基于机动动作类型识别的飞行员飞行训练质量自动评估方法

doi:10.12305/j.issn.1001-506X.2023.12.23

摘要/Abstract

摘要：

飞行训练质量评估是飞行员日常飞行训练的重要组成部分, 针对传统飞行训练质量评估方法评估效率低、主观性强等问题, 提出了一种基于机动动作识别的自动评估方法。首先, 通过构建采用多尺度特征提取的深度残差网络(multi-scale feature extraction deep residual network, MSDRN), 实现了战术机动动作的准确识别, 克服了传统机动动作识别方法识别准确度低的问题。然后, 针对不同类型的战术机动动作, 构建了飞行质量评估指标体系, 在具体评估时, 根据动作识别结果自动选择评估指标, 并采用基于博弈论的组合赋权法计算并获得评估结果, 由此构建了基于机动动作类型识别的飞行员飞行训练质量自动评估方法。所构建的动作识别方法相较于利用传统的一维卷积神经网络(one-dimensional convolutional neural network, 1D-CNN)、残差神经网络(residual neural network, ResNet)构建的分类器, 识别准确度分别提升了16.2%和3.5%, 识别计算时间相比ResNet缩短了23.9%。所提出的整个飞行训练质量自动评估方法摆脱了飞行员训练过程对飞行教官的严重依赖, 实现了飞行训练质量的自动化快速评估, 且评估结果更具科学性, 从而为飞行训练质量评估提供了一种全新的方法。

关键词: 机动识别, 多尺度特征, 深度残差网络, 博弈论, 飞行训练评估

Abstract:

The quality evaluation of flight training is an important part of pilots'daily flight training. Aiming at the problems of low efficiency and strong subjectivity of traditional flight training quality evaluation methods, an automatic evaluation method based on maneuver action recognition is proposed. Firstly, by constructing a multi-scale feature extraction deep residual network (MSDRN), accurate recognition of tactical maneuver actions is achieved, which overcomes the problem of low recognition accuracy in traditional maneuver action recognition methods. Secondly, aiming at different types of tactical maneuver action, the flight quality evaluation index system is constructed. During the specific evaluation process, the evaluation index is automatically selected according to the action recognition results, and the evaluation results are calculated and obtained by the combination weighting method based on the game theory. Thus, the automatic evaluation method of pilot flight training quality based on maneuver action type recognition is constructed. Compared with the classifiers constructed by traditional one-dimensional convolutional neural network (1D-CNN) and residual neural network (ResNet), the established action recognition method has improved recognition accuracy by 16.2% and 3.5% respectively, and reduced the recognition computation time by 23.9% compared with ResNet. The proposed overall automatic evaluation method for the whole flight training quality gets rid of the serious dependence of the pilot on the flight instructor in the training process, which realizes the automatic and rapid evaluation of the flight training quality. The evaluation results are more scientific, providing a new method for the flight training quality evaluation.

Key words: maneuver recognition, multi-scale feature, deep residual network (DRN), the game theory, flight training evaluation

中图分类号:

V323

朱立成, 孙青, 端军红, 庞敏. 基于机动动作类型识别的飞行员飞行训练质量自动评估方法[J]. 系统工程与电子技术, 2023, 45(12): 3932-3940.

Licheng ZHU, Qing SUN, Junhong DUAN, Min PANG. Automatic evaluation method of pilot flight training quality based on maneuver action type recognition[J]. Systems Engineering and Electronics, 2023, 45(12): 3932-3940.

