CCP-YOLO: An Improved YOLOv11n Algorithm for Steel Surface Defect Detection

Area of Science:

• Materials Science
• Computer Vision
• Artificial Intelligence

Background:

• Steel surface defect detection faces challenges in multi-scale feature extraction, feature fusion, and spatial detail preservation.
• Existing methods often struggle with heterogeneous feature integration and loss of critical spatial information.

Purpose of the Study:

• To propose CCP-YOLO, an improved algorithm for steel surface defect detection based on YOLOv11n.
• To enhance multi-scale feature extraction, heterogeneous feature fusion, and spatial detail retention in defect detection.

Main Methods:

• Introduced a Multi-Scale Dilated Reparameterization module (C3k2_MSD) for superior multi-scale feature extraction.
• Implemented an Interactive Adaptive Feature Fusion Module (IAFM) for effective heterogeneous feature integration.
• Utilized a Progressive Shared-Weight Context Aggregation (PSWCA) module and a Wise-Inner-MPDIoU fusion loss function to preserve spatial details and improve bounding box regression.

Main Results:

• CCP-YOLO achieved an mAP50 of 80.2% on the NEU-DET dataset, a 4.1% improvement over YOLOv11n.
• Demonstrated a recall of 0.754 with 2.7M parameters and 6.6 GFLOPs, achieving 133.14 FPS inference speed.
• Achieved a 3.7% mAP50 improvement on the GC10-DET dataset, confirming robust performance.

Conclusions:

• CCP-YOLO effectively addresses key challenges in steel surface defect detection, enhancing accuracy and computational efficiency.
• The algorithm shows significant potential for practical industrial deployment in quality control and inspection.