High-dimensional features and limited labeled training samples often lead to dimensionality disaster for hyperspectral image classification. Semisupervised learning has shown great significance in hyperspectral image processing. A semisupervised classification algorithm based on spatial-spectral clustering () was proposed. In the proposed framework, spatial information extracted by Gabor filter was first stacked with spectral information. After that, an active learning (AL) algorithm was used to select the most informative unlabeled samples. Then a probability model based support vector machine combined with the technique was used to predict the labels of the selected unlabeled data. The proposed algorithm was experimentally validated on real hyperspectral datasets, indicating that the proposed framework can utilize the unlabeled data effectively and achieve high accuracy compared with state-of-the-art algorithms when small labeled data are available.