一种基于机器视觉的漆包线表面缺陷检测与分类方法
By combining frequency domain analysis and phase sampling with convolutional neural network parameterization, the problem of misjudgment of surface scratches on enameled wires was solved, achieving high-precision defect detection and classification, reducing the false detection rate and improving the stability and adaptability of detection.
Patent Information
- Authority / Receiving Office
- CN · China
- Patent Type
- Patents(China)
- Current Assignee / Owner
- SHANDONG HUAWU ELECTRIC CO LTD
- Filing Date
- 2025-11-24
- Publication Date
- 2026-07-17
AI Technical Summary
Existing image analysis methods struggle to distinguish between genuine spiral scratches and pseudo-stripe signals on the surface of enameled wires in high-speed sampling environments, resulting in a high misjudgment rate. Furthermore, convolutional neural networks have difficulty generalizing to different line speeds, exposure times, or variations in wire spin angular velocity.
By acquiring single-pixel time series of line scan images, performing frequency domain transformation to determine the main peak index, calculating the helical bispectral locking degree, performing phase sampling and convolutional neural network parameterization based on the helical bispectral locking degree, optimizing the training process, and using the power function of the helical bispectral locking degree as sample weights, high-precision automatic identification and classification of surface defects of enameled wire can be achieved.
It achieves high-precision automatic identification and classification of real defects on the surface of enameled wire under different line speeds, exposure and lighting conditions, reducing the false detection rate and maintaining the stability and versatility of the detection.
Smart Images
Figure CN121304647B_ABST