面向车窗状态检测的轻量化目标检测模型

doi:10.19678/j.issn.1000-3428.0069393

摘要/Abstract

摘要：

针对当前智慧化城市建设中安防检测技术对检测精度和检测速度的要求, 提出了一种基于改进YOLOv5s的轻量化车窗目标检测模型。该模型在保证精度的同时降低了计算复杂度, 能够部署在边缘计算平台中, 实现对车窗开闭状态的检测。首先, 在主干网络中引入FasterNet, 以减少模型的计算量和模型参数; 其次, 在特征融合阶段, 使用GSConv降低模型的复杂度, 并使用双向特征金字塔网络(BiFPN)连接特征融合网络, 以融合更多的特征信息; 最后, 使用结构化交并比(SIoU)损失函数加快收敛。实验结果表明, 改进后的算法性能在车窗数据集上相较于原算法明显得到提升, 交并比(IoU)阈值在0.5以及0.5~0.95范围内时模型的平均精度mAP@0.5和mAP@0.5:0.95分别提高了0.001和0.005, 模型的参数量大幅减少且浮点运算数(FLOPs)大幅降低, 仅为原模型的18.74%和17.09%, 推理速度提升了260%。改进后的模型部署在NVIDIA Jetson Nano平台上时有着良好的表现。

关键词: FasterNet, GSConv, 双向特征金字塔网络, 轻量化, 车窗状态检测

Abstract:

To improve the security detection technology in smart cities in terms of detection accuracy and speed, a lightweight car window target detection model based on improved YOLOv5s is proposed. The proposed model reduces computational complexity while ensuring accuracy and can be deployed in edge computing platforms to detect the open and closed states of car windows. First, FasterNet is introduced into the backbone network to reduce the computational complexity and number of model parameters. Second, in the feature fusion stage, GSConv is used to reduce the complexity of the model and a Bidirectional Feature Pyramid Network (BiFPN) is used to connect the feature fusion network and fuse more feature information. Finally, the Structured Intersection over Union (SIoU) loss function is employed to accelerate convergence. The experimental results show that the improved algorithm exhibits significant differences compared with the original algorithm on the car window dataset. The mean Average Precision (mAP@0.5) for the Intersection over Union (IoU) threshold of 0.5 and mean average precision (mAP@0.5:0.95) for IoU in the range of 0.5 to 0.95 increase by 0.001 and 0.005, respectively. The number of parameters and Floating Point Operations Per Second (FLOPs) of the model are significantly reduced to 18.74% and 17.09% of those of the original model, respectively, and the inference speed is increased by 260%. The proposed model exhibits good performance when deployed on an NVIDIA Jetson Nano.

Key words: FasterNet, GSConv, Bidirectional Feature Pyramid Network (BiFPN), lightweight, car window status detection

梅华威, 王泽洋, 苏攀, 尚虹霖, 方毅. 面向车窗状态检测的轻量化目标检测模型[J]. 计算机工程, 2025, 51(12): 285-293.

MEI Huawei, WANG Zeyang, SU Pan, SHANG Honglin, FANG Yi. Lightweight Object Detection Model for Car Window Status Detection[J]. Computer Engineering, 2025, 51(12): 285-293.

