Fabric weft density measurement based on convolutional neural network

The accurate detection of weft density is a crucial aspect of textile quality control, and the measurement accuracy directly affects the quality control level of products. Traditional image processing methods rely on experience-based feature extraction algorithms, which usually leads to insufficient generalization performance for diverse fabric structures. To overcome these limitations, the CMixFUNeXt convolutional neural network combined with an optimized gray-scale projection algorithm is proposed to achieve end-to-end weft segmentation and calculate weft density. To validate this algorithm, a standardized image acquisition device based on a transmitted light source was constructed to acquire different fabric images and train the network. A dataset containing different fabric samples was also built for model training and verification. Experimental results show that the suggested method presents good performances in accuracy and adaptability for weft density detection.