Multilevel receptive field expansion network for small object detection

Abstract Small object detection remains a bottleneck because there is little visual information about them, especially in the deep layers. To improve the detection performance of small objects, here, Swin Transformer is introduced as the model backbone network to extract rich features of small objec...

Full description

Bibliographic Details
Main Authors: Zhiwei Liu, Menghan Gan, Li Xiong, Xiaofeng Mao, Yue Que
Format: Article
Language:English
Published: Wiley 2023-06-01
Series:IET Image Processing
Subjects:
Online Access:https://doi.org/10.1049/ipr2.12799
_version_ 1797813794765799424
author Zhiwei Liu
Menghan Gan
Li Xiong
Xiaofeng Mao
Yue Que
author_facet Zhiwei Liu
Menghan Gan
Li Xiong
Xiaofeng Mao
Yue Que
author_sort Zhiwei Liu
collection DOAJ
description Abstract Small object detection remains a bottleneck because there is little visual information about them, especially in the deep layers. To improve the detection performance of small objects, here, Swin Transformer is introduced as the model backbone network to extract rich features of small objects. Then, a multilevel receptive field expansion network (MRFENet) is proposed based on the characteristics of different stages in the Swin Transformer. Specifically, a receptive field expansion block (RFEB) is designed to acquire contextual cues and extract detailed information. The RFEB is carefully designed to target the required receptive fields of different layers and further refine the features. MRFENet combined with RFEBs implements the retention of small object context cues and the acquisition of receptive fields for the adaptive detection tasks. Finally, a union loss function is designed to enhance the localization ability. Experiments on the MS COCO dataset demonstrate that the proposed MRFENet has a significant improvement against other state‐of‐the‐art methods, which further validates that MRFENet can effectively utilize small object information.
first_indexed 2024-03-13T07:58:03Z
format Article
id doaj.art-513154c93e3941e6bdec664af4813a83
institution Directory Open Access Journal
issn 1751-9659
1751-9667
language English
last_indexed 2024-03-13T07:58:03Z
publishDate 2023-06-01
publisher Wiley
record_format Article
series IET Image Processing
spelling doaj.art-513154c93e3941e6bdec664af4813a832023-06-02T03:06:38ZengWileyIET Image Processing1751-96591751-96672023-06-011782385239810.1049/ipr2.12799Multilevel receptive field expansion network for small object detectionZhiwei Liu0Menghan Gan1Li Xiong2Xiaofeng Mao3Yue Que4School of Information Engineering East China Jiaotong University Nanchang ChinaSchool of Information Engineering East China Jiaotong University Nanchang ChinaSchool of Information Engineering East China Jiaotong University Nanchang ChinaSchool of Information Engineering East China Jiaotong University Nanchang ChinaSchool of Information Engineering East China Jiaotong University Nanchang ChinaAbstract Small object detection remains a bottleneck because there is little visual information about them, especially in the deep layers. To improve the detection performance of small objects, here, Swin Transformer is introduced as the model backbone network to extract rich features of small objects. Then, a multilevel receptive field expansion network (MRFENet) is proposed based on the characteristics of different stages in the Swin Transformer. Specifically, a receptive field expansion block (RFEB) is designed to acquire contextual cues and extract detailed information. The RFEB is carefully designed to target the required receptive fields of different layers and further refine the features. MRFENet combined with RFEBs implements the retention of small object context cues and the acquisition of receptive fields for the adaptive detection tasks. Finally, a union loss function is designed to enhance the localization ability. Experiments on the MS COCO dataset demonstrate that the proposed MRFENet has a significant improvement against other state‐of‐the‐art methods, which further validates that MRFENet can effectively utilize small object information.https://doi.org/10.1049/ipr2.12799neural netsobject detectionreceptive field expansionsmall object detectiontransformer
spellingShingle Zhiwei Liu
Menghan Gan
Li Xiong
Xiaofeng Mao
Yue Que
Multilevel receptive field expansion network for small object detection
IET Image Processing
neural nets
object detection
receptive field expansion
small object detection
transformer
title Multilevel receptive field expansion network for small object detection
title_full Multilevel receptive field expansion network for small object detection
title_fullStr Multilevel receptive field expansion network for small object detection
title_full_unstemmed Multilevel receptive field expansion network for small object detection
title_short Multilevel receptive field expansion network for small object detection
title_sort multilevel receptive field expansion network for small object detection
topic neural nets
object detection
receptive field expansion
small object detection
transformer
url https://doi.org/10.1049/ipr2.12799
work_keys_str_mv AT zhiweiliu multilevelreceptivefieldexpansionnetworkforsmallobjectdetection
AT menghangan multilevelreceptivefieldexpansionnetworkforsmallobjectdetection
AT lixiong multilevelreceptivefieldexpansionnetworkforsmallobjectdetection
AT xiaofengmao multilevelreceptivefieldexpansionnetworkforsmallobjectdetection
AT yueque multilevelreceptivefieldexpansionnetworkforsmallobjectdetection