基于动态集成的遥感图像分类

鉴于多分类器集成能够获得比单个分类器更好的性能,但是对于支持向量机(support vector ma-chine,SVM),一般的集成方法很难达到效果.特提出了基于局部精度(local accuracy,LA)的动态集成算法.首先,通过多种方法产生个体分类器;其次,利用验证数据集来定义LA,LA用来衡量各个体分类器的权重以及判断是否挑选该个体分类器的标准;最后,在某研究区的遥感图像数据集上进行实验.实验结果表明,动态集成的效果要优于静态集成,特别是异类动态集成效果最好.静态集成只考虑了分类器在训练样本中的表现而没有考虑测试样本的特征,对于动态集成,可以根据测试样本在验证集上的表现来选择个体分类器,因此它展现出更好的性能.

Remote sensing images classification based on dynamic ensemble

Multi-classifiers ensemble is better performance than single classifier.However,it is difficult for normal support vector machines(SVM) ensemble to get expected result.A dynamic ensemble algorithm based on local accuracy(LA) was introduced.At first,generated several individual classifiers.Secondly,defined the LA on validation dataset.LA was used to weight the individual classifiers and decide whether to choose the individual classifier.At last some remoted sensing datasets of study site were used for experi...