作者：张青春, 王文聘, 张洪源, 张恩溥, 宁建峰

作者单位：淮阴工学院自动化学院，江苏 淮安 223001

关键词：障碍物检测;轻量化;YOLOv7-tiny;FasterNet;PConv卷积;CA注意力机制

摘要：

针对道路限高障碍物检测困难、模型复杂以及难以在嵌入式端部署等问题，提出一种基于改进轻量化YOLOv7-tiny模型的道路限高障碍物检测方法。改进模型采用更加轻量的FasterNet网络替换原有主干网络，在Neck层使用PConv卷积替代部分Conv卷积，以减少计算冗余和内存访问，从而有效降低模型的参数量和计算量。同时，引入CA注意力机制提高检测精度，并使用Focal-EIoU损失函数优化模型的收敛速度和效率。实验结果表明：相较于YOLOv7-tiny目标检测模型，改进模型在检测数据集上，mAP@0.5提高6.6%，参数量和计算量分别降低24%和20.5%，模型权重文件减少27.2%，能够在保持较高检测精度的同时，满足轻量化的需求。

Improved lightweight YOLOv7-tiny method for road height limitation obstacle detection
ZHANG Qingchun, WANG Wenpin, ZHANG Hongyuan, ZHANG Enpu, NING Jianfeng
Automation Faculty, Huaiyin Institute of Technology, Huaian 223001, China
Abstract: In response to the challenges of difficult detection of road height restrictions, complex and bulky models, and difficulties in deploying on embedded devices, a road height obstacle detection method based on an improved lightweight YOLOv7-tiny model is proposed. The improved model utilizes a lighter FasterNet network instead of the original backbone network and incorporates PConv convolution in the Neck layer to reduce computational redundancy and memory access, effectively reducing the model's parameter and computation complexity. The introduction of the CA attention mechanism enhances the detection accuracy, and the Focal-EIoU loss function optimizes the model's convergence speed and efficiency. Experimental results showed that compared to the YOLOv7-tiny object detection model, the improved model achieved a 6.6% increase in mAP@0.5 on the detection dataset, reduced parameters and computations by 24% and 20.5% respectively, and reduced the model weight file by 27.2%. This improved model successfully meets the lightweight requirement while maintaining a high detection accuracy.
Keywords: obstacle detection;lightweight;YOLOv7-tiny;FasterNet;PConv convolution;CA attention mechanism
2024, 50(5):186-192  收稿日期: 2023-07-28;收到修改稿日期: 2023-09-17
基金项目: 江苏省研究生科研与实践创新计划项目（SJCX21_1507）
作者简介: 张青春(1964-),男,江苏盱眙县人,教授,研究方向为智能检测技术、物联网应用技术、移动机器人、虚拟仪器技术等。
参考文献
[1] 石镇岳, 侯婷, 苏勇东. 改进YOLOv7的交通标志检测算法[J]. 计算机系统应用, 2023, 32 (10): 157-165.
SHI Z Y, HOU T, SU Y D. Improved YOLOv7 algorithm for traffic sign detection[J]. Computer Systems ＆ Applications, 2023, 32 (10): 157-165.
[2] 邵延华, 张铎, 楚红雨, 等. 基于深度学习的YOLO目标检测综述[J]. 电子与信息学报, 2022, 44(10): 3697-3708.
SHAN Y H, ZHANG D, CHU H Y, et al. A review of YOLO object detection based on deep learning[J]. Journal of Electronics & Information Technology, 2022, 44(10): 3697-3708.
[3] 刘浩翰, 樊一鸣, 贺怀清, 等. 改进YOLOv7-tiny的目标检测轻量化模型[J]. 计算机工程与应用,  2023, 59 (14): 166-175.
LIU H H, FAN Y M, HE H Q, et al. Improved YOLOv7-tiny's object detection lightweight model[J]. Computer Engineering and Applications, 2023, 59 (14): 166-175.
