Automatic measurement of surface parameters of composite braided fabrics based on modified U-shaped network

Braided composite materials have broad application prospects in advanced fields such as aerospace and rail transportation. Studies have shown that surface parameters of braided fabrics—such as braiding angles and pitch lengths—significantly influence their performance. However, measuring these parameters during mass production presents challenges related to automation, efficiency, and precision. Traditional image processing methods often suffer from limited accuracy, while deep learning approaches are difficult to implement, and their generalization ability remains unverified. In this paper, a measurement method based on a modified U-shaped network (denoted as U-Net) is proposed. First, a data-augmented image dataset of fabric surfaces with yarn edge annotations is constructed. Then, an enhanced U-Net model incorporating Dropout and an encoder based on the Visual Geometry Group's 16-layer (denoted as VGG16) network is applied, along with a weighted mixed loss function combining Focal Loss and Dice Loss to improve detection of weak yarn edge regions. Morphological operations and distortion rectification are employed for post-processing and noise reduction. Experimental results show that the proposed method improves measurement accuracy by up to 1.84 % across various fabric samples, demonstrating excellent stability and generalization performance.