Research On High-score Image Classification Based On Decision Tree Model

In recent years,with the development of space remote sensing satellite technology,the spectral,spatial and temporal resolutions of remote sensing images have been greatly improved.Especially,the appearance of high score images makes the extraction of object information from remote sensing images become an important means of basic data acquisition.However,how to recognize the object information in high resolution images is a hot topic in the area of remote sensing image identification and classification technology.Based on the traditional image pixel classification method,mainly based on the statistical theory and statistical analysis of the spectral characteristics of different objects so as to realize the classification,if applied in high resolution image,will appear,different spectrums of objects with the same spectrum "phenomenon,it is difficult to achieve the ideal classification of object recognition.This paper studies object-based image recognition and classification based on decision tree model.Object-oriented classification is based on the collection of multiple pixels in the image as the analysis target,breaking the traditional image classification in a single pixel as the analysis unit limitations.Based on the pixel set as the analysis target,we can take full account of the characteristics of the objects,such as the spectral features,shape features and texture features of the image.The decision tree model is used to calculate the classification features of different features,and the multi-Image classification.In this study,we use the multi-source fusion images of the Koh Bieke in the Emin County as the data source,and carry out the following research with ENVI,eCognition,ArcGis and other data processing tools:(1)The classification of detailed comparative experiments on multi-scale optimal fusion of multi-source image segmentation parameters,through experiments on the segmentation of multi-scale segmentation scale,band weight,shape factor and compactness parameter,finally determined by visual effect of multi-scale image segmentation of optimal parameters.(2)Through the classification of statistical decision tree model of different objects,the classification features of different objects to construct the feature space set,using K nearest neighbor classifier and CART decision tree classifier to achieve classification experiments,multi-scale feature object information.(3)The accuracy of object-oriented classification results and traditional minimum distance classification results are evaluated respectively,and the results show that the object-oriented classification accuracy is better than the traditional classification methods.