﻿ 基于AlexNet卷积神经网络的激光雷达飞机尾涡识别研究
 光电工程  2019, Vol. 46 Issue (7): 190082      DOI: 10.12086/oee.2019.190082

Research on aircraft wake vortex recognition using AlexNet
Pan Weijun, Duan Yingjie, Zhang Qiang, Wu Zhengyuan, Liu Haochen
Air Traffic Management Institute, Civil Aviation Flight University of China, Guanghan, Sichuan 618307, China
Abstract: In order to solve the flight safety issues threatened by wake vortex of leading aircraft, ensure air traffic safety, and improve the capacity of airdrome and airspace, an AlexNet convolutional neural network model algorithm is proposed to identify aircraft wake vortex. Combined with the detection principle of Doppler LiDAR and the classic model of Hallck-Burnham wake vortex velocity, the AlexNet neural network model was constructed to extract the image features of the wake vortex velocity images in the atmosphere and identify the aircraft wake vortex. The research shows that the model is able to accurately identify the aircraft wake vortex in the target airspace. After the network model converges, the accuracy rate reaches to 91.30%, which can effectively realize the identification work. Meanwhile, this study also demonstrates the low probability of false alarm of the AlexNet neural network in detecting wake vortex, which meets the requirement of early warning and monitoring of the aircraft wake vortex.
Keywords: wake vortex identification    AlexNet convolution neural network    target recognition    Doppler LiDAR

1 引言

2 激光雷达探测原理

 $\Delta {f_{\rm{D}}} = 2{f_0}\frac{v}{c} = \frac{2}{{{\lambda _0}}}{V_{\rm{R}}},$ (1)

 ${\Gamma _0} = \frac{{4Mg}}{{{\rm{ \mathsf{ π} }}\rho VB}},$ (2)

 ${b_0} = \frac{{\rm{ \mathsf{ π} }}}{4}B,$ (3)

 $V(r) = \frac{{{{\mathit{\Gamma}} _0}}}{{{\rm{2 \mathsf{ π} }}}} \cdot \frac{r}{{{r^2} + r_0^2}},$ (4)

 图 1 激光雷达在国内某机场实地探测图 Fig. 1 LiDAR experiment at an airport in China

 图 2 2018年9月5日国内某机场飞机尾涡速度场云图 Fig. 2 September 5, 2018, at an airport in China, the vortex velocity image of an aircraft
3 基于AlexNet的飞机尾涡识别方法 3.1 卷积神经网络

3.2 AlexNet模型

 图 3 AlexNet神经网络算法结构 Fig. 3 AlexNet algorithm structure

AlexNet神经网络模型共含8层结构(不包括池化层和局部响应归一化)，其中前5层为卷积层，后3层为全连接层，网络参数设置如表 1所示。

 Name Type Filter size Stride Padding Output size Input data Color image — — — 224×224×3 Conv1 — 7×7 4 0 55×55×96 Pool1 Max pooling 3×3 2 0 27×27×96 Conv2 — 5×5 1 2 27×27×256 Pool2 Max pooling 3×3 2 0 13×13×256 Conv3 — 3×3 1 1 13×13×384 Conv4 — 3×3 1 1 13×13×384 Conv5 — 3×3 1 1 13×13×384 Pool5 Max pooling 3×3 2 0 6×6×256 Fc6 — — — — 4049 Fc7 — — — — 4049 Fc8 — — — — 1000 Softmax with loss

AlexNet神经网络模型最后一层是由1000类输出的Softmax函数层用作分类。局部响应归一化层(local response normalization，LRN)出现在第1个及第2个卷积层后，而最大池化层出现在两个LRN层及最后一个卷积层后。线性整流函数(rectified linear unit，ReLU)激活函数则应用在这8层每一层的后面。因为AlexNet训练时使用了两块图形处理器(graphics processing unit，GPU)，因此该结构图中相关组件被拆为了两个部分[16]

