Research On Scale Selection And Classification Method Of Urban Feature Information Based On GF-2 Image

With the successful launch of GF-2 remote sensing imagery satellites,Makes China to independently acquire high resolution remote sensing image data with a resolution of 1m,providing a large amount of data support for urban construction in China.Because high-resolution image data combines the spatial features such as spectrum,shape,and texture,the traditional classification method only considers the spectral information of the image,which results in the classification accuracy being unable to meet the requirements of practical applications.Therefore,object-oriented classification methods came into being and brought good news to high-resolution image classification.The biggest difference between object-oriented classification and traditional classification methods is that the smallest unit of the analysis image is the object,not the pixel.The object is achieved through image segmentation.Selecting different scales for segmentation results in objects of different sizes,so the choice of scale parameters is particularly critical.In this paper,Chengdu City GF-2 remote sensing image is taken as the data source.After the image data is preprocessed,multi-level object-oriented classification,single-level object-oriented supervised classification and pixel-based supervised classification methods are used to extract the feature information.Analyzed and compared the accuracy index from various categories classification,and analyzed carefully the optimal scale selection problem in the process of object-oriented classification.The specific work is as follows:(1)According to the selection of scale parameters in the object-oriented classification process,the use of the ESP scale evaluation tool combined with the RMAS method to achieve the selection of the optimal object scale in urban areas.In the process of scalar primary selection,nine potential optimal measures of 80,100,150,170,185,195,220,250,and 270 were initially selected through the ESP scale evaluation tool.In the process of scale selection,by analyzing the problems in the RMAS method,a method for statistically calculating the RMAS value of objects whose neighboring objects are richer is selected and compared with the optimal scale obtained by using ESP.The obtained optimal scales are basically the same,which proves that the optimal scale can be obtained through this method.Finally,the optimal scales for shadows,buildings,vegetation,water bodies,and roads were 80,100,150,170,and 200 by combining the two methods.(2)To solve the problem of object-oriented feature information classification accuracy for GF-2 images,the five object layers of feature extraction are constructed on the optimal scale of each object,Analyze spectral features,shape features,and texture features of objects in various categories of objects at different layers of image objects,to establish the classification rules by selecting the common features that can distinguish between similar object objects and other feature objects,so as to achieve multi-level object-oriented classification.Compared with the object-oriented single-level nearest neighbor classification,the overall accuracy of the method is improved by 3.22%,and the Kappa coefficient is improved by 0.0490.Compared with the object-oriented single-level support vector machine,the overall accuracy is improved by 1.47%,and the Kappa coefficient is improved by 0.0248.which proves the superiority of the optimal scale to the multi-level classification method of objects.(3)In order to compare the advantages and disadvantages of object-oriented classification and pixel-based classification based on GF-2 images,six feature information classification method were used to classify the images in the study area.The research shows that the object-oriented classification results are superior to pixel-based classification in both effect and accuracy,the object-oriented classification accuracy index of various categories of objects is relatively balanced and the boundaries of the objects are more clear,there is no obvious phenomenon of missed points and wrong points,In particular,the distinction between roads and buildings is better.