Multisensor Analysis and Algorithm Development for Detection and Classification of Buried and Obscured Targets
Abstract
An important goal for the U.S. Army is remediating the threats of buried and obscured explosive hazards, as these devices cause uncountable deaths and injuries to both Civilians and Soldiers throughout the world. To support this goal we propose to investigate theory and algorithms for multisensor target detection that achieve high probability of detection and classification with low false-alarm-rate. The primary sensors of interest are multisensor forward-looking GPR (i.e., FLGPR plus other sensor modalities, such as video and LiDAR), side-looking (SL) GPR, acoustic, and Ku-band radar, although our methods will be applicable to other modalities as well. The objective of this project is to learn feature representations for target-in-environment responses that enable highly effective computer-aided detection and classification (CAD/CAC) of buried and obscured targets with multisensor systems. The key challenge we address is recognition of environmental contexts and discrimination between targets and clutter in those contexts. This project will focus on four main tasks: i) radar data analysis, ii) optical data analysis, iii) obscured target analysis, and iv) automated detection and classification algorithms. This project will undertake four related primary tasks and one supporting task. 1) Radar Data Analysis: Evaluate data quality to identify sensor or processing artifacts and develop processing methods to improve image metrics and prepare data for downstream CAD/CAC. Evaluate target and clutter signatures / features with an emphasis on developing an understanding of the phenomenology to identify signature and feature information for optimal CAD/CAC. Support data analysis with laboratory measurements and simulation. Produce penetrative 3D radar imagery using novel imaging algorithms incorporating various regularizations to enhance target observability. 2) Optical Data Analysis: Evaluate image data quality and develop processing an algorithmic methods to improve data quality, such as multi-frame image alignment or superresolution methods. Evaluate target and clutter signatures / features with an emphasis on developing a deep understanding of the phenomenology to identify signature and feature information that can inform CAD/CAC algorithms. Simulate target signatures, comparing to measurements and/or evaluating over a wide parameter space to increase understanding of sensor physics and associated target features. 3) Obscured Target Analysis: Investigate the impact of obscuration, cultural or natural, on target signatures and features for both radar and optical modalities. Augment collected data with simulations and/or laboratory experiments to increase understanding of multisensor physics. Develop multisensor processing and imaging methods to produce imagery and features that are more effective in the multisensor domain, e.g., induced vibration. 4) End-to-End Automated CAD/CAC: Investigate and develop feature learning algorithms based on physics knowledge gained in Tasks 1-3 and state-of-the-art machine learning methods. Develop multisensor fusion algorithms for optimal CAD/CAC of targets-in-environment. The final task of this project is comprehensive evaluation and testing of the various components of our work on government-furnished multisensor data.
Document Details
- Document Type
- DoD Grant Award
- Publication Date
- Jan 12, 2017
- Source ID
- W911NF1610241
Entities
People
- Timothy C Havens
Organizations
- Army Contracting Command
- Michigan Technological University
- United States Army