Abstract: To improve hyperspectral image classification accuracy,a classification method based on combination of spatial-spectral features and ensemble extreme learning machines is proposed in this paper.First,a spatial-spectral feature vector for each pixel is extracted using its neighboring information. Considering the strong correlation relationship between neighboring bands in a hyperspectral image,average grouping is performed for the extracted features,and a certain number of bands are first selected randomly from each interval and then combined to form a new feature with fewer dimensions.Extreme learning machine which can be trained fast is used to train a classifier.To improve the generalization performance of the learned model,several rounds of sampling are carried out based on ensemble learning theory,and several weak classifiers are trained and then combined to build a strong classifier using majority vote method.The classification experiments are performed using three typical hyperspectral image datasets,and the experimental results demonstrate that,the proposed algorithm can achieve preferable results compared with the state-of-the-art classifiers.It can achieve better classification accuracies when fewer training samples are used.The proposed algorithm has the advantages of few adjustable parameters,fast training speed,and high classification accuracy,and can be applied in many areas.

Key words: hyperspectral image classification, spatial-spectral feature, extreme learning machine, ensemble learning, feature sampling