基于IFPRM-SBLFS深度学习的飞行模式智能识别方法

doi:10.12382/bgxb.2024.0893

摘要/Abstract

摘要：

在实际监测任务中,及时有效地识别飞行模式至关重要。然而,现有的飞行模式识别方法主观性强、模式单一,限制了在复杂情况下的飞行监控能力,在实际应用中有局限性,进而导致模式边界定位不精确、识别精度低。为此提出一种基于敏感边界和长飞行序列的飞行模式智能识别方法(Intelligent Flight Pattern Recognition Method for Sensitive Boundaries and Long Flight Sequences, IFPRM-SBLFS),以对飞行模式进行智能识别。为了更好地探索多模式飞行参数的空间关系,设计自适应图嵌入,针对不同持续时间的飞行模式提出去噪深度多尺度自动编码器,以及用于减轻模型损失的分类加权焦点损失和回归联合时空交集损失。为验证所提方法的优越性,采集多架民用航班的真实参数,涵盖11种飞行模式,通过人工标注构建飞行模式数据集。仿真计算结果表明:新模型能够在连续飞行架次中自动区分不同的飞行模式,并准确提取模式边界,识别准确率达到了99.07%,且无需任何预处理或后处理;新的智能识别方法可以有效提高精确度和敏感边界的飞行模式识别效果。

关键词: 飞行模式, 深度学习, 敏感边界, 长飞行序列, 自动编码器

Abstract:

Timely and effective identification of aircraft flight patterns is crucial in monitoring task. However, the existing flight pattern recognition methods have limitations in practical applications due to strong subjectivity and single pattern, which limits the flight monitoring capability in complex situations, and in turn leads to imprecise pattern boundary positioning and low recognition accuracy. For this reason, a flight pattern intelligent recognition method based on sensitive boundaries and long flight sequences is proposed for the intelligent recognition of flight states. In order to better explore the spatial relationships of multi-modal flight parameters, an adaptive graph embedding is designed. A denoising depth multi-scale autoencoder is proposed for the flight patterns at different durations, as well as the classification-weighted focal point loss and regression-joint spatio-temporal intersection loss for mitigating model loss. In order to verify the superiority of the proposed method, the real parameters of several civil flights covering 11 flight patterns are collected, and a flight state dataset is constructed by manual labelling. The results show that the proposed model is able to automatically distinguish different flight patterns in consecutive flight sorties and accurately extract the mode boundaries without any pre-processing or post-processing, with an identification accuracy of 99.07%. The intelligent recognition method can effectively improve the recognition accuracy and the flight state recognition of sensitive boundaries.

Key words: flight pattern, deep learning, sensitive boundary, long flight sequence, autoencoder

中图分类号:

E926

孙世岩, 李琳, 朱惠民, 石章松, 梁伟阁. 基于IFPRM-SBLFS深度学习的飞行模式智能识别方法[J]. 兵工学报, 2025, 46(5): 240893-.

SUN Shiyan, LI Lin, ZHU Huimin, SHI Zhangsong, LIANG Weige. Intelligent Recognition of Flight Pattern Based on IFPRM-SBLFS Deep Learning[J]. Acta Armamentarii, 2025, 46(5): 240893-.

