基于多尺度注意力机制的无人机小目标检测算法

doi:10.12382/bgxb.2023.1124

摘要/Abstract

摘要：

针对无人机航拍图像密集度大、目标尺寸小、背景复杂等难点,提出一种基于多尺度注意力机制的小目标检测(Small target detection of BPAN-EF_C2f YOLOv8s,SBE_ YOLOv8s)算法,通过设计一种基于多尺度注意力机制的特征提取模块(EMA-Faster Block_C2f,EF_C2f),替换YOLOv8网络中的C2f模块,提高网络对小目标特征的提取能力;在特征融合网络中增加P1检测层,并设计一种跨尺度特征融合结构(Bi-Path Aggregation Network,BPAN),融合小目标特征信息;增加一个微小目标检测头,使用SIoU Loss作为边界框损失函数,提升小目标检测精度和网络收敛速度。在公开数据集VisDrone2019上进行实验验证。验证结果表明:与YOLOv8s算法相比,新算法在检测精度上提升了6.9%、mAP₅₀提升了9.1%,模型参数量减少了44.6%,检测速度为28帧/s,新算法在小目标检测领域具有一定的实用性。

关键词: 多尺度注意力机制, YOLOv8s算法, 特征提取, 跨尺度特征融合, 小目标检测

Abstract:

A SBE_YOLOv8s small target detection algorithm based on multi-scale attention mechanism is proposed for the UAV aerial images with high density,small target size and complex background.First,a feature extraction module EF_C2f(EMA-Faster Block_C2f) based on the multi-scale attention mechanism is designed to replace the C2f module in the YOLOv8 network to improve the network’s ability to extract small target features.And then a P1 detection layer is added to the feature fusion network,and a cross-scale feature fusion structure BPAN(Bi-Path Aggregation Network) is designed to fuse the small target feature information.Finally,a tiny target detection head is added,and SIoU Loss is used as the bounding-box loss function to improve the detection accuracy of small targets and the convergence speed of the network.The proposed algorithm is validated on the public dataset VisDrone2019.Compared with YOLOv8s algorithm,the proposed algorithm improves the detection accuracy by 6.9% and mAP₅₀ by 9.1%,and reduces the amount of parameters of the model is by 46.4%,and the detection speed is 28 fps.The experimental results show that the proposed algorithm has a certain degree of utility in the field of small target detection.

Key words: multi-scale attention mechanism, YOLOv8s algorithm, feature extraction, cross-scale feature fusion, small target detection

中图分类号:

TP391

冯迎宾, 郭枭尊, 晏佳华. 基于多尺度注意力机制的无人机小目标检测算法[J]. 兵工学报, 2025, 46(1): 231124-.

FENG Yingbin, GUO Xiaozun, YAN Jiahua. Small UVA Target Detection Algorithm Based on Multi-scale Attention Mechanism[J]. Acta Armamentarii, 2025, 46(1): 231124-.

图/表 15

图1 不同情况下无人机航拍图像

Fig.1 UAV aerial images under different situations

图2 SBE_YOLOv8s网络结构图

Fig.2 SBE_YOLOv8s network structure diagram

图3 EMA注意力机制原理图

Fig.3 Schematic diagram of EMA attention mechanism

图4 FasterNet_Block原理图

Fig.4 Schematic diagram of FasterNet_Block

图5 EF_C2f结构图

Fig.5 EF_C2f structure diagram

图6 VisDrone2019数据尺寸分布图

Fig.6 VisDrone2019 data size distribution graph

图7 PANet、BiFPN、BPAN结构图

Fig.7 PANet,BiFPN and BPAN structure diagrams

图8 SBE_YOLOv8s网络结构简图

Fig.8 Schematic diagram of SBE_YOLOv8s network structure

表1 实验环境配置信息

Table 1 Experimental environment configuration information

配置	版本
操作系统	Windows 10
处理器	Intel Xeon Sliver 4210 CPU @2.20GHz
显卡	NVIDIA Quadro RTX 4000
框架	Pytorch 1.13.1
开发软件	Pycharm 2020.3.12
开发语言	Python 3.8.17

表2 消融实验中各改进方法的性能测试评价指标

Table 2 Performance test evaluation results of each improved method in ablation experiments

编号	模型	P/%	mAP₅₀/%	mAP_50-95/%	Par/10⁶	O/10⁹
1	YOLOv8s	44.8	33.0	18.6	11.2	28.5
2	YOLOv8s+BPAN	49.4	40.2	23.1	7.7	123.4
3	YOLOv8s+BPAN+SIoU Loss	50.1	40.9	23.4	7.7	123.4
4	YOLOv8s+BPAN+EF_C2f	51.2	41.8	23.8	6.2	105.4
5	YOLOv8s+BPAN+SIoU Loss+EF_C2f	51.7	42.1	24.3	6.2	105.4

表3 对比实验中各模型的性能评价指标

Table 3 Comparison of the performance evaluation results of models in the experiment

