士兵和装甲车目标多尺度检测方法

姓名
邮箱
手机号码
标题
留言内容
验证码

doi:10.15918/j.tbit1001-0645.2022.022

bob手机在线登陆机电学院, 北京　100081

基金项目:国防基础科研计划资助项目（JCKY2021602B029）

详细信息

作者简介:
王建中（1963—），男，教授，E-mail:cwjzwang@bit.edu.cn

通讯作者:
王加乐（1996—），男，硕士生，E-mail:jialewbit@163.com

中图分类号:TP399
计量
- 文章访问数:124
- HTML全文浏览量:50
- PDF下载量:24
- 被引次数:0
出版历程
- 收稿日期:2022-06-18
- 录用日期:2022-08-30

Multi-Scale Detection Method for Soldier and Armored Vehicle Objects

School of Mechatronical Engineering, Beijing Institute of Technology, Beijing 100081, China

摘要

摘要:针对士兵和装甲车目标的尺度差异大以及目标距离远近造成的目标多尺度问题，以YOLOv4深度学习算法为基础，提出了一种多尺度目标检测方法. 通过针对性的数据增强方法丰富小目标样本的多样性，对输入图像进行分割预处理以提高网络输入小目标的分辨率，并基于特征金字塔网络实现大、中、小目标的分离检测，最后匹配检测结果并进行NMS处理去除冗余检测框，从而实现多尺度目标检测. 实验结果表明，本文方法在保持大目标检测效果的情况下，中、小目标的平均检测精度分别提升了1.20%和5.54%，有效提高了中、小目标的检测效果.
- 多尺度目标检测/
- 小目标检测/
- 数据增强
Abstract:A multi-scale object detection method was proposed based on YOLOv4 deep learning algorithm to solve the multi-scale problem caused by the huge-scale difference between soldiers and armored vehicles, as well as object distance. The diversity of small object samples was enriched through targeted data augmentation methods input images were segmented to improve the resolution of input small objects of network, the detection results of large, medium and small objects were separated based on the feature pyramid network, and finally the detection results were matched and NMS processing was carried out to remove the redundant detection boxes, so as to achieve multi-scale object detection. The experimental results show that the average mean precision of small and medium objects is improved by 1.20% and 5.54% respectively, while the detection effect of large objects is maintained, which effectively improves the detection effect of small and medium objects.
- mutil-scale object detection/
- small object detection/
- data augmentation

HTML全文

图 1小目标数据增强

Figure 1.Data augmentation for small object

下载: 全尺寸图片幻灯片

图 2YOLOv4网络结构

Figure 2.YOLOv4 network structure

下载: 全尺寸图片幻灯片

图 3多尺度目标检测方法

Figure 3.Multi-scale object detection method

下载: 全尺寸图片幻灯片

图 4图像分割预处理

Figure 4.Pre-processing for image segmentation

下载: 全尺寸图片幻灯片

图 5NMS处理

Figure 5.NMS processing

下载: 全尺寸图片幻灯片

图 6士兵和装甲车数据集样本示例

Figure 6.Examples of soldier and armored vehicle dataset

下载: 全尺寸图片幻灯片

图 7小目标的检测效果对比

Figure 7.Comparison of small object detection results

下载: 全尺寸图片幻灯片

图 8遮挡小目标的检测效果对比

Figure 8.Comparison of occluded small object detection results

下载: 全尺寸图片幻灯片

图 9密集小目标的检测效果对比

Figure 9.Comparison of dense small object detection results

下载: 全尺寸图片幻灯片

表 1平均检测精度

Table 1.Mean detection precision

方法	mAP_L/%	mAP_M/%	mAP_S/%
方法① 基础YOLOv4	96.62	77.26	66.98
方法② 基于小目标数据增强的YOLOv4	96.38(↓0.24)	76.49(↓0.77)	68.78(↑1.80)
方法③ 基于分割检测的YOLOv4	96.90(↑0.28)	78.45(↑1.19)	71.19(↑4.21)
方法④ 本文的多尺度目标检测方法 (方法①+方法②+方法③)	96.45(↓0.17)	78.46(↑1.20)	72.52(↑5.54)

下载: 导出CSV

参考文献 (16)

