基于空间-光谱结构稀疏编码的多波段夜视目标识别技术

项目名称： 基于空间-光谱结构稀疏编码的多波段夜视目标识别技术

项目编号： No.61501235

项目类型： 青年科学基金项目

立项/批准年度： 2016

项目学科： 无线电电子学、电信技术

项目作者： 韩静

作者单位： 南京理工大学

项目金额： 22万元

中文摘要： 夜视技术已从单一波段成像发展为多波段探测，但随着应用环境的复杂化和需求的多样性，传统信息处理技术已难以满足现代多波段夜视的探测需求，因此迫切需要发展新型信息分析的理论方法。围绕视觉稀疏分析在多光谱夜视目标感知识别中面临的问题，本项目结合多传感器夜视成像装置，拟探索研究基于空间-光谱结构稀疏编码的夜视信息理解方法，其核心思想是结合视觉稀疏响应机理和多光谱夜视数据的多维结构特征，充分分析和利用夜视数据的本质、不变信息，有效解决复杂成像环境导致的夜视背景干扰和目标多样化。主要研究内容包括局部空间/空间-光谱结构保持稀疏编码、基于多维结构稀疏编码的多光谱典型夜视目标认知识别以及高性能夜视数据异构处理模型。本项目模拟高效的生物认知系统，实现基于空间-光谱结构稀疏编码的多光谱典型夜视目标识别，其研究成果可为复杂环境下多波段夜视图像智能理解提供理论与技术支撑。

中文关键词： 多光谱夜视；空间-光谱结构；稀疏编码；核函数；视觉信息处理

英文摘要： The night vision technology has been developed from the traditional way by single-band imaging to multi-band detection. However, with the continuous development of night vision technology, the complexity of application environments and diversity of demands are driving the modern multi-band night vision detection to develop the new information processing theories and methods. Focus on the problems of visual sparse analysis around target perception and recognition in multispectral night vision images, this project combines the device of multi-sensor for night vision imaging, and intends to study the methods of night vision information understanding based on the spatial-spectral structure sparse coding. The core idea is to combine the visual sparse response mechanism and the multidimensional structure features of multi-spectral night vision data, then full analyze and use of their essence and invariant information, and finally effectively solve the interference in night vision images and target diversification caused by complex imaging environments. The major researches include: the local space/spatial-spectral structure preserved sparse coding, the typical target recognition in multispectral night vision image based on the multidimensional structure sparse coding, the heterogeneous framework of night vision data processing in high performance. This project simulates the efficient biological cognitive system and achieves the multispectral night vision target recognition based on the spatial-spectral structure sparse coding algorithm, its research results can provide theoretical and technical supports for the intelligent understanding of multi-band night vision images in complex environment.

英文关键词： Multispectral night vision;Spatial-spectral structure;Sparse coding;Kernel function;Visual information processing