CMA-YOLO11: An Advanced Algorithm for High-Precision Steel Surface Defect Detection

Steel surface defect detection is vital for ensuring steel quality. To address the low accuracy, high false-and miss-detection rates of existing algorithms, this paper presents the CMA-YOLO11 model. It designs the CSOTC module to remedy C3k2's feature-extraction flaws, integrates the MSDA and C2PSA modules to enhance multi-scale feature processing, and constructs the ASFFDCHead to improve information capture and detection accuracy. On the NEU-DET dataset, CMA-YOLO11 outperforms the baseline YOLO11n, with [email protected] and [email protected]:0.95 increased by 1.5% and 2.2% respectively, and an FPS of 100.1. When the number of parameters is only 25.0% of YOLOv8s and the computational complexity is only 44.1%, [email protected] achieves a 0.7% improvement. Also, it shows excellent generalization on multiple datasets. These accuracy gains, real-time performance, and generalization abilities make CMA-YOLO11 a highly valuable solution for industrial applications.