[1]	周蓓蓓, 刘珏. 智能化技术在精确打击体系中的应用[J]. 空天防御, 2019, 2(3): 77 − 83.doi:10.3969/j.issn.2096-4641.2019.03.013 ZHOU Beibei, LIU Jue. Application of intelligent technology in precision strike system[J]. Air& Space Defense, 2019, 2(3): 77 − 83. (in Chinese)doi:10.3969/j.issn.2096-4641.2019.03.013
[2]	邓淳方. 基于深度学习的多尺度目标检测研究[D]. 杭州: 浙江大学, 2021. DENG Chunfang. Multi-scale object detection based on deep learning[D]. Hangzhou: Zhejiang University, 2021. (in Chinese)
[3]	陈科圻, 朱志亮, 邓小明, 等. 多尺度目标检测的深度学习研究综述[J]. 软件学报, 2021, 32(4): 1201 − 1227.doi:10.13328/j.cnki.jos.006166 CHEN Keqi, ZHU Zhiliang, DENG Xiaoming, et al. Deep learning for multi-scale object detection: a survey[J]. Journal of Software, 2021, 32(4): 1201 − 1227. (in Chinese)doi:10.13328/j.cnki.jos.006166
[4]	SINGH B, DAVIS L. An analysis of scale invariance in object detection - SNIP[C] // Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition. [S. l. ]: IEEE, 2018: 3578-3587.
[5]	SINGH B, NAJIBI M, DAVIS L. SNIPER: efficient multi-scale training[J]. Advances in Neural Information Processing Systems, 2018, 31(15): 9310 − 9321.
[6]	MENG Fanjie, WANG Xinqing, SHAO Faming, et al. Fast-armored target detection based on multi-scale representation and guided anchor[J]. Defence Technology, 2020, 16(4): 922 − 932.doi:10.1016/j.dt.2019.11.009
[7]	LIN T Y, DOLLÁR P, GIRSHICK R, et al. Feature pyramid networks for object detection[J]. Journal of Visual Communication and Image Representation, 2016, 79: 103260.
[8]	LIU W, ANGUELOV D, ERHAN D, et al. SSD: single shot multibox detector[C] // Proceedings of European Conference on Computer Vision. Cham: Springer, 2016: 21-37.
[9]	孙皓泽, 常天庆, 王全东, 等. 一种基于分层多尺度卷积特征提取的坦克装甲目标图像检测方法[J]. 兵工学报, 2017, 38(9): 1681 − 1691.doi:10.3969/j.issn.1000-1093.2017.09.003 SUN Haoze, CHANG Tianqing, WANG Quandong, et al. Image detection method for tank and armored targets based on hierarchical multi-scale convolution feature extraction[J]. Acta Armamentarii, 2017, 38(9): 1681 − 1691. (in Chinese)doi:10.3969/j.issn.1000-1093.2017.09.003
[10]	周治国, 刘开元, 郑翼鹏, 等. 一种基于深度学习的高速无人艇视觉检测实时算法[J]. bob手机在线登陆学报, 2021, 41(7): 758 − 764.doi:10.15918/j.tbit1001-0645.2018.317 ZHOU Zhiguo, LIU Kaiyuan, ZHENG Yipeng, et al. A real-time algorithm for visual detection of high-speed unmanned surface vehicle based on deep learning[J]. Transactions of Beijing Institute of Technology, 2021, 41(7): 758 − 764. (in Chinese)doi:10.15918/j.tbit1001-0645.2018.317
[11]	韩子硕, 王春平, 付强. 基于深层次特征增强网络的SAR图像舰船检测[J]. bob手机在线登陆学报, 2021, 41(9): 1006 − 1014.doi:10.15918/j.tbit1001-0645.2021.004 HAN Zishuo, WANG Chunping, FU Qiang. Ship detection in SAR images based on deep feature enhancement network[J]. Transactions of Beijing Institute of Technology, 2021, 41(9): 1006 − 1014. (in Chinese)doi:10.15918/j.tbit1001-0645.2021.004
[12]	孙皓泽, 常天庆, 张雷, 等. 基于Top-down网络结构的坦克装甲目标检测[J]. 计算机仿真, 2020, 37(3): 18 − 22.doi:10.3969/j.issn.1006-9348.2020.03.006 SUN Haoze, CHANG Tianqing, ZHANG Lei, et al. Tank and armored target detection based on top-down network[J]. Computer Simulation, 2020, 37(3): 18 − 22. (in Chinese)doi:10.3969/j.issn.1006-9348.2020.03.006
[13]	BOCHKOVSKIY A, WANG C Y, LIAO H Y. YOLOv4: optimal speed and accuracy of object detection[EB/OL]. [2020-04-23]. https://arxiv.org/abs/2004.10934.
[14]	李维浩, 姚世明, 李蔚清, 等. 面向AR沙盘异地协同标绘的动作重构技术[J]. bob手机在线登陆学报, 2019, 39(12): 1298 − 1303.doi:10.15918/j.tbit1001-0645.2018.431 LI Weihao, YAO Shiming, LI Yuqing, et al. A motion reconstruction technology for distributed collaborative plotting of AR sand table[J]. Transactions of Beijing Institute of Technology, 2019, 39(12): 1298 − 1303. (in Chinese)doi:10.15918/j.tbit1001-0645.2018.431
[15]	王粉花, 黄超, 赵波, 等. 基于YOLO算法的手势识别[J]. bob手机在线登陆学报, 2020, 40(8): 873 − 879.doi:10.15918/j.tbit1001-0645.2019.030 WANG Fenhua, HUANG Chao, ZHAO Bo, et al. Gesture recognition based on YOLO algorithm[J]. Transactions of Beijing Institute of Technology, 2020, 40(8): 873 − 879. (in Chinese)doi:10.15918/j.tbit1001-0645.2019.030
[16]	吴浩民. 基于行车视频的道路交通标志识别研究与实现[D]. 南京: 南京邮电大学, 2020. WU Haomin. Research and implementation of road traffic sign recongition based on driving video[D]. Nanjing: Nanjing University of Posts and Telecommunications, 2020. (in Chinese)