基于区间重叠度的可拓模式识别方法

Extension pattern recognition method based on interval overlapping degree

针对可拓模式识别时特征的量值区间重叠会影响识别效果的问题,提出基于区间重叠度的可拓模式识别方法（Extension Patter Recognition Method Based on Interval Overlapping Degree,IOD-EPRM）。该方法利用区间重叠度可客观反映特征重要程度的特点,构造合适的区间重叠度到特征权重的映射,来反映特征在计算综合关联度中的比重,以降低量值区间重叠对识别率的影响。采用UCI数据库中iris数据和wine数据进行性能验证,结果表明,IOD-EPRM将可拓模式识别率提高了7%~9%,且具有受抽取样本数量影响小的特点。

The extension pattern recognition method based on interval overlapping degree（IOD-EPRM）is proposed to effectively solve the problem of recognition rate decrease which results from the overlapping of value ranges.In this method,making use of the property that interval overlapping degree could reflect the importance degree of the character in the extension pattern recognition,constructing appropriate mapping function from interval overlapping degree to character weight could decrease the affect of the character which is seriously overlapped.The simulation test with the data of iris and wine among UCI database prove that the recognition rate of IOD-RPRM is about 7% to 9% higher than the extension recognition method,and subjects little effect to the number of the selected data.