Automated Curvilinear Mineline Detection

This project will advance the capabilities of the Coastal Battlefield Reconnaissance and Analysis (COBRA) system for detecting minelines laid out in curvilinear configurations. Currently, the COBRA minefield detection (MFD) algorithms detect either straight mineline (patterned MFD) or randomly distributed mine (scattered MFD) configurations. The former capability typically cannot detect minelines with significant curvature, and the latter fails to utilize patterned information available with mineline configurations. Areté Associates (Areté) will develop algorithms and software for automated curvilinear mineline detection (CMD) by assimilating multiple types of evidence and evaluating the evidence with machine learning inferencing methods. The result will improve the COBRA automated MFD performance, thereby facilitating post-mission review and reducing mission timelines for the Navy.Challenges to automated CMD include overcoming the effects of numerous mine-like object (MLO) false alarms and missed MLO detections. To mitigate these issues, it will be necessary to collect and utilize multiple types of evidence from COBRA data, including: 1) MLO attributes, 2) curvilinear mineline configuration constraints, 3) natural terrain landform contexts, and 4) optional a priori information. Although some evidence may not be present and can vary in strength, by combining the evidence using machine learning methods, automated CMD performance will be optimized.