基于谱图理论的变压器区域大规模点云轻量化方法  

Large Scale Point Cloud Lightweight Method for Power Transformer Area Based on Spectral Graph Theory

作　　者：杨帆[1] 吴涛[1] 郝翰学 刁冠勋 李勇 Yang Fan;Wu Tao;Hao Hanxue;Diao Guanxun;Li Yong(State Key Laboratory of Power Transmission Equipment Technology Chongqing University,Chongqing 400044 China;State Grid Shanghai Electric Power Company,Shanghai 200010 China;The YDROBOT Company,Beijing 100080 China)

机构地区：[1]输变电装备技术全国重点实验室(重庆大学),重庆400044 [2]国网上海市电力公司,上海200010 [3]北京易达图灵科技有限公司,北京100080

出　　处：《电工技术学报》2024年第23期7528-7541,共14页Transactions of China Electrotechnical Society

基　　金：国家电网公司总部科技项目资助(5500-202017468A-0-0-00)。

摘　　要：点云在电网数字化转型中有重要应用价值。现有方法在处理变压器区域大规模点云时难以兼顾点云轻量化后的视觉效果,导致轻量化点云存在较大视觉失真。该文在点云轻量化时引入视觉失真度,以其最小作为约束,实现点云轻量化后视觉效果最优。首先基于谱图理论将原始点云转换为图信号,建立轻量化点云视觉信息损失与重采样矩阵间的函数关系;然后以视觉信息损失最小为目标函数,获得一组满足视觉信息损失最小但特征信息和均匀信息比例不同的轻量化点云;进一步,将原始点云和轻量化点云投影到几何和颜色特征域,用一个低维向量表征点云视觉效果,从而选择出视觉失真度最小的轻量化点云;最后,使用基准数据集和包含8000余万个点的变压器区域大规模点云进行了验证。结果表明:所提方法与主流的随机降采样法、体素平均法、非均匀网格法、曲率采样法相比,在轻量化点云的视觉效果方面,分别提升了57.4%、69.2%、62.2%、75.6%。The three-dimensional visualization based on point clouds has important application value in the digital and intelligent transformation of the power industry.For transformers,the actual production often obtains the point cloud of their entire operating area,and the huge data scale brings difficulties to use.Existing lightweight methods for processing large-scale point clouds always lead to significant visual distortion.Therefore,this paper proposes anew point cloud lightweight method that can take visual effects into account.Firstly,the K-nearest neighbor(KNN)algorithm establishes an edge set for the original point cloud and converts it into a graph signal.Then,the graph signal is transformed into the frequency domain through the graph Fourier transform.Further,the feature and uniformity loss expressions during point cloud lightweight are derived.After quantifying the visual information loss,an objective function was established to minimize the loss.Then,the grid search was performed on parameters k andρ,which affect the proportion of features and uniformity in the lightweight point cloud.A set of lightweight point clouds that meet the minimum visual information loss but have different proportions of features and uniformity can be obtained.A visual distortion quantification criterion was established to select the lightweight point cloud with the lowest visual distortion.The criteria project the 3D point cloud onto a feature domain composed of geometric and color features closely related to visual effects,and convert the visual effects of the point cloud into multiple histograms.Furthermore,statistical parameters quantify each feature's histograms,and the point cloud's visual effects are transformed into vectors.The visual distortion was calculated using the difference between vectors,and the lightweight point cloud with the lowest visual distortion was selected.Finally,the effectiveness of the proposed method was validated using a benchmark dataset and a large-scale point cloud of transformer areas containing ov

关 键 词：谱图理论 点云 轻量化 视觉失真 变压器区域 

分 类 号：TM507[电气工程—电器]