A lightweight fabric defect detection with parallel dilated convolution and dual attention mechanism

Area of Science:

• Textile Manufacturing
• Computer Vision
• Artificial Intelligence

Background:

• Automated fabric defect detection is crucial for textile quality control.
• Existing methods often use complex models with high computational costs, limiting real-time application.
• There is a need for efficient and lightweight models for fabric defect detection.

Purpose of the Study:

• To develop a lightweight fabric defect detection (Light-FDD) model.
• To improve detection efficiency and accuracy in textile manufacturing.
• To reduce computational costs for real-time defect detection.

Main Methods:

• Utilized the You Only Look Once v8 Nano (YOLOv8n) framework with optimizations.
• Employed an improved FasterNet architecture for feature extraction.
• Introduced a parallel dilated convolution downsampling (PDCD) block and a global context and receptive-field (GCRF) attention mechanism.
• Implemented a lightweight cross-stage partial (CSP) layer for feature fusion.

Main Results:

• The proposed Light-FDD model demonstrated superior detection accuracy compared to existing lightweight models.
• Light-FDD achieved low computational cost, enabling real-time detection.
• Experiments on public datasets validated the model's performance and effectiveness.

Conclusions:

• Light-FDD offers a balanced approach to detection performance and computational efficiency.
• The model is a viable solution for real-time fabric defect detection in textile manufacturing.
• Optimized model design strategies can enhance the effectiveness of defect detection systems.