共孔径主被动高光谱三维成像实时处理框架

Real-time processing framework of common-aperture active and passive hyperspectral 3D imaging

共孔径主被动高光谱三维成像技术是一种结合激光雷达主动探测和高光谱相机被动成像的新型遥感探测手段,通过共光路设计,降低了主被动数据配准难度,使得实时融合生成三维高光谱影像成为可能。三维高光谱成像实时处理兼具数据密集和运算密集的特点,可编程片上系统软硬件协同设计为其提供了解决方案。而目前软硬件划分多基于定性经验分析设计,难以实现定量化最优设计。针对该问题,提出了一种采用基于权重法的多目标规划模型的可编程片上系统处理框架。该处理框架利用图论模型Ncut准则开展高内聚度、低耦合度的单元分割,并对各单元的软件和硬件实现特性分别进行分析评估,最终面向应用需求,利用多目标规划模型求解最优的软硬件划分方案。利用该处理框架,针对速率优先和功耗优先两种高光谱三维实时成像应用场景,进行了软硬件划分方案的定量化求解与分析,结果表明,在速率优先设计中,相对于传统的设计,处理速率提升了43.4%,而功耗降低了53.5%。

Common-aperture active and passive hyperspectral three-dimensional imaging technology is a new remote sensing detection method,which combines active LiDAR and passive hyperspectral cameras in a single framework with shared optical systems.Thus,the difficulty of heterogeneous data registration is reduced,and the generation of the 3 D spectral image by real-time fusion becomes possible.The real-time 3 D imaging is characterized by data-intensiveness and computing-intensiveness,and the software and hardware co-design framework for system-on-a-programmable-chip provides a feasible solution to it.At present,the hardware/software partitioning is mostly derived from qualitative and empirical analysis,and it is challenging to achieve a quantitative and optimal design.A system-on-a-programmable-chip processing framework using a multi-objective programming model based on object weight was proposed to tackle this problem.In this processing framework,the graph-theory-based model with Ncut criterion was used to achieve high cohesion and low coupling functional modules partitioning.Then,the performances of functional modules with software fulfilment and hardware fulfilment were thoroughly analyzed and evaluated.Finally,aiming at the design requirement,the proposed multiobjective programming model was used for the hardware/software partitioning scheme.Two optimal hardware/software partitioning schemes based on the speed-first criterion or the power-first criterion were solved quantitatively for different scenarios.The result shows that the speed-first design overperforms an empirical design with an increase of 43.4%in processing speed,a reduction of 53.5%in power consumption.