Stacking Learning在高光谱图像分类中的应用

Application of Stacking Learning in hyperspectral image classification

高光谱图像分类研究中,集成学习能够显著地提高分类效果。但是传统的并行多分类系统对基础分类器有较高要求,即要求差异性及分类均衡。为了解决这一问题,采用Stacking Learning的堆栈式学习方式,首先使用K-Fold和交叉验证的方式进行数据分割和训练,将原始特征进行特征变换,重新构建二级特征。再使用新特征进行对Meta分类器进行训练得到判决分类器,用于样本的最后分类判断。实验结果表明,采用的Stacking Learning方法不依赖基础分类器,且相比较于传统的多分类系统具有更高的精度和良好的稳定性。

In the study of hyperspectral image classification,integrated learning can significantly improve the classification effect.However,traditional parallel multiple classifier system has higher requirements for the basic classifier,namely,the diversity of requirement and the equalization of classification.In order to solve this problem,we use Stacking Learning method.Firstly,K-Fold and cross-validation were used to segment and train the data.The original features were transformed and the secondary features were reconstructed.Further the new feature was used to train the Meta classifier to obtain a decision classifier for the final classification of the sample.Experimental results show that the Stacking Learning method does not rely on basic classifier and has higher accuracy and stability than traditional multiple classifier system.