Research On Object-Oriented Green Space Extraction Method Of High-Resolution Remote Sensing Image

Spatial,spectral and temporal resolution of remote sensing data sources for information extraction have brought a lot of convenience Only the pixel-based green information extraction method with spectral information is considered.When the spectral information is not obvious enough and the spectral information is not abundant,the extraction result is not satisfactory.This paper studies a new method based on the realization of feature selection and threshold determination automation,combined with object-oriented method,spectral features,texture features and shape features,and uses the CART decision tree to complete the high resolution remote sensing image classification,to achieve object-oriented urban green space information extraction.First of all,to solve the feature selection,classification of the characteristics of the threshold selection problem,the traditional method requires a large number of researchers experiments,which will inevitably lead to a large number of human and time loss,we study the multiresolution segmentation of high-resolution WorldView-II images by combining the spectral features,shape features and band weights with the elements of object-oriented thinking.Through quantitative comparison and analysis,we choose different split scales,Compactness factor and shape factor,the segmentation results under the optimal segmentation parameters are selected and compared with the morphological watershed segmentation results.The results show that the multiresolution segmentation method can separate the green space from other objects,especially for the high resolution remote sensing image segmentation Then,the algorithm is used to classify the rule set based on the CART decision tree.The algorithm makes use of the idea of data mining,so that the feature and threshold can be automatically determined.The rule set is used as the reference of fuzzy classification,taking into account the brightness,normalized vegetation index,all directions of homogeneity Gray-level co-occurrence matrix,aspect ratio and the band mean and other characteristics of the urban green space information extraction Finally,the accuracy of the green space information extraction is using the confusion matrix classification index.The overall accuracy of the classification is 91.98%and the Kappa coefficient is 0.9244.The method of urban green space information extraction based on WorldView-Ⅱ image is completed,In the application and future improvement,we need to consider the complexity of the WorldView-Ⅱ data processing process,the training sample number of selected redundancy and the effectiveness of the segmentation algorithm.