基于孔径分割的快照式成像光谱技术及其图像处理方法研究

Most of the common imaging spectrometers cannot capture both spatial and spectral information from a scene in a single frame. They require a form of temporal scanning process to build up the spatial-spectral cube of information. As a result, it is difficult to observe dynamic objects, such as moving objects or temporally changing sources, by these instruments. In order to simultaneously obtain the spatial and spectral information, developing snapshot imaging spectrometers becomes a research focus. Recently, a division of apeture (DoAp) snapshot imaging spectrometer has been proposed by using a light-filed camera.The light-field camera can simultaneously record two spatial dimensions and two angular dimensions information from a scene. The light-field imaging spectrometer is achieved by implementing a spectral filters array to divide the objective aperture. Therefore, the spectral information is modulated with the directional information of ligh-field and achieved the "snapshot" of the spatial-spectral datacube for a scene. However, this technique still needs to be improved and overcome its inherent low spatial resolution defect. In this research, the model and calibration methods for the DoAp-based snapshot light-field imaging spectrometer will be proposed. The image resolution will be enhanced by employing the super-resolution and image fusion algorithms. First, the the principle of the spectrometer will be studied and analyzed along with the error sources. In order to optimize the system, the error models will be established and calibration methods will be proposed to correct the errors. Then, the observation model will be modified and a super-resolution algorithm will be employed to obtain a high spatial resolution image for image fusion. The image fusion algorithm based on Fuourier transform will be studied and modified to improving the spatial resolution of the spatial-spectral data cube without loosing too much spectral information. Furthermore, computer simulation will be performed to verify the models and algorithms. Adjustments of the models and algorithms will be made according to the simulation results. Finally, an experimental setup will be designed and built up to test the feasibility of the proposed calibration methods and image processing methods in this research.

大多数成像光谱仪都需要通过某种扫描方式获取目标场景的空间和光谱信息，无法在一次曝光内获取完整图像光谱信息，因而限制了其在动态的监测等方面的应用。为了同时获取目标完整图谱信息，快照式成像光谱仪称为研究重点之一。近年来出现了一种基于孔径分割和光场成像技术的快照式成像光谱技术，结构简单紧凑、图谱数据提取相对容易。但是该技术还处于理论研究和演示验证阶段，且存在光谱图像空间分辨率低的缺陷。本项目将在研究该技术光场调制和光谱调制的基础上，分析误差源、建立误差模型，提出相关定标方法优化系统；基于光场理论和光谱图像特点，改进图像降质模型、多帧图像超分辨算法和图像融合算法，提出一种无需外部参考图像的光场光谱图像空间图像处理方法提高空间分辨率，仿真验证模型和算法；搭建原理演示装置并进行定标，利用实验数据验证算法，进而优化系统参数和算法。

光谱成像技术可以获取目标的三维图谱数据,但是传统的成像光谱仪无法一次曝光获取完整图像光谱信息，限制了其在动态目标监测领域的应用。为了同时获取目标完整图谱信息，快照式成像光谱仪称为研究重点之一，其中基于孔径分割和光场成像技术的光场光谱成像技术，将目标光谱信息与光场方向信息耦合，将多通道光谱图像信息投影到二维探测器上，目标图谱数据立方体的快照式获取。该技术处于起步阶段，缺乏相关理论和数据处理方法，因此本课题针对这一问题开展相关研究工作。.本项目将在研究目标辐射在光场光谱成像系统中传输和探测机理的基础上，建立了光场光谱成像的数理模型，并开展系统的优化设计；针对光场元件建立起耦合误差模型，提出了基于非线性最优估计的耦合参数忙盲估计算法；在分析光场光谱数据特点的基础上，提出了光场光谱成像系统的光谱标定和信息重构方法；针对谱段间存在亚像素位移的光谱数据建立新型图像降质方程，提出基于双阈值Huber范数的正则化超分辨算法获取高空间分辨率的全色图像，采用该图像与原数据立方体融合提高光谱数据立方体的空间分辨率；搭建了搭建原理演示装置开展定标和成像实验，验证了定标方法和数据处理方法的有效性。.上述研究工作的主要意义包括：首先，建立了光场光谱成像技术的理论体系，为技术研究和数据处理奠定基础；其次解决了关键元件光场成像器件的特性测定的难题，保证系统误差可测和可控；再次光谱数据标定和提取方法保证了重构光谱数据的可靠性；最后，光谱数据分辨率提高方法无需额外的全色相机，降低了系统设计的复杂度，有利于其在遥感探测等利于的应用。