光通信技术2026，Vol.50Issue(3)：22-28,7.DOI:10.13921/j.cnki.issn1002-5561.2026.03.004
面向高效DAS识别的手工-深度特征融合网络模型
Manual-deep feature fusion network model for efficient DAS recognition
摘要
Abstract
To address the challenges of cumbersome feature extraction,weak model generalization,and high computational complexity in distributed acoustic sensing(DAS)signal classification,this paper proposes a manual-deep feature fusion net-work for efficient DAS recognition.First,the DAS5C-0.5s-20k dataset is constructed.Then,80-dimensional multi-domain handcrafted features reduced by principal component analysis(PCA)are fused with 128-dimensional deep feature vectors ex-tracted by a multi-scale one-dimensional convolutional neural network(MS1DCNN),and support vector machine(SVM)is employed for classification.Experimental results demonstrate that the proposed model achieves an accuracy of 98.44%across five categories on the test set,representing improvements of 17.94%,11.25%,and 4.69%over SVM model using only hand-crafted features,one-dimensional convolutional neural network(1DCNN)model,and MS1DCNN model,respectively.关键词
分布式声学传感/特征融合网络/手工特征提取/深度特征提取Key words
distributed acoustic sensing/feature fusion network/manual feature extraction/deep feature extraction分类
信息技术与安全科学引用本文复制引用
张泽霆,梁清泉,吴迪,蒋佳骏,易高纬,王雪峰,于洋,张振荣..面向高效DAS识别的手工-深度特征融合网络模型[J].光通信技术,2026,50(3):22-28,7.基金项目
广西重点研发计划项目(桂科AB23075155)资助 （桂科AB23075155）
广西重点研发计划项目(桂科AB24010203)资助. （桂科AB24010203）
