Hyperspectral image classification with limited labeled samples is a challenging task and still an open research issue. In this article, a novel technique is presented to address such an issue by exploiting dimensionality reduction, spectral-spatial information, and classification with active learning. The proposed technique is based on two phases. Considering the importance of dimensionality reduction and spatial information for the analysis of hyperspectral images, Phase I generates the patterns corresponding to each pixel of the image using both spectral and spatial information. To this end, first, principal component analysis is used to reduce the dimensionality of an hyperspectral image, then extended morphological profiles are exploited. The spectral-spatial information based patterns generated by extended morphological profiles are used as input to the Phase II. Phase II performs the classification task guided by an active learning technique. This technique is based on a novel query function that uses uncertainty, diversity, and cluster assumption criteria by exploiting the properties of k-means clustering, K-nearest neighbors algorithm, support vector machines, and genetic algorithms. Experiments on three benchmark hyperspectral datasets demonstrate that the proposed method outperforms five state-of-the-art active learning methods.