Fabric defect detection algorithm based on dynamic enhanced texture features

The primary challenge in fabric defect detection lies in the visual similarity between defect regions and complex background textures. Although existing methods have achieved certain success in simple scenarios, detection accuracy still struggles to meet industrial production requirements under complex backgrounds. Therefore, this paper proposes a fabric defect detection network based on dynamically enhanced texture features (DETFNet). First, a Multi-scale Texture Feature Enhancement Module (MTFEM) is designed, which dynamically updates convolutional kernels to adaptively capture fine-grained texture details based on the content of the input feature maps, thereby improving the model's ability to handle complex texture backgrounds. Second, to further enhance the model's ability to extract multi-scale defect features, a Feature Fusion Attention Module (FFAM) is proposed. This module combines multi-scale feature fusion, channel attention, and local attention mechanisms to effectively integrate feature maps from different levels. Finally, a lighter GELAN module is introduced to replace the C3 module, improving parameter utilization and detection speed while maintaining detection performance. Experiments on the Alibaba Tianchi Fabric Dataset show that the proposed method achieves [email protected] and mAP of 49.2% and 25.1%, respectively, representing improvements of 2.2% and 2.6% over the baseline method. Additionally, the detection speed of the proposed method reaches 68 frames per second (fps), meeting the requirements for accuracy and real-time performance in textile industrial production.