Steel surface defect detection based on MobileViTv2 and YOLOv8

To address the issue of low detection accuracy of steel surface defects due to complex texture background interference and complex defect morphology, this paper proposes an improved YOLOv8 model based on MobileViTv2 and Cross-Local Connection for steel surface defect detection. Firstly, the lightweight MobileViTv2 network is introduced into the backbone network, which enhances the feature extraction capability of the model in complex defect shapes by combining the advantages of CNN and Transformer. Then, the designed CLC method is introduced into the neck network, which connects deep and shallow features through additional convolutional layers, further integrating defect features in the presence of complex texture background interference. Finally, the NET-DET dataset is augmented to improve the model’s robustness. Experimental results show that the mAP of the improved model is 74.1%, with a detection speed of 86.2 FPS and model memory usage of 27.5 MB. Compared to YOLOv5 and YOLOv8, the mAP of the improved model is increased by 6.5% and 4%, respectively. Compared to existing object detection models, the improved model has the characteristics of high detection accuracy and fast detection speed, meeting the requirements of industrial production for steel surface defect detection.