-
在国外,基于图像的自动目标识别技术已从早期对空目标的武器系统逐步应用到反舰和对地大型固定目标的武器系统中。在ATR技术实用化之前,美国SLAM采用人在回路方式。AGM-84 H/K(SLAM-ER)是世界上第一种进入实战应用的ATR导弹系统,SLAM和SLAM-ER是ATR技术实用化的分水岭[3]。SLAM-ER采用模板匹配算法将红外导引头实时获取图像与任务规划时存储的基准图进行比较,从场景中自动捕获目标,为驾驶员提供实时目标引导。美国的LRASM反舰导弹采用主动/被动雷达和红外成像制导体制,如图1所示,首先利用被动射频传感器的数据识别并跟踪舰船目标,然后通过弹载雷达接收的回波信号与存储的目标三维模型进行匹配识别敌方舰艇,在末段,综合雷达和红外图像数据提取目标要害点进行攻击,是自动目标识别在多模复合制导技术的典型应用[4-5]。针对地面大型固定目标,如美国的SLAM-ER空地导弹、JASSM空地导弹,英国的风暴阴影导弹,德国的KEPD-350空地导弹等,射程都超过了300 km,采用基于模板匹配技术,实现对被攻击目标的自动识别和捕获,获得10 m以内的CEP (Circular Error Probable)命中精度。在针对地面移动目标的应用中,美国GBU-53/B小直径炸弹Ⅱ (SDBⅡ)验证了从诱饵中识别出目标的能力,实现了运动目标的自动检测、捕获和跟踪,达到了3 m CEP的精度水平,为战术飞机打击移动/可重新定位目标和静止目标提供了新技术、新手段[6-8]。SPICE-250炸弹最大射程超过100 km,采用图像末制导,集成了移动目标探测和预打击自动识别能力,飞行员选择要攻击的目标分配给某一炸弹,并向目标附近发射,惯导系统进行初始制导,在接近目标区域,炸弹使用ATR模式识别目标,自主截获锁定并追踪选定目标[9],命中精度CEP小于3 m,如图2所示。在国内的防空、空空、反舰和对地固定目标打击等导弹或炸弹的图像末制导应用中,已经从完全人在回路逐渐发展到了发射前锁定、发射后不管模式,实现了基于模板匹配目标识别,自动目标捕获跟踪功能,具备了防区外远程精确打击能力。典型的ATR图像末制导应用如表1所示。
图 1 采用自动目标识别技术的LRASM反舰导弹
Figure 1. LRASM anti-ship missile with automatic target recognition technology
表 1 国外自动目标识别技术应用
Table 1. Application of automatic target recognition technology abroad
Missile model Nation Main features AGM-84 H/K SLAM-ER U.S. Air-to-surface missiles, complex ground, sea and sea-sky scenes, ground fixed、moving targets and ship targets, infrared imaging guidance, recognition technology based on template matching,
moving target detection technologyAGM-158 series JASSM & LRASM U.S. Air-to-surface missiles, complex ground, sea and sea-sky scenes, infrared imaging guidance, fixed or moving targets on land or sea, 3D model matching recognition technology,
feature-based recognition technology, and identification of key parts of targetsStorm Shadow U.K. Air-to-ground missiles, complex ground scenes, ground fixed targets, infrared imaging guidance,
based on template matching recognition technologySCALP France KEPD-350 Germany SPICE-250 Israel Air-to-surface bombs, complex ground scenes, fixed targets, time-sensitive targets, moving targets, with automatic target recognition and moving target detection functions, photoelectric imaging guidance, scene matching technology, deep learning technology, can be loaded with more than
100 target information, after launch updatable target
Application of automatic target recognition in image terminal guidance
-
摘要: 实现图像末制导导弹发射后不管和远程精确打击,自动目标识别的工程化应用是关键技术。概述了国内外精确制导武器自动目标识别的发展历程、识别方法、技术水平和应用效果等现状,分析了基于目标特征和模板匹配的识别方法与应用场景,指出了两类经工程化验证有效的自动目标识别方法,梳理了任务规划、主要执行内容、规划质量对不同识别方法的影响等自动目标识别流程。为了适应未来精确制导武器智能化发展需求,深度学习识别技术工程化应用成为了新趋势,针对解决好深度学习算法效率与应用精度的平衡问题,重点分析了网络剪枝、权值量化、低秩近似和知识蒸馏等实时加速推理关键技术;针对网络模型训练,提出了有效解决训练样本不足或军事目标样本获取困难等问题的思路。随着多波段、多模复合制导技术的广泛应用,信息融合为目标识别的工程化应用提供了新技术途径。如何适应各种复杂场景和人工主动干扰是图像末制导面临的重大挑战,阐述了在干扰条件下目标识别鲁棒性,是自动目标识别技术在图像末制导应用中需要迫切解决的工程化问题。Abstract: The engineering application of automatic target recognition is the key technology to realize the long-range and precise strike after the image terminal-guided missile is launched. The development history, identification method, technical level and application effect of automatic target recognition of precision-guided weapons at home and abroad are summarized. The recognition methods and application scenes based on target features and template matching are analysed, and two types of engineering verification effective methods are identified. The automatic target recognition method combines the automatic target recognition process, such as task planning, main execution content, and the impact of planning quality on different recognition methods. To meet the needs of intelligent development of precision guided weapons in the future, the engineering application of deep learning recognition technology has become a new trend. To solve the balance problem between the efficiency and application accuracy of deep learning algorithms, this paper focuses on the analysis of network pruning, weight quantization, and low rank. The key technologies of real-time acceleration inference such as approximation and knowledge distillation; for network model training, ideas for effectively solving problems such as insufficient training samples or difficulty in obtaining military target samples are proposed. With the wide application of multiband and multimode composite guidance technology, information fusion provides a new technical approach for the engineering application of target recognition. How to adapt to various complex scenes and artificial active interference is a major challenge for image terminal guidance. The robustness of target recognition under interference conditions is expounded, which is an engineering problem that needs to be urgently solved in the application of automatic target recognition technology in image terminal guidance.
