Hyperspectral Image Classification Based On Integrated Learning

With the rapid development of imaging spectroradiometer, the spectral resolution of data which generated by it becomes higher, the traditional data analysis algorithms can not meet the demand of hyperspectral data processing.Ensemble learning, as the field of machine learning hot research topic, introduced to the hyperspectral data processing, utilizing many learning algorithms to solve a prediction, has played a significant role.This dissertation introduces ensemble learning into hyperspectral image classification.Support vector data description ensemble and support vector machine ensemble are studied.The main contributions of our works in this dissertation are summarized as follows:1. A support vector data description ensemble classification method is presented in the dissertation.Combining with the data characteristics of hyperspectral image, this dissertation uses principal component analysis to do data dimension reduction process, and then trains support vector data description with the training samples which were randomly selected, at last, ensembles the results.Experimental results show that the ensemble method proposed can improve the accuracy of hyperspectral image classification in a certain extent.2. A support vector machine ensemble classification method is presented in the dissertation.Spatio-spectrum feature extract is used to extract the data feature, and then active learning is used to select training samples.Using support vector machine as the weak learning of AdaBoost algorithms to classify hyperspectral image.The result shows that the above ensemble method can effectively improve the classification accuracy.