Target Detection in Hyperspectral Imagery Using Sparse Learning Systems

Abstract

This project will address target detection from hyperspectral imagery using a supervised learning system. The proposed system will consist of feature extraction, feature selection, and a sparse representation-based classification. Data sparsity will be considered in feature and training sample spaces. For feature extraction, the original hyperspectral data will be decomposed into uncorrelated sparse components using principal component analysis and independent component analysis. A sparsitymodel will be built for classification and feature selection with each pixel represented as a sparse linear combination of training samples in a selected feature space. To alleviate the problem introduced by lack of sufficient training data, feature selection will be embedded in the learning procedure in order to reduce over-fitting. The problem will be formulated as a minimization of the approximation error in the feature space, with a certain sparsity level. A Laplacian prior will be added to the model to enhance spatial correlation in local neighborhoods.

Document Details

Document Type
DoD Grant Award
Publication Date
Feb 17, 2016
Source ID
W911NF1510521

Entities

People

  • Behzad Kamgar-Parsi

Organizations

  • Army Contracting Command
  • Fort Valley State University
  • Office of the Secretary of Defense

Tags

Fields of Study

  • Computer science

Readers

  • Computer Vision.
  • Neural Network Machine Learning.

Technology Areas

  • AI & ML
  • Space