A One-Stage Approach for Surface Anomaly Detection with Background Suppression Strategies

Area of Science:

• Computer Vision
• Machine Learning
• Industrial Automation

Background:

• Surface anomaly detection is critical for industrial quality control.
• Existing methods often struggle with small or subtle defects.
• A need exists for efficient and accurate one-stage detection systems.

Purpose of the Study:

• To develop a novel one-stage method for surface anomaly detection.
• To improve the capture of small targets and reduce noise.
• To minimize information loss in defect detection.

Main Methods:

• An encoder-decoder segmentation network was designed to maximize small target capture.
• Dual background suppression mechanisms were implemented for coarse and fine noise reduction.
• A parameter-free classification module was developed to prevent information loss.

Main Results:

• The proposed one-stage detector achieved state-of-the-art performance.
• High precision, recall, and f-score were demonstrated in experiments.
• The method effectively handles small targets and reduces noise patterns.

Conclusions:

• The developed one-stage method offers a highly effective solution for industrial surface anomaly detection.
• The approach demonstrates superior performance compared to existing methods.
• This technique can significantly enhance manufacturing quality control processes.