This lecture has provided an exhaustive list of platform attributes needed for evidential reasoning techniques that aim to provide an ID within a well-defined taxonomy tree (several of which were presented others can be found in [10,11,13,17]). It also has presented the formalism and real-world examples for four well-known reasoning frameworks, mostly applicable to Level 1 and 2 fusion, a partial list being: (1) fuzzy logic, in particular its use in fuzzification (pre-processing) for other means of reasoning, but also with an ESM application utilizing fuzzy combination rules and defuzzification (2) NNs, particularly useful for large sets of data, such as imagery datasets, which are easily decomposable into training, validation and test sets, with FLIR and SAR examples detailed (3) Bayesian approach (with a priori information) with an application to classifiers where an attribute stands out as a discriminator 9, such as line ship length for a SAR classifier) (4) DS approach with its variants: the original orthogonal sum and its renormalization though conflict, followed by the HOS and its novel conflict resolution through the taxonomy tree, and finally generic considerations about truncating the exponentiation of proposition (NP-hard aspect of the problem). In addition, examples were given where all of the above methods form crucial parts of a larger more versatile classifier.
展开▼