AlexNet神经网络模型主要使用到的新技术如下：1)成功使用ReLU作为CNN的激活函数，并验证了其在较深网络中的有效性，解决了Sigmoid在网络较深时的梯度弥散问题。2)训练时在最后的全连接层使用Dropout随机忽略一部分神经元，以避免模型过拟合，提高了精度。3)使用重叠的最大池化，解决了平均池化模糊化效果，提升了特征的丰富性。4)运用局部响应归一化(local response normalization，LRN)层，增强模型的泛能力。5)数据增强，减轻过拟合，提升泛化能力[17]。在本文研究中，加之其具有较少的层数，时间复杂度少，有望实现尾涡的实时探测识别功能，故选择AlexNet模型作为识别模型来训练。

3.3 飞机尾涡识别

 Name Value Observation mode RHI Detection range/m 45~915 Detection accuracy/m 30 Scan angel range/(°) 10~60 Scan step length/(°) 1

 图 4 基于AlexNet模型的飞机尾涡识别流程 Fig. 4 Wake vortex identification process based on AlexNet model
4 仿真试验 4.1 实验平台

4.2 样本处理与分类

4.3 模型训练及识别结果

 图 5 部分实验结果AlexNet对飞机尾涡的识别情况 Fig. 5 Part of experimental results identification of the wake vortex by AlexNet

5 算法分析 5.1 准确率及其质量评价s

 ${\eta _{{\rm{ACC}}}} = \frac{{{N_{{\rm{TP}}}} + {N_{{\rm{TN}}}}}}{{{N_{{\rm{TP}}}} + {N_{{\rm{TN}}}} + {N_{{\rm{FP}}}} + {N_{{\rm{FN}}}}}},$ (5)

 图 6 准确率随迭代次数变化曲线 Fig. 6 Accuracy rate with training epochs
5.2 损失函数及其质量评价

 $H(p, q) = - \sum\nolimits_x {p(x)\log q(x)}。$ (6)

