【摘要】 点云异常检测旨在从整体数据分布中识别缺陷样本，并进一步定位其在空间中偏离预期模式的异常区域．针对现有全局匹配策略难以有效捕捉集中于局部细微几何结构异常的问题，文中提出基于局部感知图重构的点云异常检测网络．首先，将点云建模为图结构，通过局部敏感的边卷积操作挖掘点云的局部结构特征，提升对局部细微异常的识别能力．然后，基于子图结构对齐策略设计局部对齐重建损失，从结构层面放大正常样本与异常样本间的差异性．此外，引入局部异常生成策略，构建异常-正常样本对，约束网络学习异常样本向正常样本的模式映射．最后，采用局部匹配检测算法计算异常样本与期望正常样本之间的差异，实现对点云异常区域的检测．实验表明，文中网络显著增强对局部细节的感知能力，在多个类别的点云异常检测任务中均取得较优性能．更多还原

【Abstract】 Point cloud anomaly detection aims to identify defective samples from the overall datadistribution and further locate the abnormal regions deviating from the expected pattern in space. Existingglobal matching strategies struggle to effectively capture anomalies concentrated in local subtle geometricstructures. To address this issue, a point cloud anomaly detection network based on local perceptiongraph reconstruction is proposed. First, the point cloud is modeled as a graph structure, and localsensitive edge convolution operations are utilized to mine local structural features to enhance theidentification ability for local subtle anomalies. Second, a local alignment reconstruction loss is designedbased on a subgraph structure alignment strategy to amplify the differences between normal and abnormalsamples at the structural level. Furthermore, a local anomaly simulation strategy is introduced toconstruct an anomaly-normal sample pair. Through this strategy, the model is constrained to learn thepattern mapping from abnormal samples to normal samples. Finally, a local matching algorithm is applied to calculate the differences between abnormal samples and expected normal samples to achieve thedetection of point cloud abnormal regions. Experimental results show that the proposed methodsignificantly enhances the perception ability for local details and achieves excellent performance onmultiple categories of point cloud anomaly detection tasks.更多还原