方法	mAP₅₀/%	mAP_50-95/%	FPS
Faster-RCNN	21.7	15.1	15
Cascade-RCNN	31.9	16.1
Light-RCNN	30.8	16.5
YOLOv3	32.1	17.5	31
YOLOv4	30.7	15.9	32
YOLOv5	31.5	16.8	121
YOLOv8	33.0	18.6	169
Ours	42.1	24.3	28

表4 本文算法在VisDrone 2019数据集上的性能评价指标

Table 4 Performance evaluation metrics of the proposed algorithm on VisDrone 2019 dataset

类别	P/%	R/%	mAP₅₀/%	mAP_50-95/%
Pedestrian	62.4	40.0	43.3	18.5
People	60.5	26.8	31.5	12.4
Bicycle	38.7	19.5	18.9	8.3
Car	72.8	80.9	82.7	51.9
Van	46.3	49.9	46.9	31.7
Truck	44.7	49.5	43.1	28.8
Tricycle	28.5	36.4	25.6	14.7
Awning-tricycle	42.7	25.4	23.8	15.2
Bus	68.7	56.3	62.3	43.3
Motor	51.7	45.5	42.5	18.5

图9 小目标检测对比效果图

Fig.9 Comparing effect graph of small target detection

表5 YOLOv8s算法与本文算法检测对比

Table 5 Comparison of detected results of YOLOv8s algorithm and the proposed algorithm

类别	图9(a)	图9(b)	图9(c)	图9(d)
Pedestrian	13	74	24	42
People	0	1	1	1
Bicycle	0	0	0	1
Car	39	44	10	8
Van	0	4	0	1
Truck	0	2	0	0
Tricycle	2	4	0	0
Awning-tricycle	1	1	0	0
Motor	1	5	1	1

