FEB-YOLOv8: A Steel Surface Defect Detection Algorithm Based on Improved YOLOv8s

In order to solve the problems of low recognition accuracy, many model parameters and large computation in the current steel surface defect detection algorithms. This paper proposes a steel strip surface defect detection algorithm based on improved YOLOv8s, called FEB-YOLOv8. Firstly, a new partial convolution is used to improve the C2f structure, which is able to effectively reduce the latency and reduce the model parameters. Secondly, an EMA attention mechanism is introduced into the end of the backbone network, which further enhances the feature extraction capability by encoding global information. Finally, a simplified version of BiFPN feature fusion is used to improve the model's ability to detect multi-scale defects. Experimental results show that compared with YOLOv8s, FEB-YOLOv8 increases mAP by 4.7% on the GC10-DET dataset, reduces the number of model parameters by 12.98
^%
, and reduces the number of floating point operations (FLOPs) by 14.58%, achieving a balance between algorithm detection accuracy and model parameters.