Information Aggregation for IED Identification with GPR, Video, and Electromagnetic Induction: Within Sensor Processing, Multi Sensor Fusion, and Large-Scale Learning
Technical Report,27 Mar 2013,26 Mar 2018
DUKE UNIV DURHAM NC DURHAM United States
Pagination or Media Count:
The fundamental objectives of this work are to use modern machine learning techniques to 1 develop new algorithms for both prescreening and object discrimination to support the HMDS program 2 assess the utility of information-theoretic approaches that we have been developing for other sponsors for consideration by the various NVESD programs, particularly the incoming forward looking and handheld programs 3 develop algorithms for any other sensors of interest to the sponsor and assess their performance. Historically, algorithm development work has included downward and forward looking GPR, IR, hyperspectral, acoustic, seismic, EMI, and video sensing modalities. For HMDS we have been carefully considering robustness issues with respect to target localization for feature extraction, modifying one of our previously developed prescreeners, and investigating convolutional neural networks as a new potentially effective processing algorithm.