Research On The Hyperspectral Image Classification Based On Ensemble Learning

Hyperspectral images contain hundreds of spectral bands,which are very rich in feature information and have been extensively studied in the field of remote sensing.Among them,the classification of hyperspectral images has always been a research hotspot in the field of remote sensing.In the field of hyperspectral image classification,the acquisition of labeled samples is difficulty.Many classification methods have been studied in order to solve the problem of insufficient labeled samples.Among them,the semi-supervised learning method between supervised learning and unsupervised learning has received great attention and research because it can make full use the information of unlabeled samples.In addition,the ensemble learning method can obtain a strong classifier by combining multiple basic classifiers,and the classification accuracy can be further improved.The multiple basic learners lack effective communication in the traditional ensemble learning classification method.In view of the above analysis,this paper proposes two classification methods of hyperspectral image based on ensemble learning:(1)An ensemble learning classification method based on multiple feature images is proposed.The main idea of this method is to obtain a variety of feature images by Gaussian filter and rolling guide filtering,and then the feature images are classified by the Support Vector Machine(SVM).The adaptive enhanced learning method is used to flexibly and organically combine the multiple kernels.The process of combination does not require complicated calculations.The experimental results are carried out on the Indian Pines,University of Pavia and Salinas datasets.The experimental results show that the proposed hyperspectral image classification method based on multiple feature images can effectively improve the classification accuracy compared to RMKL and S2MKL method.(2)A hyperspectral image classification method based on multiple kernel mutual learning is proposed.The informative unlabeled samples are selected by using the concept of information entropy.The ensemble learning method is used to combine the final classification results to further improve the classification accuracy of the hyperspectral image.First,each basic classifier uses the randomly selected training sample set for pre-training to obtain preliminary classification results.Secondly,each basic classifier calculates the entropy value of the sample according to the concept of information entropy.The samples with high-entropy value were selected.The selected informative unlabeled samples compose an undetermined sample set;then,the pseudo-labels of the selected undetermined sample set are negotiation and determined between the process of mutual learning of multiple basic classifiers,thereby,a new training samples set with pseudo-labels is formed;Finally,a new pseudo-labeled sample is added to the initial training sample set to retrain the model.The proposed method is tested on three real hyperspectral image datasets.The experimental results show that the proposed method has better classification performance than other comparison methods.