You are here

Advanced Hyperspectral Exploitation Using 3D Spatial Information

Description:

TECHNOLOGY AREA(S): Sensors 

OBJECTIVE: Develop advanced hyperspectral exploitation algorithms incorporating 3D spatial information for improved target detection and identification. 

DESCRIPTION: Hyperspectral imaging (HSI) has demonstrated utility for material classification and target detection/identification. Generally, hyperspectral exploitation algorithms operate using spectral information alone due to insufficient spatial resolution of the sensor or the lack of coincident data from another sensing modality, such as RADAR, LIDAR, or passive 3D imaging. False alarms and low detection/ID confidence can exist for certain target classes that are not well separated using spectral information alone. As sensor technology matures, more opportunities exist to collect HSI data with coincident 3D spatial information at each pixel of the HSI data cube. This information could come from LIDAR data collected over the same spatial area or with novel passive sensing modalities, such as passive 3D HSI [1]. This additional information can be used to help improve the separability of material and target classes, thereby reducing false alarms and improving ID confidence. Previous efforts in exploiting HSI with coincident LIDAR data have demonstrated benefits for material classification [2]. This effort seeks to improve and expand upon previous work with emphasis specifically on target detection and identification rather than material classification. Research should focus not only on the use of 3D spatial target information but implications for atmospheric characterization, shadow mitigation, bi-directional reflectance distribution function (BRDF) properties, and other items associated with the physics of radiative transfer. Assumptions that can be made regarding this effort include: 1) sensor viewing geometry is known along with solar geometry, 2) targets will span multiple pixels, 3) a spectral library of target and background signatures exists, 4) knowledge exists about the 3D structure of the targets of interest in the sense that the target shape is known (i.e. vehicle shape/size, etc.), 5) hyperspectral data will be in calibrated spectral radiance units, and 6) 3D spatial information available for each hyperspectral pixel with in the form of point cloud data and/or co-registered digital surface model data at roughly ½ the ground sample distance of the HSI data. Algorithms should produce a confidence measure associated with each target ID. Algorithms should demonstrate an order of magnitude decrease in false alarms with a 25% increase in ID confidence when compared with state-of-the-art spectral-only algorithms currently being used by the Air Force. Algorithms should be able to operate near real-time (within seconds or minutes of data collection) or a path demonstrated to optimize for near real-time operation using state-of-the-art processing hardware, such as graphical processing units (GPUs). 

PHASE I: It will explore and develop novel algorithms and test with synthetic data if appropriate. Testing will continue using government furnished airborne hyperspectral data and coupled 3D point-cloud data. Algorithm performance will be quantified and compared with current state-of-the-art spectral-only detection and ID algorithms. 

PHASE II: It will modify and further develop the algorithms based upon Phase 1 results. Further testing will occur using additional government-furnished data sets. The code will additionally be optimized with a hardware architecture identified for real-time or near real-time implementation. 

PHASE III: It will transition the software for incorporation into existing hyperspectral exploitation tools or other assets based upon interaction with the customer. 

REFERENCES: 

1. J. Ahlberg, et al, “Three-dimensional hyperspectral imaging technique,” Proc. SPIE, 10198, 1019805-1 – 1019805-10, (2017); 2. M. Khodadadzadeh, et al, “Fusion of Hyperspectral and LiDAR Remote Sensing Data Using Multiple Feature Learning,” IEEE JSTARS, Vol. 8, No. 6, (2015)

KEYWORDS: Target Detection, Identification, Hyperspectral Imaging, 3D Fusion 

US Flag An Official Website of the United States Government