YOLO-RDM: A high accuracy and efficient algorithm for magnetic tile surface defect detection with practical...

Area of Science:

• Materials Science
• Mechanical Engineering
• Computer Vision

Background:

• Surface defects on magnetic tiles, such as chipping and wear, degrade permanent magnet motor performance.
• Manual inspection methods for these defects are inaccurate, costly, and struggle with curved surfaces and tiny features.

Purpose of the Study:

• To develop an advanced algorithm for accurate and efficient detection of magnetic tile surface defects.
• To address the challenges of curved surfaces and small defect features in automated inspection.

Main Methods:

• Proposed the YOLO-RDM algorithm, integrating DOConv in the neck network for lightweight feature extraction.
• Enhanced the C2f module with an RPA Block incorporating a parallel attention mechanism.
• Replaced the YOLOv8 backbone with MogaNet for improved contextual information aggregation and feature learning.

Main Results:

• Achieved a mean average precision (mAP@0.5) of 95.0%, a 4.8% improvement over the original model.
• Demonstrated an inference time of less than 5.6 ms, outperforming other object detection models.
• Showcased strong recognition and generalization capabilities on the NEU metal surface defect dataset.

Conclusions:

• The YOLO-RDM algorithm significantly enhances magnetic tile defect detection accuracy and efficiency.
• The model's performance advantages make it a viable solution for practical industrial applications.
• The study validates the effectiveness of lightweight convolutions, attention mechanisms, and advanced backbone networks in defect detection.