Multiple instance ensemble learning method for high-resolution remote sensing image classification
高分辨率遥感影像分类的多示例集成学习

Multiple Instance Learning Via Embedded Instance Selection (MILES) has shown good performance in dealing with noisy training samples, but its bag prediction rule may introduce new uncertainty into the remote sensing image classification results. In order to overcome this limitation, two popular ensemble learning strategies, Bagging and AdaBoost are integrated with MILES. Two methods are proposed to constrain the uncertainty in remote sensing image classification: re-classification of coarse bags, and integration of MILES with diverse density and maximum likelihood classifier. The experimental results show that the uncertainty of remote sensing image classification can be obviously reduced by the integration of multiple instance learning with ensemble learning.