| 注册
首页|期刊导航|电焊机|基于改进的DeepLabv3+的盖面焊缝识别

基于改进的DeepLabv3+的盖面焊缝识别

夏羽 周鲁兵 吴慧峰 包志刚 闫佳 冯消冰

电焊机2026,Vol.56Issue(10):64-70,7.
✕
电焊机2026,Vol.56Issue(10):64-70,7.DOI:10.7512/j.issn.1001-2303.2026.10.08

基于改进的DeepLabv3+的盖面焊缝识别

Cover Weld Identification based on Improved DeepLabv3+

夏羽 1周鲁兵 1吴慧峰 1包志刚 1闫佳 2冯消冰2

作者信息

  • 1. 南通中集太平洋海洋工程有限公司,江苏 启东 226251
  • 2. 北京博清科技有限公司,北京 100176
  • 折叠

摘要

Abstract

Aiming at the difficulty of accurately recognizing the inflection point features of the weld surface in the cover layer of multi-layer and multi-pass automatic welding using laser vision sensors,a visual recognition method based on an im-proved DeepLabv3+model is proposed.An active light source is used to illuminate the weld surface and capture visual im-ages.MobileNetV2 replaces Xception as the lightweight backbone network to improve detection real-time performance.The Selective Kernel(SKNet)module is embedded into the Atrous Spatial Pyramid Pooling(ASPP)module,enabling the model to adaptively select convolution kernel sizes and suppress the interference of irrelevant features after concatenation.Experi-mental results show that compared with the original DeepLabv3+,the improved model achieves an inference speed increase of 26.3 frames/s and a mean intersection over union(MIoU)improvement of 3.32 percentage points,reaching 98.78%.The proposed method outperforms mainstream models including PSPNet,U-Net,and HRNet under both low-light and high-reflection conditions.

关键词

深度学习/语义分割/焊缝盖面识别/DeepLabv3+/MobileNetV2

Key words

deep learning/semantic segmentation/cover weld seam recognition/DeepLabv3+/MobileNetV2

分类

矿业与冶金

引用本文复制引用

夏羽,周鲁兵,吴慧峰,包志刚,闫佳,冯消冰..基于改进的DeepLabv3+的盖面焊缝识别[J].电焊机,2026,56(10):64-70,7.

电焊机

1001-2303

访问量0
|
下载量0
段落导航相关论文