图/表 6

参考文献 23

[1]	HU C Y, WU J Y, SUN C, et al. Intelligent temporal detection network for boundary-sensitive flight regime recognition[J]. Engineering Applications of Artificial Intelligence, 2023, 126: 106949.
[2]	王志森, 张召悦, 冯朝辉, 等. 终端区飞行轨迹聚类分析及异常轨迹识别[J]. 科学技术与工程, 2022, 29(9):3807-3813.
	WANG Z S, ZHANG Z Y, FENG C H, et al. Cluster analysis and abnormal trajectory identification of flight trajectory in terminal area[J]. Science Technology and Engineering, 2022, 29(9):3807-3813. (in Chinese)
[3]	ZHANG Q L, MOTT J H, JOHNSON M E, et al. Development of a reliable method for general aviation flight phase identification[J]. IEEE Transactions on Intelligent Transportation Systems, 2022, 23(8):11729-11738.
[4]	任济寰, 吴祥, 薄煜明, 等. 基于增强上下文信息长短期记忆网络的弹道轨迹预测[J]. 兵工学报, 2023, 44(2):462-471. doi: 10.12382/bgxb.2021.0489
	REN J H, WU X, BO Y M, et al. Ballistic trajectory prediction based on long-term and short-term memory network with enhanced contextual information[J]. Acta Armamentarii, 2023, 44(2):462-471. (in Chinese)
[5]	SHEN L, WEI Y N, WANG Y Z, et al. AFMF: time series anomaly detection framework with modified forecasting[J]. Knowledge-Based Systems, 2024, 296: 111912.
[6]	LI W H, CHEN Z Y, HE G L. A novel weighted adversarial transfer network for partial domain fault diagnosis of machinery[J]. IEEE Transactions on Industrial Informatics, 2021, 17(3): 1753-1762.
[7]	LI J, WANG Y, ZI Y Y, et al. Whitening-net: a generalized network to diagnose the faults among different machines and conditions[J]. IEEE Transactions on Neural Networks and Learning Systems, 2022, 33(10): 5845-5858.
[8]	WU J Y, ZHAO Z B, SUN C, et al. Learning from class-imbalanced data with a model-agnostic framework for machine intelligent diagnosis[J]. Reliability Engineering & System Safety, 2021, 216:107934.
[9]	LEE D H. Implementation of tactical maneuvers with maneuver libraries[J]. Chinese Journal of Aeronautics, 2020, 33(1):255-270.
[10]	WU J Y, SUN C, ZHANG C, et al. Deep clustering variational network for helicopter regime recognition in HUMS[J]. Aerospace Science and Technology, 2022, 124: 107553.
[11]	FAN Y Q, SHEN G M, XU X H, et al. Flight track pattern recognition based on few labeled data with outliers[J]. Journal of Electronic Imaging, 2021, 30(3):031204.
[12]	ALVAREZMONTOYA J, CARVAJALCASTRILLON A, SIERRA-PEREZ J. In-flight and wireless damage detection in a UAV composite wing using fiber optic sensors and strain field pattern recognition[J]. Mechanical Systems and Signal Processing, 2020, 136: 106526.
[13]	ZHOU S H, ZHOU Y L, XU Z Z, et al. The landing safety prediction model by integrating pattern recognition and Markov chain with flight data[J]. Neural Computing and Applications, 2019, 31(S1): 147-159.
[14]	JIAO Q C, WANG X L, WANG L J, et al. Audio features based ADS-CNN method for flight attitude recognition of quadrotor UAV[J]. Applied Acoustics, 2023, 211: 109540.
[15]	殷之平, 黄勇胜, 田珈玮, 等. 一种基于二维轨迹与时序重要点的飞机机动划分与识别方法[J]. 航空工程进展, 2025, 16(1):45-53.
	YIN Z P, HUANG Y S, TIAN J W, et al. An aircraft manoeuvre classification and identification method based on two-dimensional trajectories and time-series significant points[J]. Advances in Aeronautical Engineering, 2025, 16(1):45-53. (in Chinese)
[16]	张裕禄, 毕红葵, 叶泽浩, 等. 基于随机森林的HRGV滑翔段飞行模式识别[J]. 战术导弹技术, 2020(2): 1-8,21.
	ZHANG Y L, BI H K, YE Z H, et al. Flight state identification of HRGV gliding segment based on random forest[J]. Tactical Missile Technology, 2020(2): 1-8,21. (in Chinese)
[17]	LEE M, CHUNG M, SHIN Y G. Cephalometric landmark detection via global and local encoders and patch-wise attentions[J]. Neurocomputing, 2022, 470: 182-189.
[18]	HE X, ZHOU Y, ZHAO J Q, et al. Swin transformer embedding UNet for remote sensing image semantic segmentation[J]. IEEE Transactions on Geoscience and Remote Sensing, 2022, 60.DOI:10.1109/TGRS.2022.3144165.
[19]	孟浩, 田洋, 孙宇婷, 等. 全局注意力关系网络的小样本船舶识别[J]. 仪器仪表学报, 2021, 42(12):220-227.
	MENG H, TIAN Y, SUN Y T, et al. Small sample ship identification based on global attention relation network[J]. Chinese Journal of Scientific Instrument, 2021, 42(12):220-227. (in Chinese)
[20]	LINT Y, GOYAL P, GIRSHICK R, et al. Focal loss for dense object detection[J]. IEEE Transactions on Pattern Analysis & Machine Intelligence, 2017, 99:2999-3007.
[21]	WEI Z L, DING D L, ZHOU H, et al. A flight maneuver recognition method based on multi-strategy affine canonical time warping[J]. Applied Soft Computing, 2020, 95: 106527.
[22]	LEONI J, ZINNARI F, VILLA E, et al. Flight regimes recognition in actual operating conditions: a functional data analysis approach[J]. Engineering Applications of Artificial Intelligence, 2022, 114: 105016.
[23]	FAN Y Q, SHEN G M, XU X, et al. Flight track pattern recognition based on few labeled data with outliers[J]. Journal of Electronic Imaging, 2021,30: 031204.

