基于低秩分解的YOLO轻量化目标检测模型

林德铝; 刘畅; 陈琦; 曾阳; 何琨

doi:10.13890/j.issn.1000-128X.2024.01.120

您当前的位置：

首页 >

文章列表页 >

基于低秩分解的YOLO轻量化目标检测模型

智能化技术 | 更新时间：2024-08-02

- 基于低秩分解的YOLO轻量化目标检测模型
- Lightweight YOLO models for object detection based on low-rank decomposition
- 机车电传动 2024年第1期页码：138-144
- 作者机构：
  
  1.华中科技大学计算机科学与技术学院，湖北武汉 430074
  2.中车株洲电力机车研究所有限公司，湖南株洲 412001
- 作者简介：
  
  何　琨（1972—），女，博士，教授，博士生导师，主要从事深度学习、机器学习、算法设计与优化方面的研究； E-mail: brooklet60@hust.edu.cn
- 基金信息：
  
  国家自然科学基金项目(62076105)
- DOI：10.13890/j.issn.1000-128X.2024.01.120
  中图分类号： TP18
- 纸质出版日期：2024-01-10，
  
  收稿日期：2023-08-21，
  
  修回日期：2024-01-01，
- 稿件说明：
扫描看全文
林德铝, 刘畅, 陈琦, 等. 基于低秩分解的YOLO轻量化目标检测模型[J]. 机车电传动, 2024(1): 138-144.

LIN Delü, LIU Chang, CHEN Qi, et al. Low-rank decomposition based lightweight YOLO model for the object detection[J]. Electric drive for locomotives,2024(1): 138-144.
林德铝, 刘畅, 陈琦, 等. 基于低秩分解的YOLO轻量化目标检测模型[J]. 机车电传动, 2024(1): 138-144. DOI：10.13890/j.issn.1000-128X.2024.01.01.120

LIN Delü, LIU Chang, CHEN Qi, et al. Low-rank decomposition based lightweight YOLO model for the object detection[J]. Electric drive for locomotives,2024(1): 138-144. DOI：10.13890/j.issn.1000-128X.2024.01.120.

摘要

随着列车智能化程度的不断提高，许多研究探索了车载设备目标检测模型的轻量化技术，以满足在资源有限情况下的高效计算。针对当前YOLO（You Only Look Once）系列目标检测模型轻量化方法通用性不够强的问题，文章提出了一种针对YOLO系列的低秩分解参数压缩算法。首先通过预设的低秩比例系数和卷积单元的输入/输出通道数量计算低秩，然后通过对目标结构的卷积层进行Tucker分解，得到新的卷积序列，最后融合新的卷积序列，取代原有卷积层。使用公开数据集对所提出的基于低秩分解的参数压缩方法进行试验，选用了YOLOv5-l、YOLOv8-x和YOLOX-x这3种模型，在保证低秩分解后的模型检测平均精度为原模型96%的前提下，模型参数量和浮点计算量均减少了约40%，同时图像检测速度能达到原模型的150%左右。此外，可视化结果显示，该方法压缩过的模型与原模型在相同图像上的关注区域基本相同。试验结果表明，文章提出的方法可以有效地对单阶段YOLO系列目标检测模型进行轻量化压缩，提高模型在车载设备上的可用性；同时，所做工作对轨道交通领域自动驾驶场景下的其他模型的轻量化处理也具有重要的借鉴意义。

Abstract

As train intelligence continues to advance

numerous studies have emerged to explore lightweight techniques for object detection models on onboard equipment

for the purpose of improving calculation efficiency amidst limited resources. This paper proposed a parameter compression algorithm based on low-rank decomposition for the You Only Look Once (YOLO) series of object detection models

aiming to overcome the limited versatility of current lightweight treatment methods for these models. Initially

calculations were conducted to determine a low rank using the preset low-rank ratio coefficient along with the number of input and output channels of convolution units. Subsequently

a new convolution sequence was obtained by performing the Tucker decomposition to the convolutional layer of the target structure. Lastly

a new convolution sequence was fused to replace the original convolutional layer. Experiments were conducted on the proposed parameter compression method based on low-rank decomposition

using public datasets and three models

i.e.

YOLOv5-l

YOLOv8-x

and YOLOX-x. While ensuring an average detection accuracy of the models based on low-rank decomposition over 96% compared to the original models

the number of parameters and float-point calculations of these models reduced by about 40%. The experiments resulted in a nearly 50% improvement in image detection speed. Further visual displays illustrated similar receptive fields on the same images between the compressed and original models. The experimental results show that the proposed method is effective in lightweight compression of single-stage object detection models of the YOLO series and can enhance their usability for onboard equipment. In addition

this study serves as a valuable reference for the lightweight treatment of other models in the automatic train operation scenarios of rail transit field.

关键词

深度学习低秩分解轨道交通自动驾驶模型轻量化目标检测

Keywords

deep learninglow-rank decompositionautomatic train operationlightweight treatment of modelobject detection

references

REN Shaoqing, HE Kaiming, GIRSHICK R, et al. Faster R-CNN: towards real-time object detection with region proposal networks[J]. IEEE transactions on pattern analysis and machine intelligence, 2017, 39(6): 1137-1149.

