Detection of Steel Surface Defects Based on Optimised YOLOv9 Mode

In the production and processing of steel, numerous factors, such as process control, equipment condition, and raw material quality, can influence the occurrence of surface defects. These defects may present as cracks, inclusions, scratches, and other visible imperfections. Such flaws not only detract from the visual quality of steel but also significantly impact its mechanical properties and service life. Therefore, timely and accurate detection of surface defects is crucial for ensuring steel quality and improving product qualification rates. This paper proposes a steel surface defect detection algorithm based on YOLOv9. First, the InceptionV1 module is utilized in the backbone network to replace the RepNCSPELAN4 module, enhancing the understanding of image content and enriching the expressiveness of features. Second, the Dysample upsampling operator is introduced in the neck network, effectively improving image resolution by intelligently adjusting sampling points to avoid distortion and enhance detail performance. Finally, incorporating the P6 detection layer into the head network enables the model to better recognize small-scale targets while maintaining high accuracy for larger ones. This approach allows for a more comprehensive and detailed detection of steel surface defects. Experimental results demonstrate that the enhanced model achieves average precision and accuracy of 78.7% and 78.4%, respectively, representing improvements of 2.3% and 3.3% over the benchmark model. The findings of this study have substantial practical implications for enhancing industrial product quality and production efficiency.