输入	层	过滤器in×out	输出尺寸C×L	范围限制
输入	层	过滤器in×out	44×256	范围限制
	分段截断	d=11, s=1, p=5	(44×22)×256
自适应图嵌入	图形卷积1	22×128,O=4	(44×128)×256
	图形卷积2	128×64,O=4	(44×64)×256
	线性层1	2006×256	256×256
	变压器模块1	r=1	256×256
	变压器模块2	r=1	256×256	[0,4)
	变压器模块3	r=1	256×128	[4,8)
变压器编码器	变压器模块4	r=1	256×64	[8,16)
	变压器模块5	r=2	256×32	[16,32)
	变压器模块6	r=2	256×16	[32,64)
	变压器模块7	r=2	256×8	[64,128)
	变压器模块8	r=2	256×4	[128,∞)
	卷积层1	256×256, k=4,s=1,p=1	[256×4,256×8,…,256×256]
飞行模式分类器	卷积层2	256×256,k=4,s=1,p=1	[256×4,256×8,…,256×256]
	卷积层3	256×10,k=4,s=1,p=1	[10×4,10×8,…,10×256]
	卷积层4	256×256,k=4,s=1,p=1	[256×4,256×8,…,256×256]
	卷积层5	256×256,k=4,s=1,p=1	[256×4,256×8,…,256×256]
边界回归器	卷积层6	256×256,k=4,s=1,p=1	[256×4,256×8,…,256×256]
	卷积层7	256×256,k=4,s=1,p=1	[256×4,256×8,…,256×256]
	卷积层8	256×256,k=4,s=1,p=1	[6×4,6×8,…,6×256]

输入	层	过滤器in×out	输出尺寸C×L	范围限制
输入	层	过滤器in×out	44×256	范围限制
	分段截断	d=11, s=1, p=5	(44×22)×256
自适应图嵌入	图形卷积1	22×128,O=4	(44×128)×256
	图形卷积2	128×64,O=4	(44×64)×256
	线性层1	2006×256	256×256
	变压器模块1	r=1	256×256
	变压器模块2	r=1	256×256	[0,4)
	变压器模块3	r=1	256×128	[4,8)
变压器编码器	变压器模块4	r=1	256×64	[8,16)
	变压器模块5	r=2	256×32	[16,32)
	变压器模块6	r=2	256×16	[32,64)
	变压器模块7	r=2	256×8	[64,128)
	变压器模块8	r=2	256×4	[128,∞)
	卷积层1	256×256, k=4,s=1,p=1	[256×4,256×8,…,256×256]
飞行模式分类器	卷积层2	256×256,k=4,s=1,p=1	[256×4,256×8,…,256×256]
	卷积层3	256×10,k=4,s=1,p=1	[10×4,10×8,…,10×256]
	卷积层4	256×256,k=4,s=1,p=1	[256×4,256×8,…,256×256]
	卷积层5	256×256,k=4,s=1,p=1	[256×4,256×8,…,256×256]
边界回归器	卷积层6	256×256,k=4,s=1,p=1	[256×4,256×8,…,256×256]
	卷积层7	256×256,k=4,s=1,p=1	[256×4,256×8,…,256×256]
	卷积层8	256×256,k=4,s=1,p=1	[6×4,6×8,…,6×256]

