<p>Textiles have become an indispensable part of daily life. However, defects are difficult to avoid during the production process. If not detected promptly, these defects can lead to material and economic losses. Currently, deep learning methods for defect detection are widely used, but researchers have primarily focused on accuracy while neglecting detection speed. To address the practical needs of defect detection in factories, we propose a textile defect detection method called YOLOv8n-SSSL. First, we introduce the innovative C2f-Star module, which combines the multi-scale information capture capabilities of C2f with the efficient depthwise separable convolution design of the Star Block, reducing the model’s computational complexity and parameter count. Second, we incorporate the parameter-free SimAM (simple attention module) to accelerate network extraction and increase the integration of small-object feature information. Next, we improve the loss function by using SIoU to minimize the impact of pixel shifts on small objects. Finally, we introduce a lightweight shared convolutional detection head, which significantly reduces model complexity by sharing convolution operations across multi-scale feature maps. In this experiment, eight common types of textile defects were selected from two defect datasets for training, and the proposed YOLOv8n-SSSL was validated. The results demonstrated a reduction of 2.5 giga floating point operations per second (GFLOPs) and achieved mAP of 84.39%. The inference speed reached 114.8 frames per second (FPS) on the A100 GPU. To verify the generalization ability of YOLOv8n-SSSL in defect detection, further experiments were conducted on the NEU surface defect dataset and the fabric defect dataset with more complex backgrounds, achieving mAPs of 79.67% and 88.64%, respectively. The experiments confirm that YOLOv8n-SSSL outperforms existing textile defect detection methods, maintaining high accuracy while meeting real-time detection requirements.</p>