降质靶标检测算法

doi:10.12382/bgxb.2023.0180

摘要/Abstract

摘要：

装甲车辆动态性能考核中的立靶成像测试环节,靶标检测的准确性与武器装备鉴定及定型的精度息息相关。针对靶标图像对比度低、可辨识度低等降质问题,提出一种基于改进YOLOv5的降质靶标检测算法:使用多分支分组卷积结构配合深度、逐点卷积搭建主干特征提取网络,降低网络参数计算量,提高网络的检测速度;引入表征注意力机制,增强靶标的表征能力;在网络输出层,引入3分支空间特征融合,利用低层特征图的细粒度特征信息与高层特征图丰富的语义信息组合,保留降质靶标图像的细节、边缘语义信息;实验结果表明:在靶标数据集中,所提算法的检测精度mAP达到90.88%,检测速度达到52.74帧/s,能在降质环境下够高效、精准地完成动态性能考核中立靶成像测试环节中的靶标检测部分。

关键词: 靶标, 降质图像, 目标检测, 特征融合, 注意力机制

Abstract:

The target detection accuracy in the imaging test phase of armored vehicle dynamic performance assessment is closely related to the precision of weapon equipment identification and qualification. To address the degradation issues such as low target image contrast and poor discernibility, a degraded target detection algorithm based on improved YOLOv5 is proposed. The proposed algorithm utilizes a multi-branch grouping convolutional structure combined with deep and pointwise convolutions to construct a backbone feature extraction network, thus reducing the computational complexity of network parameters and improving the detection speed. The representation attention mechanism is introduced to enhance the representation capability of the targets. At the network output layer, a three-branch spatial feature fusion is introduced to combine the fine-grained feature information from low-level feature maps and the rich semantic information from high-level feature maps, preserving the details and edge semantic information of degraded target images. Experimental results demonstrate that, in the target dataset, the proposed algorithm achieves a detection accuracy of 90.88% in terms of mean average precision (mAP) and a detection speed of 52.74 fps. It can efficiently and accurately complete the target detection phase in the imaging test of dynamic performance assessment.

Key words: target, degradedimage, target detection, feature fusion, attention mechanism

中图分类号:

TJ811

刘鹏, 熊泽宇, 景文博, 冯萱, 张俊豪, 刘桐伯, 吴雪妮, 夏璇, 万琳琳, 赵海丽. 降质靶标检测算法[J]. 兵工学报, 2024, 45(6): 2065-2075.

LIU Peng, XIONG Zeyu, JING Wenbo, FENG Xuan, ZHANG Junhao, LIU Tongbo, WU Xueni, XIA Xuan, WAN Linlin, ZHAO Haili. Degrad Target Detection Algorithm[J]. Acta Armamentarii, 2024, 45(6): 2065-2075.

图/表 19

图1 单幅图像增强效果

Fig.1 Single image enhancement

图2 YOLOE网络模型结构

Fig.2 YOLOE network model structure

图3 深度逐点卷积结构

Fig.3 Deep point-by-point convolution structure

图4 卷积块结构对比图

Fig.4 Comparison diagram of convolutional block structures

图5 表征注意力机制示意图

Fig.5 Schematic representation of attention mechanism

图6 通道注意力子模块

Fig.6 Channel attention sub-module

图7 空间注意力子模块

Fig.7 Spatial attention sub-module

图8 三分支自适应空间特征融合

Fig.8 Three-branch adaptive spatial feature fusion

图9 EIoU示意图

Fig.9 EIoU diagram

表1 初始参数设置

Table 1 Initial parameter setting

参数	数值
初始学习率设置	0.005
学习率调整算法	余弦退火算法
训练周期	100epoch
梯度下降优化算法	SGD算法
批量处理样本数量	16
IoU阈值	0.5
置信度阈值	0.5

