Designing and Implementation of the Nonsymmetrical Hierarchical Model for Solving Complicated Problems of Pattern Recognition

Aiming at solving complicated problems of pattern recognition, such as recognition and real-time forecasting of catastrophic storms, a intelligent three-level model with nonsymmetrical is proposed, which could progressively recognize one object from coarse to fine, then identify and compound the results. The data-driven bottom level is committed to rough classification, the feature-driven internal level is aimed at discerning the local sub-problem, and the rule-driven top level works at the integration of multiple information. As a whole, the rules are used throughout the model. The space-time blend problems are disassembled into local sub-problems in time and space, and then these sub-problems are solved by using corresponding time and space rules respectively, so the complex problems become easily to be dealt with. Experimental results show that through using this model the accuracy rate of the identification and prediction for storms is improved, the missing rate and vacancy rate are dropped. In addition, it can give comprehensive identification and prediction for the weather with much-type storms.