This paper introduces a novel semi-supervised tri-training classification algorithm based on diversity measurement for hyperspectral imagery. In this algorithm, three measures of diversity, i.e., double-fault measure, disagreement metric and correlation coefficient, are applied to select the optimal classifier combination from different classifiers, e.g., support vector machine (SVM), multinomial logistic regression (MLR), extreme learning machine (ELM) and k-nearest neighbor (KNN). Then, unlabeled samples are selected using an active learning (AL) method, and consistent results of any other two classifiers combined with a spatial neighborhood information extraction strategy are employed to predict their labels. Moreover, a multi-scale homogeneity (MSH) method is utilized to refine the classification result with the highest accuracy in the classifier combination, generating the final classification result. Experiments on three real hyperspectral data indicate that the proposed approach can effectively improve classification performance.