基于StyIeGAN2-ADA的多尺度小样本感知模型研究

打开文本图片集

中图分类号：TP391.4 文献标识码：A 文章编号：2096-4706（2026）03-0026-08

Research on Multi-scale Few-shot Perception Model Based on StyleGAN2-ADA

Chunyu'， ZHANG Xinyan'，WU Jing'，CHEN Enginering， ， 55oo25， ， ， ，）

Abstract：As an importantcomponent intangible cultural heritage，Miaoembroidery has crucial significance fordigital protectionworkthroughtheidentificationitspattefeaturestheclassifcationitscraftsmanshiptechnques.However， traditional methods struggle toaccuratelydistinguishcomplex stitchvariations texture features， existingDeepLeang models haveobvious deficiencies in cro-scale feature corelationmodeling，which seriouslyrestrictsclasificationaccuracy. To addressteabove problems，amulti-scalefew-shot perception model is proposed.StyleGAN-ADA is adopted to generate diverseMiaoembroiderysamplesfordataaugmentation.ASPPis introducedattheendthedownstreamResNet1Olnetwork tostrengthen globalfeaturefusion.Meanwhile，aimAMisembedded，themodel'sabilitytodistinguishcomplexptesis significantly improved through adaptive weighting feature maps.The proposed model exhibits superior performance on both self-built datasets descriptive texture datasets.

Keywords：Miao embroidery; data augmentation; image classification;ASPP; SimAM

0 引言

苗绣是中国传统民族手工艺中具有悠久历史和独特文化价值的艺术形式之一，近年来备受国家的重视和研究者的关注[1-2]。（剩余11392字）