暗环境下红外目标检测跟踪方法研究

doi:10.12382/bgxb.2024.0081

摘要/Abstract

摘要：

针对暗环境动态特征轮廓模糊、盲区遮挡情况,高效准确地检测跟踪动态目标特征,对灾害救援、搜寻跟踪具有实际意义。为实现暗环境下模糊轮廓特征的有效检测跟踪,提出一种时空关联机制的红外目标实时检测深度学习网络(Spatial Local Dynamic You Only Look Once Version 8,SLD-YOLOv8),设计非局部自适应Non-local模块和空间通道卷积关联模块,对原YOLOv8网络的瓶颈层Bottleneck CSP进行优化。为有效提取深层空间多尺度表征信息,增加用于小目标检测的160×160检测层和动态检测头,较好地提升暗环境中目标跟踪的边界回归性能,并实时有效地推理出目标特征的相对深度位置信息。实验结果表明,改进后的红外目标检测算法对暗环境下的动态特征检测具有较好的鲁棒性和准确性,其平均精度评估指标mAP_0.5和mAP_0.5:0.95比原模型提高了5.6%和4.5%,证明了新算法对暗环境目标跟踪的有效性。

关键词: 暗环境, 深度学习, 注意力机制, 目标跟踪检测, 非局部域机理, 动态检测头

Abstract:

The efficient and accurate detection and tracking of dynamic target features in dark environments,where contours are blurred and occlusions are present,hold practical significance for disaster relief,search and tracking operations.To effectively detect and track the blurred contour features in dark environments,an improved real-time infrared target tracking and detection algorithm is proposed.This algorithm,based on the deep learning network (Spatial Local Dynamic You Only Look Once,SLD-YOLOv8),incorporates a non-local adaptive module and a spatial channel convolution (SCC) correlation module to optimize the Bottleneck CSP of YOLOv8 network for better feature extraction.A dedicated 160×160 detection layer and a dynamic head are introduced for the improved detection of small-scale targets and the enhanced boundary regression capabilities in low-light scenarios,enabling the accurate real-time inference of relative target position.Experimental validation shows that the proposed algorithm has good robustness and accuracy in detecting the dynamic features in dark environments.The average precision evaluation metrics mAP_0.5 and mAP_0.5:0.95 of this model are increased by 5.6% and 4.5%,respectively,compared to the original model,affirming its effectiveness of tracking the targets in dark environments.

Key words: dark environments, deep learning, attention mechanism, target detection and tracking, non-local, dynamic head

刘辉, 李明益, 韩立金, 刘宝帅. 暗环境下红外目标检测跟踪方法研究[J]. 兵工学报, 2025, 46(8): 240081-.

LIU Hui, LI Mingyi, HAN Lijin, LIU Baoshuai. Research on Infrared Target Detection and Tracking in Dark Environments[J]. Acta Armamentarii, 2025, 46(8): 240081-.

图/表 17

图1 BottleneckCSP网络结构

Fig.1 The structure of BottleneckCSP Network

图2 YOLOv8 网络结构

Fig.2 The structure of YOLOv8 network

图3 SLD-YOLOv8目标检测网络结构

Fig.3 The structure of SLD-YOLOv8 target detection network

图4 空间重组单元

Fig.4 Spatial reconstruction unit

图5

Fig.5 CRU Channel reconstruction unit

图6 Non-local模块

Fig.6 Non-local module

图7 改进BottleneckCSP网络结构

Fig.7 Improved Bottleneck CSP structure

图8 改进Dyhead 模块

Fig.8 Improved Dyhead module

图9 Mosaic数据增强方法

Fig.9 Methods of Mosaic data enhancement

表1 正负样本关系

Table 1 The relationship of positive and negative samples

真实情况	预测情况
真实情况	正例	反例
正例	TP(真正例)	FN(假反例)
反例	FP(假正例)	TN(真反例)

