highlight performance,but strong reverberation interference also does.This thesis investigates the corresponding suppress reverberation approaches in answer to the problem of improving target highlights and suppressing reverberation.One may argue that reverberation and target echo satisfy low rank and sparsity in the time-frequency domain,respectively,based on a theoretical analysis of the differences in energy distribution between the two in the timefrequency domain.On the basis of this,a low rank sparse decomposition-based time-frequency filtering technique for reverberation suppression was presented.The time-frequency domain distribution properties of target echo signals and reverberation were confirmed using experimental data and numerical simulation studies.Furthermore,it has been established that the time-frequency filtering technique may successfully separate target echo from reverberation in the time-frequency domain,increasing the signal-to-reverberation ratio.On the other hand,in order to accomplish spatial filtering,studies have been done on suppress reverberation technology for acoustic parametric array narrow beam transmission.Pre modulation of transmission signals in parametric arrays is a key technology for the application of parametric arrays.Based on the Berktay far-field solution,a Kalman recursive filtering modulation algorithm was proposed to improve the performance of the parametric array from the perspective of signal processing.The numerical simulation results and experimental data results have verified that this algorithm can reduce the distortion of parametric signals,thereby effectively improving the performance of target highlight enhancement.Then,the beam domain target echo highlight enhancement technology under reverberation background was studied.Both theory and experiment have proven that the highlight characteristics of complex underwater targets vary with angle changes.In many cases,the highlights from a single angle cannot serve as the basis for target recognition.In order to enhance two-dimensional target echo highlights and obtain the spatiotemporal distribution of target highlights,this thesis conducts research on beam domain target highlight enhancement.The reverberation in the beam domain can seriously interfere with target highlights and even form false target highlights.This thesis adopts the method of joint processing of multi frame beamspace data,and based on the potential low rank structure of reverberation in multi frame data,transforms the problem of enhancing two-dimensional target highlights under reverberation background into a low rank matrix recovery problem.Based on this,a twodimensional echo highlight extraction technique for beam domain reverberation suppression using iterative weighted least squares method is proposed.The experimental data has verified that this method can effectively achieve the separation of target highlights and reverberation,improve the signal-to-reverberation ratio,but the convergence of this method is difficult to guarantee.Therefore,a highlight enhancement method based on alternating multipliers is proposed,which achieves the restoration of low rank matrices from the perspective of convex optimization.The experimental data verified the convergence and good performance of the method.In response to the problem that the above method heavily relies on parameter selection,a highlight enhancement method based on variational sparse Bayesian robust principal component analysis is proposed in the Bayesian framework.The parameters of this method are treated as random variables and processed through a combination of prior distribution and Bayesian inference.This means that the method does not require a large amount of parameter adjustment and can automatically infer the parameters involved in the model through closed equations.The experimental data validated the effectiveness of this method.Furthermore,to address the issue of inter frame uncorrelated components affecting the performance.Under the existing Bayesian framework,mixed Gaussian modeling is applied to non low rank parts of multi frame data to remove interference from inter frame uncorrelated reverberation.Further improved the performance of enhancing two-dimensional target highlights in the beam domain.The effectiveness of the proposed method was verified through field experimental data.Finally,this thesis is inspired by synthetic aperture sonar and proposes a spatial projection domain target echo highlight extraction technique based on backward projection algorithm.By using the backward projection algorithm,the target echoes received from multiple angles are mapped into the projection domain,achieving the mapping of multi angle target highlights on the same plane.Aiming at the problem of chaotic and unsmooth amplitude spectrum of echo envelope,as well as the presence of sharp parts in backward projection images that affect highlight extraction,a one-dimensional Kalman filtering based approach is proposed αSmoothing methods for filters.Numerical simulation experiments and experimental data have verified that the proposed method can obtain more target scattering information compared to a single angle.更多还原