基于UWB信号的目标识别关键技术研究

发布时间：2018-07-07 17:16

本文选题：超宽带 + 目标识别　；参考：《北京邮电大学》2014年博士论文

【摘要】：超宽带(Ultra-Wideband, UWB)无线通信技术具有穿透叶簇对隐蔽目标进行检测和识别的能力,应用潜力巨大。如何从UWB信号中提取叶簇隐蔽目标的特征并对其进行识别是需要研究的重要关键技术。本文选题来源于国家自然科学基金等项目,具有重要的理论意义和应用价值。本文针对基于UWB信号的目标识别技术进行了深入研究,主要完成了以下具有创新性的研究成果：针对目标特征的有效提取问题,本文提出了基于稀疏表示的目标特征提取算法。首先,构造一个冗余完备字典,该字典的基函数可以从正弦函数、小波函数等变换基中选择得到,也可以从测量目标的UWB信号中学习得到；然后,在该冗余完备字典上求解目标测量UWB信号的稀疏表示；最后,通过对稀疏分解系数进行分析处理来提取目标的稀疏特征。基于叶簇覆盖目标数据集的验证表明,所提取的基于稀疏表示的目标特征具有很好的可分性,能够显著提高目标识别的性能。针对基于支持向量机(Support Vector Machine, SVM)的目标识别性能受该算法参数影响较大,并且传统的SVM参数选择方法易陷入局部极值的问题,提出了两种改进的粒子群优化算法,来优化选择SVM参数。在改进的粒子群优化算法中,算法的控制参数可随着算法的进化自适应调整,并且在每次迭代中分别引入混沌搜索和差分进化搜索过程,从而兼顾全局搜索和局部搜索,以提高算法的收敛速度和搜索性能。进一步提出了基于改进算法优化SVM的叶簇覆盖目标识别方法。实验结果表明,改进的算法收敛速度快、搜索能力强,能有效提高目标识别性能。针对基于SVM的目标识别性能与核函数及其参数的选择有很大关系,本文构造了基于小波核函数的小波支持向量机(WSVM),并提出了一种混合量子粒子群优化算法(HQPSO),来对WSVM的参数进行优化选择。在HQPSO算法中,采用量子比特对粒子位置进行编码,用量子旋转门实现对粒子最优位置的搜索；并且每次迭代过程中包含一个局部增强搜索过程,通过量子门旋转机制,可加速每个粒子朝着当前代局部最优与全局最优解的方向进行进化。实验验证表明,借助于HQPSO算法出色的搜索能力,可以有效提高基于WSVM的目标识别性能。针对多种场景下多类目标的识别问题,提出了一种基于稀疏表示的多场景下多目标的识别方法。该方法分为两个步骤：首先,从每类目标在不同场景下测量的UWB信号中学习出两个冗余完备字典,一个用于目标类型识别,另一个用于目标场景识别；然后基于这两个冗余完备字典求解目标测量UWB信号的稀疏表示,从中提取目标的类型信息和所在场景的信息。与基本的稀疏分类相比,本文提出的方法能够有效提高目标识别性能,同时能够显著提高目标识别的效率。论文最后对全文研究工作进行了总结,并对基于UWB的目标特征提取与识别问题的研究进行了展望。
[Abstract]:Ultra - wideband ( UWB ) wireless communication technology has the capability of detecting and identifying hidden targets by penetrating leaf clusters , and has great application potential . How to extract the characteristics of leaf cluster hidden targets from UWB signals and identify them is an important key technology to be studied . This thesis comes from the national natural science funds and other projects , and has important theoretical significance and application value .

In this paper , aiming at the target recognition technology based on UWB signal , this paper has completed the following innovative research results :

Firstly , a redundant complete dictionary is constructed . The base function of the dictionary can be selected from the transformation groups such as sine function , wavelet function and so on , and can also be learned from the UWB signal of the measurement target ;
then , solving the sparse representation of the target measurement UWB signal on the redundant complete dictionary ;
Finally , the sparse feature of the target is extracted by analyzing the sparse decomposition coefficient . Based on the validation of the leaf cluster coverage target data set , the extracted target feature based on sparse representation has very good scalability , which can significantly improve the performance of the target recognition .

In this paper , two improved particle swarm optimization algorithms are proposed to optimize the selection of SVM parameters . In the improved particle swarm optimization algorithm , we propose two improved particle swarm optimization algorithms to optimize the selection of SVM parameters . In the improved particle swarm optimization algorithm , we propose two improved particle swarm optimization algorithms to improve the convergence speed and search performance of the algorithm .

In this paper , a wavelet support vector machine ( WSVM ) based on wavelet kernel function is constructed , and a hybrid quantum particle swarm optimization algorithm is proposed to optimize the parameters of WSVM .
Experimental results show that the performance of target recognition based on WSVM can be effectively improved by means of the excellent searching ability of the QPSO algorithm .

Aiming at the problem of multi - object recognition under various scenes , a multi - scene recognition method based on sparse representation is proposed . The method is divided into two steps : firstly , two redundant complete dictionaries are learned from UWB signals measured under different scenes from each type of target , one is used for target type recognition , and the other is used for target scene recognition ;
Compared with the basic sparse classification , the proposed method can effectively improve the target recognition performance and improve the efficiency of the target recognition .

Finally , the thesis summarizes the work of full - text research , and looks forward to the research of target feature extraction and recognition based on UWB .
【学位授予单位】：北京邮电大学
【学位级别】：博士
【学位授予年份】：2014
【分类号】：TN925

【参考文献】