Research On 3D Point Cloud Denoising Classification Method Based On Density Clustering And Deep Learning

3D point cloud data can overcome the limitations of traditional 2D image data,provide 3D spatial information including object shape,position,and size,and support various 3D data processing and visualization technologies.It has become an important data source for industries such as computer vision,autonomous driving,terrain surveying,and human intelligence.With the widespread use of various 3D laser scanners,the collection of point cloud data has become more convenient and popular,thus promoting the expansion of the application range of point cloud data.The processing of 3D point cloud data has become one of the current popular research directions.Point cloud denoising and classification are the foundation of point cloud data processing,which affects the subsequent reconstruction of point cloud models.Therefore,it is an important research direction.The discrete and disordered nature of point clouds makes high-precision denoising and classification of point clouds extremely challenging.This thesis conducts research on the above issues,mainly divided into two aspects: point cloud denoising based on density clustering and point cloud classification based on deep learning.The specific content is as follows:(1)In terms of point cloud denoising,this thesis proposes a point cloud denoising method based on density clustering to address the problem of discrete and chaotic points in point cloud data.Firstly,the adaptive DBSCAN algorithm is used to cluster and denoise the large-scale point cloud noise,obtaining the coarse denoised point cloud data.Then,the moving least squares method is used to smooth and denoise the small-scale point cloud noise.By classifying and processing noise at different scales,noise in point cloud data can be effectively filtered out.Compared to a single filtering method,it can simultaneously preserve the details of point cloud features and effectively remove noise.The experimental results show that compared with other filtering methods through quantitative experiments,the method proposed in this paper has a higher noise removal rate,effectively reducing noise interference in point cloud data,improving the accuracy and reliability of point cloud data,and laying the foundation for subsequent classification of point cloud data.(2)In terms of point cloud classification,due to the non structural and irregular nature of point clouds,existing point cloud classification algorithms fail to effectively describe the local feature information of point clouds.Based on this issue,this thesis proposes a point cloud classification method based on adaptive graph convolution.In the process of graph convolution,adaptive convolution kernels are used to dynamically extract the feature information of each point,and the maximum pooling function and average pooling function are used to aggregate and then concatenate the features,achieving effective extraction and aggregation of point cloud features.The experimental results show that the average accuracy of the proposed method on the ModelNet40 dataset is 90.6%,and the overall accuracy is 93.3%.Classification experiments were also conducted on the ScanObjectNN dataset,and the average accuracy of the classification network structure is 76.1%,with an overall accuracy of79.2%.Compared with current mainstream network models,its accuracy is higher and its robustness is stronger.In summary,this thesis focuses on the denoising and classification aspects of 3D point cloud data processing.By using this method to denoise and classify point cloud data,the application efficiency of point cloud data can be further improved,thereby promoting its application in various industries.