无人机航拍军事车辆实时检测及定位算法

doi:10.12382/bgxb.2024.0518

摘要/Abstract

摘要：

针对地面高价值军事车辆难以实时观测定位的问题,提出无人机航拍军事车辆实时检测及定位算法。创建真实作战场景下的多类型多尺度航拍军事车辆目标检测数据集Armed_vehicle;在轻量级神经网络模型YOLOX-Tiny中引入大卷积核注意力模块,采用边框回归函数,达到85.82%的检测准确率,且在检测速率上具有优势;创新性地提出一种基于无人机可见光图像的单目视觉定位算法,在百米飞行高度上的目标平均定位误差为3.69m,表明所提算法能准确地获取地面目标的地理位置,具有良好的综合性能与应用前景。

关键词: 无人机航拍, 目标检测, 目标定位, 军事车辆

Abstract:

With the goal of addressing the challenges of real-time observation and localization of high-value military vehicles on the ground, a real-time algorithm for the detection and localization of military vehicles in aerial photography is proposed. An Armed_vehicle dataset for the detection of multi-type and multi-scale military vehicles in an actual combat environment in aerial photography is established. The detection accuracy reaches 85.82% and the detection efficiency is higher by introducing a large kernel attention (LKA)module into the lightweight neural network model YOLOX-Tiny and using the SIoU edge regression function. A monocular visual localization algorithm based on the visible light images from the unmanned aerial vehicles (UAVs) is proposed. The average target localization error is 3.69m at a flight altitude of 100 meters. It indicates that the proposed algorithm can accurately obtain the geographical location of ground targets and has good comprehensive performance and application prospects.

Key words: drone aerial photography, target detection, target localization, military vehicle

中图分类号:

TP391.4

姚雨, 宋春林, 邵江琦. 无人机航拍军事车辆实时检测及定位算法[J]. 兵工学报, 2024, 45(S1): 354-360.

YAO Yu, SONG Chunlin, SHAO Jiangqi. Real-time Detection and Localization Algorithm for Military Vehicles in Drone Aerial Photography[J]. Acta Armamentarii, 2024, 45(S1): 354-360.

图/表 14

图1 总体架构图

Fig.1 Overall architecture diagram

图2 YOLOX-Tiny模型结构

Fig.2 YOLOX-Tiny model structure

图3 改进后的YOLOX-Tiny主干网络

Fig.3 Improved YOLOX Tiny backbone network

图4 无人机相机视场

Fig.4 Drone camera field of view

图5 航拍图像像素点的空间几何关系

Fig.5 Spatial geometric relationship of pixels in aerial image

图6 Armed_vehicle数据集

Fig.6 Armed_vehicle dataset

表1 Armed_vehicle数据集基本信息

Table 1 Basic information of the Armed_vehicle dataset

数据集属性名称	数据集属性描述
数据集名称	航拍军事车辆数据集Armed_vehicle
图像分辨率	608×608
图像数量	1 884
军事车辆数量	3 408
图像类型	RGB、灰度
军事车辆种类	坦克、装甲车、运输车、野战皮卡、炮弹发射车
军事车辆状态	正常、损伤、损毁
作战场景	城市、山地、河岸、荒漠、草原

图7 视角测量参照

Fig.7 Perspective measurement reference

图8 拍图像标注

Fig.8 Aerial image annotation

图9 改进前后训练损失和平均精准度变化图

Fig.9 Changes in training loss and average accuracy before and after improvement

图10 改进前后检测效果对比

Fig.10 Comparison of detection effects before and after improvement

表2 本文算法和其他目标检测算法对比结果

Table 2 Comparison of the detected results of the proposed algorithm and other detection algorithms

算法	mAP/%	FPS	Params/10⁶	Weights/MB
SSD	87.31	11	24.7	188
CenterNet	89.67	12	32.7	124
YOLOv3_MobileNetV3	81.39	26	29.1	88.6
YOLOv5-S	82.28	20	7.3	28.1
YOLOX-S	85.66	19	9.0	34.2
YOLOX-Tiny	82.59	21	5.1	19.3
本文算法	85.82	21	5.1	19.3

图11 单目视觉定位算法定位效果

Fig.11 Positioning effect of monocular visual localization algorithm

表3 目标定位算法精度验证

Table 3 Accuracy verification of target localization algorithm

参数	目标车辆											总和
参数	A	B	C	D	E	F	G	H	I	J	K	总和
验证数目	51	45	27	66	79	25	88	69	39	43	33	565
最大误差/m	7.19	8.65	6.23	6.76	8.60	8.51	9.76	11.75	8.03	10.59	9.06	11.75
平均误差/m	3.25	3.44	3.77	3.55	2.99	2.84	3.88	4.49	4.05	4.32	3.79	3.69

