SBIR/STTR Award attributes
The Navy’s AN/DVS-1 Coastal Battlefield Reconnaissance and Analysis (COBRA) system conducts unmanned aerial tactical reconnaissance in the littoral battlespace for detection and localization of minefields and obstacles in the surf zone and beach zone prior to an amphibious assault. COBRA collects multispectral UAV-based overhead imagery which must be examined for targets of interest: mines. Since mines exhibit specific spectral signatures and sizes, and are typically placed in linear patterns at regular spacing, e.g., “minelines,” there are algorithms currently in use to exploit these characteristics. Existing mine detection strategies universally assume they are placed in strictly linear patterns, so if mines are instead placed in a curvilinear pattern, the algorithm performance degrades and false alarms increase. An approach is needed that allows for the detection of non-linear arrangements of mines. The team of In-Depth Engineering (IEC) and Miami University (MU) proposes a solution that combines conventional machine learning approaches for mine-like object (MLO) detection with a Markov Chain Monte Carlo (MCMC) approach that searches for a broad class of mineline patterns, including curvilinear patterns. The proposed approach combines MLO detection with mine placement strategy so that mineline recognition can be used to improve the performance of MLO detection.
