军事知识服务平台
基于特征融合的焊缝区域点云精准识别
摘要 为提升结构光视觉引导下焊缝区域识别的准确性并保持方法稳健,本文提出一种融合局部形态学特征与空间位置特征的自适应焊缝识别方法.首先,基于曲率估计和空间距离计算构建点云数据的二维特征空间,并采用高斯混合模型进行点云数据的无监督迭代学习,实现焊缝区域特征点提取.其次,结合层次密度聚类算法和体素连通块分析,完成焊缝区域的精确分割与边界细化.最后,实验结果表明,该方法对环形焊缝 V 型坡口区域进行识别时,精确率、召回率与 F1 分数分别达到 99.38%、88.41%和 93.57%,综合性能优于对比方法,如随机抽样一致性结合密度聚类算法.该方法的边界定位误差小于0.3 mm,且参数测试表明其参数敏感度低、稳定性强,能够满足实际生产应用需求.
Abstract:
To enhance the accuracy of weld region identification under structured light vision guidance while maintaining method robustness,this paper proposes an adaptive weld identification method fusing local morphological and spatial position features.First,a fused feature space for point cloud data is constructed based on curvature estimation and spatial distance calculation,and the Gaussian mixture model(GMM)is adopted for unsupervised iterative learning of point cloud data to realize the extraction of feature points in the weld region.Second,combining with the hierarchical density-based spatial clustering of applications with noise(HDBSCAN)and voxel connectivity block analysis,the precise segmentation and boundary refinement optimization of the weld region are completed.Finally,experimental results demonstrate that this method achieves a precision of 99.38%,a recall of 88.41%,and an F1 score of 93.57%in identifying V-groove regions in circumferential welds,outperforming comparison methods such as random sample consensus combined with density-based spatial clustering of applications with noise(RANSAC-DBSCAN).Furthermore,the boundary localization error is less than 0.3 mm;parameter tests verify that the proposed method has low parameter sensitivity and strong stability,which meets practical production requirements.
年,卷(期): 2026, 43(4)
分类号: TP274+.5
机标分类号: TP391.41 F323.8 X51
在线出版日期: 2026年9月20日