面向森林监护的敏捷高光谱影像获取方法研究

Forest monitoring and forest fire prevention are essential for forest resources management and conservation. Hyperspectral data is an important information resourse for forest management, which is an indispensable aid for forest monitoring. High spectrum for technology develops rapidly, but it is still unable to meet the demand of the forest resource management and reservation, such as acquiring spectral video at real time. Aiming at the disadvantages of the classical spectral imaging methods, that traditional method's speed of information acquisition is slow, and data is large, and compressed spectral imaging method's decoding speed and extraction speed in specific spectrum are slow, the agile spectral image acquisition method is proposed. From three aspects, the mathematical model, the design and optoelectronic implementation of observation matrix, and the research of the inherent characteristics of forest spectral data, the project will study the compressed spectral imaging method whose spectrum should be selectable, or compression ratio should be selectable, based upon the framework of compressed sensing. It is worth to mention that the agile spectral imaging method can balance the demand of high spectral resolution and high video frame rate in different application requirements. The project will study the computing spectral imaging method thoroughly. First, the project takes the optical element as the coding or calculation operation with the spectral information of scene, and uses mathematical language to describe them. Furthermore, it considers the whole spectral imaging process to build the mathematical model, thus the validity of the theory of agile grabbing hyperspectral image/video can be given from the mathematical model and the optical implementation.

森林监护对于森林资源的管理与保护至关重要。高光谱数据是森林管理的重要影像信息，是森林监护不可缺少的辅助工具。虽然高光谱获取技术发展迅速，但仍然无法满足森林监护中光谱影像获取速度的需求。针对经典的光谱成像方法中的不足，比如传统方法信息采集速度慢、数据量大，而压缩光谱成像方法的解码速度慢、特定谱段提取速度慢，我们提出了敏捷光谱影像获取方法。本课题基于压缩感知框架，从敏捷光谱成像的数学模型、观测矩阵的设计与光电实现、森林光谱数据内在特征挖掘三个方面，研究谱段可选择、压缩率可选择的压缩光谱成像方法。值得一提的是，该方法可以平衡不同应用背景下对高光谱分辨率、高视频帧速率的要求。课题深入研究计算光谱成像方法，将光学元件看成对场景光谱信息的编码、计算，用数学语言进行描述，综合考虑整个光谱成像流程，从数学模型与光学原理上保证敏捷光谱影像数据的获取。

高光谱数据是森林管理的重要影像信息。针对推扫式光谱成像方法中的不足，比如信息采集速度慢、数据量大等，我们提出了敏捷光谱影像获取方法。研究内容主要围绕编码感知场景，从提高光通量、减少成像面传感器阵列数目等角度研究敏捷光谱影像获取方法,设计了多狭缝光谱成像方法和编码分光成像方法。设计的多狭缝光谱成像方法，在保证光谱数据质量（高分辨率）的情况下，有效地降低了曝光次数、数据的存储传输量；多狭缝光谱成像模式下，对场景中各个谱段、各道狭缝是独立解码的，既可以减少下传的谱段，又提高了解码的速度。提出的编码分光成像模式的成像面的像元数目小于所采集的光谱数据的光谱维度，不再由提高传感器阵列的密度来提高光谱分辨率，而是由提高成像系统的信息采样率来提高光谱分辨率；编码分光成像模式下，不仅可以在图像质量和传输时间之间权衡，以确定合适的样本数目进行下传, 而且其稀疏域的研究，仅需要研究特定目标（应用领域所关心的目标）的光谱信号的稀疏域。敏捷光谱影像获取方法，在系统实现上借鉴了计算成像方法，在采集场景光谱信号过程中完成光谱信号的压缩。不仅减少了数据压缩环节，而且还使得设计成像系统更为灵活，提供针对森林监护不同场景要求而设计相应的成像系统的新思路。