Airborne LiDAR Point Cloud Classification Based On Probabilistic Graph Model

Airborne Li DAR point cloud data has the characteristics of complicated scene,rich target,uneven density and noise,which makes the existing three-dimensional scene object extraction and identification method low in the degree of automation,and the recognition accuracy.Aiming at the problem of low target recognition rate in the current three-dimensional scene,this paper extends the strategy of multi-source data fusion based on the characteristics of integrated information aerial image and airborne point cloud data,and analyzes the characteristics of point cloud classification.In the paper,we present a Bayesian network model which integrates aerial image spectral information assistant point cloud classification,and classifies the three-dimensional scene into ground,low vegetation,high vegetation and building.And with the continuous development of remote sensing technology,many scholars are committed to use remote sensing technology in power line monitoring,and airborne LiDAR has its unique advantages and has also been widely used in electricity patrol line.Aiming at the problem of low degree of automation of point cloud classification in the current power line field,a Markov random field model based on the posterior probability of random forest is proposed,which is used for point cloud classification of power line scene.In this paper,we analyze the LiDAR point cloud of urban and power line patrol scene in the case of comprehensive analysis of the characteristics and scene of airborne LiDAR point cloud data.The main contents are as follows,(1)For the LiDAR point cloud data in the urban area,this paper extracts the geometric features used to describe the features of the points on the basis of analyzing the characteristics of the point cloud data and summarizing the predecessors' experience.The multi-scale segmentation of the aerial image is implemented,and the corresponding image information is extracted from the image object,and the corresponding spectral point is assigned to the corresponding point.Based on the mutual information theory,the paper analyzes the dependence of the point cloud geometric features and the characteristics of the image object.Based on this,the optimal Bayesian network model is constructed to describe the joint probability distribution of the point cloud classification feature vector for automatic classification and information extraction of point cloud.(2)For the LiDAR point cloud of the power line patrol scene,the multi-scale visual classification feature is constructed with the spatial pyramid theory to describe the geometric shape information of the spatial point and its neighborhood.The probability of observing the data is described by the random forest classifier,And then builds a multi-tagging energy function based on the Markov random field model to establish the prior probabilities of the scrupulous context information.Finally,the multi-marker graph technique is used to minimize the energy function to complete the classification tag optimization.The experimental results demonstrate that it can effectively improve the classification accuracy of all kinds of objects,and the classification accuracy can reach more than 90% by combining the image object information into the cloud classification process.The classification result is optimized by considering the spatial relation based on the optimization of MRF model,.Compared with other traditional classifiers,the Bayesian network model and the MRF model proposed in this paper are superior to other conventional classifiers in most data.