Improving Spectral Image Classification through a Custom Data Sparsing and Pixel Clustering Algorithm
- Nevada National Security Site, Mission Support and Test Services, LLC
- Horizon GeoImaging, LLC
The Underground Nuclear Explosion Signatures Experiment (UNESE) seeks to characterize non-prompt observables from underground nuclear explosions (UNE). As part of this effort, we evaluated the ability of Digital Globe’s World View-3 (WV3) to detect and map UNE signatures. WV3 has proven capabilities in the detection of large-scale, anomalous, man-made features; however, it has relatively limited spectral and spatial resolution as compared to airborne hyperspectral imaging. These limitations impede standard image classifiers from identifying UNE signatures that are spatially small and lack distinct spectral features. In order to improve image classification results, we developed a custom algorithm that increases the true positive rate, while decreasing the false positive rate. Our algorithm’s flexible front-end allows the user to choose the method used to sparse the data. These methods include band-ratio optimization, correlation coefficient threshold, or Adaptive Coherence Estimator (ACE) threshold. The resulting sparsed pixels are clustered into like-objects. A novel weighted-sum of ACE and correlation coefficient scores is used to classify each object based on the average spectrum of its constituent pixels. WV3 and AVIRIS data of Cuprite, Nevada were used as a verification data set. This algorithm improves overall pixel classification accuracy over standard classifiers by up to 14%.
- Research Organization:
- Nevada National Security Site (NNSS), North Las Vegas, NV (United States)
- Sponsoring Organization:
- USDOE National Nuclear Security Administration (NNSA), Office of Defense Nuclear Nonproliferation
- DOE Contract Number:
- DOE/NV/03624-0336
- OSTI ID:
- 1486883
- Report Number(s):
- DOE/NV/03624-0336
- Resource Relation:
- Conference: American Geophysical Union Fall Meeting, Washington, DC, December 10–14, 2018
- Country of Publication:
- United States
- Language:
- English
Similar Records
Hyperspectral vegetation identification at a legacy underground nuclear explosion test site
Classification of multispectral or hyperspectral satellite imagery using clustering of sparse approximations on sparse representations in learned dictionaries obtained using efficient convolutional sparse coding