A lightweight steel surface defect detection model based on YOLOv9

In addressing the problems of low detection accuracy and high model complexity in current deep learning-based steel surface defect detection algorithms, a steel surface defect detection model based on an improved YOLOv9 is put forward. This model augments the general efficient layer aggregation network by adding four large separable kernel attention modules, thereby enhancing the detection accuracy of minor defects in complex backgrounds. To lower model complexity, an ultra-lightweight dynamic up-sampling operator is integrated, making the network more lightweight and resource-efficient. Moreover, omni-dimensional dynamic convolutions are used to replace traditional convolutions in the YOLOv9 network to further improve defect detection performance. Experimental results show that the proposed model attains a mean average precision (mAP) of 78.9% on the NEU-DET steel defect detection dataset, surpassing the benchmark YOLOv9 model by 4.8%. The detection speed is increased by 5 frames per second (FPS), while model parameters and computational complexity are reduced by 1.1% and 0.6% respectively. These enhancements make our model highly suitable for steel surface defect detection in industrial environments with limited hardware resources.