受制于嵌入式平台的性能和资源制约,基于深度学习的车辆检测算法在部署时面临网络参数量过大、模型复杂、移植困难等问题。提出一种基于MobileNetv3网络的YOLOv3改进目标检测算法,使用轻量级MobileNetv3网络替换传统主干特征提取网络Darknet53,修改FPN特征金字塔为FPN+PAN结构,同时引入注意力机制以提高算法的检测精度。在计算机平台和瑞芯微RV1126嵌入式平台上的实验结果表明,改进后的YOLOv3算法模型减小50%,检测精度提升0.85%,推理时间缩短50%。
Abstract
Restricted by the performance and resources of embedded platform,the vehicle detection algorithm based on deep learning is faced with problems such as large number of network parameters,complex model and difficult migration during deployment.In this paper,an improved target detection algorithm of YOLOv3 based on MobileNetv3 network is proposed.The light-weight MobileNetv3 network is used to replace the traditional backbone feature extraction network Darknet53,the FPN feature pyramid is modified to FPN+PAN structure,and the attention mechanism is introduced to improve the detection accuracy of the algorithm.The experimental results on computer platform and RV1126 embedded platform show that the improved YOLOv3 algorithm reduces the model size by 50%,improves the detection accuracy by 0.85%,and improves the inference time by 50%.
关键词
MobileNetv3 /
YOLOv3 /
实时车辆检测 /
注意力机制
Key words
MobileNetv3 /
YOLOv3 /
real-time vehicle detection /
attention mechanism
{{custom_sec.title}}
{{custom_sec.title}}
{{custom_sec.content}}
参考文献
[1] Redmon J,Divvala S,Girshick R,et al.You Only Look Once: Unified, Real-Time Object Detection[C]//Computer Vision & Pattern Recognition.IEEE,2016.
[2] 顾晋,罗素云.基于改进的YOLOv3车辆检测方法[J].农业装备与车辆工程,2021,59(7):6.
[3] 王硕,王孝兰,王岩松,等.基于改进tiny-YOLOv3的车辆检测方法[J].计算机与数字工程,2021,49(8):6.
[4] 王玲,张松,王鹏,等.基于MS-YOLOv3的车辆目标实时检测算法[J].计算机应用与软件,2021,38(10):7.
[5] Howard A G,Zhu M,Chen B,et al.MobileNets:Efficient Convolutional Neural Networks for Mobile Vision Applications[EB/OL].[2022-11].https://blog.csdn.net/weixin_36670529/article/details/102768196.
[6] 张达为,刘绪崇,周维,等.基于改进YOLOv3的实时交通标志检测算法[J].计算机应用,2022,42(7):8.
[7] Redmon J,Farhadi A.YOLOv3:An Incremental Improvement[J].arXiv e-prints,2018.
[8] He K,Zhang X,Ren S,et al.Deep Residual Learning for Image Recognition[C]//2016 IEEE Conference on Computer Vision and Pattern Recognition (CVPR).IEEE,2016.
[9] Redmon J,Farhadi A.YOLO9000:Better, Faster, Stronger[C]//IEEE Conference on Computer Vision & Pattern Recognition.IEEE,2017:6517-6525.
[10] Howard A,Sandler M,Chu G,et al.Searching for MobileNetV3[EB/OL].[2022-11].https://blog.csdn.net/hzhj2007/article/details/79731888.
[11] Jie H,Li S,Gang S,et al.Squeeze-and-Excitation Networks[J].IEEE Transactions on Pattern Analysis and Machine Intelligence,2017(99).
[12] Liu S,Qi L,Qin H,et al.Path Aggregation Network for Instance Segmentation[C]//2018 IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR).IEEE,2018.
基金
*辽宁省自然科学基金(2015020103)。
PDF(2018 KB)
