面向战场复杂场景的增强识别方法

doi:10.12305/j.issn.1001-506X.2026.07.05

摘要/Abstract

摘要：

针对战场复杂环境（小目标、烟雾遮挡、过曝/欠曝、小样本）导致目标识别精度低的问题，提出一种基于目标相关性诱导与时空域混合丢弃的增强识别方法，以提升复杂场景下的数据质量与识别性能。算法基于事件流数据，通过在脉冲神经网络（spiking neural network，SNN）的时空域上同时进行相关性传播将目标相关性映射至特征图。首先利用相关性分数对特征加权，从而优化数据表征；随后对多事件流前景进行尺度缩放与无遮挡数据融合，同步融合标签；最终通过二次相关性加权与事件重要性自适应丢帧生成增强数据。基于SNN在4个数据集的目标识别实验均取得最好结果，尤其相比未增强基线，算法在N-Caltech101上准确率提升9.8%，战场环境提升6.14%。结果表明，该算法可有效利用事件时空特性，规避传统方法的目标遮挡、噪声干扰及特征丢失问题，显著提升SNN泛化能力与识别任务鲁棒性，为战场复杂环境下的智能装备提供可靠技术支持。

关键词: 事件数据, 数据增强, 脉冲神经网络, 相关性诱导, 时空域混合丢弃

Abstract:

To address the issue of low target recognition accuracy caused by complex battlefield scenarios （small targets, smoke occlusion, over/under-exposure, few-shot）, an enhanced recognition method based on target correlation induction and spatiotemporal domain mixup drop is proposed, aiming to improve data quality and recognition performance in complex scenarios. The algorithm based on event stream data by propagating target correlations across both spatiotemporal domains in spiking neural network （SNN）, mapping these correlations to feature maps. Firstly, it optimizes data representation through correlation-score-based feature weighting. Subsequently, multi-event stream foregrounds undergo scale transformation and occlusion-free mixup with synchronized label fusion. Finally, augmented data is generated through secondary correlation weighting and adaptive frame-dropping based on event importance. Target recognition experiments conduct with SNN across four datasets achieve the best results. Specifically, the algorithm improves accuracy by 9.8% on N-Caltech101 and 6.14% in battlefield scenarios compared to non-enhanced baselines. Results demonstrate that the algorithm effectively exploits spatiotemporal domain event characteristics, mitigates traditional method of target occlusion, noise interference, and feature loss, significantly enhancing SNN’s generalization capability and recognition task robustness. This provides reliable technical support for intelligent equipment operating in complex battlefield scenarios.

Key words: event data, data augmentation, spiking neural network （SNN）, correlation induction, spatiotemporal domain mixup drop

中图分类号:

TP 181

张霄嵩, 王鼎衡, 李昊飞, 王一朋, 张岳, 刘保荣, 贾世诚. 面向战场复杂场景的增强识别方法[J]. 系统工程与电子技术, 2026, 48(7): 2172-2183.

Xiaosong ZHANG, Dingheng WANG, Haofei LI, Yipeng WANG, Yue ZHANG, Baorong LIU, Shicheng JIA. Enhanced recognition method for complex battlefield scenarios[J]. Systems Engineering and Electronics, 2026, 48(7): 2172-2183.

