# Welding Seam Recognition and Trajectory Planning Based on Deep Learning in Electron Beam Welding

## TL;DR

This paper introduces a deep learning method to improve welding seam recognition in electron beam welding, enhancing accuracy and real-time performance.

## Contribution

The novel approach combines YOLOv11-seg with adaptive Canny edge detection and UFO-ViT attention for better weld recognition in challenging environments.

## Key findings

- The optimized model achieves an average precision (mAP) of 77.4%, a 9-percentage-point improvement over the baseline.
- The system operates at 20 FPS, meeting real-time requirements with a trajectory deviation of less than 3 mm.

## Abstract

To address challenges in weld recognition during vacuum electron beam welding caused by dark environments and metal reflections, this study proposes an improved hybrid algorithm combining YOLOv11-seg with adaptive Canny edge detection. By incorporating the UFO-ViT attention mechanism and optimizing the network architecture with the EIoU loss function, along with adaptive threshold setting for the Canny operator using the Otsu method, the recognition performance under complex conditions is significantly enhanced. Experimental results demonstrate that the optimized model achieves an average precision (mAP) of 77.4%, representing a 9-percentage-point improvement over the baseline YOLOv11-seg. The system operates at 20 frames per second (FPS), meeting real-time requirements, with the generated welding trajectories showing an average length deviation of less than 3 mm from actual welds. This approach provides an effective pre-weld visual guidance solution, which is a critical step towards the automation of electron beam welding.