Textile Fingerprinting for Dismount Analysis in the Visible, Near, and Shortwave Infrared Domain
Abstract
The ability to accurately and quickly locate an individual, or a dismount, is useful in a variety of situations and environments. A dismount's characteristics such as their gender, height, weight, build, and ethnicity could be used as discriminating factors. Hyperspectral imaging (HSI) is widely used in e orts to identify materials based on their spectral signatures. More speci cally, HSI has been used for skin and clothing classi cation and detection. The ability to detect textiles (clothing) provides a discriminating factor that can aid in a more comprehensive detection of dismounts. This thesis demonstrates the application of several feature selection methods (i.e., support vector machines with recursive feature reduction, fast correlation based lter) in highly dimensional data collected from a spectroradiometer. The classi cation of the data is accomplished with the selected features and arti cial neural networks. A model for uniquely identifying ( ngerprinting) textiles are designed, where color and composition ard deternimed in order to ngerprint a speci c textile. An arti cial neural network is created based on the knowledge of the textile's color and composition, providing a uniquely identifying ngerprinting of a textile. Results show 100% accuracy for color and composition classi cation, and 98% accuracy for the overall textile ngerprinting process.
Document Details
- Document Type
- Technical Report
- Publication Date
- Mar 01, 2014
- Accession Number
- ADA601950
Entities
People
- Jennifer S. Yeom
Organizations
- Air Force Institute of Technology