Multiple Classifier Combination for Hyperspectral Remote Sensing Image Classification

Multiple classifier combination is used to hyperspectral remote sensing image classification in this paper, and some classifier combination algorithms are experimented. Based on a brief introduction to multiple classifier system and general algorithms, a modified evidence combination algorithm is adopted to handle evidence with high inconsistency, and a hierarchical multiple classifier system is designed to integrate the advantages of multiple classifiers. Using the OMIS hyperspectral remote sensing image as the study data, training samples manipulation approaching including boosting and bagging, together with parallel and hierarchical combination schemes are experimented. Mahalanobis distance classifier, MLPNN, RBFNN, SVM and J4.8 decision tree are selected as member classifiers. In the multiple classifier combination scheme based on training samples, both boosting and bagging can enhance the classification accuracy of any individual classifier, and boosting performs a bit better than bagging when the same classifier is used. In classification ensemble using multiple classifier combination approaches, both parallel combination using modified evidence theory and hierarchical classifier system can obtain higher accuracy than any individual member classifiers. So it can be concluded that multiple classifier combination is suitable for hyperspectral remote sensing image classification.