Research On Multiscale Segmentation Technology Of Land Cover Images Based On National Census Geography In Karst

The national census geography is the basic work of obtaining the information of the geographical conditions,such as the surface nature,the ecology and the basic situation of human activities.It is an important basis for formulating and implementing national development strategy and planning,optimizing the spatial development pattern of land and resources allocation.Karst area of Guizhou has complex terrain and deep ravines.There are many problems in automatic interpretation of remote sensing images,such as low accuracy and difficulty.The first interpretation work of the National Geographic census is mainly based on manual interpretation,which is time-consuming and laborious and inefficient.According to the needs of contents and objects in census of geographical conditions in Karst area,how to improve the accuracy and speed of remote sensing image interpretation,and get high-quality geographic information.it is the Research Institute of article.In this paper,we combine LBP operator to improve the mean shift segmentation algorithm,and research the multiscale segmentation technology.The improved mean shift algorithm is used for image initial segmentation,and the mean variance method is used to determine the optimal segmentation scale of each object category.The results of multiscale segmentation are optimized by spectral difference method,and the optimized image segmentation results are applied to the extraction of ground information in the research area.The main conclusions of this paper are as follows:(1)A mean shift algorithm based on LBP texture feature is proposed.By analyzing the advantages and disadvantages of the traditional mean shift algorithmand the feature of LBP texture,LBP texture features are added to the multidimensional eigenvector of the mean shift algorithm,and all kinds of bandwidth parameters are set and the minimum heterogeneity area is merged,and the improved segmentation results are obtained.The experimental results show that compared to the segmentation result of the traditional mean shift algorithm and eCognition software,the improved algorithm shows good segmentation effect in the area with rich texture.(2)The multiscale segmentation technology based on high resolution remote sensing image is optimized.In this paper,the existing multiscale segmentation technology is analyzed.Firstly,the mean shift algorithm added to LBP texture feature is applied to initial segmentation.Then we use the mean variance method to determine the optimal segmentation scale of all kinds of ground objects,and get the multiscale segmentation results of all kinds of ground objects.Finally,the spectral segmentation method is used to optimize the segmentation results and get the final multiscale segmentation results.(3)Multiscale segmentation experiment and application research in Karst area were carried out.In order to verify the effectiveness of the multiscale segmentation technology proposed in this paper,WorldView-2 image data of the study area were selected as data sources,and multiscale segmentation experiments of all kinds of objects were carried out.According to the experimental segmentation results,we extract multi-scale information from the research area,and the total accuracy of surface information extraction reaches 84.24%.