Hyperspectral Image Classification Based On Deep Learning Ensemble

Hyperspectral image(HSI)classification technology has always been one of the hot issues in the field of remote sensing.With the further development of hyperspectral imaging systems,the spatial resolution of hyperspectral data will be higher,the information will be more abundant,and the traditional hyperspectral data classification Technology cannot efficiently utilize spatial and spectral information of hyperspectral data.In order to effectively combine the spatial information and spectral information of hyperspectral data to improve the classification accuracy,this paper studies the hyperspectral data classification method based on deep ensemble learning.The main contents of this article include:Firstly,it introduces the basic knowledge of hyperspectral data,deep learning and ensemble learning,including the characteristics of hyperspectral data,and the classification of hyperspectral data.It introduces several traditional methods of classification of hyperspectral data in detail,and has learned and implemented several based on the traditional hyperspectral data classification algorithm,the shortcomings of these traditional classification methods are analyzed.Next,in order to improve the accuracy of hyperspectral data classification,this paper uses a hyperspectral data classification method based on deep learning ensemble.The specific method is to randomly extract the spectral dimension of hyperspectral data,and use the deep convolutional neural network and Res Net model as an individual.The sub-classifier finally determines the class of the last sample by performing the majority vote of all sub-classifiers,and obtains better classification accuracy than the traditional classification method.In order to further improve the classification accuracy,the purpose of transfer learning has effectively achieved this goal.The learning weight of the trained deep convolutional neural network model is transferred to another deep convolutional neural network model,which effectively improves the training speed of the model and classification accuracy.Finally,this paper studied an algorithm for classifying hyperspectral data based on ensemble learning and random forest.First,the hyperspectral data is randomly selected in the spectral dimension to establish a forest,followed by a random extraction of the spatial dimension in the forest to form a tree.The deep convolutional neural network model is also used as an individual sub-classifier.Weight transfer is performed by means of ensemble learning,and the majority of voting methods determine the final class.It is proved from experiments that this method is a powerful solution to the classification of hyperspectral data than the traditional classification method.