Remote Sensing Image Classification Method Based On Object Oriented And Ensemble Learning

The rapid development of remote sensing technology, we can get richinformation on the Earth’s surface, especially remote sensing sensor platformtechnology continues to progress. The emergence of high resolution remote sensingimages, extends the degree of awareness of the nature of things. Compared to mediumand low resolution remote sensing images, high resolution remote sensing imageperformance richer imaging features of the spectrum, shape, texture a nd contextsemantic information. How to handle and apply these data, we are faced withchallenges and to be resolved the problem. For high resolution remote sensing imageinformation extraction, the use of traditional pixel based spectral brightness valuemethod will not only reduce the accuracy of the information extraction, and couldlead to spatial data redundancy, waste of resources.Therefore, object oriented and ensemble learning ideas applied to high resolutionremote sensing image classification, such as QuickBird high resolution remotesensing image as the main data source, analysis and verification with respect to thetraditional classification methods, object oriented and ensemble learning greatpotential in high resolution remote sensing image classification. The same time, onthe basis of previous research work, the image segmentation algorithm to improve itsuncertainty analysis and choose the best segmentation scale.The results of this study are as follows:1) The three aspects of object oriented remote sensing image classificationmethod has the image segmentation, image feature selection and object orientedclassification. Existing segmentation algorithm based on watershed transformsegmentation method has been improved so that it can be applied to high resolutionremote sensing image information extraction. In high resolution remote sensing imageclassification, decision trees, support vector machine and artificial neural networkclassification method, and experiments show that the object oriented classificationmethod is superior to traditional pixel level classification. 2) The theoretical foundation of the ensemble learning methods and the mainalgorithm, described in detail Bagging, Boosting and voting method ensemblelearning methods. The experimental results show multiple classifiers ensemblemethod is superior to single classifier in remote sensing image classification.