图10 YOLOv8 mAP_0.5精度对比结果

Fig.10 Comparison results of YOLOv8 mAP_0.5 accuracy

图11 YOLOv8 mAP_0.5:0.95精度对比结果

Fig.11 Comparison results of YOLOv8 mAP_0.5:0.95 accuracy

图12 YOLOv8 loss回归曲线对比结果

Fig.12 Comparison results of YOLOv8 loss regression curves

图13 模型跟踪检测对比结果

Fig.13 Comparison results of model tracking

图14 模型输出坐标信息结果(左为实时检测, 右为检测坐标)

Fig.14 Results of the model outputting the coordinate information (Left:real-time detection,Right:detected coordinates)

表2 模型消融对比实验

Table 2 Ablation comparison experiment of models

模型				P/%	R/%	mAP_0.5/%	推理时间/ms
B	SC	NL	Dy	P/%	R/%	mAP_0.5/%	推理时间/ms
P P	P			80.4	74.6	78.6	7.6
P P	P			82.6	77.8	82.5	7.8
P		P		81.8	75.3	78.5	9.4
P			P	81.7	76.4	79.6	6.8
P	P	P		83.7	79.5	82.4	8.7
P	P	P	P	85.3	80.7	83.7	7.5

表3 模型定量评估实验

Table 3 Quantitative evaluation of model experimental results

方法	主干网络	mAP_0.5/%		mAP_0.5:0.95/%		推理时间/ms
方法	主干网络	D1	D2	D1	D2	D1	D2
DETR^[20]	ResNet-101	76.4	80.4	56.4	62.2	17.6	42.4
Deformable DETR^[21]	ResNet-101	77.3	78.6	56.3	61.8	15.8	40.6
FasterR-CNN^[22]	ResNet-101	72.8	74.5	50.9	59.6	23.3	38.2
ATSS^[23]	ResNet-101	76.9	78.2	56.9	60.5	16.4	46.3
YOLOv8n	CSPDarknet53	78.9	81.7	58.1	62.8	8.3	21.9
YOLOv8s	CSPDarknet53	79.4	82.2	59.4	64.7	9.1	24.6
YOLOv8m	CSPDarknet53	80.3	84.4	60.3	66.3	9.5	32.7
YOLOv8l	CSPDarknet53	83.4	85.3	61.2	67.2	12.8	36.8
YOLOv8x	CSPDarknet53	85.7	88.3	63.8	68.7	17.4	39.6
SLD-YOLOv8n	Modified CSP	84.5	85.3	62.6	66.8	7.8	15.2
SLD-YOLOv8s	Modified CSP	84.9	85.9	62.7	67.5	8.4	16.8
SLD-YOLOv8m	Modified CSP	85.4	87.4	63.7	68.1	8.6	19.4
SLD-YOLOv8l	Modified CSP	88.3	89.6	65.4	70.4	10.7	23.7
SLD-YOLOv8x	Modified CSP	91.1	92.7	70.8	71.3	15.4	30.2