-
表 1 国外自动目标识别技术应用
Table 1. Application of automatic target recognition technology abroad
Missile model Nation Main features AGM-84 H/K SLAM-ER U.S. Air-to-surface missiles, complex ground, sea and sea-sky scenes, ground fixed、moving targets and ship targets, infrared imaging guidance, recognition technology based on template matching,
moving target detection technologyAGM-158 series JASSM & LRASM U.S. Air-to-surface missiles, complex ground, sea and sea-sky scenes, infrared imaging guidance, fixed or moving targets on land or sea, 3D model matching recognition technology,
feature-based recognition technology, and identification of key parts of targetsStorm Shadow U.K. Air-to-ground missiles, complex ground scenes, ground fixed targets, infrared imaging guidance,
based on template matching recognition technologySCALP France KEPD-350 Germany SPICE-250 Israel Air-to-surface bombs, complex ground scenes, fixed targets, time-sensitive targets, moving targets, with automatic target recognition and moving target detection functions, photoelectric imaging guidance, scene matching technology, deep learning technology, can be loaded with more than
100 target information, after launch updatable target -
[1] 王颖, 唐倩, 马菁汀, 等. 2020年精确制导武器制导技术发展综述[J]. 飞航导弹, 2021(1): 31-38. doi: 10.16338/j.issn.1009-1319.20200849 [2] 赵丹辉, 王俊敏, 张宏宇, 等. 精确制导武器导航与末制导技术发展综述[J]. 飞航导弹, 2021(2): 64-70. doi: 10.16338/j.issn.1009-1319.20200157 [3] Zhang Dongqing, Zhang Chunxue, Wen Suli. Research on the application of automatic target recognition on missile [J]. Tactical Missile Technology, 2010(5): 1-6. (in Chinese) [4] Fan Jinxiang, Liu Jia. Challenges and thinking for the precision guidance ATR intelligentization [J]. Aero Weaponry, 2019, 26(1): 30-38. (in Chinese) [5] Meng Bo. Analysis and consideration of American long range anti-ship missile [J/OL]. Command Control & Simulation, (2022-02-24) [2022-06-23]. https://kns.cnki.net/kcms/detail/32.1759.TJ.20220223.1334.009.html. (in Chinese) [6] Li Jiatai, Chen Guoliang, Ma Shanbin, et al. Development of light photoelectric guided missile [J]. Air & Space Defense, 2021, 4(4): 30-38. (in Chinese) [7] Yin Ximei, Kang Yanqing. Development trend of UAV-borne precision guided bomb technology [J]. Ordnance Industry Automation, 2021, 40(9): 92-96. (in Chinese) [8] 中国航天科工集团第三研究院三一〇所. 精确制导武器领域科技发展报告[M]. 北京: 国防工业出版社, 2018. [9] Fan Jinxiang, Chen Jinghua. New development of american airborne weapons [J]. Aero Weaponry, 2020, 27(5): 13-22. (in Chinese) [10] Liu Yunfeng, Yang Xiaogang, Qi Naixin, et al. Forward-looking ATR model based on template matching [J]. Ordnance Industry Automation, 2016, 35(7): 4-6. (in Chinese) [11] Xie Chunsi, Liu Zhiying, Sang Yu. Target recognition model of ship-to-land missile based on feature matching [J]. Systems Engineering and Electronics, 2021, 43(8): 2244-2253. (in Chinese) [12] Ma Xiaoping, Zhao Liangyu. An overview of infrared seeker key technologies at home and abroad [J]. Aero Weaponry, 2018(3): 3-10. (in Chinese) [13] Liu Shijian, Jin Lu. A survey on algorithms for automatic target recognition [J]. Electronics Optics & Control, 2016, 23(10): 1-7. (in Chinese) [14] Sun Fangwei, Li Chengyang, Xie Yongqiang, et al. Review of deep learning applied to occluded object detection [J/OL]. Journal of Frontiers of Computer Science and Technology, (2022-03-24) [2022-06-23]. https://kns.cnki.net/kcms/detail/11.5602.TP.20220323.2003.002.html. (in Chinese) [15] Liu Yang, Xie Yongqiang, Li Zhongbo, et al. Research progress of deep learning