参考文献 20

[1]	朱家辉, 苏维均, 于重重, 等. 基于RGB图像的坦克损伤目标三维检测研究与应用[J]. 火力与指挥控制, 2022, 47(4):169-175.
	ZHU J H, SHU W J, YU C C, et al. Research and application of tank damage target 3D detection based on RGB image[J]. Fire Control and Command Control, 2022, 47(4):169-175. (in Chinese)
[2]	童雪东. 面向无人作战车辆的图像敏感目标检测技术的研究与实现[D]. 南京: 南京理工大学, 2019.
	TONG X D. Research and implementation of image sensitive target detection technology for unmanned combat vehicles[D]. Nanjing: Nanjing University of Science and Technology, 2019. (in Chinese)
[3]	褚文杰. 基于YOLOv5的坦克装甲车辆目标检测关键技术的研究[D]. 北京: 北京交通大学, 2022.
	CHU W.J. Research on key technologies for tank armored vehicle target detection based on YOLOv5[D]. Beijing: Beijing Jiaotong University, 2022. (in Chinese)
[4]	DU X L, SONG L Q, LÜ Y N, et al. A lightweight military target detection algorithm based on improved YOLOv5[J]. Electronics, 2022, 1120: 3263.
[5]	GE Z, LIU S T, WANG F, et al. YOLOX: exceeding YOLO series in 2021: arXiv:2107.08430[R]. Ithaca,NY, US: Cornell University, 2021:2107.08430.
[6]	GUO M H. Visual attention network[J]. Computational Visual Media, 2023, 9(4): 733-752.
[7]	GEVORGYAN Z SIoU loss: more powerful learning for bounding box regression:arXiv:2205.12740[R].Ithaca,NY,US: Cornell University, 2022:2205.12740.
[8]	ZHAO F Y, ZHANG J W, ZHANG G Q. FFEDet:fine-grained feature enhancement for small object detection[J]. Remote Sensing, 2024, 16(11):143-146.
[9]	LÜ Z L, AN W. HDR-YOLO:adaptive object detection in haze, dark, and rain scenes based on YOLO[J]. International Journal of Pattern Recognition and Artificial Intelligence, 2024, 38(5):278-284.
[10]	ZENG Y Y, ZHOU Z H, YU Y. Study of YOLOX target detection method based on stand-alone self-attention[J]. Academic Journal of Computing & Information Science, 2022, 5(12):572-580.
[11]	SUN Y, ZHENG W X, DU X, et al. Underwater small target detection based on YOLOX combined with mobileViT and double coordinate attention[J]. Journal of Marine Science and Engineering, 2023, 11(6):1129-1133.
[12]	YANG M D, YUAN W H, XU G J. YOLOX target detection model can identify and classify several types of tea buds with similar characteristics[J]. Scientific Reports, 2024, 14(1): 2855-2855. doi: 10.1038/s41598-024-53498-y pmid: 38310143
[13]	ZHANG S Y. Fusion network for small target detection based on YOLO and attention mechanism[J]. Optoelectronics Letters, 2024, 20(6):372-378.
[14]	LI S, TAO T, ZHANG Y, et al. YOLOv7-CS: a YOLO v7-based model for lightweight bayberry target detection count[J]. Agronomy, 2023, 13(12):399-405.
[15]	TIAN Y N, WANG S H, LI E, et al. MD-YOLO:multi-scale dense YOLO for small target pest detection[J]. Computers and Electronics in Agriculture, 2023, 13(2):89-94.
[16]	CHANG Y L, LI D, GAO Y L, et al. An improved YOLO model for UAV fuzzy small target image ietection[J]. Applied Sciences, 2023, 13(9):632-639.
[17]	ZHOU Z H, YU X G, WANG X K. Object detection in drone video based on recurrent motion attention[J]. Pattern Recognition Letters, 2024,18356-63.
[18]	ZHANG H L. AC-YOLOv5:an improved algorithm for small target face detection[J]. EURASIP Journal on Image and Video Processing, 2024, 2024(1):327-331.
[19]	TOIVOLA J, MOILANEN S, TERVOKOSKI J, et al. Remote intelligent perception system for multi-object detection[J]. Frontiers in Neurorobotics, 2024, 18(6):1398703-1398703.
[20]	DUAN K W, BAI S, XIE L X, et al. CenterNet++ for object detection[J]. IEEE Transactions on Pattern Analysis and Machine Intelligence, 2024, 46(5):3509-3521.