基于复眼DRA阵列的仿生偏振导航罗盘信息层次化编解码研究

Aiming at resolving the key scientific issue of acquiring compass information from sky polarization pattern in the research of bionic polarization navigation, this project is going to construct a novel bionic model system that integrates visual perception, neural computation and information encoding and decoding based on our previous researches. With the mechanisms of polarization vision in animals as research basis and the acquisition and processing of polarization information as main research line, this project consists of four tasks. Firstly, we are going to build a new simulation model for describing the distribution features of the signal source in line with biological visual logic by introducing the theoretical framework of polarization distance. Secondly, we will build a perception model to reveal the internal relationships between polarization information in different regions of sky pattern based on the characteristics of the polarization-sensitive ommatidium and the topological distribution of the whole ommatidial array in the dorsal rim area of compound eyes. After that, we are going to establish a neural network model that encodes and decodes compass information hierarchically based on the response features of local neurons in different levels and the hierarchical structure of the overall polarization-sensitive nervous system. At last, functional verifications and application experiments of the cascaded model system will be carried out under simulation and measured datasets. The results obtained could promote the further development of the theoretical research and practical application of bionic polarization navigation and provide new theories and methods for solving the problems of autonomous navigation under conditions of weak or no satellite signal.

针对仿生偏振导航中关于利用天空偏振模式获取罗盘信息的关键科学问题，本项目以生物的偏振视觉机理为仿生依据，以偏振信息的获取与处理为探索主线，在课题组的前期相关工作基础上开展“视觉感知-神经计算-信息编解码”融合统一的仿生模型研究。分析天空偏振模式的分布规律，利用偏振距离理论框架建立符合生物视觉逻辑的信号源表征模型；分析生物复眼DRA中单个小眼的偏振感光特性及整体小眼阵列的拓扑分布特征，建立可以映射天空模式中各区域间内在联系的偏振信息结构化感知模型；分析偏振敏感神经系统中各级局部神经元的偏振信息响应特性及整体神经系统的层级结构特征，建立可以实现罗盘信息层次化编解码的神经网络模型；最后，从仿真与实测两个方面对级联后的模型系统完成功能验证与应用实验分析。本项目成果可推动仿生偏振导航理论研究与实际应用的进一步发展，为解决弱/无卫星信号条件下的自主导航问题提供新理论、新方法。

本课题针对仿生偏振导航中关于利用天空偏振模式获取罗盘信息的关键科学问题，以生物的偏振视觉机理为仿生依据，以偏振信息的获取与处理为探索主线，开展“视觉感知-神经计算-信息编解码”融合统一的仿生模型和信号与信息处理方法研究。在信号源天空偏振模式的表征模型方面，充分研究并分析了其内在的分布规律，建立了基于生物偏振视觉理论的天空偏振模式解析模型，提出了一种新的天空偏振模式解析模型修正方法，并研究了基于机器学习的异常天空偏振模式信息恢复算法；在仿生偏振信息结构化感知模型方面，以生物解剖学的相关研究成果为基础，深入探讨单个小眼及DRA阵列的特殊光学结构与功能的联系，建立了多通道的偏振距离信息感知模型，提出了自适应的偏振信息结构化感知模型；在层次化偏振罗盘信息编解码神经网络模型方面，分析了局部及完整偏振敏感神经系统的响应特性及层级结构特征，建立了POL-ANN偏振信息编码模型，提出并验证了偏振罗盘信息的解码方法；在此基础上，设计了一种新的时序天空偏振模式信息分布式自动采集系统，提出了一种自适应的天空偏振模式信息获取方法，并从仿真与实测两个方面开展了验证实验。本项目成果可推动仿生偏振导航理论研究与实际应用的进一步发展，为解决弱/无卫星信号条件下的自主导航问题提供新理论、新方法。除此之外，针对仿生偏振视觉的研究成果在伪装目标检测、偏振去雾、水下目标检测等多个领域均具有广泛的应用前景。