Abstract: The traditional ensemble learning algorithm was a combination of the individual classifiers, which couldn't reflect the differences of individual classifiers and effectively improve the recognition rate of ensemble classifier. A measurement method based on Normalized Mutual Information was proposed, and the Hungarian algorithm was used to match the labeled vectors of individual classifiers. On the basis of that, a AP (Affinity Propagation) clustering integrated strategy was proposed based on component data. Compared with the classical algorithm on the remote sensing image, the results showed that the proposed algorithm has certain advantages in classification performance.