Scholarlink

<p>Fabric defect detection is a crucial direction in the field of industrial inspection. Due to various constraints associated with local deployment, lightweight detection models have become indispensable. In recent years, the YOLO series of detectors has demonstrated significant potential in this domain. However, challenges such as complex backgrounds and small defects in real-world production environments often lead to suboptimal detection accuracy. To address these issues, this paper proposes a lightweight LF-YOLO algorithm. Based on YOLOv8 as the backbone network, we integrate a Contextual Feature Integrator (CFI), which employs grouped dilated convolutions to perform dual feature extraction on multi-layer fused features, achieving multi-scale information fusion and effectively tackling the problem of fabric detection in complex backgrounds. Additionally, we introduce a Tiny-Detect (TD) layer to enhance the model’s perception of minute defects by generating small-scale feature maps. Furthermore, we design an Enhanced Depthwise Separable Convolution (EDSC), which supplements an additional pointwise convolution to improve inter-channel feature mixing, optimizing the feature extraction process while reducing computational overhead. This leads to an efficient and lightweight network structure. Experimental results on the ZJU-Leaper dataset demonstrate that LF-YOLO not only maintains the advantages of being a lightweight model but also achieves excellent detection speed and accuracy. Moreover, it performs exceptionally well in scenarios involving complex backgrounds and small-object detection tasks. Therefore, the proposed algorithm exhibits great potential in the field of fabric defect detection.</p>

错误：搜索内容不能为空，请输入英文关键词
错误：关键词超出字数限制，请精简
高级检索

Lighter, faster, more focused YOLO fabric defect detection model

  • 期刊论文

  • 发表日期：

    2025年9月3日

查看全文
  • Hao Wu,
  • Jun Liu,
  • Hao Liu,
  • Jiuzhen Liang

摘要

Fabric defect detection is a crucial direction in the field of industrial inspection. Due to various constraints associated with local deployment, lightweight detection models have become indispensable. In recent years, the YOLO series of detectors has demonstrated significant potential in this domain. However, challenges such as complex backgrounds and small defects in real-world production environments often lead to suboptimal detection accuracy. To address these issues, this paper proposes a lightweight LF-YOLO algorithm. Based on YOLOv8 as the backbone network, we integrate a Contextual Feature Integrator (CFI), which employs grouped dilated convolutions to perform dual feature extraction on multi-layer fused features, achieving multi-scale information fusion and effectively tackling the problem of fabric detection in complex backgrounds. Additionally, we introduce a Tiny-Detect (TD) layer to enhance the model’s perception of minute defects by generating small-scale feature maps. Furthermore, we design an Enhanced Depthwise Separable Convolution (EDSC), which supplements an additional pointwise convolution to improve inter-channel feature mixing, optimizing the feature extraction process while reducing computational overhead. This leads to an efficient and lightweight network structure. Experimental results on the ZJU-Leaper dataset demonstrate that LF-YOLO not only maintains the advantages of being a lightweight model but also achieves excellent detection speed and accuracy. Moreover, it performs exceptionally well in scenarios involving complex backgrounds and small-object detection tasks. Therefore, the proposed algorithm exhibits great potential in the field of fabric defect detection.