【摘要】 针对旋转森林算法(rotation forest,RF)处理遥感影像分类时容易出现过拟合现象,以及极限学习算法(extreme learning machine,ELM)泛化性能较差问题,提出一种将旋转森林与极限学习相结合(RF-ELM)的影像分类算法。该方法首先用旋转森林算法对基分类器进行训练,然后利用极限学习算法作为基分类器解决旋转森林中存在的过拟合问题。通过利用Landsat-8遥感影像分别对比RF、ELM、Bag-ELM和RF-ELM进行分类实验。结果表明,所提出的集成方法比RF、ELM单一算法具有更高的分类精度,相比Bag-ELM具有更高泛化能力,有效改善了分类过拟合现象,计算效率也继承了ELM快速运算的特点。更多还原

【Abstract】 Aiming at rotation forest algorithm is prone to overfitting and limit learning algorithm has poor generalization performance when dealing with remote sensing image classification,an image classification algorithm combining rotational forest with extreme learning was proposed.Firstly,this method applied rotation forest algorithm to train the base classifier.Then the limit learning algorithm was used as the base classifier to solve the overfitting problem in the rotating forest.By using Landsat-8 remote sensing images,it compared RF,ELM,Bag-ELM and RF-ELM.The results show that the proposed method has higher classification accuracy than RF and ELM;compared with Bag-ELM,it has higher generalization ability which effectively improves the classification over-fitting phenomenon,and the calculation efficiency also inherits the characteristics of ELM fast calculation.更多还原