Imaging Science Thesis Defense: End-to-End Systems Limitations of Hyperspectral Target Detection using Parametric Models and Subpixel Lattice Targets for Validation

Imaging Science Ph.D. Defense
End-to-End Systems Limitations of Hyperspectral Target Detection using Parametric Models and Subpixel Lattice Targets for Validation

Chase Cañas
Imaging Science Ph.D. Candidate
Rochester Institute of Technology                                                        

Register for Zoom here

Abstract:
Hyperspectral target detection applies algorithms on high-dimensional images to identify rare targets, or objects of interest, within cluttered scenes of various backgrounds. Computational demands of processing hyperspectral datasets, along with their limited availability in both public and private sectors, impose challenges to assess system-level limitations of detection. This research presents a methodology to quantify end-to-end limitations through statistical modeling across thousands of target detection scenarios, each with unique combinations of system parameters. The objective is to identify specific parametric thresholds or regions where detection is limited, based on surface “elbows” and contours in the detection response function derived from model outputs. The model considers subpixel targets, where an object’s signal occupies an area smaller than a pixel. Since subpixel targets are spatially unresolved, validating subpixel models is challenging with conventional datasets. To address this, a lattice-based target was developed and deployed in a UAV data collection, producing a novel dataset with empirical observations of approximately 300 subpixel samples for each constant fill fraction (0.2, 0.4, 0.6, 0.8). To investigate limitations, eight parameters across the imaging chain (scene, atmosphere, sensor, and processing) were considered in combination with four qualitative scenes (urban, rural, forest, and desert). Approximately 70,000 unique scenarios were generated, with scalar detection outputs stored in a multidimensional array. These detection outputs were derived from a proposed metric, the log-weighted area under the ROC curve (wAUC), which evaluates overall performance while emphasizing small false alarm rates (e.g. PFA < 0.001). Results include correlation coefficients to assess sensitivities between system parameters and detection, where the relative magnitudes of coefficients reveal which parameters “drive” detection under various conditions. Additionally, numerical methods for computing the Laplacian, which represents the divergence of the gradient, are used to measure how the wAUC detection surface bends or curves to locate “elbows.” This method is also applied to contours of the detection surface, where negative Laplacian values across sampled contours are minimized. The results provide insights into parametric thresholds and regions in which detection (wAUC) is limited for a given remote sensing configuration.

Intended Audience:
All are Welcome!

To request an interpreter, please visit myaccess.rit.edu