计算层析成像光谱仪高精度实时振动模糊谱图复原技术研究

Owing to the ability of acquiring whole data cube in one shot, Computed Tomography Imaging Spectrometer (CTIS) has wonderful potential for military reconnaissance, environmental monitoring, and on-line measurement. However, the spectral images will be blurred due to the vibration and jitter of the platform, which will lead to spatial resolution degradation and spectral curve distortion. As a result, a mistake will be made in targets recognition and classification, which becomes an urgent problem to be solved. However, traditional image restoration methods can not satisfy well with the demand of high accuracy, speed and reliability. In previous study, we found the high-energy zero-order projecting image of CTIS has same image motion with spectral image. Therefore, this project proposals a new method to detect the image shift of zero-order beam images of CTIS to restore blurred spectral images. A high speed image detector will be used to take several short-exposure-time zero-order images, and images shifts will be calculated out. After that, the blur function of the spectral images will be resolved. When the blurred spectral images are reconstructed out near real time using multi-core GPU based parallel acceleration method, an accurate image deconvolution algorithm will be adopted to restore the blurred spectral images. Ultimately, an accurate spectral will be retrieved. This method would be accurate, reliable, and high speed and would solve the image blur of CTIS imaging spectrometer introduced by platform jitter in poor working conditions, which will improve the spectral accuracy. Using our method, this instrument will play an important role in aeronautics and astronautics usage and industry application.

计算层析成像光谱仪（CTIS）具有独到的单次曝光全谱成像能力，在军事侦察、环境监测和在线检测等领域具有重要应用前景。然而实际应用中，平台振动会引起CTIS光谱图像的严重退化，造成光谱的失真，影响目标识别和分类精度，成为迫切需要解决的问题。而传统图像复原方案在精度、实时性和可靠性等方面都难以满足要求。项目研究发现，CTIS的高亮度0级投影图像和光谱图像具有像移一致性，因此提出基于0级投影像移测量的CTIS谱图复原新方法，利用高速相机探测0级投影的多帧短曝光图像，并解算帧间像移，然后构建出平台振动下的长曝光时间谱图退化函数，在采用多核GPU并行加速近实时重构出模糊谱图后，通过高精度反卷积算法复原出无失真的光谱图像，最终获得准确的光谱曲线。该方法速度快、精度高、成本低，克服了恶劣环境下的光谱失真问题，对进一步推动该仪器在航空航天和工业现场等领域的广泛应用有重要的意义。

计算层析型成像光谱仪（CTIS）可以以视频帧率获取光谱图像，因此在空天遥感、目标跟踪和工业检测方面有重要应用前景，然而不可避免的平台振动会导致CTIS图像数据的模糊，造成光谱失真。.本项目针对“平台振动下的CTIS光谱曲线失真”问题，(1)研究了CTIS成像光谱仪的投影原理，分析了非衍射0级光和其它各级次衍射光的光学特性，发现它们之间有几何分布的相似性和能量分布的差异性，设计了基于0级光像移测量的CTIS平台振动探测机制，并提出采用具有PIV功能的高帧频相机，是实现本项目所提方案的理想探测器，最终搭建了基于Andor Neo 5.5的CTIS成像光谱系统；(2)利用像移测量实验装置，比较了相位相关、互信息、联合变换相关、SIFT和嵌入式光流计算的像移测量精度和速度，实现了0.1像素、100帧/s以上的像移测量，在此基础上实现了基于多帧短曝光像移矢量的模糊核构建；(3)分析了数据立方体重构的实现方案，对基于迭代算法的重建过程的每一步都进行了优化，将原来需要大矩阵相乘的关键步骤转化为矩阵相加，使各部分可计算、可并行，最终采用GPU进行任务分配和加速，在数据立方体为100×100×60，探测器为2560×2160的条件下，重构时间缩短到2秒，极大地提高了速度，为实时光谱数据获取奠定了基础；(4)搭建了基于二维位移台的谱图复原实验平台，通过像移测量、图像复原、立方体重构和真实曲线比对等环节，验证了本方案能够有效提高振动条件下的光谱曲线准确度。.本项目通过机制提出、系统搭建、算法改进和实验验证，验证了项目所提出的CTIS谱图复原方案可以从原理上克服 CTIS 在恶劣环境下的光谱失真问题，并具有测量精度高、成本低、实时性好等优点，可为计算层析成像光谱仪在航天探测、目标跟踪等领域的实用化奠定基础。