摘要
针对现有直觉模糊核匹配追踪算法采用部分样本进行训练和停机策略而导致学习机泛化能力下降的缺陷,结合集成学习的思想,提出了一种基于直觉模糊核匹配追踪集成的目标识别方法。该算法通过采用样本扰动和参数扰动的二重扰动策略产生子学习机,并利用多数投票法对其识别结果进行融合,从而提高了集成学习机的分类精度和泛化性能。实验结果表明,与传统方法相比,该方法可获得更优的识别效果,有效提高了识别精度,并能避免采样学习带来的不稳定性。
Considering that the generalization of the learning machine performed poorly in the present intuitionistic fuzzy kernel matching pursuit algorithm(IFKMP) due to its training method and stopping criteria, a new recognition method based on intuitionistic fuzzy kernel matching pursuit ensemble(IFKMPE) was proposed by introducing the idea of ensemble learning. In IFKMPE, the double perturbation strategy including sample and parameter perturbation was applied to generate the sub-learning machine, the recognition results were fused by the principle of majority voting, and therefore both the classify accuracy and generation ability were enhanced. Simulation results show the new algorithm IFKMPE performs better in terms of recognition accuracy and stability of sample learning compared with the traditional ones.
出处
《通信学报》
EI
CSCD
北大核心
2015年第10期165-171,共7页
Journal on Communications
基金
国家自然科学基金资助项目(61272011
61309022
61402517)~~
关键词
直觉模糊集
核匹配追踪
集成学习
目标识别
intuitionistic fuzzy set
kernel matching pursuit
ensemble learning
target recognition