| Objective: ||Investigate advanced methods for detecting specific materials in hyperspectral image data based on realistic temporal changes between successive observations.
|| Description: ||The ability to remotely detect military targets hidden in background clutter such as foliage and urban areas is of great interest to the Air Force. By exploiting specific signature content of targets relative to background clutter materials, hyperspectral imaging presents one promising method for achieving this capability. Unfortunately, the great diversity of materials in realistic background areas often results in false alarm rates that are higher than desired. By imaging areas successively over time, change detection presents a promising discrimination method for detecting changes in target state (e.g., insertion, deletion, or movement of targets) with a lower false alarm rate. Prior research has provided a proof of concept of hyperspectral change detection, but has been generally limited in three ways: 1) they are based on a simple, space-invariant, affine model of background change, 2) they rely on fine, subpixel registration between successive images that is difficult to achieve in practice, and 3) they generally detect anomalous change without regard to the specific spectral character of the change.
The focus of this effort is to explore novel change detection methods that specifically address the aforementioned limitations of the current state of the art. The signal processing methods should be based on generalized phenomenological models that capture possible space-variant and nonlinear temporal changes between hyperspectral images (e.g., changes in shadowing), and should support high clutter rejection even in the presence of unknown mis-registration between images. Furthermore, the detection of specific target state changes is of prime interest to this topic as opposed to nonspecific, anomalous changes. Such signature-based methods should be able to incorporate both laboratory and in situ spectral signature data.
|| ||PHASE I: Establish a theoretical foundation for a novel hyperspectral change detection process addressing the limitations of the current state of the art. Develop a baseline algorithm and perform proof-of-concept experiments using realistic hyperspectral imagery to establish the method effectiveness.
|| || ||PHASE II: Refine the baseline processing algorithm into a robust hyperspectral change detection process for hyperspectral imagery from various sources. Thoroughly evaluate the detection performance as a function of operating conditions. Implement the algorithm on a computing platform that supports turn-key operation by a moderately trained image analyst.
|| ||DUAL USE COMMERCIALIZATION: Military application: Detection of difficult military targets, including camouflage, concealment, and urban clutter Commercial application: Civil remote sensing applications such as climate change monitoring and agricultural remote sensing, as well as law enforcement and border surveillance applications.
|| References: ||1. A. Schaum and A. Stocker, “Hyperspectral change detection and supervised matched filtering based on covariance equalization,” Proceedings of the SPIE, Vol. 5425, pp. 77-90, 2004.
2. A. Schaum and A. Stocker, “Linear chromodynamics models for hyperspectral target detection,” Proceedings of the IEEE Aerospace Conference, February 2003.
|Keywords: ||hyperspectral, change detection, remote sensing, target detection|