Fabric defect detection algorithm based on improved YOLOv8s

In the fabric manufacturing industry, various unfavorable factors often lead to fabric defects. However, current systems face issues such as multi-scale defects and loss of information on tiny objects, resulting in low target detection accuracy and high missed detection rates. Based on the aforementioned problems, this paper proposes an improved YOLOv8s algorithm to detect fabric defects. By integrating the DCNv3 operator into the C2f module of YOLOv8s's backbone network, the C2f-DCNv3 module is successfully constructed. This improvement enhances the model's ability to model geometric transformations and contextual information, thereby better capturing spatial deformations and contextual details. Experiments conducted on the Tianchi Fabric Defect Detection dataset reveal that compared to the original YOLOv8s, the algorithm proposed in this paper achieves a 3.14% improvement in the [email protected] metric. The algorithm presented in this paper achieves efficient and accurate defect recognition, demonstrating superior performance across various fabric types and complex backgrounds.