摘要： 在按照“测试-选择”方法设计多分类器系统时，从超量生成的候选分类器集中选取一个最优子集是关键环节之一。基于此，定义一个组合适宜度概念，提出一种新的分类器选择方法。将该方法用于高光谱遥感数据分类实验中，并从具有27个候选的分类器集中挑选子集。实验结果表明，该方法在选择效率和识别精度方面具有优势，能保证所选子集的泛化能力。

关键词: 组合适宜度, 分类器选择, 高光谱数据

Abstract: Selecting an optimal subset of classifiers from overproduced candidates is a key step when the “test-and-select” method is employed to design a multiple classifier system. This paper therefore defines a new concept——Degree of Combination Fitness(DCF) and presents a new DCF-based classifier selection method. In its application to experiments of hyperspectral data classification, the new method selects a subset from a pool of twenty-seven candidate classifiers. Results show that in comparison to other popular methods, the proposed method has advantages both in efficiency and recognition accuracy. Besides, it can guarantee the generalization ability of the selected set as well.

Key words: Degree of Combination Fitness(DCF), classifier selection, hyperspectral data

中图分类号: 

• TP391.4