Automated Curvilinear Mineline Detection

The objective of this project is to develop algorithms and real-time software for detecting curvilinear minelines to address the issue of meeting Coastal Battlefield Reconnaissance and Analysis (COBRA) Block I minefield detection (MFD) performance requirements for minefields with curvilinear minelines. To meet this objective, Areté’s approach is to use computational geometry and graph-theoretic (CG-GT) algorithms to detect candidate minelines, along with machine learning algorithms that fuse available types of evidence for false alarm mitigation. Areté will develop these algorithms, validate them, and implement them in the COBRA Block I software architecture paradigm such that they can be inserted easily into the COBRA Block I -3 Real Time Processor (RTP). The delivered software will provide the Navy with a capability for improved COBRA MFD performance in cases where curvilinear minelines are present. During Phase I of this project, Areté created statistically significant datasets for developing and testing algorithms and developed a software framework for investigating algorithms. This effort provided a basis for creating curvilinear minefield detection (CMFD) algorithms based on CG-GT methods. With these algorithms, Areté demonstrated a CMFD capability that achieved ~66% reduction in false alarm density in COBRA imagery with curvilinear minelines compared with the current COBRA patterned MFD (PMFD) and scattered MFD (SMFD) algorithms at the same probability of detection. Based on the Phase I results Areté has concluded that using CG-GT and machine learning methods for CMFD is a promising and feasible approach that should be continued in a Phase II effort. In particular, Areté recommends that the Phase I algorithms be extended and optimized, particularly with regard to false alarm mitigation. Further, the algorithms need to be validated on large datasets to reveal the limits of their performance for various conditions. Finally, the algorithms should be implemented in C++ and CUDA, optimized for real-time operation, and validated to prepare for insertion into the COBRA Block I -3 RTP in a Phase III effort.