图10 锚框锚框聚类图

Fig.10 Anchor frame clustering

图11 目标检测算法性能对比图

Fig.11 Comparison chart of performances of various target detection algorithms

表2 目标检测算法实验结果对比

Table 2 Test results of various target detection algorithms

序号	目标检测算法	图片分辨率/像素	mAP/%	F₁/%	FPS/(帧·s)
1	FASTER R-CNN	600×600	79.51	69.81	15.16
2	SSD	600×600	64.91	60.91	25.23
3	NanoDet	320×320	70.12	61.82	60.82
4	Cascade R-CNN	640×640	83.48	80.18	20.18
5	YOLOv5	640×640	83.02	80.71	49.12
6	YOLOE	640×640	90.88	88.92	52.74
7	YOLOX	640×640	86.55	86.12	52.60

表3 降质靶标测试结果对比

Table 3 Comparison of degrad target test results

表4 YOLOE模型消融实验对比

Table 4 Results of YOLOE model ablation experiments

模型	改进策略	mAP/%	FPS/ (帧·s^-1)	Parameters/ M
A	YOLOv5	83.02	49.12	7.60
B	A+主干网络改进	82.86	56.25	6.34
C	B+表征注意力机制	85.24	54.68	6.52
D	C+三分支自适应空间特征融合	88.91	52.74	7.05
E	D+损失函数改进	90.88	52.74	7.05

图12 注意力机制效果对比

Fig.12 Comparison of feature fusion maps

图13 干扰情况下注意力机制效果对比

Fig.13 Comparison of the effects of attention mechanisms in the presence of interference