 图 7 损失函数输出值随迭代次数变化曲线 Fig. 7 Output value of loss function with training epochs

6 结论

 [1] Hallock J N, Holzäpfel F. A review of recent wake vortex research for increasing airport capacity[J]. Progress in Aerospace Sciences, 2018, 98: 27-36. [Crossref] [2] Gerz T, Holzäpfel F, Darracq D. Commercial aircraft wake vortices[J]. Progress in Aerospace Sciences, 2002, 38(3): 181-208. [Crossref] [3] Frehlich R, Sharman R. Maximum likelihood estimates of vortex parameters from simulated coherent Doppler Lidar data[J]. Journal of Atmospheric and Oceanic Technology, 2005, 22(2): 117-130. [Crossref] [4] Choroba P.Comprehensive study of the wake vortex phenomena to the assessment of its incorporation to ATM for safety and capacity improvements[D].Slovakia: The University of Zilina, 2006. [5] Wei Z Q.The Research on modeling and simulation on flow field and safety spacing for wake vortex[D].Tianjin: Civil Aviation University of China, 2008. 魏志强.尾涡流场及安全间隔的建模与仿真计算研究[D].天津: 中国民航大学, 2008. [6] K pp F, Rahm S, Smalikho I. Characterization of aircraft wake vortices by 2-μm pulsed Doppler lidar[J]. Journal of Atmospheric and Oceanic Technology, 2004, 21(2): 194-206. [Crossref] [7] Holzäpfel F, Gerz T, K pp F, et al. Strategies for circulation evaluation of aircraft wake vortices measured by lidar[J]. Journal of Atmospheric and Oceanic Technology, 2003, 20(8): 1183-1195. [Crossref] [8] Barbaresco F, Jeantet A, Meier U.Wake vortex detection & monitoring by X-band doppler radar: paris orly radar campaign results[C]//Proceedings of 2007 IET International Conference on Radar Systems, Edinburgh, UK, 2007. [9] Barbaresco F, Meier U. Radar monitoring of a wake vortex:electromagnetic reflection of wake turbulence in clear air[J]. Comptes Rendus Physique, 2010, 11(9): 54-67. [Crossref] [10] Profeta A, Rodriguez A, Clouse H S. Convolutional neural networks for synthetic aperture radar classification[J]. Proceedings of SPIE, 2006, 9843: 98430M. [Crossref] [11] Wu Y H, Hu Y H, Dai D C, et al. Research on the technique of aircraft wake vortex detection based on 1.5μm doppler lidar[J]. Acta Photonica Sinica, 2011, 40(6): 811-817. 吴永华, 胡以华, 戴定川, 等. 基于1.5μm多普勒激光雷达的飞机尾涡探测技术研究[J]. 光子学报, 2011, 40(6): 811-817 [Crossref] [12] Li C, Liu J W, Zhao P E, et al. Correction method of tilt wind field of mobile wind lidar[J]. Laser Technology, 2017, 41(3): 385-390. 李策, 刘俊伟, 赵培娥, 等. 机动型激光测风雷达倾斜风场修正算法研究[J]. 激光技术, 2017, 41(3): 385-390 [Crossref] [13] Arel I, Rose D C, Karnowski T P. Deep machine learning-a new frontier in artificial intelligence research[Research Frontier][J]. IEEE Computational Intelligence Magazine, 2010, 5(4): 13-18. [Crossref] [14] Yuan Q Z, Wei S J, Luo N. Research on SAR satellite target recognition system based on deep learning neural network[J]. Aerospace Shanghai, 2017, 34(5): 46-53. 袁秋壮, 魏松杰, 罗娜. 基于深度学习神经网络的SAR星上目标识别系统研究[J]. 上海航天, 2017, 34(5): 46-53 [Crossref] [15] Yin B C, Wang W T, Wang L C. Review of deep learning[J]. Journal of Beijing University of Technology, 2015, 41(1): 48-59. 尹宝才, 王文通, 王立春. 深度学习研究综述[J]. 北京工业大学学报, 2015, 41(1): 48-59 [Crossref] [16] Krizhevsky A, Sutskever I, Hinton G E.ImageNet classification with deep convolutional neural networks[C]//Proceedings of the 25th International Conference on Neural Information Processing Systems, South Lake Tahoe, USA: 2012: 1097-1105. [17] Zhang S H.Research on facial landmark localization based on deep convolutional neural network[D].Wuhan: Huazhong University of Science and Technology, 2016. 张少华.基于深度卷积神经网络的人脸基准点定位研究[D].武汉: 华中科技大学, 2016. [18] Hu Y H, Wu Y H. Study on the characteristic of aircraft wake vortex and lidar detection technique[J]. Infrared and Laser Engineering, 2011, 40(6): 1063-1069. 胡以华, 吴永华. 飞机尾涡特性分析与激光探测技术研究[J]. 红外与激光工程, 2011, 40(6): 1063-1069 [Crossref] [19] Duan M, Wang G P, Niu C Y. Method of small sample size image recognition based on convolution neural network[J]. Computer Engineering and Design, 2018, 39(1): 224-229. 段萌, 王功鹏, 牛常勇. 基于卷积神经网络的小样本图像识别方法[J]. 计算机工程与设计, 2018, 39(1): 224-229 [Crossref] [20] Dai W C, Jin L X, Li G N, et al. Real-time airplane detection algorithm in remote-sensing images based on improved YOLOv3[J]. Opto-Electronic Engineering, 2018, 45(12): 180350. 戴伟聪, 金龙旭, 李国宁, 等. 遥感图像中飞机的改进YOLOv3实时检测算法[J]. 光电工程, 2018, 45(12): 180350 [Crossref] [21] Pan W J, Zhang Q Y, Zhang Q, et al. Identification method of aircraft wake vortex based on doppler lidar[J]. Laser Technology, 2019, 43(2): 233-237. 潘卫军, 张庆宇, 张强, 等. 多普勒激光雷达的飞机尾涡识别方法[J]. 激光技术, 2019, 43(2): 233-237 [Crossref]