https://www.ecice06.com/CN/Y2025/V51/I12/285

图/表 14

图1 YOLOv5网络结构

Fig.1 YOLOv5 network structure

图2 改进的YOLOv5网络结构

Fig.2 Improved YOLOv5 network structure

图3 深度可分离卷积

Fig.3 Depthwise separable convolution

图4 PConv结构

Fig.4 PConv structure

图5 FasterNet块

Fig.5 FasterNet block

图6 GS bottleneck和VOV-GSCSP

Fig.6 GS bottleneck and VOV-GSCSP

图7 PANet和BiFPN

Fig.7 PANet and BiFPN

图8 数据集中4种类别图像示例

Fig.8 Examples of 4 categories of images in dataset

图9 Jetson Nano上运行结果

Fig.9 Running results on Jetson Nano

参考文献 28

1	DENG S G , ZHAO H L , FANG W J , et al. Edge intelligence: the confluence of edge computing and artificial intelligence. IEEE Internet of Things Journal, 2020, 7 (8): 7457- 7469. doi: 10.1109/JIOT.2020.2984887
2	LV Z H , CHEN D L , LOU R R , et al. RETRACTED: intelligent edge computing based on machine learning for smart city. Future Generation Computer Systems, 2021, 115, 90- 99. doi: 10.1016/j.future.2020.08.037
3	HUANG K Q , TAN T N . Vs-star: a visual interpretation system for visual surveillance. Pattern Recognition Letters, 2010, 31 (14): 2265- 2285. doi: 10.1016/j.patrec.2010.05.029
4	LI H, LI J, WEI H, et al. Slim-neck by GSConv: a better design paradigm of detector architectures for autonomous vehicles[EB/OL]. [2024-01-21]. https://www.semanticscholar.org/paper/Slim-neck-by-GSConv:-A-better-design-paradigm-of-Li-Li/728ddffea17d18cf765a38fe526908716d6f2a0e.
5	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 (CVPR). Washington D. C., USA: IEEE Press, 2020: 10781-10790.
6	GEVORGYAN Z. SIoU loss: more powerful learning for bounding box regression[EB/OL]. [2024-01-21]. https://arxiv.org/abs/2205.12740.
7	KRIZHEVSKY A , SUTSKEVER I , HINTON G E . ImageNet classification with deep convolutional neural networks. Communications of the ACM, 2017, 60 (6): 84- 90. doi: 10.1145/3065386
8	SIMONYAN K, ZISSERMAN A. Very deep convolutional networks for large-scale image recognition[EB/OL]. [2024-01-21]. https://www.semanticscholar.org/paper/Very-Deep-Convolutional-Networks-for-Large-Scale-Simonyan-Zisserman/eb42cf88027de515750f230b23b1a057dc782108
9	SZEGEDY C, LIU W, JIA Y Q, et al. Going deeper with convolutions[C]//Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition (CVPR). Washington D. C., USA: IEEE Press, 2015: 1-9.
10	GIRSHICK R, DONAHUE J, DARRELL T, et al. Rich feature hierarchies for accurate object detection and semantic segmentation[C]//Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition. Washington D. C., USA: IEEE Press, 2014: 580-587.
11	GIRSHICK R. Fast R-CNN[C]//Proceedings of the IEEE International Conference on Computer Vision (ICCV). Washington D. C., USA: IEEE Press, 2015: 1440-1448.
12	REN S Q , HE K M , GIRSHICK R , et al. Faster R-CNN: towards real-time object detection with region proposal networks. IEEE Transactions on Pattern Analysis and Machine Intelligence, 2017, 39 (6): 1137- 1149. doi: 10.1109/TPAMI.2016.2577031
13	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 (CVPR). Washington D. C., USA: IEEE Press, 2016: 779-788.
14	REDMON J, FARHADI A. YOLO9000: better, faster, stronger[C]//Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition (CVPR). Washington D. C., USA: IEEE Press, 2017: 6517-6525.
15	REDMON J, FARHADI A. YOLOv3: an incremental improvement[EB/OL]. [2024-01-21]. https://arxiv.org/abs/1804.02767.
16	BOCHKOVSKIY A, WANG C Y, LIAO H Y M. YOLOv4: optimal speed and accuracy of object detection[EB/OL]. [2024-01-21]. https://www.semanticscholar.org/paper/YOLOv4%3A-Optimal-Speed-and-Accuracy-of-Object-Bochkovskiy-Wang/2a6f7f0d659c5f7dcd665064b71e7b751592c80e.
17	KRIZHEVSKY A , SUTSKEVER I , HINTON G E . ImageNet classification with deep convolutional neural networks. Communications of the ACM, 2017, 60 (6): 84- 90. doi: 10.1145/3065386
18	SIFRE L, MALLAT S. Rigid-motion scattering for texture classification[EB/OL]. [2024-01-21]. https://arxiv.org/abs/1403.1687.
19	HOWARD A G, ZHU M, CHEN B, et al. MobileNets: efficient convolutional neural networks for mobile vision applications[EB/OL]. [2024-01-21]. https://arxiv.org/abs/1704.04861.
20	SANDLER M, HOWARD A, ZHU M L, et al. MobileNetV2: inverted residuals and linear bottlenecks[C]//Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition. Washington D. C., USA: IEEE Press, 2018: 4510-4520.
21	HOWARD A, SANDLER M, CHEN B, et al. Searching for MobileNetV3[C]//Proceedings of the IEEE/CVF International Conference on Computer Vision (ICCV). Washington D. C., USA: IEEE Press, 2019: 1314-1324.
22	ZHANG X Y, ZHOU X Y, LIN M X, et al. ShuffleNet: an extremely efficient convolutional neural network for mobile devices[C]//Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition. Washington D. C., USA: IEEE Press, 2018: 6848-6856.
23	MA N N, ZHANG X Y, ZHENG H T, et al. ShuffleNet V2: practical guidelines for efficient CNN architecture design[C]//Computer Vision - ECCV 2018. Cham, Switzerland: Springer International Publishing, 2018: 122-138.
24	HAN K, WANG Y H, TIAN Q, et al. GhostNet: more features from cheap operations[C]//Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR). Washington D. C., USA: IEEE Press, 2020: 1577-1586.
25	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 (CVPR). Washington D. C., USA: IEEE Press, 2023: 12021-12031.
26	何雨, 田军委, 张震, 等. YOLOv5目标检测的轻量化研究. 计算机工程与应用, 2023, 59 (1): 92- 99.
	HE Y , TIAN J W , ZHANG Z , et al. Lightweight Research of YOLOv5 Target Detection. Computer Engineering and Applications, 2023, 59 (1): 92- 99.
27	张宝朋, 康谦泽, 李佳萌, 等. 轻量化的YOLOv4目标检测算法. 计算机工程, 2022, 48 (8): 206- 214. doi: 10.19678/j.issn.1000-3428.0062216
	ZHANG B P , KANG Q Z , LI J M , et al. Lightweight YOLOv4 target detection algorithm. Computer Engineering, 2022, 48 (8): 206- 214. doi: 10.19678/j.issn.1000-3428.0062216
28	舒腾辉, 丛屾. 基于改进YOLOv5s的轻量化车辆目标检测算法. 黑龙江大学工程学报(中英俄文), 2023, 14 (4): 33- 39.
	SHU T H , CONG S . Lightweight vehicle target detection algorithm based on improved YOLOv5s. Journal of Engineering of Heilongjiang University, 2023, 14 (4): 33- 39.