贺德强, 江洲, 陈基永, 等. 基于深度卷积神经网络的铁路接触网鸟窝检测方法研究[J]. 机车电传动, 2019(4): 126-130.

HE Deqiang, JIANG Zhou, CHEN Jiyong, et al. Research on detection of bird nests in overhead catenary based on deep convolutional neural network[J]. Electric drive for locomotives, 2019(4): 126-130.

林珊, 邓树, 谌小莉, 等. 基于Mask-RCNN的牵引供电设备红外智能诊断[J]. 机车电传动, 2022(4): 55-61.

LIN Shan, DENG Shu, CHEN Xiaoli, et al. Intelligent infrared diagnosis method for traction power supply equipment based on Mask-RCNN[J]. Electric drive for locomotives, 2022(4): 55-61.

REDMON J, DIVVALA S, GIRSHICK R, et al. You only look once: unified, real-time object detection[C]//IEEE. 2016 IEEE Conference on Computer Vision and Pattern Recognition (CVPR). Las Vegas: IEEE, 2016: 779-788.

REDMON J, FARHADI A. YOLO9000: better, faster, stronger[C]//IEEE. 2017 IEEE Conference on Computer Vision and Pattern Recognition (CVPR). Honolulu: IEEE, 2017: 6517-6525.

REDMON J, FARHADI A. YOLOv3: an incremental improvement[EB/OL]. (2018-04-08) [2023-07-10]. https://arxiv.org/abs/1804.02767https://arxiv.org/abs/1804.02767.

BOCHKOVSKIY A, WANG C Y, LIAO H Y M. YOLOv4: optimal speed and accuracy of object detection[EB/OL]. (2020-04-23) [2023-07-10]. https://arxiv.org/abs/2004.10934https://arxiv.org/abs/2004.10934.

JOCHER G, STOKEN A, BOROVEC J, et al. Ultralytics/yolov5: v4.0-nn.SiLU()activations, weights & biases logging, pytorch hub integration[EB/OL]. (2021-01-05) [2023-10-22]. https://zenodo.org/record/4418161https://zenodo.org/record/4418161.

GE Zheng, LIU Songtao, WANG Feng, et al. YOLOX: exceeding YOLO series in 2021[EB/OL]. (2021-08-06) [2023-10-13]. https://arxiv.org/abs/2107.08430https://arxiv.org/abs/2107.08430.

林军, 潘文波, 游俊, 等. 基于多传感器融合的视觉感知困难样本挖掘方法[J]. 机车电传动, 2021(6): 93-99.

LIN Jun, PAN Wenbo, YOU Jun, et al. Hard example mining method for visual perception based on multi-sensor fusion[J]. Electric drive for locomotives, 2021(6): 93-99.

周兆安, 李树枝. 基于深度学习的弓网异常状态在线检测研究[J]. 机车电传动, 2022(5): 135-143.

ZHOU Zhaoan, LI Shuzhi. Research on online inspection of pantograph and catenary based on deep learning[J]. Electric drive for locomotives, 2022(5): 135-143.

李昂, 孙士杰, 张朝阳, 等. 改进YOLOv5s的轨道障碍物检测模型轻量化研究[J]. 计算机工程与应用, 2023, 59(4): 197-207.

LI Ang, SUN Shijie, ZHANG Chaoyang, et al. Research on lightweight of improved YOLOv5s track obstacle detection model[J]. Computer engineering and applications, 2023, 59(4): 197-207.

卢艳东, 李积英, 王筱婷. 一种基于改进YOLOv3-tiny的轻量级轨道紧固件检测算法[J]. 铁道标准设计, 2022, 66(7): 41-47.

LU Yandong, LI Jiying, WANG Xiaoting. A lightweight track fastener detection algorithm based on improved YOLOv3-tiny[J]. Railway standard design, 2022, 66(7): 41-47.

WANG C Y, MARK LIAO H Y, WU Y H, et al. CSPNet: a new backbone that can enhance learning capability of CNN[C]//IEEE. 2020 IEEE/CVF Conference on Computer Vision and Pattern Recognition Workshops (CVPRW). Seattle: IEEE, 2020: 1571-1580.

LIN T Y, MAIRE M, BELONGIE S, et al. Microsoft COCO: common objects in context[C]//Springer. Computer Vision-ECCV 2014. Zurich: Springer, 2014: 740-755.

TAGIEW R, KÖPPEL M, SCHWALBE K, et al. OSDaR23: open sensor data for rail 2023[EB/OL]. (2023-05-04) [2023-10-13]. https://arxiv.org/abs/2305.03001https://arxiv.org/abs/2305.03001.

浏览量

下载量

CSCD

CNKI被引量

文章被引用时，请邮件提醒。

提交

工具集

关联资源

基于三维点云与图像融合的轨道交通场景行人检测方法

轨道交通车载端到端语音合成

基于改进yolov4-tiny的地铁受电弓燃弧检测方法

基于深度学习的弓网异常状态在线检测研究

基于Mask-RCNN的牵引供电设备红外智能诊断