High-resolution Remote Sensing Image Building Extraction Based On CART Decision Tree

With the rapid development of remote sensing technology,high-resolution image data has grown rapidly.Obtaining real-time,high-precision,and accurate ground-based information from images has become the key to remote sensing image interpretation.Buildings are the main features in urban areas and the main sites for human settlement activities.It is of great significance to quickly and accurately obtain building information from images in the study of changes in human life activities.The research on building extraction is still a hot topic in the interpretation of remote sensing images.In this paper,the buildings in the urban area are extracted according to the geometric features and texture features of the building and the decision tree classification method.For the problem of building extraction,this paper studies the area of Wuhan as experimental data.The following three aspects were studied and several key elements in the object-oriented supervised classification extraction method were introduced.The first was how to obtain the affected objects,and the more accurate and appropriate image objects were used to reflect the similarity and differentness of the same kind of objects.The dissimilarity of objects;the second is how to combine the feature combinations required for training samples.The appropriate feature combination is the key to the classification extraction.Unreasonable or less feature combinations will reduce the extraction accuracy,too many feature combinations,and will affect the classification and extraction.The third is how to choose the appropriate classification method.Different classification methods have their own advantages and disadvantages.The appropriate classification method can improve the accuracy of building extraction.The main work of this paper is as follows.Firstly,the related theoretical and technological researches have introduced the theory and technology of image segmentation,feature combination,classification method,and decision tree classification method.Second,using the same experimental data,the optimal segmentation scale With the combination of optimal features,different object-oriented classification methods were used to classify the experimental data,and it was concluded that the decision tree classification accuracy was higher than other classification methods in high-resolution imagery.Information extraction has a high stability.Finally,using the decision tree algorithm combined with the geometric features and texture features of the building to carry out experiments on building extraction,the importance of building geometric features and texture features in building extraction is proved.This paper proposes the method of training the sample features after training to construct the classification rules and training after the sample training to improve the efficiency and accuracy of building classification rules.The extraction of buildings has a certain significance for the study of urban cultural changes.