基于迁移学习和注意力机制的伪装图像分割

TN957.52; 不同于常规目标,伪装目标特征模糊、尺度信息复杂多变、检测和分割难度更高.在现有伪装数据集基础上,提出了一种结合迁移学习和有效通道注意力的UNet网络伪装图像分割方法.首先,针对伪装目标特征模糊难以有效提取的问题,在UNet的下采样和上采样过程中,引入一种有效通道注意力机制,在不增加网络参数的同时,提高有效区域的特征权重;并将在ImageNet预训练好的视觉几何组(visual geometry group,VGG)系列网络迁移到UNet网络中,实现特征迁移和参数共享,提高模型的泛化能力,降低训练效果对数据集的依赖,减少训练成本;在训练过程中引入FocalLoss函数,增加难...

Full description

Saved in:
Bibliographic Details
Published in系统工程与电子技术 Vol. 44; no. 2; pp. 376 - 384
Main Authors 吴涛, 王伦文, 朱敬成
Format Journal Article
LanguageChinese
Published 国防科技大学电子对抗学院,安徽合肥230037 01.02.2022
Subjects
伪装图像;图像分割;注意力机制;迁移学习
Online AccessGet full text
ISSN1001-506X
DOI10.12305/j.issn.1001-506X.2022.02.03

