On Image Ｒecognition Using Bidirectional Feature Pyramid and Deep Neural Network
Object recognition is one of the fundamental tasks in the area of computer vision. The development of deep neural networks advances the object recognition. Nonetheless，multi-scale object recognition still remains to be a challenging task. The feature pyramid is a promising technology to address the...
Saved in:
| Main Authors: | , |
|---|---|
| Format: | Article |
| Language: | zho |
| Published: |
Harbin University of Science and Technology Publications
2021-04-01
|
| Series: | Journal of Harbin University of Science and Technology |
| Subjects: | |
| Online Access: | https://hlgxb.hrbust.edu.cn/#/digest?ArticleID=1940 |
| Tags: |
Add Tag
No Tags, Be the first to tag this record!
|
| Summary: | Object recognition is one of the fundamental tasks in the area of computer vision. The development of deep neural networks advances the object recognition. Nonetheless，multi-scale object recognition still remains to be a challenging task. The feature pyramid is a promising technology to address the multi-scale object recognition. However，the existing feature pyramid-based object recognition schemes usually employed a top-down pathway， which cannot improve the recognition of large-scale objects. To address this issue，a novel bidirectional enhanced feature pyramid-based object recognition scheme is proposed. The proposed scheme can improve the precisions of both large-scale and small-scale object recognition by enabling the semantic information enhancement from both top to down and down to top. The experiment results showed that the proposed scheme can improve the mean average precision by at least 0. 7% in PASCAL VOC dataset and outperformed all the baselines in MS COCO dataset. These findings verified the effectiveness of the proposed scheme. |
|---|---|
| ISSN: | 1007-2683 |