Heterogeneous Multisensor Buried Target Detection Using Spatiotemporal Feature Learning
Abstract
The proposed research will investigate theory and algorithms for multi-sensor buried target detection that achieve high probability of detection and classification with low false-alann-rate. This project will undertake four tasks: 1) Investigate and develop spatiotemporal feature learning methods for effective detection and classification. Two approaches that will initially be investigated are deep learning and multiple kernel learning. 2) Investigate and develop multi-sensor fusion methods that combine sensor modalities for optimal detection perfonnance. The project will focus on sensor registration, heuristic fusion such as human- and bio-mimetic approaches, and non-linear fusion. 3) Develop spatiotemporal coherence processing for feature detection in space-time coherence information. The 5-dimensional space-time coherence space presents unique challenges and opportunities for feature learning. The project will focus on space-time manifold learning and coherence compression. 4) Investigate and develop physics-based features aimed at discriminating potential targets based on depth. Ground attenuation provides a unique property for feature learning in forward-looking coherent sensing. The project will investigate the physics of the air-ground interface to develop features by which surface clutter and buried targets can be discriminated. It will utilize simulated and archival data to explore the feature learning in environments with realistic variability. Government-furnished testing data will be used, as available, for test and evaluation purposes and, subsequently, for further development. Test and evaluation results will be analyzed according to target type and emplacement.
Document Details
- Document Type
- DoD Grant Award
- Publication Date
- Jan 12, 2017
- Source ID
- W911NF1610017
Entities
People
- Timothy C Havens
Organizations
- Army Contracting Command
- Michigan Technological University
- United States Army