YOLOv8-WBE: A Fabric Defect Detection Algorithm Integrating Frequency Domain Features

Due to the small size of fabric defects and the possibility that the detected features may resemble the background characteristics of the fabric, models often struggle to correctly distinguish between the target and the background. To address this issue, we propose a fabric defect detection algorithm integrating frequency domain features—YOLOv8-WBE. First, in the feature extraction stage, a frequency domain downsampling module (WaveDown) is incorporated. By using wavelet transformation to improve the downsampling operation within the deep neural network, this module helps suppress aliasing issues commonly encountered in fabric detection. Second, in the Neck part, a Bidirectional Feature Pyramid Network (BiFPN) structure is adopted to achieve cross-scale spatial fusion, enhancing the model’s ability to learn detailed and multi-scale features of fabric defects. Lastly, an Efficient Multi-scale Attention mechanism (EMA) is introduced for multi-scale feature learning and cross-spatial interaction, further improving the detection accuracy of fabric defects. Experimental results show that the improved model achieves a mAP of 96.3%, a 2.3% increase compared to the original YOLOv8n, meeting the detection requirements of textile enterprises.