图14 特征融合可视化效果对比

Fig.14 Comparison of feature fusion visualisation effects

图15 损失函数对比图

Fig.15 Comparison of bounding box loss functions

参考文献 20

[1]	穆原子, 丁力军, 陈锋. 武器装备试验靶标建设与发展研究[J]. 国防科技, 2013, 34(3): 40-44.
	MU Y Z, DING L J, CHEN F. Research on construction and development for armament test target[J]. National Defense Science & Technology, 2013, 34(3): 40-44. (in Chinese)
[2]	武治国, 刘伟宁. 抗光线干扰的复杂背景下运动靶标实时跟踪技术[J]. 仪器仪表学报, 2014(增刊1): 34-38.
	WU Z G, LIU W N. Motion target real-time tracking technology of light anti-interference under thecomplex background[J]. Chinese Journal of Scientific Instrument, 2014(S1): 34-38. (in Chinese)
[3]	顾玲嘉. 活动靶标图像处理技术研究[D]. 长春: 长春理工大学, 2005.
	GU L J. Image processing technology research of movable target[D]. Changchun: Changchun University of Science and Technology, 2005. (in Chinese)
[4]	于博文, 吕明. 改进的YOLOv3算法及其在军事目标检测中的应用[J]. 兵工学报, 2022, 43(2): 345-354. doi: 10.3969/j.issn.1000-1093.2022.02.012
	YU B W, LV M. Improved YOLOv3 algorithm and its application to military target detection[J]. Acta Armamentarii, 2022, 43(2): 345-354. (in Chinese)
[5]	杨传栋, 刘桢, 石胜斌. 基于CNN的弹载图像目标检测方法研究[J]. 战术导弹技术, 2019, 196(4):85-92.
	YANG C D, LIU Z, SHI S B. Research on object detection for missile-borne image based on CNN[J]. Tactical Missile Technology, 2019, 196(4):85-92. (in Chinese)
[6]	张欣, 张永强, 何斌, 等. 基于YOLOv4-tiny的遥感图像飞机目标检测技术研究[J]. 光学技术, 2021, 47(3): 344-351.
	ZHANG X, ZHANG Y Q, HE B, et al. Research on remote sensing image aircraft target detection techonlogy based on YOLOv4-tiny[J]. Optical Technology, 2021, 47(3): 344-351. (in Chinese)
[7]	REDMON J, DIVVALA S, GIRSHICK R, et al. You only look once: unified, real-time object detection[C]∥Proceedings of the 2016 IEEE Computer Vision & Pattern Recognition. Las Vegas, NV,US:IEEE, 2016:779-788.
[8]	ZHU X, LYU S, WANG X, et al. TPH-YOLOv5: Improved YOLOv5 based on transformer prediction head for object detection on drone-captured scenarios[C]∥Proceedings of the 2021 IEEE/CVF International Conference on Computer Vision Workshops. Montreal, BC, Canada:IEEE, 2021:2778-2788.
[9]	YUN S D, HAN D Y, CHUN S H, et al. CutMix: regularization strategy to train strong classifiers with localizable features[C]∥Proceedings of 2019 IEEE/CVF International Conference on Computer Vision. Seoul, Korea:IEEE, 2019: 6023-6032.
[10]	HE K M, ZHANG X Y, REN S Q, et al. Deep residual learning for image recognition[C]∥Proceedings of the 2016 IEEE Conference on Computer Vision and Pattern Recognition.Las Vegas, NV, US:IEEE, 2016,770-778
[11]	CHOLLET F. Xception: deep learning with depthwise separable convolutions[C]∥Proceedings of the 2017 IEEE Conference on Computer Vision and Pattern Recognition.Honolulu, HI, US:IEEE, 2017:1800-1807.
[12]	JIE H, LI S, GANG S. Squeeze-and-excitation networks[C]∥Proceedings of the 2018 IEEE/CVF Conference on Computer Vision and Pattern Recognition.Salt Lake City, UT, US:IEEE, 2017:2011-2023.
[13]	WOO S, PARK J C, LEE J Y, et al. CBAM: Convolutional block attention module[C]∥ Proceedings of the 15thEuropean Conference on Computer Vision.Berlin, Germany:Springer-Verlag, 2018: 3-19.
[14]	HE K M, ZHANG X Y, REN S Q, et al. Spatial pyramid pooling in deep convolutional networks for visual recognition[J]. IEEE Transactions on Pattern Analysis and Machine Intelligence, 2015, 37(9): 1904-1916. doi: 10.1109/TPAMI.2015.2389824 pmid: 26353135
[15]	鞠默然, 罗江宁, 王仲博, 等. 融合注意力机制的多尺度目标检测算法[J]. 光学学报, 2020, 40(13): 132-140.
	JU M R, LUO J N, WANG Z B, et al. Multi-scale target detection algorithm based on attention mechanism[J]. Acta Optica Sinica, 2020, 40(13):132-140. (in Chinese)
[16]	ARTHUR D, VASSILVITSKII S. K-means++: the advantages of careful seeding[C]∥Proceedings of the Eighteenth Annual ACM-SIAM Symposium on Discrete Algorithms. New Orleans, LA, US: Society for Industrial and Applied Mathematics, 2007: 1027-1035.
[17]	LIU W, ANGUELOV D, ERHAN D, et al. SSD: Singleshot multibox detector[C]∥ Proceedings of the 3rd European Conference on Computer Vision.Amsterdam, The Netherlands: Springer Cham, 2016, 99(5):21-37.
[18]	REN S, HE K, GIRSHICK R, et al. Faster R-CNN: towards real-time object detection with region proposal networks: NIPS[J]. IEEE Transactions on Pattern Analysis & Machine Intelligence, 2017, 39(6):1137-1149.
[19]	CAI Z W, VASCONCELOS N. Cascade R-CNN: delving into high quality object detection[C]∥Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition. Salt Lake City, UT, US: IEEE, 2018:6154-6162.
[20]	SELVARAJU R R, COGSWELL M, DAS A, et al. Grad-CAM: visual explanations from deep networks via gradient-based localization[C]∥Proceedings of the IEEE International Conference on Computer Vision.Venice, Italy:IEEE,2017:618-626.