Image analysis – Applications – Target tracking or detecting
Reexamination Certificate
1999-04-08
2001-08-28
Johns, Andrew W. (Department: 2621)
Image analysis
Applications
Target tracking or detecting
Reexamination Certificate
active
06282301
ABSTRACT:
BACKGROUND OF THE INVENTION
1. Field of the Invention
The present invention generally relates to hyperspectral image sensors, i.e. sensors wherein each pixel consists of a plurality of separately detected components each within predefined adjacent narrow frequency bands that cover the visible, infrared and far infrared spectrums.
2. Description of Prior Art
In prior art surveillance and reconnaissance applications, hyperspectral target identification technologies have been extensively investigated. The majority of the work has been focused on techniques using the Reed and Xiaoli (RX) multispectral detection algorithm, Linear Mixing Model (LMM) (or Spectral Unmixing) and spectral clustering. These are designed to detect sub-pixel or low spatial resolution targets (tens of pixels on targets) using spectral signatures and spatial properties of targets. Each of these techniques presents significant limitations in real-time surveillance and reconnaissance applications. RX target identification techniques assume targets have some known continuous spatial patterns with describable statistical properties; it also assumes gradual change in the spatial domain. However, edges (where the spectral signatures are highly correlated with man-made material spectral signatures) tend to produce high false alarm rates (FARs). Further RX, in its statistical modeling and hypothesis testing, quite often does not represent the target or background characteristics adequately. Spectral Clustering depends on the cluster initialization parameters, such as threshold values and numbers of iterations; which may not result in the desired number of clusters. Reinitialization may cause unacceptable rates of convergence, hence, spectral clustering techniques are not feasible for real-time detection. LMM assumes that the spectral signature of the pixel in the scene is a linear combination of spectral signatures for pure endmembers; such as sand, water, grass, trees, etc. Even if pure endmember spectral signatures can be obtained, the spectral signature of the pixel in the scene does not represent only the linear combination of pure endmembers, but also system and atmospheric parameters which alter spectral information. The signature is also dependent on lighting conditions, clutter and, worse yet, sensor artifacts which affect apparent radiance. Spectral clustering techniques present greater difficulties when operated in real-time scenarios, since they depend on an iterative clustering algorithm of indeterminate length which may or may not converge to produce usable clusters. Thus, spectral clustering is not dependable for real-time operation.
SUMMARY OF THE INVENTION
In the ARES method of target identification algorithms are provided that are based on the adaptive real-time endmembers selection. This method suppresses the clutter by correlating a test pixel with the clutter (not with targets of interest as traditionally done). Since clutter dominates 90% or more of the scene image, only a small percentage of man-made materials are left in the image as detections. This approach offers a robust technique for the selection of endmembers in real-time by letting the scene dictate its own endmembers. These in turn, fully represent scene characteristics. No assumptions have to be made about either target or background.
REFERENCES:
patent: 5337053 (1994-08-01), Dwyer
patent: 5502688 (1996-03-01), Recchione et al.
patent: 5598355 (1997-01-01), Derou et al.
patent: 5644508 (1997-07-01), McNary et al.
patent: 5798942 (1998-08-01), Danchick et al.
patent: 5805106 (1998-09-01), Baum
patent: 5805742 (1998-09-01), Whitsitt
patent: 6023235 (2000-02-01), Sauer
patent: 6075891 (2000-06-01), Burman
patent: 6208752 (2001-03-01), Palmadesso et al.
Bashore Alain L.
Holford John E.
Johns Andrew W.
Lee Milton W.
Nakhjavan Shervin
LandOfFree
Ares method of sub-pixel target detection does not yet have a rating. At this time, there are no reviews or comments for this patent.
If you have personal experience with Ares method of sub-pixel target detection, we encourage you to share that experience with our LandOfFree.com community. Your opinion is very important and Ares method of sub-pixel target detection will most certainly appreciate the feedback.
Profile ID: LFUS-PAI-O-2456282