To address the current steel surface defect detection algorithms in practical applications involving low detection accuracy, an efficient and highly accurate strip steel surface defect detection algorithm, DEW-YOLO, is proposed in this paper. Firstly, by combining the advantages of deformable convolutional networks (DCNs), this paper innovates the C2F module in YOLOv8 and proposes a C2f_DCN module that can flexibly sample features to enhance the abilities of learning and expressing defect features of different sizes and shapes. Secondly, the explicit visual center (EVC) is introduced into the backbone network, which enhances feature extraction capabilities and adaptability and enables the model to better adjust features at different levels and scales. Finally, the original loss function is replaced with the Wise-IoU (WIoU) loss function to accurately measure the similarity between the target frames and improve the defect detection performance of the model. The experimental results on the NEU-DET dataset demonstrate that the algorithms proposed in this paper achieved a mean average precision (mAP) of 80.3% in steel surface defect detection tasks, which was a 3.9% improvement over the original YOLOv8 model. The model’s inference speed reached 91 frames per second (FPS). DEW-YOLO effectively enhances the accuracy of steel defect detection and better satisfies industrial inspection requirements.

Aiming at the existing bearing defect detection algorithms with low accuracy, large number of parameters and computation, this paper proposes an efficient and lightweight bearing surface defect detection algorithm FBS-YOLO based on YOLOv8. Firstly, FasterNet replaces the original feature extraction network of YOLOv8, and uses Partial Convolution (PConv) to reduce redundant computation and memory access. Secondly, the fusion of weighted Bidirectional Feature Pyramid Network (BiFPN) in Neck network, which removes less efficient feature transmission nodes in the process of multi-scale feature fusion to achieve a higher level of fusion, improves the fusion efficiency of features at different scales. Finally, the advantages of Switchable Atrous Convolution (SAConv) are introduced to innovate the CSP Bottleneck with the two convolutions (C2f) module in the original model Neck network, and SAConv is combined with C2f (C2f_SAConv) to from a more flexible module adapted to the features of different scales is proposed to enhance the feature extraction and processing capability of the model. The experimental results show that the algorithm FBS-YOLO proposed in this paper achieves a mAP of 91.4% in the bearing defect detection task, which is 2.8% higher than that of the original YOLO8 model, and the number of parameters and computation volume are reduced by 39.8% and 41.9%, respectively, and the model inference speed can be up to 161 fps. The algorithm meets the light-weight requirements of industrial detection deployment while maintaining high accuracy, effectively achieving a balance between model lightweight and performance, and providing new ideas for end-to-end industrial deployment.