图/表 17

图1

图2

图3

表1

表2

表3

表4

表5

表6

图4

图5

表7

表8

表9

表10

表11

图6

参考文献 31

1	杨俊, 谢寿生. 基于模糊支持向量机的飞机飞行动作识别[J]. 航空学报, 2005, 26 (6): 84- 88.
	YANG J , XIE S S . Fuzzy support vector machines based recognition for aeroplane flight action[J]. Acta Aeronautica et Astronautica Sinica, 2005, 26 (6): 84- 88.
2	TIAN F, ZHANG T, MENG G L, et al. Intelligent recognition of fighter' s maneuver based on fuzzy control algorithm[C]//Proc. of the 4th International Conference on Instrumentation & Measurement, 2014: 548-589.
3	徐西蒙, 杨任农, 于洋, 等. 基于运动分解和H-SVM的空战目标机动识别[J]. 控制与决策, 2020, 35 (5): 1265- 1272. doi: 10.13195/j.kzyjc.2018.1210
	XU X M , YANG R N , YU Y , et al. Target maneuver recognition in air combat based on motion decomposition and H-SVM[J]. Control and Decision, 2020, 35 (5): 1265- 1272. doi: 10.13195/j.kzyjc.2018.1210
4	YANG Z , SUN Z X , PIAO H Y , et al. Online hierarchical re-cognitionn method for target tactical intention in beyond-visual-range air combat[J]. Defence Technology, 2022, 18 (8): 1349- 1361. doi: 10.1016/j.dt.2022.02.001
5	WEI Z L , DING D L , ZHOU H , et al. A flight maneuver re-cognition method based on multi-strategy affine canonical time warping[J]. Applied Soft Computing Journal, 2020, 95, 106527. doi: 10.1016/j.asoc.2020.106527
6	张建业, 李学仁, 倪世宏. 飞行成绩评定及管理系统[J]. 空军工程大学学报(自然科学版), 2001, 2 (1): 70- 73. doi: 10.3969/j.issn.1009-3516.2001.01.020
	ZHANG J Y , LI X R , NI S H . A kind of applied system for assessing and managing flying score[J]. Journal of Air Force Engineering University, 2001, 2 (1): 70- 73. doi: 10.3969/j.issn.1009-3516.2001.01.020
7	LONG Z, XU K J, YIN H, et al. Flight operation quality assessment model based on the fuzzy logic theory[C]//Proc. of the 10th International Conference on Intelligent Systems and Knowledge Engineering, 2016: 99-103.
8	柳忠起, 袁修干, 樊瑜波. 基于BP神经网络的飞行绩效评价模型[J]. 北京航空航天大学学报, 2010, 36 (4): 403- 406. doi: 10.13700/j.bh.1001-5965.2010.04.010
	LIU Z Q , YUAN X G , FAN Y B . Pilot performance evaluation model based on BP neural networks[J]. Journal of Beijing University of Aeronautics and Astronautis, 2010, 36 (4): 403- 406. doi: 10.13700/j.bh.1001-5965.2010.04.010
9	姚裕盛, 徐开俊. 基于BP神经网络的飞行训练品质评估[J]. 航空学报, 2017, 38 (S1): 24- 32.
	YAO Y S , XU K J . Quality assessment of flight training based on BP neural network[J]. Acta Aeronautica et Astronautica Sinica, 2017, 38 (S1): 24- 32.
10	刘浩, 王昊, 孟光磊, 等. 基于动态贝叶斯网络和模糊灰度理论的飞行训练评估[J]. 航空学报, 2021, 42 (8): 250- 261.
	LIU H , WANG H , MENG G L , et al. Flight training evalu-ation based on dynamic Bayesian networkand fuzzy gray theory[J]. Acta Aeronautica et Astronautica Sinica, 2021, 42 (8): 250- 261.
11	李鸿利, 单征, 郭浩然. 基于MDTW的飞行动作识别算法[J]. 计算机工程与应用, 2015, 51 (9): 267- 270.
	LI H L , SHAN Z , GUO H R . Flight action recognition algorithm based on MDTW[J]. Computer Engineering and Applications, 2015, 51 (9): 267- 270.
12	LUGHOFER E . Evolving multi-label fuzzy classifier[J]. Information Sciences, 2022, 597, 1- 23.
13	SUN P , YANG L M . Low-rank supervised and semi-supervised multi-metric learning for classification[J]. Knowledge-based Systems, 2022, 236, 107787.
14	SUN C, SHRIVASTAVA A, SINGH S, et al. Revisiting unreasonable effectiveness of data in deep learning era[C]// Proc. of the IEEE International Conference on Computer Vision, 2017: 843-852.
15	LECUN Y , BOTTOU L , BENGIO Y , et al. Gradient-based learning applied to document recognition[J]. Proceedings of the IEEE, 1998, 86 (11): 2287- 2324.
16	HE K M, ZHANG X Y, REN S Q, et al. Deep residual learning for image recognition[C]//Proc. of the IEEE Conference on Computer Vision and Pattern Recognition, 2016: 770-778.
17	HONG F, LIU C W, GUO L J, et al. Underwater acoustic target recognition with ResNet18 on ShipsEar dataset[C]//Proc. of the IEEE 4th International Conference on Electronics Technology, 2021.
18	ZHANG K , TANG B P , DENG L , et al. A hybrid attention improved ResNet-based fault diagnosis method of wind turbines gearbox[J]. Measurement, 2021, 179 (10): 109491.
19	SONG H, ZHOU Y, JIANG Z Q, et al. ResNet with global and local image features, stacked pooling block, for semantic segmentation[C]//Proc. of the IEEE/CIC International Conference on Communications, 2018. DOI: 10.1109/ICCChina.2018.8641146.
20	WU Z , SHEN C , VAN-DEN-HENGEL A . Wider or deeper: revisiting the ResNet model for visual recognition[J]. Pattern Recognition, 2019, 90, 119- 133.
21	SZEGEDY C, LIU W, JIA Y Q, et al. Going deeper with convolutions[C]//Proc. of the IEEE Conference on Computer Vision and Pattern Recognition, 2015. DOI: 10.1109/CVPR.2015.7298594.
22	张慧敏. 战机自主与协同飞行训练智能化评估方法研究[D]. 沈阳: 沈阳航空航天大学, 2020.
	ZHANG H M. Research on intelligent evaluation method for autonomous and cooperative flight training of warplanes[D]. Shenyang: Shenyang Aerospace University, 2020.
23	ZHENG Y W , GAO L , LI S , et al. A comprehensive evaluation model for full-chain CCUS performance based on the analytic hierarchy process method[J]. Energy, 2022, 239, 122033.
24	SHI Z F, HE W, SHI J, et al. Reliability evaluation of power system with photovoltaic generation based on multi level cross entropy method[C]// Proc. of the International Conference on Renewable Power Generation, 2020.
25	CHEN Z , TIAN K . Optimization of evaluation indicators for driver's traffic literacy: an improved principal component ana-lysis method[J]. Sage Open, 2022, 12 (2): 2158244.
26	郭金玉, 张忠彬, 孙庆云. 层次分析法的研究与应用[J]. 中国安全科学学报, 2008, 18 (5): 148- 153.
	GUO J Y , ZHANG Z B , SUN Q Y . Study and applications of analytic hierarchy process[J]. China Safety Science Journal, 2008, 18 (5): 148- 153.
27	DIAKOULAKI D , MAVROTAS G , PAPAYANNAKIS L . Determining objective weights in multiple criteria problems: the critic method[J]. Computers and Operations Research, 1995, 22 (7): 763- 770.
28	LIU B , HUANG J J , MCBEAN E , et al. Risk assessment of hybrid rain harvesting system and other small drinking water supply systems by game theory and fuzzy logic modeling[J]. The Science of the Total Environment, 2020, 708 (3): 134436.
29	MASTERS D, LUSCHI C. Revisiting small batch training for deep neural networks[EB/OL]. [2023-03-10]. https://arxiv.org/pdf/1804.07612.pdf.
30	LOSHCHILOV I, HUTTER F. Fixing weight decay regularization in Adam[EB/OL]. [2023-03-10]. https://arxiv.org/abs/1711.05101v1.
31	SRIVASTAVA N , HINTON G , KRIZHEVSKY A , et al. Dropout: a simple way to prevent neural networks from overfitting[J]. Journal of Machine Learning Research, 2014, 15 (1): 1929- 1958.

