Intelligent Fabric Pattern Detection and Design Based on Mask R-CNN

This research introduces an automated fabric pattern detection method using Mask R-CNN, enhancing accuracy and automation for the textile and fashion industries. The method begins with data preprocessing to build a high-quality dataset, employing the ResNet-50 backbone, Region Proposal Network (RPN), and ROI alignment for model development. The model's generalization is boosted through hyperparameter optimization and multi-scale training, supported by hardware acceleration to improve real-time processing. Mask R-CNN achieves 95% accuracy, 90.48% recall, and 92.7% F1 score, significantly surpassing Faster R-CNN and YOLO, which exhibit lower F1 scores and processing speeds. Despite a lower frame rate compared to YOLO, Mask R-CNN's robust performance under complex and low-light conditions recommends it for high-precision tasks in fabric pattern detection, with potential for future efficiency and application enhancements through continued technological advancement.