One of the key goals of strengthening maritime security is to increase maritime domain awareness, involving a combination of intelligence, surveillance, and operational information to build as complete a picture as possible to assess the threats and vulnerabilities in the maritime realm. Maintaining coherent situation awareness is essential for making informed timely decisions aimed at detecting and deferring threat and assessing the impact of those decisions more effectively. The problem of threat identification is complicated by number and types, sometimes unknown, of RF emitters in the littoral environments where the features used for their classification are highly multidimensional, possibly noisy, corrupt, and with large intra-class variations. Due to these input feature characteristics, existing algorithms are ineffective for dealing with complex unreliable and uncertain multi-dimensional multi-source data streams. We propose to confront the challenge of processing these data streams by designing an adaptive context-dependent multi-layer hybrid fusion process engine that combines heuristic and connectionist approaches to feature extraction, selection, and classification
Keywords: Neural Networks, Neural Networks, Context-Dependence, Reinforcement Learning, Data Quality, Multi-Model Multi-Sensor Data Fusion, Transferable Belief Model, Pattern Classifica
