This paper presents an improved YOLOv8 algorithm for steel surface defect detection, aiming to enhance the accuracy and efficiency of defect detection. The key contributions include: 1. **Deformable Convolution (DCNv2)**: Introducing DCNv2 into the C2F module of YOLOv8 to enhance the model's ability to capture complex shapes and irregular target features, improving the detection accuracy of small and complex defects. 2. **Bi-Directional Feature Pyramid Network (BiFPN)**: Replacing the original PAFPN with BiFPN to better integrate multi-scale feature information, particularly for small or complex defects, further improving detection accuracy. 3. **BiFormer Attention Mechanism**: Integrating BiFormer into the backbone network to enhance the model's focus on defect details, allowing for more adaptive attention and computational resource allocation. 4. **Wise-IoUv3 Loss Function**: Replacing the original CIoU loss with WIoUv3 to improve the accuracy of bounding box regression, especially for low-quality data, and enhancing the model's robustness and generalization. The experimental results show that the improved model achieves a mean Average Precision (mAP) of 84.8% on the NEU-DET dataset, a significant improvement of 6.9% over the original YOLOv8 model. The model demonstrates superior performance in detecting various types of steel surface defects, including cracks and rolled-in scales, while maintaining real-time performance suitable for industrial applications. The proposed improvements are validated through ablation and comparative experiments, demonstrating their effectiveness and broad applicability in small target detection tasks.This paper presents an improved YOLOv8 algorithm for steel surface defect detection, aiming to enhance the accuracy and efficiency of defect detection. The key contributions include: 1. **Deformable Convolution (DCNv2)**: Introducing DCNv2 into the C2F module of YOLOv8 to enhance the model's ability to capture complex shapes and irregular target features, improving the detection accuracy of small and complex defects. 2. **Bi-Directional Feature Pyramid Network (BiFPN)**: Replacing the original PAFPN with BiFPN to better integrate multi-scale feature information, particularly for small or complex defects, further improving detection accuracy. 3. **BiFormer Attention Mechanism**: Integrating BiFormer into the backbone network to enhance the model's focus on defect details, allowing for more adaptive attention and computational resource allocation. 4. **Wise-IoUv3 Loss Function**: Replacing the original CIoU loss with WIoUv3 to improve the accuracy of bounding box regression, especially for low-quality data, and enhancing the model's robustness and generalization. The experimental results show that the improved model achieves a mean Average Precision (mAP) of 84.8% on the NEU-DET dataset, a significant improvement of 6.9% over the original YOLOv8 model. The model demonstrates superior performance in detecting various types of steel surface defects, including cracks and rolled-in scales, while maintaining real-time performance suitable for industrial applications. The proposed improvements are validated through ablation and comparative experiments, demonstrating their effectiveness and broad applicability in small target detection tasks.