target detection algorithm in security field [J]. Communications Technology, 2021, 54(9): 2063-2073. (in Chinese) [16] Zhao Xiaodong, Zhang Xunying, Che Jun, et al. Research on optimization algorithm of terminal guidance target recognition for precision-guided weapons [J]. Aeronautical Science & Technology, 2022, 33(1): 126-134. (in Chinese) [17] Li Kequan, Chen Yan, Liu Jiachen, et al. Survey of deep learning-based object detection algorithms [J]. Computer Engineering, 2022, 48(7): 0062725. (in Chinese) [18] 中国航天科工集团第三研究院三一〇所. 自主系统与人工智能领域科技发展报告[M]. 北京: 国防工业出版社, 2018. [19] Wang Jianzhong, Wang Hongfeng, Liu Hongyang, et al. A deep learning detection method for soldier target based on few samples [J]. Transactions of Beijing Institute of Technology, 2021, 41(6): 629-635. (in Chinese) [20] Huang Pan, Yang Xiaogang, Lu Ruitao, et al. Data augmentation method of infrared ship target based on spatial association [J]. Infrared and Laser Engineering, 2021, 50(12): 20210281. (in Chinese) doi: 10.3788/IRLA20210281 [21] Chen Xianzhi, Luo Zhenbao, Yang Xu, et al. Design of information processing system for photoelectric seeker [J]. Infrared and Laser Engineering, 2020, 49(9): 20200312. (in Chinese) doi: 10.3788/IRLA20200312 [22] Bahy R M. An efficient enhanced automatic target recognition approach [J]. Journal of Physics: Conference Series, 2021, 1721(1): 012057. doi: 10.1088/1742-6596/1721/1/012057 [23] Li Liangfu, Chen Weidong, Gao Qiang, et al. Deep learning-based intelligent target recognition technology for electro-optical system [J/OL]. Acta Armamentarii, (2022-03-12) [2022-06-23]. https://kns.cnki.net/kcms/detail/11.2176.TJ.20220310.1821.026.html. (in Chinese) [24] Shen Ying, Huang Chunhong, Huang Feng, et al. Research progress of infrared and visible image fusion technology [J]. Infrared and Laser Engineering, 2021, 50(9): 20200467. (in Chinese) doi: 10.3788/IRLA20200467 [25] Chen Yang, Wang Shifeng, Du Kaiyue, et al. A Image Pixel Level Fusion Based on Fast Guided Filter [J]. Journal of Changchun University of Science and Technology (Natural Science Edition), 2018, 41(6): 11-15. (in Chinese) [26] Xu Yunfei, Zhang Duzhou, Wang Li, et al. Lightweight feature fusion network design for local feature recognition of non-cooperative target [J]. Infrared and Laser Engineering, 2020, 49(7): 20200170. (in Chinese) doi: 10.3788/IRLA20200170 [27] Wang Hongmei, Wang Xiaoge, Wang Xiaoyan. Research on target detection under complex background based on deep learning [J/OL]. Control and Decision, (2021-09-01) [2022-06-23]. https://doi.org/10.13195/j.kzyjc.2021.0686. (in Chinese) [28] Niu Deqing, Wu Youli, Xu Yang, et al. Quantificagion modeling for environmental complexity under point source infrared decoy interference [J]. Infrared and Laser Engineering, 2020, 49(2): 20204003. (in Chinese) [29] Lu Xiao, Liang Xiaogeng, Jia Xiaohong. Study on intelligent counter-countermeasures of infrared imaging air-to-air missiles [J]. Infrared and Laser Engineering, 2021, 50(4): 20200240. (in Chinese) doi: 10.3788/IRLA20200240 [30] Li Chuanliang, Li Taorui, Wang Chaozhe. Influence of parameters of decoy on jamming effectiveness of surface source infrared decoy [J]. Infrared and Laser Engineering, 2021, 50(8): 20210006. (in Chinese) doi: 10.3788/IRLA20210006 [31] Chen Faling, Ding Qinghai, Luo Haibo, et al. Anti-occlusion real time target tracking algorithm employing spatio-temporal context [J]. Infrared and Laser Engineering, 2021, 50(1): 20200105. (in Chinese) doi: 10.3788/IRLA20200105