Filamentous target segmentation of weft micro-CT image based on U-Net

Area of Science:

• Materials Science
• Computer Science
• Textile Engineering

Background:

• Textile fabric inspection is crucial in manufacturing, with increasing adoption of advanced technologies.
• Micro-computed tomography (Micro CT) imaging is emerging as a valuable tool for textile material inspection.

Purpose of the Study:

• To propose and evaluate a U-Net based deep learning method for automatic segmentation of weft in micro-CT images of textile materials.
• To enhance the accuracy and efficiency of textile defect detection through automated image analysis.

Main Methods:

• Acquisition of weft micro-CT images using X-ray micro-CT scanning.
• Manual segmentation of target objects to create a high-accuracy textile material CT image dataset.
• Development of a modified U-Net model incorporating an attention mechanism and adjusted modules (encoder, decoder, loss function).

Main Results:

• The proposed algorithm demonstrated superior segmentation performance.
• The developed U-Net model achieved a Dice similarity coefficient of 0.843.
• The method enables accurate automatic segmentation of filamentous objects in textile micro-CT images.

Conclusions:

• The integration of deep learning and micro-CT technology offers a promising direction for industrial detection in the textile industry.
• The proposed segmentation method provides an effective solution for automated inspection of textile materials.
• Further research can explore the application of this technique in various industrial inspection scenarios.