一种自适应子融合集成多分类器方法

融合集成方法已经广泛应用在模式识别领域,然而一些基分类器实时性能稳定性较差,导致多分类器融合性能差,针对上述问题本文提出了一种新的基于多分类器的子融合集成分类器系统。该方法考虑在度量层融合层次之上通过对各类基多分类器进行动态选择,票数最多的类别作为融合系统中对特征向量识别的类别,构成一种新的自适应子融合集成分类器方法。实验表明,该方法比传统的分类器以及分类融合方法识别准确率明显更高,具有更好的鲁棒性。

An adaptive sub-fusion integration classification methodLI Min1 PAN Ying1 LI Hua1 CHENG Mao-hua2

<sub>Fusion integration method has been widely used in the field of pattern recognition. However, some base classifiers have poor real-time performance stability, </sub><sub>which c</sub><sub>auses poor performance of multiple classifiers. A new multi-classifier-based sub-fusion integration classification is proposed for the above problems. </sub><sub>T</sub><sub>his method considers the dynamic selection of various classifiers at the level of measurement layer fusion, </sub><sub>t</sub><sub>he category with the highest number of votes is the category identified by the feature vector in the fusion system</sub><sub> to constitute a new adaptive sub-fusion integration classifier method</sub><sub>. Experiments show that this method is significantly more accurate than conventional classifiers and classification fusion methods and has better robustness.</sub>