图/表 16

图1

图2

图3

表1

表2

表3

表4

图4

图5

图6

表5

图7

图8

图9

图10

图11

参考文献 27

1	张亚丽, 冯伟, 全英汇, 等. 基于多源遥感图像多级协同融合的舰船识别算法[J]. 系统工程与电子技术, 2024, 46 (2): 407- 418.
	ZHANG Y L, FENG W, QUAN Y H, et al. Ship recognition algorithm based on multi-level collaborative fusion of multi-source remote sensing image[J]. Systems Engineering and Electronics, 2024, 46 (2): 407- 418.
2	姜杰, 张立民, 刘凯, 等. 基于任务对齐学习的红外船舶目标检测方法[J]. 系统工程与电子技术, 2025, 47 (1): 34- 40.
	JIANG J, ZHANG L M, LIU K, et al. Infrared ship target detection method based on task alignment learning[J]. Systems Engineering and Electronics, 2025, 47 (1): 34- 40.
3	霍永青, 彭启琮. 高动态范围图像及反色调映射算子[J]. 系统工程与电子技术, 2012, 34 (4): 820- 826.
	HUO Y Q, PENG Q Z. High dynamic range images and reverse tone mapping operators[J]. Systems Engineering and Electronics, 2012, 34 (4): 820- 826.
4	GALLEGO G, DELBRUCK T, ORCHARD G M, et al. Event-based vision: a survey[J]. IEEE Trans. on Pattern Analysis and Machine Intelligence, 2020, 44 (1): 154- 180.
5	王宝宇, 李汉棠, 陈希用, 等. 融合渐进空间金字塔的YOLOv8芒果目标检测算法研究[J]. 激光与光电子学进展, 2025, 62 (14): 183- 190.
	WANG B Y, LI H T, CHEN X Y, et al. Research on mango detection algorithm in YOLOv8 with integrated progressive spatial pyramid[J]. Laser & Optoelectronics Progress, 2025, 62 (14): 183- 190.
6	李扬, 李现国, 陈莲, 等. FDLIE-YOLO: 频域增强的端到端低照度图像目标检测方法[J]. 红外与激光工程, 2025, 54 (1): 266- 278.
	LI Y, LI X G, CHEN L, et al. FDLIE-YOLO: frequency domain enhanced end-to-end low-light image target detection method[J]. Infrared and Laser Engineering, 2025, 54 (1): 266- 278.
7	陈红阳, 曾上游, 赵俊博, 等. 基于Retinex改进的低照度图像增强网络[J]. 计算机应用与软件, 2025, 42 (1): 158- 166.
	CHEN H Y, ZENG S Y, ZHAO J B, et al. Improved low-light image enhancement network based on Retinex[J]. Computer Applications and Software, 2025, 42 (1): 158- 166.
8	宋建辉, 刘鑫, 庄爽, 等. 面向无人驾驶的多任务环境感知算法研究[J]. 电子测量与仪器学报, 2025, 39 (1): 122- 132. doi: 10.13382/j.jemi.B2407565
	SONG J H, LIU X, ZHUANG S, et al. Research on multi-task environment perception algorithm for unmanned driving[J]. Journal of Electronic Measurement and Instrumentation, 2025, 39 (1): 122- 132. doi: 10.13382/j.jemi.B2407565
9	伍谷馨, 胡异丁, 杨栋. 基于数据增强的深度学习声学场景分类算法[J]. 现代信息科技, 2024, 8 (23): 76- 81. doi: 10.19850/j.cnki.2096-4706.2024.23.016
	WU G X, HU Y D, YANG D, et al. Deep learning acoustic scene classification algorithm based on data enhancement[J]. Modern Information Technology, 2024, 8 (23): 76- 81. doi: 10.19850/j.cnki.2096-4706.2024.23.016
10	ZHANG H, CISSE M, DAUPHIN Y N, et al. Mixup: beyond empirical risk minimization [EB/OL]. [2025-04-27]. https: //arxiv.org/abs/1710.09412.
11	LI Y, KIM Y, PARK H, et al. Neuromorphic data augmentation for training spiking neural networks[C]//Proc. of the European Conference on Computer Vision, 2022: 631−649.
12	BENDIG K, SCHUSTER R, STRICKER D. Shapeaug: occlusion augmentation for event camera data[EB/OL]. [2025-04-27]. https: //arxiv.org/abs/2404.11718.
13	SHEN G, ZHAO D, ZENG Y. Eventmix: an efficient data augmentation strategy for event-based learning[J]. Information Sciences, 2023, 644, 119170. doi: 10.1016/j.ins.2023.119170
14	YUN S, HAN D, OH S G, et al. Cutmix: regularization strategy to train strong classifiers with localizable features[C]//Proc. of the IEEE/CVF International Conference on Computer Vision, 2019: 6023-6032.
15	GU F Q, SNG W C, HU X K, et al. Eventdrop: data augmentation for event-based learning [EB/OL]. [2025-04-27]. https: //arxiv.org/abs/2106.05836.
16	胡一凡, 李国齐, 吴郁杰, 等. 脉冲神经网络研究进展综述[J]. 控制与决策, 2021, 36 (1): 1- 26. doi: 10.13195/j.kzyjc.2020.1006
	HU Y F, LI G Q, WU Y J, et al. Spiking neural networks: a survey on recent advances and new directions[J]. Control and Decision, 2021, 36 (1): 1- 26. doi: 10.13195/j.kzyjc.2020.1006
17	WU Y J, DENG L, LI G Q, et al. Spatio-temporal backpropagation for training high-performance spiking neural networks[J]. Frontiers in NeuroScience, 2018, 12, 331. doi: 10.3389/fnins.2018.00331
18	SIRONI A, BRAMBILLA M, BOURDIS N, et al. HATS: histograms of averaged time surfaces for robust event-based object classification[C]//Proc. of the IEEE Conference on Computer Vision and Pattern Recognition, 2018: 1731-1740.
19	GARRICK O, AJINKYA J, COHEN G K, et al. Converting static image datasets to spiking neuromorphic datasets using saccades[J]. Frontiers in Neuroscience, 2015, 9, 437.
20	LI H M, LIU H C, JI X Y, et al. Cifar10-DVS: an event-stream dataset for object classification[J]. Frontiers in Neuroscience, 2017, 11, 309. doi: 10.3389/fnins.2017.00309
21	BACH S, BINDER A, MONTANON G, et al. On pixel-wise explanations for non-linear classifier decisions by layer-wise relevance propagation[J]. PLOS ONE, 2015, 10 (7): e0130140. doi: 10.1371/journal.pone.0130140
22	LIU Y H, WANG X J. Spike-frequency adaptation of a generalized leaky integrate-and-fire model neuron[J]. Journal of Computational NeuroScience, 2001, 10 (1): 25- 45. doi: 10.1023/A:1008916026143
23	MONTAVON G, LAPUSCHKIN S, BINDER A, et al. Explaining nonlinear classification decisions with deep taylor decomposition[J]. Pattern Recognition, 2017, 65, 211- 222. doi: 10.1016/j.patcog.2016.11.008
24	LEE J R, KIM S, PARK I, et al. Relevance-cam: your model already knows where to look[C]//Proc. of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, 2021: 14944−14953.
25	KIM Y, PANDA P. Visual explanations from spiking neural networks using inter-spike intervals[J]. Scientific Reports, 2021, 11 (1): 19037. doi: 10.1038/s41598-021-98448-0
26	KINGMA D P, BA J. Adam: a method for stochastic optimization [EB/OL]. [2025-04-27]. https://arxiv.org/abs/1412.6980.
27	ZHANG J Q, DONG B, ZHANG H W, et al. Spiking transformers for event-based single object tracking[C]//Proc. of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, 2022: 8801−8810.

