| 基于特征图裁剪的高铁周界入侵实时检测算法 |
| |
| 引用本文： | 王玮,朱力强.基于特征图裁剪的高铁周界入侵实时检测算法[J].铁道学报,2019(9):74-80. |
| |
| 作者姓名： | 王玮 朱力强 |
| |
| 作者单位： | 北京交通大学机械与电子控制工程学院;北京交通大学载运工具先进制造与测控技术教育部重点实验室 |
| |
| 基金项目： | 国家重点研发计划(2016YFB1200103);中国铁路总公司科技研究开发计划(2017T001-B) |
| |
| 摘 要： | 为解决高速铁路周界入侵高可靠度检测的难题,研究基于深度学习的异物入侵实时检测算法。针对深度卷积神经网络存在内存占用量大和检测耗时长的问题,提出以特征图L1范数为准则的递归裁剪算法,逐步降低模型计算量及储存空间,同时将检测准确率保持在较高水平。在基于ImageNet数据库和铁路场景数据库的测试中,该算法可以将VGG16模型压缩约660倍并加速计算4.4倍,而损失的检测准确率分别只有1.2%和0.25%。研究结果表明,基于特征图L1范数的裁剪准则普遍适用于任何具有卷积运算的神经网络结构中,性能优于现有基于卷积核L1范数的裁剪准则。
|
| 关 键 词： | 侵限检测 卷积神经网络 模型压缩 |
Real-time Intrusion Detection Algorithm for High-speed Railway Based on Feature Map Pruning |
| |
| WANG Wei,ZHU Liqiang.Real-time Intrusion Detection Algorithm for High-speed Railway Based on Feature Map Pruning[J].Journal of the China railway Society,2019(9):74-80. |
| |
| Authors: | WANG Wei ZHU Liqiang |
| |
| Institution: | (School of Mechanical,Electronic and Control Engineering,Beijing Jiaotong University,Beijing 100044,China;Key Laboratory of Vehicle Advanced Manufacturing,Measuring and Control Technology,Ministry of Education, Beijing Jiaotong University,Beijing 100044,China) |
| |
| Abstract: | WANG Wei;ZHU Liqiang(School of Mechanical,Electronic and Control Engineering,Beijing Jiaotong University,Beijing 100044,China;Key Laboratory of Vehicle Advanced Manufacturing,Measuring and Control Technology,Ministry of Education, Beijing Jiaotong University,Beijing 100044,China) |
| |
| Keywords: | intrusion detection convolutional neural network(CNN) model compression |
| 本文献已被 CNKI 维普 等数据库收录！ |
|
