Weft density measurement of fabrics based on high-low frequency multi-scale feature enhancement network

Weft density is a core indicator for evaluating fabric quality. Aiming at the problems of traditional image processing methods in measuring fabric weft density, such as poor adaptability to different types, low accuracy, and poor performance on complex fabric textures, a fabric weft density detection method based on the High-Low Frequency Multi-Scale Feature Enhancement Network (HLFMSUNext) is proposed. Firstly, a standardized fabric acquisition system based on transmitted light source is designed to collect fabric images, and Labelme is used for data annotation to construct a dataset containing different fabric samples. Secondly, the HLFMSUNext network is designed, which uses the High-Low Frequency Adaptive Enhancement (HLFAE) module and Convolutional Multi-Scale Attention (CMEA) module to deeply extract fabric texture features and achieve accurate segmentation of fabric wefts. Finally, the Hough transform and local multi-directional projection method are used to process the weft information predicted by the network model, and the fabric weft density value is calculated. The results show that the proposed method has good detection accuracy and strong variety adaptability when detecting the weft density of different types of fabrics.