船舶格子间焊接机器人焊缝智能识别方法  

Intelligent Recognition of Weld Seams of Ship's Grid Space Welding Robot

作　　者：丁楚航 刘明雍[1] 刘建峰 DING Chuhang;LIU Mingyong;LIU Jianfeng(National Elite Institute of Engineering,Northwestern Polytechnical University,Xi'an 710072,China;Shanghai Waigaoqiao Shipbuilding Co.,Ltd.,Shanghai 200137,China)

机构地区：[1]西北工业大学国家卓越工程师学院,西安710072 [2]上海外高桥造船有限公司,上海200137

出　　处：《船舶工程》2025年第3期115-123,共9页Ship Engineering

摘　　要：[目的]为解决船舶格子间焊缝识别和追踪存在的特征丢失、误检或漏检等问题,提升焊缝检测效率和准确性,开发专门针对船舶格子间表面的焊缝图像数据集,[方法]提出一种基于深度学习算法的智能焊缝辨识方法。通过一种并行下采样替换最上层下采样的方法对图像进行预处理,以降低其分辨率并剔除多余的信息,进而提升信息的表达效率;在网络的跳跃连接环节引入一种基于坐标的注意力模块,以此增强网络特征的提取能力,进一步提升分割的准确性。在收集的焊缝图像集上对改进的U-Net模型的有效性进行测试。[结果]结果显示,与原模型相比,改进的U-Net模型在图像语义分割任务中的m IoU和Acc分别增长1.92%和0.3%。[结论]改进的U-Net模型在船舶格子间焊缝检测任务中表现出更高的分割精度和效率,有效解决了特征丢失、误检或漏检等问题,为船舶焊缝检测提供了更可靠的解决方案。[Purpose] The difficulties of identifying and tracking weld seams in the grid space of ships often limits the efficiency of inspection work. [Method] A dedicated dataset of weld seam images for the ship surface grid space has been developed and an intelligent weld seam identification technology based on deep learning algorithms has been introduced. The technology first adopts a parallel down-sampling strategy to replace the top-level down-sampling method for image pre-processing, reducing its resolution and eliminating redundant information, thereby improving the efficiency of information expression;then, the coordinate attention module is introduced in the skip connection of the network to enhance the feature extraction capability of the network, further improving the accuracy of segmentation;finally, tests are conducted on the collected weld seam images. [Result] The results reveal that compared with the original model, the improved U-Net model achieve an increase of 1.92% in mIoU and 0.3% in Acc in the image semantic segmentation task. [Conclusion] The improved U-Net model demonstrates higher segmentation accuracy and efficiency in the task of weld seam detection in ship grid compartments, effectively addressing issues such as feature loss, false detection, and missed detection, and providing a more reliable solution for ship weld seam detection.

关 键 词：图像处理 船舶格子间 深度学习 U-Net模型 焊缝视觉跟踪 

分 类 号：U671.99[交通运输工程—船舶及航道工程]