Remote Seneing Image Classification Based On Ada Boost Integrated Classifier

In recent years,with the rapid development of remote sensing technology,more and more high-performance sensors have been applied.These sensors have improved the spatial resolution and spectral resolution of remote sensing images.Hyperspectral data is an excellent data source because of its high spectral resolution.In addition,Hyperspectral data has the advantages to identify the features by combining the spectral and space information.It is one of the commonly used data sources for the classification of features.In classification work,the complexity of the actual problem leads to the difficulty of directly training strong classifiers(such as neural networks).Generally speaking,weak classifiers(such as decision trees)are easy to train,but it can't satisfy the requirements for accuracy.In recent years,with the development of machine learning technology,especially the continuous development of integrated learning technology,it has become a possible idea to solve the classification problem by using a combination model of multiple weak classifiers.AdaBoost(adaptive boosting)is one of the widely used methods in the field of integrated learning.The integration strategy of the algorithm is to build a weighted combination model.Applying integrated learning technology to remote sensing image classification has certain feasibility and research value.This paper is based on the AdaBoost integration method,using classification regression tree and convolutional neural network as the basic classifier for hyperspectral image classification.The main work is as follows:(1)Based on the AdaBoost method,the sample weights and decision tree weights are introduced into the integrated model.The overall classification accuracy on the two data sets is 0.8670 and 0.7982 that is better than comparative methods.(2)Training sample expansion by the Pixel Pair method to avoid insufficient training caused by small amount of sample data.Based on the AdaBoost method,the convolutional neural network is integrated and the overall classification accuracy on the two data sets is 0.8284 and 0.8396 that is better than comparative methods.