摘要
由于实际应用中丝织物瑕疵样本的稀缺性和异常区域的细微性,现有方法可能对瑕疵区域不敏感从而导致定位错误.因此,为了借助Transformer注意力机制识别细微特征的优势,本文旨在建立一种适合丝织物瑕疵检测任务的动态注意力共享机制,以增强模型对细微纹理的捕捉能力.具体来说,本文提出了一种基于混合多头注意力重建网络的丝织物瑕疵检测模型:首先,使用预训练Transformer编码器提取丝织物图像特征;然后,使用带有MeanDropout的瓶颈层以减少模型对重复特征的依赖;此外,提出一种混合多头注意力机制(mix mutil-head attention,MMHA)和LlamaMLP相结合的解码器以协助注意力头动态选择适当的丝织物特征,从而提升对关键细微纹理的关注;最后,通过解码器松散重建多层特征的组合,以实现瑕疵检测和定位.在真实公开数据集上进行了大量的实验.实验结果表明,所提方法在两种数据集上的图像级指标分别提升了2.1%和0.6%,并且在像素级指标上分别达到了96.0%和60.5%,取得了领先的性能.
Due to the scarcity of defective silk fabric samples in real-world applications and the subtle nature of anomaly regions,existing methods may fail to detect defect areas accurately,leading to localization errors.To leverage the Transformer attention mechanism’s ability to capture fine-grained features,this study proposes a dynamic attentionsharing mechanism tailored for silk fabric defect detection tasks,enhancing the model’s ability to capture subtle textures.Specifically,a silk fabric defect detection framework based on a mix multi-head attention(MMHA)reconstruction network is proposed.First,a pre-trained Transformer encoder is used to extract features from silk fabric images.Next,a bottleneck layer with MeanDropout is employed to reduce the model’s reliance on redundant features.Furthermore,a decoder combining MMHA with LlamaMLP is introduced to facilitate the dynamic selection of relevant silk fabric features,thus improving attention to critical fine-grained textures.Finally,the decoder loosely reconstructs combinations of multi-level features to achieve defect detection and localization.Extensive experiments conducted on publicly available datasets demonstrate that the proposed method achieves state-of-the-art performance in silk fabric defect detection.Experimental results show that the proposed method improves image-level indicators by 2.1%and 0.6%on two datasets,respectively,and achieves leading performance with pixel-level indicators of 96.0%and 60.5%.
作者
胡蓉
马浩然
李炜
刘伟霞
李佐勇
HU Rong;MA Hao-Ran;LI Wei;LIU Wei-Xia;LI Zuo-Yong(Fujian Provincial Key Laboratory of Big Data Mining and Applications,Fujian University of Technology,Fuzhou 350118,China;School of Computer Science and Mathematics,Fujian University of Technology,Fuzhou 350118,China;Xiamen Air Traffic Management Station of CAAC,Xiamen 361006,China;Fujian Provincial Key Laboratory of Information Processing and Intelligent Control,Minjiang University,Fuzhou 350108,China;School of Computer and Data Science,Minjiang University,Fuzhou 350108,China)
出处
《计算机系统应用》
2025年第8期149-158,共10页
Computer Systems & Applications
基金
国家自然科学基金面上项目(62471207)
福建省自然科学基金重点项目(2024J02029)。