Cover

Abstract TN957.52; 不同于常规目标,伪装目标特征模糊、尺度信息复杂多变、检测和分割难度更高.在现有伪装数据集基础上,提出了一种结合迁移学习和有效通道注意力的UNet网络伪装图像分割方法.首先,针对伪装目标特征模糊难以有效提取的问题,在UNet的下采样和上采样过程中,引入一种有效通道注意力机制,在不增加网络参数的同时,提高有效区域的特征权重;并将在ImageNet预训练好的视觉几何组(visual geometry group,VGG)系列网络迁移到UNet网络中,实现特征迁移和参数共享,提高模型的泛化能力,降低训练效果对数据集的依赖,减少训练成本;在训练过程中引入FocalLoss函数,增加难挖掘样本权重,提高对困难样本关注度;最后通过解码网络得到分割结果.在CHAMELEON、CAMO和COD10K数据集上进行了测试,相比原始算法,性能指标有显著提升.
AbstractList TN957.52; 不同于常规目标,伪装目标特征模糊、尺度信息复杂多变、检测和分割难度更高.在现有伪装数据集基础上,提出了一种结合迁移学习和有效通道注意力的UNet网络伪装图像分割方法.首先,针对伪装目标特征模糊难以有效提取的问题,在UNet的下采样和上采样过程中,引入一种有效通道注意力机制,在不增加网络参数的同时,提高有效区域的特征权重;并将在ImageNet预训练好的视觉几何组(visual geometry group,VGG)系列网络迁移到UNet网络中,实现特征迁移和参数共享,提高模型的泛化能力,降低训练效果对数据集的依赖,减少训练成本;在训练过程中引入FocalLoss函数,增加难挖掘样本权重,提高对困难样本关注度;最后通过解码网络得到分割结果.在CHAMELEON、CAMO和COD10K数据集上进行了测试,相比原始算法,性能指标有显著提升.
Author 朱敬成
吴涛
王伦文
AuthorAffiliation 国防科技大学电子对抗学院,安徽合肥230037
AuthorAffiliation_xml – name: 国防科技大学电子对抗学院,安徽合肥230037
Author_FL ZHU Jingcheng
WU Tao
WANG Lunwen
Author_FL_xml – sequence: 1
  fullname: WU Tao
– sequence: 2
  fullname: WANG Lunwen
– sequence: 3
  fullname: ZHU Jingcheng
Author_xml – sequence: 1
  fullname: 吴涛
– sequence: 2
  fullname: 王伦文
– sequence: 3
  fullname: 朱敬成
BookMark eNo9j8tKAzEYhbOoYK19C1fCjH-SSTNZSvEGBTcK7komk5QOMgWjeNmJVVoRCy6toCCoCG4UBKvoy0ymfQwrinDgwLc4H2cKFdJWqhGaweBjQoHNJX7T2tTHANhjUNnwCRDiwzi0gIr_eBKVrW1GwDDlDHhQRMJdD7LB-ejrcHj_7p7usrcbd3GWvzzk7Z477edXA9d5HV62s4_H0e2x63-6o57rnLju8zSaMHLT6vJfl9D64sJaddmrrS6tVOdrnsXAqMdYHGnKuZKMhxWiIhwaExOhZRjqUAacGSYwhyAQAhtjAiWwEpQqpjmPVExLaPZ3d1emRqaNetLa2UrHxvredkPtxweJ_TkLBIDSb1pQYwM
ClassificationCodes TN957.52
ContentType Journal Article
Copyright Copyright © Wanfang Data Co. Ltd. All Rights Reserved.
Copyright_xml – notice: Copyright © Wanfang Data Co. Ltd. All Rights Reserved.
DBID 2B.
4A8
92I
93N
PSX
TCJ
DOI 10.12305/j.issn.1001-506X.2022.02.03
DatabaseName Wanfang Data Journals - Hong Kong
WANFANG Data Centre
Wanfang Data Journals
万方数据期刊 - 香港版
China Online Journals (COJ)
China Online Journals (COJ)
DatabaseTitleList
DeliveryMethod fulltext_linktorsrc
Discipline Engineering
DocumentTitle_FL Camouflage image segmentation based on transfer learning and attention mechanism
EndPage 384
ExternalDocumentID xtgcydzjs202202003
GroupedDBID -0Y
2B.
4A8
5XA
5XJ
92E
92I
93N
ABJNI
ACGFS
ALMA_UNASSIGNED_HOLDINGS
CCEZO
CUBFJ
CW9
PSX
TCJ
TGP
U1G
U5S
ID FETCH-LOGICAL-s1053-55dbe377ca57862cb18ffd29ea88e8a475f5917044991fff4c91c933c5e77bcd3
ISSN 1001-506X
IngestDate Thu May 29 04:00:30 EDT 2025
IsPeerReviewed false
IsScholarly true
Issue 2
Keywords 伪装图像;图像分割;注意力机制;迁移学习
Language Chinese
LinkModel OpenURL
MergedId FETCHMERGED-LOGICAL-s1053-55dbe377ca57862cb18ffd29ea88e8a475f5917044991fff4c91c933c5e77bcd3
PageCount 9
ParticipantIDs wanfang_journals_xtgcydzjs202202003
PublicationCentury 2000
PublicationDate 2022-02-01
PublicationDateYYYYMMDD 2022-02-01
