Welding Defect Segmentation System Dataset Overview

Dataset Information

Dataset Name

• Name: test2

Dataset Categories

• Number of Classes: 3
• Class Names: [Bad Weld, Defect, Good Weld]

Dataset Description

• Purpose: Train and improve a YOLOv8‑seg welding defect segmentation system.
• Goal: Enhance accuracy and efficiency of welding defect detection.
• Class Details:
  • Bad Weld: Visible defects caused by improper welding processes, such as pores, cracks, or lack of fusion.
  • Defect: Subtle imperfections that may affect weld quality, e.g., surface irregularities or uneven weld width.
  • Good Weld: Standard‑compliant samples used as positive examples for model learning.

Dataset Construction

• Sample Diversity: Ensure balanced quantity and variety across classes, covering different welding conditions, materials, and parameters.
• Annotation Process: High‑precision image annotation tools were used for detailed classification and segmentation of each welding image.
• Data Augmentation: Includes image rotation, scaling, flipping, brightness and contrast adjustments to increase diversity.

Dataset Scale

• Number of Images: 1,100

Dataset Applications

• Objective: Train an efficient welding defect segmentation system to boost automation in defect detection.
• Expected Impact: Achieve breakthroughs in instance segmentation accuracy and speed, advancing welding technology and intelligent manufacturing.