I. Introduction
The classification accuracy of a classifier is directly influenced by the quality of the training data used [1]. However, real-world data often suffer from noise [2], [3]. Labeling training instances is a costly and rather subjective task that usually induces some labeling errors in the training set [4], [5]. In particular, mislabeled training data are inevitable in remote sensing [6]. Mislabeled training data can lead to poor classification performances in supervised classifiers, including ensemble classifiers. Classification is one of the major tasks in remote sensing information processing [7]. However, the presence of noise in remote sensing imagery degrades the interpretation ability of the data [8], [9].