SIA OpenIR  > 光电信息技术研究室
Adaptive feature fusion with attention mechanism for multi-scale target detection
Ju MR(鞠默然)1,2,2,3,4; Luo JN(罗江宁)5; Wang ZB(王仲博)1,2,2,3,4; Luo HB(罗海波)1,2,2,4
Department光电信息技术研究室
Source PublicationNeural Computing and Applications
ISSN0941-0643
2020
Pages1-13
Indexed BySCI ; EI
EI Accession number20202808929412
WOS IDWOS:000547247700004
Contribution Rank1
KeywordDeep learning Target detection Adaptive feature fusion Attention mechanism
Abstract

To detect the targets of different sizes, multi-scale output is used by target detectors such as YOLO V3 and DSSD. To improve the detection performance, YOLO V3 and DSSD perform feature fusion by combining two adjacent scales. However, the feature fusion only between the adjacent scales is not sufficient. It hasn’t made advantage of the features at other scales. What is more, as a common operation for feature fusion, concatenating can’t provide a mechanism to learn the importance and correlation of the features at different scales. In this paper, we propose adaptive feature fusion with attention mechanism (AFFAM) for multi-scale target detection. AFFAM utilizes pathway layer and subpixel convolution layer to resize the feature maps, which is helpful to learn better and complex feature mapping. In addition, AFFAM utilizes global attention mechanism and spatial position attention mechanism, respectively, to learn the correlation of the channel features and the importance of the spatial features at different scales adaptively. Finally, we combine AFFAM with YOLO V3 to build an efficient multi-scale target detector. The comparative experiments are conducted on PASCAL VOC dataset, KITTI dataset and Smart UVM dataset. Compared with the state-of-the-art target detectors, YOLO V3 with AFFAM achieved 84.34% mean average precision (mAP) at 19.9 FPS on PASCAL VOC dataset, 87.2% mAP at 21 FPS on KITTI dataset and 99.22% mAP at 20.6 FPS on Smart UVM dataset which outperforms other advanced target detectors.

Language英语
WOS SubjectComputer Science, Artificial Intelligence
WOS KeywordRECOGNITION
WOS Research AreaComputer Science
Citation statistics
Document Type期刊论文
Identifierhttp://ir.sia.cn/handle/173321/27369
Collection光电信息技术研究室
Corresponding AuthorLuo HB(罗海波)
Affiliation1.Shenyang Institute of Automation, Chinese Academy of Sciences, Shenyang
2.Liaoning 110016, China
3.Institutes for Robotics and Intelligent Manufacturing, Chinese Academy of Sciences, Shenyang
4.University of Chinese Academy of Sciences, Beijing
5.100049, China
6.Key Laboratory of Opt-Electronic Information Processing, Chinese Academy of Sciences, Shenyang
7.The Key Laboratory of Image Understanding and Computer Vision, Shenyang Liaoning
8.110016, China
9.McGill University, Montreal, QC H3A 0G4, Canada
Recommended Citation
GB/T 7714
Ju MR,Luo JN,Wang ZB,et al. Adaptive feature fusion with attention mechanism for multi-scale target detection[J]. Neural Computing and Applications,2020:1-13.
APA Ju MR,Luo JN,Wang ZB,&Luo HB.(2020).Adaptive feature fusion with attention mechanism for multi-scale target detection.Neural Computing and Applications,1-13.
MLA Ju MR,et al."Adaptive feature fusion with attention mechanism for multi-scale target detection".Neural Computing and Applications (2020):1-13.
Files in This Item:
File Name/Size DocType Version Access License
Adaptive feature fus(14426KB)期刊论文出版稿开放获取CC BY-NC-SAView Application Full Text
Related Services
Recommend this item
Bookmark
Usage statistics
Export to Endnote
Google Scholar
Similar articles in Google Scholar
[Ju MR(鞠默然)]'s Articles
[Luo JN(罗江宁)]'s Articles
[Wang ZB(王仲博)]'s Articles
Baidu academic
Similar articles in Baidu academic
[Ju MR(鞠默然)]'s Articles
[Luo JN(罗江宁)]'s Articles
[Wang ZB(王仲博)]'s Articles
Bing Scholar
Similar articles in Bing Scholar
[Ju MR(鞠默然)]'s Articles
[Luo JN(罗江宁)]'s Articles
[Wang ZB(王仲博)]'s Articles
Terms of Use
No data!
Social Bookmark/Share
File name: Adaptive feature fusion with attention mechanism for multi-scale target detection.pdf
Format: Adobe PDF
All comments (0)
No comment.
 

Items in the repository are protected by copyright, with all rights reserved, unless otherwise indicated.