Land Cover Classification And Mapping Based On Segmentation Optimized Hierarchical Object Classification System

Land cover and its changes are the basic data reflecting the interrelation and influence of human beings and the natural environment,revealing geographical phenomena,and studying earth sciences and global changes.Quick and accurate access to land cover data is important for many aspects such as natural resource management,environmental research,and urban planning.With the development of remote sensing technology,object-based image analysis methods have gradually become the main method for obtaining land cover data.Image segmentation is a key step in object image analysis.The quality of segmentation results directly affects the classification accuracy of land cover.Optimizing image segmentation parameters can improve the segmentation quality of image objects and improve the final classification accuracy.However,different land cover types usually have different optimized segmentation parameters.How to make full use of the optimal segmentation parameters of each category to establish a segmentation classification hierarchy system,reduce the dependence on the operator's personal experience during the construction process,improve the applicability of the construction method,and achieve high-precision land cover mapping,which is an object-oriented image analysis method.There is a problem to be solved.This paper proposes an object-oriented land cover classification method based on optimized partition classification hierarchy.Based on the optimal segmentation parameters of each type of coverage,the method analyzes the separability of each coverage class based on the minimum segmentation unit,and uses the decision tree to explore the relationship of each coverage class in the hierarchical network.Based on the analysis results,a multi-scale segmentation classification hierarchy system is constructed.Each layer corresponds to a type of cover,and the corresponding optimal segmentation object is used for information extraction.After traversing all layers,the land cover classification result is obtained,and accurate and efficient land cover classification is realized.In this paper,based on the inconsistent evaluation method to obtain the optimal segmentation parameters of each type of coverage,the feature optimization algorithm combined with filtering and encapsulation is applied to optimize the feature space,and based on the minimum segmentation unit,the sample data of each cover category is selected.Combining class separation matrix and J48 decision tree,the size of each class separability is analyzed,and a multi-scale segmentation classification hierarchy system is constructed.Four high-resolution remote sensing images of WorldView II,WorldView III,GF-1 and GF-2 are used as data sources.For the four typical research areas of rural,township,urban and urban-rural integration,a hierarchical system is constructed for image-based objects.The land cover is classified.In order to verify the effectiveness of the multi-scale segmentation classification hierarchy proposed in this paper,a hierarchical ranking method is used to construct a hierarchical system for comparative analysis.Experimental results show that:(1)Construct a multi-scale segmentation classification hierarchy system,and extract information on the optimal segmentation layer of each coverage class,which can reduce the segmentation error and improve the classification accuracy.The method proposed in this paper overcomes the problem of constructing the partition classification hierarchy system in the existing research and relies on the individual experience of the operator,and realizes the rapid and high-precision automatic classification and mapping of land cover.(2)In order to avoid the influence of irrelevant features and redundant features on classification efficiency and classification accuracy,this paper explores a feature selection method combining filter and wrap by studying various feature optimization algorithms.The method combines the characteristics of the two methods.Firstly,the filtering algorithm is used for feature pre-selection to reduce the initial feature space dimension.Then the wrap method is used to further optimize the pre-selected results,and the feature subset with better classification performance is obtained to realize the feature space optimization.(3)This paper analyzes the influence of the number of features selected in constructing a single tree and the number of decision trees in the forest on the classification efficiency and accuracy in the random forest algorithm,and optimizes the parameters based on the Weka software platform design experimental scheme.The problem of reduced classification accuracy due to unreasonable parameter setting of the classifier is avoided.(4)The method of constructing the hierarchical classification hierarchy based on minimum segmentation unit and decision tree proposed in this paper has better generalization ability.In different types of research areas,the proposed method can improve the classification effect and accuracy.