【作者】 闫莉萍；

【导师】 周东华；

【作者基本信息】 清华大学 ， 控制科学与工程， 2006， 博士

【摘要】 信息融合技术具有可以提高系统的可靠性和稳定性、扩大空间和时间覆盖范围与改善尺度等优点,因而,在众多的军事和非军事领域都存在着非常广泛的应用。状态融合估计和图像融合是其中两个研究热点,本文的研究工作也主要在这两方面展开。在状态融合估计方面,针对多速率传感器同步采样的时变线性动态系统、多速率传感器异步时不变和时变线性动态系统,分别提出了不同的状态融合估计算法。对多传感器同步采样系统,采用分块,无反馈分布式,以及有反馈分布式结构等方法,分别对最高采样率下的状态进行了有效的估计;对时不变异步采样多传感器系统,采用多尺度建模、尺度递归分层融合的策略,得到了方差的迹最小意义下状态的线性无偏最优估计;而对于多速率时变异步采样的多传感器数据融合问题,则通过状态和观测的分块与扩维,将其在形式上转化为单速率同步多传感器观测系统,进而运用Kalman滤波和分布式融合结构得出方差最小意义下状态的最优估计。通过具体应用实例,仿真验证了所提出算法的有效性。在图像融合方面,分别研究了多传感器多分辨率图像融合算法以及图像融合结果的性能评估问题。在算法方面,给出了问题的数学描述,采用二维多尺度Kalman滤波的方法融合具有不同分辨率的观测图像;对上述融合图像依据其熵的大小进行加权融合并进行边缘修正,最终将不同传感器观测的具有不同分辨率的图像进行了有机融合。通过多组图像融合的实验,验证了算法的有效性,并利用多种性能评价指标对其进行了分析。图像的性能评估是图像融合领域的研究难点之一。从信息论的角度和人的视觉感应系统原理出发,本文依次提出了综合熵、归一化互信息熵和归一化视觉感应信息熵等性能评价指标。实验结果表明,由于归一化视觉感应信息熵兼顾了信息的充分转移和人的视觉感应原理,因此,在图像的性能评估方面具有独特的优势。更多还原

【Abstract】 Information fusion has been widely applied in many military and non-military fields because it has the advantages of improving the reliability and the stability, expanding the space and the time covering scope, and improving scale etc. State fusion estimation and image fusion are two important topics of information fusion, which are discussed in this dissertation.State fusion estimation is studied first. Aimed at multirate linear time-varying dynamic systems with synchronous sampling, multirate linear time-invariant dynamic systems with asynchronous sampling, and multirate linear time-varying dynamic systems with asynchronous sampling, the related data fusion state estimation algorithms are proposed, respectively. For the synchronous sampling system, the blockwised, the distributed fusion structure with feedback and without feedback are used to obtain the effective state fusion estimation. For the asynchronous sampling time-invariant system, through multiscale modeling and scale recursion, the optimal state fusion estimation is generated in the sense of minimizing the traces of the estimation error covariances. For the asynchronous sampling time-varying system, by augmentation of the state and measurements and by dividing them into proper blocks, the multirate asynchronous sampling system is formalized into a synchronous sampling system with single sampling rate, therefore, by use of Kalman filter and distributed data fusion structure, the optimal state fusion estimation in the sense of linear minimum variance is achieved. The effectiveness of the proposed algorithms is shown through computer simulations.On image fusion, multisensor multiresolutional image fusion algorithms, and the performance evaluation of image fusion results are studied, respectively. For the fusion of multisensor multiresolution images, the system equations are formulated first, and then by use of 2D Kalman filter, the images that have different resolutions taken by the same sensor are fused. Later, by use of entropy to construct the weight, the weighted average image fusion method is used to fuse the former fused images. The final fused image is obtained by modifying its edges. Multiple experiments aimed at multiple groups of images that have different performances show the effectiveness of the proposed image fusion algorithm. Performance evaluation is still an open problem in the field of image fusion. In view of information theory and human perception system, respectively and collaborately, three performance evaluation indices, including collective entropy, normalized information entropy and normalized human perception entropy, are proposed successively. Experiments show that the normalized human perception entropy is predominant among the existing performance evaluation indices, because it pays attention to both the completeness of information transmission and the characteristics of human perception.更多还原