Ground Object Extraction Of High Resolution Remote Sensing Image Based On Mark Clustering Point Process

With the increasing of the spatial resolution,objects in remote sensing image are revealed both with great geometrical precision and a high level of thematic detail,which can benefit the accurate extraction of the objects and cause the difficulties for modeling the image with respect to the geometric structures and data features.Firstly,the issue concerning the use of the geometrical precision is how to describe the geometric structures of specific objects so that they are suitable for their extraction.Secondly,the significant amount of geometrical details presenting in a very high resolution remote sensing image increases the internal spectral variability of each object and decreases the spectral variability between different objects and their background,which lead to the difficulty of modeling image itself.To motivate the above problems,the paper proposes the strategy combining mark clustering point process and statistical modeling to address the task of locating and identifying an unknown number of objects of different types in a very high resolution remote sensing image.The main tasks can be summarized as follows.(a)The point process theory is used to model the geometric structure of the object.This paper proposed a simple polygon construction method,which is incorporated into the theoretical framework of the clustering point process,and the simple polygon constructed by the cluster point in the mark clustering point process is used to describe the spatial distribution and geometric shape of the object.(b)Building the object extraction model combining geometry structure and observation data.Using the statistical consistency of the ground object,the observation data in the background area,and the statistical difference between the ground object and the background environment,the ground object data model is established,and the ground object extraction model is established by combining the geometry and data model.(c)Designing the samplers to estimate the optimal configuration of objects.Based on the model of the reversible jump Markov chain Monte Carlo(RJMCMC),an optimal simulation algorithm is designed and implemented.According to the object extraction model,five moving operations are designed,including adding polygon,deleting polygon,adding polygon node,deleting polygon node and merging polygon operation.(d)To validate the applicability and effectiveness of the proposed strategy,some algorithms for extracting objects are developed under the framework.Oil spill target extraction from Synthetic Aperture Radar(SAR)image,island and lake target extraction from multispectralimage,building and tree target extraction from Light Detection And Ranging(Li DAR)point cloud data.(e)In order to verify the effectiveness of the proposed algorithm,the evaluation methods such as extension area evaluation and confusion matrix evaluation are used to quantitatively evaluate the experimental results,and the visual interpretation method is used to qualitatively evaluate the experimental results.The experimental results show that the proposed algorithm can effectively extract the ground objects of high-resolution remote sensing images with arbitrary geometry.There are 81 figures,12 tables and 159 references.