系统工程与电子技术 ›› 2019, Vol. 41 ›› Issue (4): 772-779.doi: 10.3969/j.issn.1001-506X.2019.04.11

• 传感器与信号处理 • 上一篇    下一篇

基于卷积神经网络的低空风切变类型识别

熊兴隆1, 陈楠1, 李永东1, 马愈昭1, 李猛2, 冯帅3   

  1. 1. 中国民航大学天津市智能信号与图像处理重点实验室, 天津 300300;
    2. 中国民航大学民航空管研究院, 天津 300300; 3. 中国民航大学工程技术训练中心, 天津 300300
  • 出版日期:2019-03-20 发布日期:2019-03-20

Type recognition of low level wind shear based on convolutional neural network

XIONG Xinglong1, CHEN Nan1, LI Yongdong1, MA Yuzhao1, LI Meng2, FENG Shuai3   

  1. 1. Tianjin Key Laboratory for Advanced Signal Processing, Civil Aviation University of China, Tianjin 300300, China;
    2. Key Laboratory of Operation Programming and Safety Technology of Air Traffic Management, Civil Aviation University of China, Tianjin 300300, China;  3. Engineering Technical Training Center, Civil Aviation University of China, Tianjin 300300, China
  • Online:2019-03-20 Published:2019-03-20

摘要:

针对激光雷达低空风切变信号图像的类型识别问题,提出了一种基于深度卷积神经网络(deep convolutional neural network,DCNN)的多层特征提取及自适应融合算法。该方法可以有效解决网络逐层训练过程中信息丢失的问题。首先,采用DCNN提取低空风切变信号图像的各层网络特征,并将各特征进行L2范数标准化实现同趋化。其次,将其以多通道图像形式输入单层CNN进行自适应融合,将融合特征送入支持向量机进行分类识别。结果表明,采用所提算法进行低空风切变图像类型识别的平均识别率为98.1%,与其他4种算法相比均有提升。所提算法能更有效地实现低空风切变信号图像类型识别。

Abstract:

This paper proposes an algorithm for multilayer feature extraction and adaptive fusion based on deep convolutional neural network (DCNN) to solve the problem of image type recognition of the lowlevel wind shear signal scanned by the laser radar. The method effectively makes up for the information loss in the process of layerbylayer network training. First of all, the characteristics of the network layers of lowlevel wind shear signal image are extracted by using DCNN modeling and the characteristics are L2 norm standardized. Then the L2 norm standardized characteristics are put into singlelayer CNN for adaptive fusion in the form of multichannel image and the fusion characeristics are sent into the support vector machine for classification recognition. The results show that the average recognition rate of image type recognition of the low level wind shear signal obtained by using the proposed algorithm is 98.1%, being improved compared with the other four algorithms. The algorithm can effectively realize image type recognition of the low level wind shear signal.