图10 SBE_YOLOv8算法应用在4种复杂场景中的效果图

Fig.10 SBE_YOLOv8 algorithm applied in four complex scenarios

参考文献 29

[1]	İBROŞKA B, ÖZPEYNIRCI S, ÖZPEYNIRCI Ö. Multiple traveling salesperson problem with drones:general variable neighborhood search approach[J]. Computers and Operations Research, 2023,160:106390.
[2]	HU X, ASSAAD R H. The use of unmanned ground vehicles(mobile robots) and unmanned aerial vehicles(drones) in the civil infrastructure asset management sector:applications,robotic platforms,sensors,and algorithms[J]. Expert Systems with Applications, 2023,232:120897.
[3]	LI S J, WANG L Z, LI J, et al. Image classification algorithm based on improved AlexNet[J]. Journal of Physics:Conference Series. IOP Publishing, 2021, 1813(1):012051.
[4]	惠康华, 杨卫, 刘浩翰, 等. 基于YOLOv5的增强多尺度目标检测方法[J]. 兵工学报, 2023, 44(9):2600-2610. doi: 10.12382/bgxb.2022.1147
	HUI K H, YANG W, LIU H H, et al. Enhanced multi-scale target detection method based on YOLOv5[J]. Acta Armamentarii, 2023, 44(9):2600-2610. (in Chinese) doi: 10.12382/bgxb.2022.1147
[5]	GIRSHICK R. Fast R-CNN[C]//Proceedings of the 2015 IEEE International Conference on Computer Vision.Santiago, Chile:IEEE,2015:1440-1448.
[6]	REN S Q, HE K M, GIRSHICK R B, et al. Faster R-CNN:towards real-time object detection with region proposal networks[C] //Proceedings of the International Conference on Neural Information Processing System.Cambridge,MA, US: MIT Press,2015:91-99.
[7]	CAI Z W, VASCONCELOS N. Cascade R-CNN:high quality object detection and instance segmentation[J]. IEEE Transactions on Pattern Analysis and Machine Intelligence, 2021, 43(5):1483-1498.
[8]	MITTAL P, SHARMA A, SINGH R. A feature pyramid based multi-stage framework for object detection in low-altitude UAV images[J]. International Journal on Artificial Intelligence Tools, 2022, 31(2):2250028.
[9]	王殿伟, 胡里晨, 房杰, 等. 基于改进Double-Head R-CNN的无人机航拍图像小目标检测算法[J]. 北京航空航天大学学报, 2024, 50(7):2141-2149.
	WANG D W, HU L C, FANG J, et al. Small object detection algorithm based on improved Double-Head RCNN for UAV aerial images[J]. Journal of Beijing University of Aeronautics and Astronautics, 2024, 50(7):2141-2149. (in Chinese)
[10]	REDMON J, DIVVALA S, GIRSHICK R, et al. You only look once:unified,real-time object detection[C]//Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition.Washington,D.C., US:IEEE,2016:779-788.
[11]	LIU W, ANGUELOV D, ERHAN D, et al. SSD:single shot multibox detector[C]//Proceedings of the 14th European Conference on Computer Vision. Amsterdam,the Netherlands: Springer,2016:21-37.
[12]	SHANG J C, WANG J S, LIU S B, et al. Small target detection algorithm for UAV aerial photography based on improved YOLOv5s[J]. Electronics, 2023, 12(11):2434.
[13]	KIM M, JEONG J, KIM S. ECAP-YOLO:efficient channel attention pyramid YOLO for small object detection in aerial image[J]. Remote Sensing, 2021, 13(23):4851.
[14]	NIE H J, PANG H L, MA M Y, et al. A Lightweight remote sensing small target image detection algorithm based on improved YOLOv8[J]. Sensors, 2024, 24(9):2952.
[15]	WOO S, PARK J, LEE J Y, et al. CBAM:convolutional block attention module[C]//Proceedings of the 15th European Conference on Computer Vision.Munich, Germany:Springer,2018:3-19.
[16]	ZHANG Q L, YANG Y B. SA-Net:shuffle attention for deep convolutional neural networks[C]//Proceedings of 2021 IEEE International Conference on Acoustics,Speech and Signal Processing.Toronto,ON, Canada:IEEE,2021:2235-2239.
[17]	OUYANG D, HE S, ZHANG G, et al. Efficient multi-scale attention module with cross-spatial learning[C]//Proceedings of 2023 IEEE International Conference on Acoustics,Speech and Signal Processing.Rhodes Island, Greece:IEEE,2023:1-5.
[18]	CHEN J R, KAO S H, HE H, et al.Run, don’t walk:chasing higher FLOPS for faster neural networks[C]//Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition.Vancouver,BC, Canada:IEEE,2023:12021-12031.
[19]	ZHU P F, WEN L Y, DU D W, et al. Detection and tracking meet drones challenge[J]. IEEE Transactions on Pattern Analysis and Machine Intelligence, 2021, 44(11):7380-7399.
[20]	张智, 易华挥, 郑锦. 聚焦小目标的航拍图像目标检测算法[J]. 电子学报, 2023, 51(4):944-955. doi: 10.12263/DZXB.20220313
	ZHANG Z, YI H H, ZHENG J. Focusing on small objects detector in aerial images[J]. Acta Electronica Sinica, 2023, 51(4):944-955. (in Chinese) doi: 10.12263/DZXB.20220313
[21]	WANG K X, LIEW J H, ZOU Y T, et al. PANet:few-shot image semantic segmentation with prototype alignment[C]// Proceedings of the IEEE/CVF International Conference on Computer Vision.Seoul, Korea:IEEE,2019:9197-9206.
[22]	TAN M X, PANG R M, LE Q V. EfficientDet:scalable and efficient object detection[C]//Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition.Seattle,WA, US:IEEE,2020:10781-10790.
[23]	GEVORGYAN Z. SIoU loss:more powerful learning for bounding box regression[R/OL]. Ithaca,NY, US: Cornell University, 2022.
[24]	LI X, WANG W H, WU L J, et al. Generalized focal loss:learning qualified and distributed bounding boxes for dense object detection[J]. Advances in Neural Information Processing Systems, 2020,33:21002-21012.
[25]	冒国韬, 邓天民, 于楠晶. 基于多尺度分割注意力的无人机航拍图像目标检测算法[J]. 航空学报, 2023, 44(5):326738. doi: 10.7527/S1000-6893.2021.26738
	MAO G T, DENG T M, YU N J. Measurement of the engraving process of projectiles[J]. Acta Aeronautica et Astronautica Sinica, 2023, 44(5):326738. (in Chinese)
[26]	李校林, 刘大东, 刘鑫满, 等. 改进YOLOv5的无人机航拍图像目标检测算法[J]. 计算机工程与应用, 2024, 60(11):204-214. doi: 10.3778/j.issn.1002-8331.2307-0171
	LI X L, LIU D D, LIU X M, et al. Target detection algorithm of UAV aerial image based on improved YOLOv5[J]. Computer Engineering and Applications, 2024, 60(11):204-214. (in Chinese) doi: 10.3778/j.issn.1002-8331.2307-0171
[27]	LIU H Y, SUN F Q, GU J, et al. SF-YOLOv5:a lightweight small object detection algorithm based on improved feature fusion mode[J]. Sensors, 2022, 22(15):5817.
[28]	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, LÜ M. Improved YOLOv3 algorithm and its application in military target detection[J]. Acta Armamentarii, 2022, 43(2):345-354. (in Chinese) doi: 10.3969/j.issn.1000-1093.2022.02.012
[29]	奉志强, 谢志军, 包正伟, 等. 基于改进YOLOv5的无人机实时密集小目标检测算法[J]. 航空学报, 2023, 44(7):327106. doi: 10.7527/S1000-6893.2022.27106
	FENG Z Q, XIE Z J, BAO Z W, et al. Real-time dense small object detection algorithm for UAV based on improved YOLOv5[J]. Acta Aeronautica et Astronautica Sinica, 2023, 44(7):327106. (in Chinese)