衡量标准及平均值	MACTW^[21]	FDA+C-Means^[22]	SSGAN^[23]	IFPRM-SBLFS (GSCE loss)	CWFPL	精确度/%	召回率/%	F1分数
TT-Acc	0.6257	0.8279	0.8672	0.9130	0.9637
TIoU	0.5384	0.6965	0.7251	0.7478	0.8401
其他状态	0.8469	0.8993	0.8349	0.9637	0.9852	99.13	96.51	97.80
起飞	0.0000	0.1689	0.0375	0.5879	0.7649	97.24	95.37	96.30
爬升与滚转	0.0000	0.6768	0.2498	0.7638	0.7588	99.45	88.42	93.61
半滚倒转	0.0000	0.4639	0.3665	0.5869	0.7742	98.62	88.94	93.53
懒八字	0.4672	0.5724	0.5347	0.7968	0.8749	99.09	91.66	95.23
滚转	0.5738	0.7235	0.7352	0.8426	0.9120	99.57	90.83	95.00
爬升和转弯	0.4985	0.8840	0.8279	0.8549	0.9436	99.14	91.72	95.29
平飞	0.0000	0.7426	0.6429	0.6642	0.9752	99.32	91.68	95.35
下降转弯	0.5648	0.8019	0.7846	0.7954	0.9472	99.76	90.35	94.82
着陆	0.0000	0.5243	0.6999	0.7921	0.9020	99.33	91.89	95.47
平均值	0.5902	0.6458	0.5714	0.7648	0.8838	99.07	91.74	95.26

衡量标准及平均值	MACTW^[21]	FDA+C-Means^[22]	SSGAN^[23]	IFPRM-SBLFS (GSCE loss)	CWFPL	精确度/%	召回率/%	F1分数
TT-Acc	0.6257	0.8279	0.8672	0.9130	0.9637
TIoU	0.5384	0.6965	0.7251	0.7478	0.8401
其他状态	0.8469	0.8993	0.8349	0.9637	0.9852	99.13	96.51	97.80
起飞	0.0000	0.1689	0.0375	0.5879	0.7649	97.24	95.37	96.30
爬升与滚转	0.0000	0.6768	0.2498	0.7638	0.7588	99.45	88.42	93.61
半滚倒转	0.0000	0.4639	0.3665	0.5869	0.7742	98.62	88.94	93.53
懒八字	0.4672	0.5724	0.5347	0.7968	0.8749	99.09	91.66	95.23
滚转	0.5738	0.7235	0.7352	0.8426	0.9120	99.57	90.83	95.00
爬升和转弯	0.4985	0.8840	0.8279	0.8549	0.9436	99.14	91.72	95.29
平飞	0.0000	0.7426	0.6429	0.6642	0.9752	99.32	91.68	95.35
下降转弯	0.5648	0.8019	0.7846	0.7954	0.9472	99.76	90.35	94.82
着陆	0.0000	0.5243	0.6999	0.7921	0.9020	99.33	91.89	95.47
平均值	0.5902	0.6458	0.5714	0.7648	0.8838	99.07	91.74	95.26

统计指标	数值
总参总量/M	12.37
模型大小/MB	45.23
浮点运算(FLOPs)/M	566.24
每个样本的计算时间/ms	1.72
连续提取的计算时间/ms	968.53