Laser Curve Extraction of Wheelset Based on Deep Learning Skeleton Extraction Network

Area of Science:

• Computer Vision
• Machine Learning
• Metrology

Background:

• Accurate laser fringe center extraction is crucial for geometric analysis.
• Traditional methods face challenges in speed and stability.
• Skeleton extraction is key for shape analysis.

Purpose of the Study:

• To propose a new deep learning algorithm for laser fringe center extraction.
• To enhance the accuracy and efficiency of laser stripe analysis.
• To address limitations of existing laser curve center extraction techniques.

Main Methods:

• Developed a hierarchical skeleton network (LuoNet) with a three-level encoder-decoder architecture.
• Incorporated YE Module interconnections between encoder and decoder levels.
• Utilized deep learning for skeleton extraction on wheelset laser curve datasets.

Main Results:

• The LuoNet algorithm achieved an F1_score of 0.714 on the wheelset laser curve dataset.
• Demonstrated significantly shorter running times compared to traditional algorithms.
• Exhibited higher accuracy and more stable extraction results.

Conclusions:

• LuoNet provides a fast, accurate, and stable method for laser fringe center extraction.
• The proposed deep learning approach outperforms traditional algorithms.
• This advancement benefits wheelset metrology and geometric shape analysis.