A Study Of Application Of Rough Set Theory In Data Fusion On The Decision Level

Nowadays data fusion technology has become a bright spot or a hot spot in IT and its application includes many fields, such as military and civil ones. For data will be missing by all kinds of reasons in the transiting procedure and minimum attribute reduction problem has been proved a NP-hard one, finding a effective and practical data complement and attribute reduction algorithm always are tempting puzzles to many researchers. Though we have some comparatively mellow algorithms to resolve the two problems, they all have some flaws. Because of the rough set theory have some particular advantages than other tools, such as vagueness set and D-S evidence theory, so this thesis selects it to study the decision level's data fusion.Firstly, this thesis deeply analyzes a classical data complement algorithm and it's improved algorithms, finds many flaws of them, which includes the competency of the filling data can't be controlled, the application scope of the algorithms is restricted, etc. This thesis, based on valued tolerance relation, presents an improved algorithm by introducing a tolerance threshold and a normalized distance function to give a further improvement. The new algorithm improves the depicting approach of two piece of data's comparability, and avoids the unilateral description of two piece of data's comparability. It reduces the cost of time and space computation, realizes the control ability of filling qualification, and enlarges the application scope of the old algorithms. Secondly, by validating and analyzing, this thesis finds a flaw in an attribute reduction algorithm, which cause the algorithm can't effectively deal with the attribute reduction when the cardinal number of core equals zero. This thesis not only analyzes the source of the flaw but also points out that the flaw widely exists in many attribute reduction algorithm. Aimming at the flaw this thesis introduces hypo-core and real-core concepts and a series of their concerning theorems. Based on these concepts and theorems this thesis improved the attribute reduction algorithm. The new algorithm amends the flaw in the original, enlarges the application scope.For the quality of decision rules depends on the data complement and attribute reduction algorithms of the decision level of data fusion, the two improved algorithms hopely have better performances than the the originals.