PublicationDate_xml – month: 02
  year: 2022
  text: 2022-02-01
  day: 01
PublicationDecade 2020
PublicationTitle 系统工程与电子技术
PublicationTitle_FL Systems Engineering and Electronics
PublicationYear 2022
Publisher 国防科技大学电子对抗学院,安徽合肥230037
Publisher_xml – name: 国防科技大学电子对抗学院,安徽合肥230037
SSID ssib051375074
ssib002263377
ssib001102898
ssib057620160
ssib023168126
ssib023646287
ssj0042237
Score 2.3215091
Snippet TN957.52;...
SourceID wanfang
SourceType Aggregation Database
StartPage 376
Title 基于迁移学习和注意力机制的伪装图像分割
URI https://d.wanfangdata.com.cn/periodical/xtgcydzjs202202003
Volume 44
hasFullText 1
inHoldings 1
isFullTextHit
isPrint
link http://utb.summon.serialssolutions.com/2.0.0/link/0/eLvHCXMwnR1Na9RAdKgtiB7ET_ymYOeYmsxkkpnjzDZLEevFFnorm6-KhxXsFnRvYpUqYqFHKygIKoIXBcEq-mc2bX-G702ym0hFqpcwzLy8z51974U3bwiZymWWYiclR8Ru5vg5E04ss8RR4M0D8HdBZk_Izd0IZhf8a4ticezQZqNqabUXTyf9P54r-R-rwhzYFU_J_oNlR0hhAsZgX3iCheF5IBvTSFDVpkbTyMenjGgkqWlDeEijkOqQGoMweobqwMIoql37FqOyRaOAGk61xIGEoLKNS1JTZXBGtSxmmAGcASJUGsEQT4tqjbQ0p1JYhIaayALzIR5JZWAHihrWDIIRFdBF3iyHysIbYFhYtuFFY6lIKxHQhbGoBFGu5VZjiUbJpB4VJ1tGXGp8KxksmnolRFwV3pZVB6wDorAGKUX27ABoWf2AFOVFo8NPI5BVu7-VmQyln6GRogpUxSrdK6_BKeDzcbK2xn6xYNBGE5VvqSYwYAbTMdgQdjJCpaLOIkvXSg2cgkEkAy1CwueWTW4qf4MVbcK1lzmOHFLZELPaeKzhXXgYNAIVXt6tt88HAg1hnSBSmB5RmEYN2f60vPb9o4rMe73l5H7av72CUG7ZQHeChSGWPkzombnrN-sgG2PSRpIOATzn9WlmhleheXVQjzcUBKxOAoTHIUqtkw5IeBm2ORzGTz4ErPZKpCHjh8nUUKyrfxHKnrrr5p3uciNAnD9OjlWZ3aQut-kJMta_dZIcbfT7PEVU8Wp7sP187-eD3Xffio9vB19fF5vPdj6_31nbKJ5u7bzcLta_7L5YG3z_sPfmUbH1o3i4Uaw_Lp58Ok0W2tF8a9ap7i5xViBj4Y4QaZyBVpIOuMSAJbEn8zxlKutImcmOH4pcKC90fR8StDzP_UR5ieI8EVkYxknKz5Dx7p1udpZM8jyROSQaiWDKB6PLIPbCACL31Eu8LM3PkSuV4EvVf9PK0n5znj8Q1AVypN5IF8l47-5qdgmi7l58ufoZ_ALMIp9f
linkProvider EBSCOhost
openUrl ctx_ver=Z39.88-2004&ctx_enc=info%3Aofi%2Fenc%3AUTF-8&rfr_id=info%3Asid%2Fsummon.serialssolutions.com&rft_val_fmt=info%3Aofi%2Ffmt%3Akev%3Amtx%3Ajournal&rft.genre=article&rft.atitle=%E5%9F%BA%E4%BA%8E%E8%BF%81%E7%A7%BB%E5%AD%A6%E4%B9%A0%E5%92%8C%E6%B3%A8%E6%84%8F%E5%8A%9B%E6%9C%BA%E5%88%B6%E7%9A%84%E4%BC%AA%E8%A3%85%E5%9B%BE%E5%83%8F%E5%88%86%E5%89%B2&rft.jtitle=%E7%B3%BB%E7%BB%9F%E5%B7%A5%E7%A8%8B%E4%B8%8E%E7%94%B5%E5%AD%90%E6%8A%80%E6%9C%AF&rft.au=%E5%90%B4%E6%B6%9B&rft.au=%E7%8E%8B%E4%BC%A6%E6%96%87&rft.au=%E6%9C%B1%E6%95%AC%E6%88%90&rft.date=2022-02-01&rft.pub=%E5%9B%BD%E9%98%B2%E7%A7%91%E6%8A%80%E5%A4%A7%E5%AD%A6%E7%94%B5%E5%AD%90%E5%AF%B9%E6%8A%97%E5%AD%A6%E9%99%A2%2C%E5%AE%89%E5%BE%BD%E5%90%88%E8%82%A5230037&rft.issn=1001-506X&rft.volume=44&rft.issue=2&rft.spage=376&rft.epage=384&rft_id=info:doi/10.12305%2Fj.issn.1001-506X.2022.02.03&rft.externalDocID=xtgcydzjs202202003
thumbnail_s http://utb.summon.serialssolutions.com/2.0.0/image/custom?url=http%3A%2F%2Fwww.wanfangdata.com.cn%2Fimages%2FPeriodicalImages%2Fxtgcydzjs%2Fxtgcydzjs.jpg