Lightweight YOLO models for object detection based on low-rank decomposition

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 l...

Full description

Saved in:
Bibliographic Details
Main Authors: LIN Delü, LIU Chang, CHEN Qi, ZENG Yang, HE Kun
Format: Article
Language:zho
Published: Editorial Department of Electric Drive for Locomotives 2024-01-01
Series:机车电传动
Subjects:
Online Access:http://edl.csrzic.com/thesisDetails#10.13890/j.issn.1000-128X.2024.01.120
Tags: Add Tag
No Tags, Be the first to tag this record!
_version_ 1849728792289869824
author LIN Delü
LIU Chang
CHEN Qi
ZENG Yang
HE Kun
author_facet LIN Delü
LIU Chang
CHEN Qi
ZENG Yang
HE Kun
author_sort LIN Delü
collection DOAJ
description 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.
format Article
id doaj-art-cd7ae7f2e22e4015a7592095a62b7607
institution DOAJ
issn 1000-128X
language zho
publishDate 2024-01-01
publisher Editorial Department of Electric Drive for Locomotives
record_format Article
series 机车电传动
spelling doaj-art-cd7ae7f2e22e4015a7592095a62b76072025-08-20T03:09:25ZzhoEditorial Department of Electric Drive for Locomotives机车电传动1000-128X2024-01-0113814450543555Lightweight YOLO models for object detection based on low-rank decompositionLIN DelüLIU ChangCHEN QiZENG YangHE KunAs 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.http://edl.csrzic.com/thesisDetails#10.13890/j.issn.1000-128X.2024.01.120deep learninglow-rank decompositionautomatic train operationlightweight treatment of modelobject detection
spellingShingle LIN Delü
LIU Chang
CHEN Qi
ZENG Yang
HE Kun
Lightweight YOLO models for object detection based on low-rank decomposition
机车电传动
deep learning
low-rank decomposition
automatic train operation
lightweight treatment of model
object detection
title Lightweight YOLO models for object detection based on low-rank decomposition
title_full Lightweight YOLO models for object detection based on low-rank decomposition
title_fullStr Lightweight YOLO models for object detection based on low-rank decomposition
title_full_unstemmed Lightweight YOLO models for object detection based on low-rank decomposition
title_short Lightweight YOLO models for object detection based on low-rank decomposition
title_sort lightweight yolo models for object detection based on low rank decomposition
topic deep learning
low-rank decomposition
automatic train operation
lightweight treatment of model
object detection
url http://edl.csrzic.com/thesisDetails#10.13890/j.issn.1000-128X.2024.01.120
work_keys_str_mv AT lindelu lightweightyolomodelsforobjectdetectionbasedonlowrankdecomposition
AT liuchang lightweightyolomodelsforobjectdetectionbasedonlowrankdecomposition
AT chenqi lightweightyolomodelsforobjectdetectionbasedonlowrankdecomposition
AT zengyang lightweightyolomodelsforobjectdetectionbasedonlowrankdecomposition
AT hekun lightweightyolomodelsforobjectdetectionbasedonlowrankdecomposition