02339nas a2200313   4500000000100000008004100001653000900042653001900051653002500070653001800095653002600113653003400139653002100173653002100194653002800215653002100243653002300264653002000287653001800307653002800325653001700353653002400370100001400394700001300408245017200421856015800593300001400751520126000765                                       d10aAIGC10aArts computing10aContrastive Learning10aGlobalisation10aInnovative approaches10aIntangible cultural heritages10aLearning systems10amachine learning10aMachine learning models10aMachine-learning10aModel construction10aNeural networks10aPaper cutting10aQuantitative evaluation10aTrajectories10aTrajectory analysis1 aZiyi Wang1 aYong Liu00aQuantitative Evaluation of Intangible Cultural Heritage Inheritance: Stroke Trajectory Analysis and Machine Learning Model Construction for Paper-Cutting Craftsmanship  uhttps://www.scopus.com/inward/record.uri?eid=2-s2.0-105012115661&doi=10.1109%2FAIITA65135.2025.11048123&partnerID=40&md5=48e68ddb5558db71ac17c698d1a91dfe  a1332-13363 aWith the acceleration of globalization, intangible cultural heritage (ICH) faces new challenges, particularly in the crisis of craftsmanship inheritance. Effectively quantifying and assessing artisans  skills is crucial for preserving traditional craftsmanship. Therefore, this study proposes an innovative approach based on papercutting art, utilizing stroke trajectory analysis combined with machine learning models to quantitatively evaluate the skills of papercutting inheritors. Specifically, high-quality stroke trajectory images of papercutting art are collected using web crawling techniques. Degraded images are then generated through threshold processing and pixel swapping, followed by the use of a diffusion model to synthesize defective stroke trajectory images. Based on these data, we design the BruNet model, which extracts multi-space features through convolutional neural networks (CNNs) and multi-layer perceptrons (MLPs). These features are then fused using a feature fusion module and trained with cross-entropy and contrastive learning losses. Experimental results demonstrate that BruNet achieves significant performance improvements over state-of-the-art models, providing novel insights and technical support for ICH preservation.