层索引	层类型	参数描述	特征数据尺寸
层索引	层类型	参数描述	N-Caltech101	N-Cars和CIFAR10-DVS
1	卷积	64通道，3×3核	128×128→128×128	48×48→48×48
2	卷积	64通道，3×3核	128×128→128×128	48×48→48×48
3	最大池化	2×2核, 步长2	128×128 → 64×64	48×48 → 24×24
4	卷积	128通道, 3×3核	64×64 → 64×64	24×24 → 24×24
5	卷积	128通道, 3×3核	64×64 → 64×64	24×24 → 24×24
6	最大池化	2×2核, 步长2	64×64 → 32×32	24×24 → 12×12
7	卷积	256通道, 3×3核	32×32 → 32×32	12×12 → 12×12
8	卷积	256通道, 3×3核	32×32 → 32×32	12×12 → 12×12
9	卷积	256通道, 3×3核	32×32 → 32×32	12×12 → 12×12
10	卷积	256通道, 3×3核	32×32 → 32×32	12×12 → 12×12
11	最大池化	2×2核, 步长2	32×32 → 16×16	12×12 → 6×6
12	卷积	512通道, 3×3核	16×16 → 16×16	6×6 → 6×6
13	卷积	512通道, 3×3核	16×16 → 16×16	6×6 → 6×6
14	卷积	512通道, 3×3核	16×16 → 16×16	6×6 → 6×6
15	卷积	512通道, 3×3核	16×16 → 16×16	6×6 → 6×6
16	最大池化	2×2核, 步长2	16×16 → 8×8	6×6 → 3×3
17	卷积	512通道, 3×3核	8×8 → 8×8	3×3 → 3×3
18	卷积	512通道, 3×3核	8×8 → 8×8	3×3 → 3×3
19	卷积	512通道, 3×3核	8×8 → 8×8	3×3 → 3×3
20	卷积	512通道, 3×3核	8×8 → 8×8	3×3 → 3×3
21	最大池化	2×2核, 步长2	8×8 → 4×4	3×3 → 1×1
22	展平	空间维度展平	4×4 → 16	1×1 → 1
23	全连接	输入尺寸×512 → 4096	16×512 → 4096	1×512 → 4096
24	全连接	4096 → 4096	4096 → 4096	4096 → 4096
25	全连接	4096 → 输出类别数	4096 → 101	4096 → 2/10

数据集	增强方式	准确率
	不增强	94.92
	NDA	90.10
N-Cars	Eventmix	96.29
	EventDrop	95.46
	本文方法	98.02
	不增强	75.70
	NDA	83.70
N-Caltech101	Eventmix	79.40
	EventDrop	74.30
	本文方法	85.50
	不增强	78.85
	NDA	79.60
CIFAR10-DVS	Eventmix	81.45
	EventDrop	77.73
	本文方法	82.33

数据集	增强方式	准确率
	加权增强	95.00
N-Cars	加权增强+无遮挡mixup	96.31
	加权增强+自适应drop	97.00
	加权增强	77.30
N-Caltech101	加权增强+无遮挡mixup	83.70
	加权增强+自适应drop	79.60
	加权增强	79.10
CIFAR10-DVS	加权增强+无遮挡mixup	82.25
	加权增强+自适应drop	79.70

数据集	增强方式	准确率
战场环境事件数据集	不增强	80.33
	NDA	81.55
	Eventmix	82.58
	EventDrop	81.92
	本文的方法	86.47

[1]	黄思佳, 宋纯锋, 李璇. 基于可变尺度先验框的声呐图像目标检测[J]. 系统工程与电子技术, 2024, 46(3): 771-778.
[2]	宋婷, 兀泽朝, 高艾, 袁建平. 基于CycleGAN的月表图像数据增强方法[J]. 系统工程与电子技术, 2023, 45(10): 3041-3048.
[3]	王彩云, 吴钇达, 王佳宁, 马璐, 赵焕玥. 基于改进的CNN和数据增强的SAR目标识别[J]. 系统工程与电子技术, 2022, 44(8): 2483-2487.