参考文献 23

[1]	ZHAO M J, LI W, LI L, et al. Single-frame infrared small-target detection:a survey[J]. IEEE Geoscience and Remote Sensing Magazine, 2022, 10(2):87-119.
[2]	LI R H, SHEN Y S. YOLOSR-IST:a deep learning method for small target detection in infrared remote sensing images based on super-resolution and YOLO[J]. Signal Processing, 2023, 208:108962.
[3]	KOU R K, WANG C P, FU Q, et al. Detection model and performance evaluation for the infrared search and tracking system[J]. Applied Optics, 2023, 62(2):398-410.
[4]	ARKIN E, YADIKAR N, XU X B, et al. A survey:object detection methods from CNN to transformer[J]. Multimedia Tools and Applications, 2023, 82(14):21353-21383.
[5]	魏居尚, 綦秀利, 尹成祥, 等. 未知建筑空间中自主搜索任务规划[J]. 火力与指挥控制, 2024, 49(7):156-161.
	WEI J S, QI X L, YIN C X, et al. An autonomous search planning method for UAV in unknown architectural space[J]. Fire Control and Command Control, 2024, 49(7):156-161. (in Chinese)
[6]	刘建华, 尹国富, 黄道杰, 等. 基于特征融合的可见光与红外图像目标检测[J]. 激光与红外, 2023, 53(3):394-401.
	LIU J H, YIN G F, HUANG D J, et al. Object detection in visible light and infrared images based on feature fusion[J]. Laser and Infrared, 2023, 53(3):394-401. (in Chinese)
[7]	王强, 吴乐天, 李红, 等. 基于双支网络协作的红外弱小目标检测[J]. 兵工学报, 2023, 44(10):3165-3176. doi: 10.12382/bgxb.2022.0605
	WANG Q, WU L T, LI H, et al. An infrared small target detection method via dual network collaboration[J]. Acta Armamentarii, 2023, 44(10):3165-3176. (in Chinese) doi: 10.12382/bgxb.2022.0605
[8]	CHEN Y H, ZHANG G P, MA Y J, et al. Small infrared target detection based on fast adaptive masking and scaling with iterative segmentation[J/OL]. IEEE Geoscience and Remote Sensing Letters, 2022, 19:1-5.
[9]	ZHANG F, LIN S L, XIAO X Y, et al. Global attention network with multiscale feature fusion for infrared small target detection[J]. Optics & Laser Technology, 2024, 168:110012.
[10]	JING J Q, JIA B, HUANG B Q, et al. YOLO-D:dual-branch infrared distant target detection based on multi-level weighted feature fusion[C]// Proceedings of International Conference on Neural Information Processing. Singapore: Springer, 2023:140-151.
[11]	SHEN L Y, LANG B H, SONG Z X. DS-YOLOv8-based object detection method for remote sensing images[J]. IEEE Access, 2023, 11:125122-125137.
[12]	ZHOU J J, ZHANG B H, YUAN X L, et al. YOLO-CIR:the network based on YOLO and ConvNeXt for infrared object detection[J]. Infrared Physics & Technology, 2023, 131:104703.
[13]	WANG C Y, LIAO H Y M, WU Y H, et al. CSPNet:a new backbone that can enhance learning capability of CNN[C]// Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition. Seattle,WA,US: IEEE Computer Society and CVF, 2020:390-391.
[14]	ZHANG H, ZU K K, LU J, et al. EPSANet:an efficient pyramid squeeze attention block on convolutional neural network[C]// Proceedings of the Asian Conference on Computer Vision. Kyoto,Japan: Springer, 2022:1161-1177.
[15]	LI J F, WEN Y W, HE L H. SCConv:spatial and channel reconstruction convolution for feature redundancy[C]// Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition. Vancouver,Canada: IEEE, 2023:6153-6162.
[16]	MEI Y Q, FAN Y C, ZHOU Y Q. Image super-resolution with non-local sparse attention[C]// Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition. Wshington,D.C.,US: IEEE, 2021:3517-3526.
[17]	LIU F C, GAO C Q, CHEN F, et al. Infrared small and dim target detection with transformer under complex backgrounds[J]. IEEE Transactions on Image Processing, 2023, 32:5921-5932.
[18]	DAI X Y, CHEN Y P, XIAO B, et al. Dynamic head:unifying object detection heads with attentions[C]// Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition. Wshington,D.C.,US: IEEE, 2021:7373-7382.
[19]	MEI Y Q, FAN Y C, ZHANG Y L, et al. Pyramid attention network for image restoration[J]. International Journal of Computer Vision, 2023, 131(12):3207-3225.
[20]	CARION N, MASSA F, SYNNAEVE G, et al. End-to-end object detection with transformers[C]// Proceedings of European Conference on Computer Vision. Cham,Switzerland: Springer, 2020:213-229.
[21]	ZHU X Z, SU W J, LU L W, et al. Deformable DETR:deformable transformers for end-to-end object detection:arXiv:2010.04159[R]. Ithaca,NY,US: Cornell University, 2020:2010.04159.
[22]	DAI X B, HU J P, ZHANG H M, et al. Multi-task faster R-CNN for night time pedestrian detection and distance estimation[J]. Infrared Physics & Technology, 2021, 115:103694.
[23]	ZHANG S F, CHI C, YAO Y Q, et al. Bridging the gap between anchor-based and anchor-free detection via adaptive training sample selection[C]// Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition. Seattle,WA,US: IEEE, 2020:9759-9768.