Target Detection and Scene Classification with VNIR/SWIR Spectral Imagery
Abstract
Spectral imagery provides a new resource in remote sensing, which can be used for defeating camouflage, concealment and detection, as well as terrain categorization. A new sensor, the Night Vision Imaging Spectrometer (NVIS), provides VNIR/SWIR (0.4-2.5 m) spectra, which are used to here to study such applications. NVIS has a nominal GSD of 0.5- 1.5 meters in operational modes utilized for this work, which make the data well suited for studying mapping and classification algorithms. Data taken at Pt. A.P. Hill on April 29, 2000 are studied here. A Principal Components Transformation was performed on the NVIS data. From this new data set, target spectra were collected for use in classification algorithms. The NVIS data was converted from radiance to reflectance in two different ways: Empirical Line Method and Internal Average Relative Reflectance. Using this data, various standard algorithms were performed. It was found that while none of the algorithms correctly classified all of the selected targets, the Mahalanobis Distance and Mixture Tuned Matched Filter(TM) algorithms were the most successful.
Document Details
- Document Type
- Technical Report
- Publication Date
- Sep 01, 2000
- Accession Number
- ADA384999
Entities
People
- David R. Perry
Organizations
- Naval Postgraduate School