一种改进YOLOv5算法的伪装目标检测方法

彭锐晖; 赖杰; 孙殿星; 李莽; 颜如玉; 李雪

doi:10.11999/JEIT231170

一种改进YOLOv5算法的伪装目标检测方法

doi: 10.11999/JEIT231170

彭锐晖^{1, 2},
赖杰^2, ,,
孙殿星^{2, 3},
李莽²,
颜如玉²,
李雪²

1.
哈尔滨工程大学信息与通信工程学院哈尔滨 150001
2.
哈尔滨工程大学青岛创新发展基地青岛 266000
3.
海军航空大学信息融合研究所烟台 264001

基金项目: 航天科技集团稳定支持项目(ZY0110020009)，国防科技重点实验室基金项目(2023-JCJQ-LB-016)

详细信息

作者简介:
彭锐晖：男，博士，副教授，研究方向为信息感知及应用、电磁隐身材料与目标特性

赖杰：男，硕士生，研究方向为伪装目标检测、多源信息融合

孙殿星：男，博士，副教授，研究方向为信号与数据处理、信息融合

李莽：男，硕士生，研究方向为钙钛矿复合吸波材料

颜如玉：女，硕士生，研究方向为雷达隐身设计

李雪：女，硕士生，研究方向为深度学习、非显著性目标检测

通讯作者:
赖杰　laijie@hrbeu.edu.cn

中图分类号: TN911.73
计量
- 文章访问数: 81
- HTML全文浏览量: 47
- PDF下载量: 19
- 被引次数: 0
出版历程
- 收稿日期: 2023-10-30
- 修回日期: 2024-03-24
- 网络出版日期: 2024-04-07

A Camouflaged Target Detection Method with Improved YOLOv5 Algorithm

PENG Ruihui^{1, 2},
LAI Jie^{2
, ,},
SUN Dianxing^{2, 3},
LI Mang²,
YANG Ruyu²,
LI Xue²

1.
College of Information and Communication Engineering, Harbin Engineering University, Harbin 150001, China
2.
Qingdao Innovation and Development Center, Harbin Engineering University, Qingdao 266000, china
3.
Insitute of Information Fusion, Naval Aeronautical University, Yantai 264001, China

Funds: China Aerospace Science and Technology Corporation Stabilization Support Project (ZY0110020009), The Defense Science and Technology Key Laboratory Fund Project (2023-JCJQ-LB-016)

摘要

摘要: 为了深入挖掘伪装目标特征信息含量、充分发挥目标检测算法潜能，解决伪装目标检测精度低、漏检率高等问题，该文提出一种多模态图像特征级融合的伪装目标检测算法(CAFM-YOLOv5)。首先，构建伪装目标多波谱数据集用于多模态图像融合方法性能验证；其次，构建双流卷积通道用于可见光和红外图像特征提取；最后，基于通道注意力机制和空间注意力机制提出一种交叉注意力融合模块，以实现两种不同特征有效融合。实验结果表明，模型的检测精度达到96.4%、识别概率88.1%，优于YOLOv5参考网络；同时，在与YOLOv8等单模态检测算法、SLBAF-Net等多模态检测算法比较过程中，该算法在检测精度等指标上也体现出巨大优势。可见该方法对于战场军事目标检测具有实际应用价值，能够有效提升战场态势信息感知能力。
- 伪装目标检测 /
- 多波谱数据集 /
- 注意力机制 /
- 可见光图像 /
- 红外图像
Abstract: To comprehensively explore the information content of camouflaged target features, leverage the potential of target detection algorithms, and address issues such as low camouflage target detection accuracy and high false positive rates, a camouflage target detection algorithm named CAFM-YOLOv5 (Cross Attention Fusion Module Based on YOLOv5) is proposed. Firstly, a camouflaged target multispectral dataset is constructed for the performance validation of the multimodal image fusion method; secondly, a dual-stream convolution channel is constructed for visible and infrared image feature extraction; and finally, a cross-attention fusion module is proposed based on the channel-attention mechanism and spatial-attention mechanism in order to realise the effective fusion of two different features.Experimental results demonstrate that the model achieves a detection accuracy of 96.4% and a recognition probability of 88.1%, surpassing the YOLOv5 baseline network. Moreover, when compared with unimodal detection algorithms like YOLOv8 and multimodal detection algorithms such as SLBAF-Net, our algorithm exhibits superior performance in detection accuracy metrics. These findings highlight the practical value of our method for military target detection on the battlefield, enhancing situational awareness capabilities significantly.
- Camouflaged target detection /
- Multispectral datasets /
- Attention mechanisms /
- Visible images /
- Infrared images