模块名称	输出维度	参数数量
卷积层(1×7@64) 批标准化	[64, 1, 7] [64]	448
卷积层2~11(1×3@64) 批标准化2~11	[64, 64, 3] [64]	12 288
卷积层2~12(1×3@64) 批标准化2~12	[64, 64, 3] [64]	12 288
卷积层2~21(1×3@128) 批标准化2~21	[128, 64, 3] [128]	24 576
卷积层2~22(1×3@128) 批标准化2~22	[128, 128, 3] [128]	49 152
卷积层2~31(1×3@256) 批标准化2~31	[256, 128, 3] [256]	98 304
卷积层2~32(1×3@256) 批标准化2~32 平均池化	[256, 256, 3] [256] [256, 128, 1]	196 608 32 768
全连接层	[6, 195 072]	1.17e+06

机动动作	标签
斤斗	1
斜斤斗	2
半滚倒转	3
半斤斗翻转	4
半滚倒转-斤斗	5
半滚倒转-半斤斗翻转	6

机动动作	标签
斤斗	进入速度, 载荷大小, 顶点速度
斜斤斗	进入速度, 顶点速度, 顶点坡度, 载荷大小
半滚倒转	进入速度, 改出速度
半斤斗翻转	进入速度, 顶点速度
半滚倒转-斤斗	同半滚倒转和斤斗指标
半滚倒转-半斤斗翻转	同半滚倒转和半斤斗翻转指标

机动动作	评估指标标准	优秀	良好	合格	不合格
斤斗	进入速度(840 km/h)	±40	±80	±120	±160
	载荷大小(4.5 g)	±0.2	±0.4	±0.6	±0.8
	顶点速度(300 km/h)	>390	+60	+30	< 300
斜斤斗	顶点坡度(35°)	±3	±5	±7	±10
	进入速度(840 km/h)	±40	±80	±120	±160
	载荷大小(4.5 g)	±0.2	±0.4	±0.6	±0.8
	顶点速度(300 km/h)	>390	+60	+30	< 300
半滚倒转	进入速度(500 km/h)	±20	±40	±60	±80
半滚倒转	改出速度(850 km/h)	±40	±80	±120	±160
半斤斗翻转	进入速度(800 km/h)	±40	±80	±120	±160
半斤斗翻转	顶点速度(350 km/h)	>440	+60	+30	< 350
半滚倒转-斤斗	该组合动作中半滚倒转阶段与半滚倒转机动动作评估标准一致, 斤斗阶段与斤斗机动动作评估标准一致
半滚倒转-半斤斗翻转	该组合动作中半滚倒转阶段与半滚倒转机动动作评估标准一致, 半斤斗翻转阶段与半斤斗翻转机动动作评估标准一致

标度	含义
1	两个因素相比, 同样重要
3	一个因素比另一个因素稍显重要
5	一个因素比另一个因素明显重要
7	一个因素比另一个因素强烈重要
9	一个因素比另一个因素极端重要
2, 4, 6, 8	上述两相邻判断的中值
上列标度的倒数	B与A相比是A与B相比的倒数