CSS-YOLO: A Steel Surface Defect Detection Algorithm Based on an Improved YOLO11n

Abstract

To address the issues of missed detections and insufficient localization accuracy caused by background noise interference, significant scale variations, and irregular geometric shapes of steel surface defects in real industrial scenarios, this paper proposes CSS-YOLO, an improved steel surface defect detection model based on YOLO11n. First, a Cross Stage Partial-Partially Transformer Block (CSP-PTB) is designed to jointly model local texture information and global contextual features through convolutional and Transformer branches, thereby enhancing feature representation capability while effectively controlling computational cost. Second, a lightweight Shared Separated-Convolution Head (SSC-Head) is proposed, employing a “shared convolution + independent Batch Normalization” architecture to reduce parameter redundancy while enhancing adaptability to multi-scale features. Finally, the Shape-IoU loss function is introduced to improve the localization accuracy of bounding boxes for complex defects by jointly considering target scale and shape information. Experimental results demonstrate that CSS-YOLO achieves 80.1% mAP@0.5 and 45.9% mAP@0.5:0.95 on the NEU-DET dataset, representing improvements of 3.4 and 1.3 percentage points over YOLO11n, respectively. Furthermore, the effectiveness of the proposed method is validated on the GC10-DET dataset, demonstrating its strong generalization ability and robustness.