Design And Experiment Of High Spatial Resolution Remote Sensing Image Information Extraction Based On Hierarchical Strategy

With the spatial resolution of remote sensing images becoming higher and higher,the study on how to extract all kinds of topographic information efficiently and accurately from high-resolution remote sensing images not only promotes the development of remote sensing technology itself,but also has important practical significance for the urbanization and modernization process being promoted in our country.Object-oriented high-resolution remote sensing information extraction has become the main technical means of current applications.However,there are still many problems to be solved urgently in object-oriented information extraction.Firstly,for images with different sizes of objects,different types of objects have their appropriate segmentation scales.Using a unified segmentation scale hierarchy can easily lead to the phenomenon of "over-segmentation" and "under-segmentation" of objects,which will directly reduce the accuracy of image extraction;secondly,ignoring the cross-influence between different objects in information extraction,directly extracting information of all types of objects will also be recessive.At last,the existing research mainly focuses on the extraction algorithm,but the research on the extraction strategy is not enough.In view of the shortcomings of the existing research,this paper takes Wanzhou District of Chongqing as the research area to study the object-oriented information extraction from the extraction strategy,studies and summarizes the hierarchical strategy of object-oriented information extraction,and uses this strategy to carry out information extraction experiments.The main research contents and results are as follows:(1)Research and implementation of scale hierarchy strategy.An optimal segmentation scale selection method based on prior knowledge is designed and implemented.According to the gray level similarity and texture similarity of the reference object and the segmentation object,the segmentation scale evaluation function is constructed,and the optimal segmentation scale of different objects is determined by calculating the value of the evaluation function.This method is compared with the maximum area method and the mean variance method through experiments to verify the effectiveness of this method.(2)Research and implementation of the strategical strategy of land feature types.This paper designs a hierarchical information extraction method which takes into account the characteristics of various terrain features.The idea starts with the water body with high separability and uses some rules to divide the whole image into water body and non-water body.Then,in the subsequent classification process,the extracted water body categories are not considered,and all the objects are extracted layer by layer until all the objects are extracted.In this way,the redundancy of data is avoided,the remaining land classes on the image are less and less,and the separation of the remaining land classes is more and more easy,so that the potential of data can be fully tapped.(3)Design and experiment of high-resolution remote sensing image information extraction combined with hierarchical strategy.The hierarchical strategy designed in this paper includes two aspects: scale hierarchy and feature type hierarchy.Firstly,the fractal network evolutionary segmentation algorithm and the optimal segmentation scale selection method are applied to obtain the segmentation results under the optimal scale.Then,based on the CART decision tree,the hierarchical extraction of feature types is realized.Finally,the nearest neighbor classification and rule classification methods are integrated to optimize the extraction.Result is obtained to realize the extraction of terrain information in the study area.(4)Comparison of extraction schemes.The accuracy of nearest neighbor classification in supervised classification,rule classification based on CART decision tree and classification results combined with hierarchical strategy are evaluated by using confusion matrix.The results show that the overall accuracy and KAPPA coefficients of the extraction results based on hierarchical strategy are better than those based on nearest neighbor classification and rule classification.The overall accuracy of ground feature information extraction based on hierarchical strategy is 11.3% and 2% higher than that of the nearest classification information extraction and rule classification information extraction,respectively.The KAPPA coefficient is 40% and 8.3% better than that of the nearest classification method and rule classification respectively.This shows that the hierarchical strategy proposed in this paper can effectively improve the classification accuracy and is more suitable for high resolution.Geographic information extraction from rate remote sensing images.This paper constructs a hierarchical strategy from two perspectives of scale stratification and terrain type stratification,and applies this strategy to the process of terrain information extraction,forming a process of remote sensing image information extraction based on hierarchical strategy.By comparing with several commonly used object-oriented information extraction schemes,the technical ideas in this paper can effectively improve the accuracy of object classification extraction.