[1]	王舒梦, 徐慧英, 朱信忠, 黄晓, 宋杰, 李毅. 基于改进YOLOv8n的航拍轻量化小目标检测算法: PECS-YOLO[J]. 计算机工程, 2025, 51(9): 280-293.
[2]	宋杰, 徐慧英, 朱信忠, 黄晓, 陈晨, 王泽宇. 基于YOLOv8改进的跌倒检测算法: OEF-YOLO[J]. 计算机工程, 2025, 51(7): 127-139.
[3]	陈梓延, 王晓龙, 何迪, 安国成. 基于改进YOLOv8的轻量化车辆检测网络[J]. 计算机工程, 2025, 51(5): 314-325.
[4]	许华杰, 郑力文, 张品, 秦远卓. 基于多维注意力模块的轻量化混凝土裂缝检测方法[J]. 计算机工程, 2025, 51(5): 351-360.
[5]	周思瑜, 徐慧英, 朱信忠, 黄晓, 盛轲, 曹雨淇, 陈晨. 基于改进YOLOv8n的手机屏幕瑕疵检测算法: PGS-YOLO[J]. 计算机工程, 2025, 51(5): 326-339.
[6]	王泽宇, 徐慧英, 朱信忠, 黄晓, 梁佳杰, 李琛. 基于改进YOLOv8的轻量化鱼苗检测算法: FD-YOLO[J]. 计算机工程, 2025, 51(4): 327-338.
[7]	袁亚剑, 毛力. 一种增强前景的轻量级交通标志检测模型[J]. 计算机工程, 2025, 51(3): 54-63.
[8]	张元, 吕德芳, 孟建军, 祁文哲. 基于双注意力和GSSN轻量化的钢轨扣件缺陷检测[J]. 计算机工程, 2025, 51(2): 289-299.
[9]	刘圣杰, 何宁, 王鑫, 于海港, 韩文静. 基于轻量级高分辨率网络的人体姿态估计算法[J]. 计算机工程, 2025, 51(2): 278-288.
[10]	谢鑫刚, 芦照烜, 梁静坤. 基于YOLO多路径融合的轻量化水下珊瑚检测算法[J]. 计算机工程, 2025, 51(12): 381-393.
[11]	马荣贵, 黄训燕, 董世浩. 基于改进YOLOv8的实时坑槽检测算法[J]. 计算机工程, 2025, 51(11): 226-234.
[12]	许德刚, 王双臣, 尹柯栋, 王再庆. 改进YOLOv8的城市车辆目标检测算法[J]. 计算机工程, 2025, 51(11): 377-391.
[13]	张倡倡, 吕卫东, 蔡子杰, 刘炎奎. 基于域泛化的轻量化图像分类算法[J]. 计算机工程, 2025, 51(1): 182-189.
[14]	阳丽莎, 李茂军, 胡建文, 王鼎湘. 基于改进YOLOv7-tiny的带钢表面缺陷检测算法[J]. 计算机工程, 2025, 51(1): 208-215.
[15]	饶日昕, 王怡文, 曾砺志, 童心恬, 赵海涛. 面向废旧电缆检测的轻量化网络模型[J]. 计算机工程, 2024, 50(8): 22-30.

选择文件类型/文献管理软件名称

选择包含的内容