Abstract: Nowadays,most SVM-based remote sensing classification methods are challenged by incorrectly selecting parameters values and the small sample problems．This paper proposed a novel ensemble model with semisupervised SVM (EPS3VM) to address the problem of remote sensing images classification．The key characteristics of this approach are to 1)self-adaptive mutation particle swarm optimizer is introduced to improve the generalization performance of the SVM classifier (PSVM),2)self-training semisupervised learning method that leverages large amounts of relatively inexpensive unlabeled data is presented to produce a number of semisupervised classifiers (PS3VM)．Then by the weighted voting method,these classifiers are combined so as to improve the generalization ability of the classification model．In order to reduce the impact of this issue by incorrect labels,Gustafson-kessel fuzzy clustering algorithm (GKclust) is used for selecting the useful points from the unlabeled set．The effectiveness of the proposed classification approach is de-monstrated for identifying different land cover regions in multispectral remote sensing imagery．In particular,the perfor-mance of the EPS3VM is compared with PSVM and PS3VM in terms of classification accuracy and kappa coefficient．On an average,the EPS3VM model yields an overall accuracy of 96.88% against 88.48% for PSVM and outperformed PS3VM in terms of overall accuracy (by about 5%)．The obtained results clearly confirm the effectiveness and robustness of the EPS3VM approach to the remote sensing land cover classification.

Key words: Support vector machines,Semisupervised learning,Ensemble learning,Gustafson-kessel fuzzy clustering,Land cover,Classification