Application Research Of Surface Point Cloud Hole Repair Based On Dynamic Weighted Combination Model

When UAV aerial photogrammetry obtains three-dimensional point clouds on the surface,it will be subject to various non-resistant factors,leading to the generation of point cloud holes,which will directly affect the production accuracy of relevant surveying and mapping products,and thus affect the application value of UAV point cloud data.In the early data processing work,these point cloud holes need to be repaired.At present,various models of machine learning have been widely used in the repair of point cloud holes,but it is difficult to guarantee the repair accuracy of holes in complex terrain with only a single repair model.Therefore,a more adaptable and more accurate repair method is needed.Based on the principle of linear combination,this paper proposes a method of repairing point cloud holes based on the dynamic weighted combination model.The main contents of the paper are as follows:1.Construct the point cloud hole repair model optimized by Sparrow search algorithm.The sparrow search algorithm has a strong optimization ability.By simulating the foraging process of sparrow population,model parameters of BP neural network,least square support vector machine and extreme learning machine were optimized,and each group of models before and after optimization was applied to repair a surface cavity area.After optimization,the repair precision and stability of the three groups of models have been greatly improved.2.The dynamic weighted composite repair model is constructed.According to the principle of linear combination,the BP neural network optimized by Sparrow search algorithm,least square support vector machine and extreme learning machine were combined linearly to construct the optimal weighted combination model and inverse variance combination model.Based on the inverse variance combination model,a dynamic weighted combination repair model is proposed,and three groups of weighted combination models are applied to the repair of a surface cavity area.Compared with three groups of single repair models and two groups of non-dynamic combined repair models,the dynamic weighted combined repair model has stronger stability and higher repair accuracy,and its adaptability is also significantly improved.3.The feasibility of the dynamic weighted combined repair model in the patch of UAV surface point cloud holes is verified.In this paper,the UAV measured data of an excavated mountain area in Guilin city was used for verification.The residual error,mean absolute error,mean square error and root mean square error of three groups of single repair models,two groups of non-dynamic weighted combination models and dynamic weighted combination models were compared and analyzed,and the repair effect of dynamic weighted combination model was demonstrated.The experimental results show that in the hole repair of UAV measured data,the mean square error of the optimized BP neural network,least squares support vector machine and extreme learning machine is 0.143 m,0.152 m and 0.190 m,respectively.The average root-mean-square error of the optimal weighted combination model and the reciprocal variance combination model is 0.100 m and 0.114 m,respectively,while the average root-mean-square error of the dynamic weighted combination model is0.081 m,which verifies that the overall repair accuracy of the model is better than that of the single repair model and the non-dynamic weighted combination model.It is more suitable for repairing point cloud holes in complex terrain of UAV.