HTML全文

图 1 YOLOv5算法网络结构图

下载: 全尺寸图片幻灯片

图 2 CAFM-YOLOv5网络结构图

下载: 全尺寸图片幻灯片

图 3 CAFM模块结构图

下载: 全尺寸图片幻灯片

图 4 通道注意力机制结构图

下载: 全尺寸图片幻灯片

图 5 空间注意力机制结构图

下载: 全尺寸图片幻灯片

图 6 多波谱伪装目标数据集

下载: 全尺寸图片幻灯片

图 7 不同方法在测试集上的检测结果

下载: 全尺寸图片幻灯片

图 8 损失函数变化曲线

下载: 全尺寸图片幻灯片

表 1 检测结果分类及其含义表

检测结果	含义
TP	将正类预测为正类数
FP	将负类预测为正类数
TN	将负类预测为负类数
FN	将正类预测为负类数

下载: 导出CSV

表 2 本文方法各项指标

模型	数据集	Parameters	Size(MB)	Precision(%)	Recall(%)	mAP@0.5:0.95(%)	RCR(%)	FPS(帧/s)
YOLOv5	可见光	7012822	14.4	85.1	78.2	34.5	83.5	50
YOLOv5	红外	7012822	14.4	93.3	84.1	47.7	66.9	53
CAFM-Net	可见光、红外	11557128	23.6	96.4	93.8	57.2	88.1	48

下载: 导出CSV

表 3 多种算法检测精度性能对比

模型	数据集	特征提取骨干	图像输入尺寸	Parametrs	mAP@0.5:0.98(%)
Faster-Rcnn	可见光	ResNet50	640×640	7864320	34.1
Faster-Rcnn	红外	ResNet50	640×640	7864320	36.3
SSD	可见光	VGG-16	640×640	7235175	43.9
SSD	红外	VGG-16	640×640	7235175	47.6
YOLOv3	可见光	Darknet-53	416×416	6501172	34.1
YOLOv3	红外	Darknet-53	416×416	6501172	39.1
YOLOv4	可见光	CSPDarknet-53	416×416	6396314	27.8
YOLOv4	红外	CSPDarknet-53	416×416	6396314	34.1
YOLOv8	可见光	CSPDarknet-53	640×640	11125971	41.1
YOLOv8	红外	CSPDarknet-53	640×640	11125971	51.1
MHA-YOLOv5	可见光	CSPDarknet-53	640×640	7704906	53.3
MHA-YOLOv5	红外	CSPDarknet-53	640×640	7704906	52.0
DETR	可见光	Resnet50	800×800	42991616	53.6
DETR	红外	Resnet50	800×800	42991616	53.4

下载: 导出CSV

表 4 多光谱数据集上多模态检测算法的结果

模型	数据集	图像输入尺寸	Parameters	mAP@0.5:0.95(%)
SLBAF-Net	可见光、红外	640×640	419430	20.7
CFT	可见光、红外	640×640	44879052	50.4

下载: 导出CSV

参考文献(17)

