The scarcity of labelled training data as well as uneven class distribution among the limitedly available labelled data have posed a critical issue in supervised hyperspectral remote sensing image classification. Semisupervised methods can be an easy solution to this critical problem. However, traditional self-training based semi-supervised approaches often give poor classification results in high dimensional multiclass classification problems. This paper proposes a novel efficient one-class classifier ensemble based self-training approach for semisupervised classification of hyperspectral remote sensing images with limited labelled data. The proposed method initially trains an ensemble of locally specialized one-class classifiers independently by using the dimensionally reduced spectral feature vectors of the available labelled samples. The trained one-class classifiers are then used to extend the labelled set by iterative addition of high quality unlabelled samples to it through the exploitation of both spectral and spatial information. The classifiers are then retrained with the extended dataset in a batchwise fashion. The procedure is repeated until an adequate quantity of labelled samples are generated. Finally, a supervised multiclass classifier is trained on the extended dataset for the final image classification purpose. Experimental results on two benchmark hyperspectral images verify the effectiveness of the proposed method over supervised and traditional self-training based semisupervised pixelwise classification in terms of different classification measures.