红外偏振与光强图像的拟态仿生融合模型研究

Fusion of infrared polarization and intensity images with different imaging physical mechanism and high complementary is a novel means to improve infrared detection performance and infrared countermeasures. At present the main problem is fusion models cannot change according to differences between two types of images, leading to low fusion effect even fusion failure; the reason is fusion models lack the ability to perceive the differences, and cannot adjust fusion strategy based on cognitive differences. Therefore, in this project, fusion of infrared polarization and intensity images is taken as research object, and mimic octopus simulating behavior and function of a variety of creatures via perceiving external threat is regarded as bionic basis, and fusion theory driven by differences, possibility theory, and mimicry bionics are employed as research techniques. Specific contents as follows: i) basic principle of image mimic bionic fusion: perception and mimicry mechanism of mimic octopus, characteristics and function of mimic fusion, mechanism driven by different features, and mimic transformation theorem, etc.;ii) fusion model of two types of images with many forms derived by perception/cognition: class set building of image different features, structure performance analysis of the mimic model, class set building of structured fusion algorithms, multi set-valued mapping between two types of class sets, and derived strategy of polymorphous substructure algorithm set/class set, etc.;iii) model validation method. Thus, the project explores new fusion theory and method to meet the dynamic change demand of multi-source heterogeneous infrared images by changing model forms.

将成像机理不同、互补性很强的红外偏振与光强图像融合是提高红外探测性能与红外对抗水平的新手段。目前主要问题是融合模型无法根据两类图像差异而变化，造成融合效果低、甚至融合失效；原因是模型缺乏感知图像间差异信息能力、不能根据差异变化调整融合策略等。为此，本项目以红外偏振与光强图像的融合为对象，以拟态章鱼感知外界威胁模拟多种生物的行为和功能为仿生依据，以差异驱动融合理论、可能性理论、拟态仿生学等为手段，研究1）图像拟态仿生融合基本原理：拟态章鱼的感知与拟态机理、拟态融合特征功能、差异特征驱动机制、拟态变换定理等；2）通过感知/认知派生多种形态的两类图像拟态仿生融合模型：图像差异特征类集构建、被拟态模型结构性能分析、结构化融合算法类集构建、两种类集间多集值映射、多形态子结构算法集/类集派生策略等；3）模型验证方法。从而，探索出通过改变模型形态满足多源异类红外图像动态变化需求的融合新理论、新方法。

将成像机理不同、互补性很强的红外偏振与光强图像融合是提高红外探测性能与红外对抗水平的新手段。当前，两类图像融合主要依据图像差异特征的先验知识事先确定融合算法。事实上，由于场景的动态变化，两类图像的差异特征很难提前有效确定；且图像的差异特征类型、差异幅度是随机变化的。因此，当前基于“先验假设”的融合模型缺乏感知图像间差异信息能力、不能根据差异变化动态调整融合策略，造成融合效果低、甚至失效的问题。.为此，本项目根据拟态章鱼的拟态特性提出了两类图像的拟态仿生融合，研究了图像拟态仿生融合原理，构建了两类图像的拟态仿生融合模型，实现了依据图像差异特征变化动态调整融合策略。主要研究内容和结果如下：.（1）提出了拟态仿生融合原理：研究了拟态章鱼的感知能力和拟态变换驱动机理，给出了拟态融合的概念、特征和功能，探索了差异特征驱动机制，研究了面向图像融合的拟态变换定理，为拟态仿生融合模型的构建提供了理论基础。.（2）构建了拟态仿生融合模型并进行了验证：研究了两类图像差异特征类集的构建，分析了被拟态模型的结构性能，构建了结构化融合算法类集，建立了两种类集间的多集值映射；分析了融合算法的派生策略，研究了算法间的协同和嵌接关系对融合效果的影响，构建了差异特征驱动的多融合算法协同嵌接融合模型，并给出了4种实现方法：1）基于差异特征指数响应函数和非负矩阵的多类融合算法并行协同融合方法，前者的平均能量、对比度和空间频率指标分别提高了42.8%、30.9%和47.5%，后者的相应指标分别提高了4%、35.3%和45.2%；2）基于差异特征的非下采样剪切波与多尺度引导滤波相结合的内嵌融合方法和基于二维全变分分解的内嵌互补融合方法，前者的平均能量、对比度、空间频率和相关性差异指标分别提高了26.5%、5%、2%和13.6%，后者的相应指标分别提高了15.5%、14.9%、48.3%和12.5%。该模型通过图像差异特征变化动态选择融合策略提升了融合性能，为高性能红外成像探测技术及图像融合探索了新理论、新方法。