Research on Fabric Defect Detection Algorithm Based on Improved YOLOv8s

Fabric defect detection plays a pivotal role in the textile industry. To tackle challenges like the wide range of fabric defect types, their small sizes, and the need for higher detection accuracy, we propose an enhanced YOLOv8s-based algorithm tailored for fabric defect detection. To boost the precision of detecting small defects, we've introduced a multi-scale feature fusion mechanism. This mechanism innovatively fuses CA (channel attention) and MSAM (multi-scale attention module) to create the CMSA (channel and multi-scale attention module). The CMSA is seamlessly integrated into the final layer of the YOLOv8s feature fusion network, addressing the issue of misalignment between classification and localization tasks. Additionally, it enriches contextual information for feature fusion, bolstering the model's multi-scale feature extraction capabilities and enhancing sensitivity to small targets. Experiments on the fabric defect dataset from the Alibaba Cloud Tianchi Competition reveal that our proposed algorithm outperforms the original YOLOv8s, achieving a 2.4% improvement in the [email protected] metric.