Hyperspectral Target Signature and Background Clutter Modeling in Tree Regions

In this project we will develop innovative techniques for modelinghyperspectral target signatures and background clutter in tree regions.We will focus on the modeling of deep hide targets that are obscuredbetween 50% and 100% by tree canopies. The measured sensor spectrumfor a target under a tree depends on the target's geometric and physicalproperties, the three-dimensional tree canopy architecture, and theoptical properties of the leaves and soil. We will use DIRSIG and PROSPECTto construct spectral subspaces to examine target/background contrast andtarget signature variability for targets under trees over ranges oftree and leaf parameters as well as over the environmental conditions.The analysis will allow us to gain insight into the fundamental physicalprocesses that govern signatures for targets under trees as well as toestablish the range of conditions over which targets under trees canbe recognized reliably using hyperspectral images. We will also developphysics-based spectral/spatial models for background clutter in treeregions. These models can be used for automated tree region segmentationand for the estimation of tree and leaf parameters from image data.The accuracy of the new modeling techniques will be evaluated by comparingthe generated models with hyperspectral image data and by analyzing theperformance of algorithms that utilize the new models.The new algorithms can be used in systems for surveillance, reconnaissance,and targeting applications. Related algorithms have important commercialapplications in areas such as biomedicine, mining, and agricultural andenvironmental monitoring.