[1]	SINGH S K, DHAWALE C A, and MISRA S. Survey of object detection methods in camouflaged image[J]. IERI Procedia, 2013, 4: 351–357. doi: 10.1016/j.ieri.2013.11.050.
[2]	王荣昌, 王峰, 任帅军, 等. 基于双流融合网络的单兵伪装偏振成像检测[J]. 光学学报, 2022, 42(9): 0915001. doi: 10.3788/AOS202242.0915001. WANG Rongchang, WANG Feng, REN Shuaijun, et al. Polarization imaging detection of individual camouflage based on two-stream fusion network[J]. Acta Optica Sinica, 2022, 42(9): 0915001. doi: 10.3788/AOS202242.0915001.
[3]	LE T N, NGUYEN T V, NIE Zhongliang, et al. Anabranch network for camouflaged object segmentation[J]. Computer Vision and Image Understanding, 2019, 184: 45–56. doi: 10.1016/j.cviu.2019.04.006.
[4]	FAN Dengping, JI Gepeng, SUN Guolei, et al. Camouflaged object detection[C]. IEEE/CVF Conference On Computer Vision And Pattern Recognition, Seattle, USA, 2020: 2774–2784. doi: 10.1109/CVPR42600.2020.00285.
[5]	FAN Dengping, JI Gepeng, ZHOU Tao, et al. PraNet: Parallel reverse attention network for polyp segmentation[C]. 23rd International Conference on Medical Image Computing and Computer Assisted Intervention–MICCAI 2020, Lima, Peru, 2020: 263–273. doi: 10.1007/978-3-030-59725-2_26.
[6]	TANKUS A and YESHURUN Y. Convexity-based visual camouflage breaking[J]. Computer Vision and Image Understanding, 2001, 82(3): 208–237. doi: 10.1006/cviu.2001.0912.
[7]	BHAJANTRI N U and NAGABHUSHAN P. Camouflage defect identification: A novel approach[C]. 9th International Conference on Information Technology, Bhubaneswar, India, 2006: 145–148. doi: 10.1109/ICIT.2006.34.
[8]	ZHANG Wei, ZHOU Qikai, LI Ruizhi, et al. Research on camouflaged human target detection based on deep learning[J]. Computational Intelligence and Neuroscience, 2022, 2022: 7703444. doi: 10.1155/2022/7703444.
[9]	赖杰, 彭锐晖, 孙殿星, 等. 融合注意力机制与多检测层结构的伪装目标检测[J]. 中国图象图形学报, 2024, 29(1): 134–146. doi: 10.11834/jig.221189. LAI Jie, PENG Ruihui, SUN Dianxing, et al. Detection of camouflage targets based on attention mechanism and multi-detection layer structure[J]. Journal of Image and Graphics, 2024, 29(1): 134–146. doi: 10.11834/jig.221189.
[10]	刘珩, 冉建国, 杨鑫, 等. 基于DETR的迷彩伪装目标检测[J]. 现代电子技术, 2022, 45(17): 41–46. doi: 10.16652/j.issn.1004-373x.2022.17.008. LIU Heng, RAN Jianguo, YANG Xin, et al. Camouflage target detection based on detection transformer[J]. Modern Electronics Technique, 2022, 45(17): 41–46. doi: 10.16652/j.issn.1004-373x.2022.17.008.
[11]	YADAV D, ARORA M K, TIWARI K C, et al. Detection and identification of camouflaged targets using hyperspectral and LiDAR data[J]. Defence Science Journal, 2018, 68(6): 540–546. doi: 10.14429/dsj.68.12731.
[12]	HU Jianghua, CUI Guangzhen, and QIN Lie. A new method of multispectral image processing with camouflage effect detection[C]. Proceedings of SPIE 9675, AOPC 2015: Image Processing and Analysis, Beijing, China, 2015: 967510. doi: 10.1117/12.2199206.
[13]	CHENG Xiaolong, GENG Keke, WANG Ziwei, et al. SLBAF-net: Super-lightweight bimodal adaptive fusion network for UAV detection in low recognition environment[J]. Multimedia Tools and Applications, 2023, 82(30): 47773–47792. doi: 10.1007/s11042-023-15333-w.
[14]	FANG Qingyun, HAN Depeng, and WANG Zhaokui. Cross-modality fusion transformer for multispectral object detection[J]. arXiv: 2111.00273, 2021. doi: 10.48550/arXiv.2111.00273. (查阅网上资料,不确定文献类型及格式是否正确,请确认) .
[15]	MA Jiayi, MA Yong, and Li Chang. Infrared and visible image fusion methods and applications: A survey[J]. Information Fusion, 2019, 45: 153–178. doi: 10.1016/j.inffus.2018.02.004.
[16]	聂茜茜, 肖斌, 毕秀丽, 等. 基于超像素级卷积神经网络的多聚焦图像融合算法[J]. 电子与信息学报, 2021, 43(4): 965–973. doi: 10.11999/JEIT191053. NIE Xixi, XIAO Bin, BI Xiuli, et al. Multi-focus image fusion algorithm based on super pixel level convolutional neural network[J]. Journal of Electronics & Information Technology, 2021, 43(4): 965–973. doi: 10.11999/JEIT191053.
[17]	GEVORGYAN Z. SIoU loss: More powerful learning for bounding box regression[J]. arXiv: 2205.12740, 2022. doi: 10.48550/arXiv.2205.12740. (查阅网上资料,不确定文献类型及格式是否正确,请确认) .