An Ensemble of Classifiers Based on Positive and Unlabeled Data in One-Class Remote Sensing Classification
Ran Liu,Wenkai Li,Xiaoping Liu,Xingcheng Lu,Tianhong Li,Qinghua Guo
DOI: https://doi.org/10.1109/jstars.2017.2789213
IF: 4.715
2018-01-01
IEEE Journal of Selected Topics in Applied Earth Observations and Remote Sensing
Abstract:One-class remote sensing classification refers to the situations when users are only interested in one specific land type without considering other types. The positive and unlabeled learning (PUL) algorithm, which trains a binary classifier from positive and unlabeled data, has been shown to be promising in one-class classification. The implementation of PUL by a single classifier has been investigated. However, implementing PUL using multiple classifiers and creating classifier ensembles based on PUL have not been studied. In this research, we investigate the implementations of PUL using several classifiers, including generalized linear model, generalized additive model, multivariate adaptive regression splines, maximum entropy, backpropagation neural network, and support vector machine, as well as three ensemble methods based on majority vote, weighted average, and weighted vote combination rules. These methods are applied in classifying the urban areas from four remote sensing imagery of different spatial resolutions, including aerial photograph, Landsat 8, WorldView-3, and Gaofen-1. Experimental results show that classifiers can successfully extract the urban areas with high accuracies, and the ensemble methods based on weighted average and weighted vote generally outperform the individual classifiers on different datasets. We conclude that PUL is a promising method in one-class remote sensing classification, and the classifier ensemble based on PUL can significantly improve the accuracy.