Undersea Object Recognition and Anomaly Detection via Manifold Mapping - N00014-21-S-SN07
Abstract
This effort focuses on applying recent research developed at the University of Rhode Island to detect and classify underwater obstac les (e.g., mine detection, etc.) by UXVs. By combining machine learning with the evolving and advanced discipline of Computational Topology, a new approach will be demonstrated that better identifies anomalies and objects in the ocean than existing methods. We present CLAM (Clustered Learning of Approximate Manifolds), a fast, hierarchical clustering technique that learns a manifold in a Ba nach space defined by a distance metric. CLAM induces a graph from the cluster tree, based on overlapping clusters determined by sev eral geometric and topological features. On these graphs, we implement CHAODA (Clustered Hierarchical Anomaly and Outlier Detection Algorithms), exploring various properties of the graphs and their constituent clusters to compute scores of anomalousness. Prelimina ry research results show this approach supports anomaly detection and object detection, outperforms other state-of-the art approache s, and scales in O(n log n) time. To demonstrate our approach is applicable for use in real-time environments such as in UXVs, prot otype software previously developed in Python will be redesigned and coded in Rust a relatively new computer language designed for safety-critical systems with runtime performance comparable to software written in C.
Document Details
- Document Type
- DoD Grant Award
- Publication Date
- Sep 07, 2021
- Source ID
- N629092112058
Entities
People
- Thomas Santos
Organizations
- Office of Naval Research
- Rite-Solutions (United States)
- United States Navy