[4] WEN H, HONG G, BING N Y, et al. Tool wear state recognition under imbalanced data based on WGAN-GP and lightweight neural network ShuffleNet[J]. Journal of Mechanical Science and Technology, 2022, 36(10): 4993-5009.
[5] LI H, LI J, WEI H, et al. Slim-neck by GSConv: A better design paradigm of detector architectures for autonomous vehicles[J]. arXiv preprint arXiv: 2206.02424, 2022.
[6] 章芮宁, 闫坤, 叶进. 轻量化YOLO-v7的数显仪表检测及读数[J]. 计算机工程与应用, 2024, 60 (8): 192-201.
ZHANG R N, YAN K, YE J. Lightweight YOLO-v7 for digital instrumentation detection and reading[J]. Computer Engineering and Applications, 2024, 60 (8): 192-201.
[7] 秦子豪, 雷鸣, 宋文广, 等. 基于轻量化深度学习模型的安全帽检测方法[J]. 科学技术与工程, 2022, 22(14): 5659-5665.
QIN Z H, LEI M, SONG W G, et al. Helmet detection method based on lightweight deep learning model[J]. Science Technology and Engineering , 2022, 22(14): 5659-5665.
[8] 吴珺, 董佳明, 刘欣, 等. 注意力优化的轻量目标检测网络及应用[J]. 智能系统学报, 2023, 18 (3): 506-516.
WU J, DONG J M, LIU X, et al. Lightweight object detection network and its application based on the attention optimization[J]. CAAI Transactions on Intelligent Systems, 2023, 18 (3): 506-516.
[9] 梁秀满, 肖寒. 基于SDD-YOLO的轻量级带钢缺陷实时检测算法[J/OL]. 中国测试, 1-8[2023-05-26]. http://kns.cnki.net/kcms/detail/51.1714.TB.20230109.1648.002.html.
LIANG X M, XIAO H. Lightweight strip defect real-time detection algorithm based on SDD-YOLO[J/OL]. China Measurement & Test, 1-8[2025-05-26]. http://kns.cnki.net/kcms/detail/51.1714.TB.20230109.1648.002.html.
[10] 王淑青, 鲁濠, 鲁东林, 等. 基于轻量化人工神经网络的PCB板缺陷检测[J]. 仪表技术与传感器, 2022(5): 98-104.
WANG S J, LU H, LU D L, et al. PCB board defect detection based on lightweight artificial neural network[J]. Instrument Technique and Sensor, 2022(5): 98-104.
[11] CHEN J, KAO S, 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, 2023.
[12] LI B, CHEN Y H, XU H. Fast vehicle detection algorithm based on lightweight YOLO7-tiny[J]. arXiv preprint arXiv: 2304.06002, 2023.
[13] 季堂煜, 赵倩, 余文涛, 等. 基于增强小目标特征提取的PCB板缺陷检测模型[J]. 仪表技术与传感器, 2023(4): 87-92.
JI T Y, ZHAO Q, YU W T, et al. PCB board defect detection model based on enhanced small-target feature extraction[J]. Instrument Technique and Sensor , 2023(4): 87-92.
[14] HOU Q, ZHOU D, FENG J . Coordinate Attention for efficient mobile Network design[J]. Neurocomputing, 2021: 13713-13722.
[15] ZHANG Y F, REN W, ZHANG Z, et al. Focal and efficient IOU loss for accurate bounding box regression[J]. Neurocomputing, 2022, 506: 146-157.
[16] 张青春, 吴峥, 周玲, 等. 基于改进 YOLOv5 车辆与行人目标识别方法的研究[J]. 中国测试, 2023, 49 (7): 82-88.
ZHANG Q C, WU Z, ZHOU L, et al. Research on vehicle and pedestrian target detection method based on improved YOLOv5[J].China Measurement ＆ Test, 2023, 49 (7): 82-88.