Real-time steel core defect detection using enhanced lightweight object detection algorithm with data augmentation

Steel core defect detection encounters significant challenges, including limited sample data, inadequate real-time performance, and ambiguous defect characteristics, which severely restrict the engineering application of detection systems. To tackle these issues, this paper presents a real-time MSC YOLO defect detection model. This paper first collects data using a steel wire rope core conveyor belt flaw detection device with the model number ZSX-127D and proposes a data augmentation method called Cycle Consistency Generative Adversarial Network and Image Synthesis (CycleGAN IS) to address the issue of sample scarcity. Second, based on the YOLOv11 algorithm, the paper designs a CSP Bottleneck with 2 Convolutions and Kernel-size specified with multi-order gated aggregation module (C3k2_Moga). Through the innovative design of spatial aggregation blocks and channel aggregation blocks, this module enhances the accuracy of locating and identifying defects in steel cores. Additionally, the paper introduces an Spatial and Channel reconstruction Convolution-Batch Normalization-Sigmoid activation (SBS) convolution architecture, which reduces computational complexity and improves the detection of small, concealed steel core defects. Finally, the paper replaces the traditional sampling operator with a Content-Aware ReAssembly of FEatures (CARAFE) operator to comprehensively capture the complete picture of defects and their environmental context. Extensive experiments conducted on the self-constructed Steel Core Defect (SCD) dataset demonstrate that the MSC YOLO model reduces its parameter count to 2.14 million and compresses its model size to 4.3 MB, while achieving an accuracy rate of 97.7% and a mean Average Precision (mAP) of 78.0% at IoU thresholds of 0.5–0.95. Additionally, the model boasts a processing speed of up to 327.7 frames per second (FPS). The MSC YOLO model proposed in this paper achieves a lightweight and high-speed performance while maintaining high accuracy, providing an effective solution for steel core defect detection in practical engineering applications.