Multi-Scale Feature Selective Matching Network for Object Detection
Numerous deep learning-based object detection methods have achieved excellent performance. However, the performance on small-size object detection and positive and negative sample imbalance problems is not satisfactory. We propose a multi-scale feature selective matching network (MFSMNet) to improve...
| Main Authors: | , , , |
|---|---|
| Format: | Article |
| Language: | English |
| Published: |
MDPI AG
2023-06-01
|
| Series: | Mathematics |
| Subjects: | |
| Online Access: | https://www.mdpi.com/2227-7390/11/12/2655 |
| _version_ | 1827736585036103680 |
|---|---|
| author | Yuanhua Pei Yongsheng Dong Lintao Zheng Jinwen Ma |
| author_facet | Yuanhua Pei Yongsheng Dong Lintao Zheng Jinwen Ma |
| author_sort | Yuanhua Pei |
| collection | DOAJ |
| description | Numerous deep learning-based object detection methods have achieved excellent performance. However, the performance on small-size object detection and positive and negative sample imbalance problems is not satisfactory. We propose a multi-scale feature selective matching network (MFSMNet) to improve the performance of small-size object detection and alleviate the positive and negative sample imbalance problems. First, we construct a multi-scale semantic enhancement module (MSEM) to compensate for the information loss of small-sized targets during down-sampling by obtaining richer semantic information from features at multiple scales. Then, we design the anchor selective matching (ASM) strategy to alleviate the training dominated by negative samples caused by the imbalance of positive and negative samples, which converts the offset values of the localization branch output in the detection head into localization scores and reduces negative samples by discarding low-quality anchors. Finally, a series of quantitative and qualitative experiments on the Microsoft COCO 2017 and PASCAL VOC 2007 + 2012 datasets show that our method is competitive compared to nine other representative methods. MFSMNet runs on a GeForce RTX 3090. |
| first_indexed | 2024-03-11T02:12:25Z |
| format | Article |
| id | doaj.art-73d83101bbb544c8b996843c8ff486b4 |
| institution | Directory Open Access Journal |
| issn | 2227-7390 |
| language | English |
| last_indexed | 2024-03-11T02:12:25Z |
| publishDate | 2023-06-01 |
| publisher | MDPI AG |
| record_format | Article |
| series | Mathematics |
| spelling | doaj.art-73d83101bbb544c8b996843c8ff486b42023-11-18T11:27:57ZengMDPI AGMathematics2227-73902023-06-011112265510.3390/math11122655Multi-Scale Feature Selective Matching Network for Object DetectionYuanhua Pei0Yongsheng Dong1Lintao Zheng2Jinwen Ma3School of Information Engineering, Henan University of Science and Technology, Luoyang 471023, ChinaSchool of Information Engineering, Henan University of Science and Technology, Luoyang 471023, ChinaSchool of Information Engineering, Henan University of Science and Technology, Luoyang 471023, ChinaDepartment of Information and Computational Sciences, School of Mathematical Sciences and LMAM, Peking University, Beijing 100871, ChinaNumerous deep learning-based object detection methods have achieved excellent performance. However, the performance on small-size object detection and positive and negative sample imbalance problems is not satisfactory. We propose a multi-scale feature selective matching network (MFSMNet) to improve the performance of small-size object detection and alleviate the positive and negative sample imbalance problems. First, we construct a multi-scale semantic enhancement module (MSEM) to compensate for the information loss of small-sized targets during down-sampling by obtaining richer semantic information from features at multiple scales. Then, we design the anchor selective matching (ASM) strategy to alleviate the training dominated by negative samples caused by the imbalance of positive and negative samples, which converts the offset values of the localization branch output in the detection head into localization scores and reduces negative samples by discarding low-quality anchors. Finally, a series of quantitative and qualitative experiments on the Microsoft COCO 2017 and PASCAL VOC 2007 + 2012 datasets show that our method is competitive compared to nine other representative methods. MFSMNet runs on a GeForce RTX 3090.https://www.mdpi.com/2227-7390/11/12/2655deep learningobject detectionselective matchtingpositive and negative sample imbalance |
| spellingShingle | Yuanhua Pei Yongsheng Dong Lintao Zheng Jinwen Ma Multi-Scale Feature Selective Matching Network for Object Detection Mathematics deep learning object detection selective matchting positive and negative sample imbalance |
| title | Multi-Scale Feature Selective Matching Network for Object Detection |
| title_full | Multi-Scale Feature Selective Matching Network for Object Detection |
| title_fullStr | Multi-Scale Feature Selective Matching Network for Object Detection |
| title_full_unstemmed | Multi-Scale Feature Selective Matching Network for Object Detection |
| title_short | Multi-Scale Feature Selective Matching Network for Object Detection |
| title_sort | multi scale feature selective matching network for object detection |
| topic | deep learning object detection selective matchting positive and negative sample imbalance |
| url | https://www.mdpi.com/2227-7390/11/12/2655 |
| work_keys_str_mv | AT yuanhuapei multiscalefeatureselectivematchingnetworkforobjectdetection AT yongshengdong multiscalefeatureselectivematchingnetworkforobjectdetection AT lintaozheng multiscalefeatureselectivematchingnetworkforobjectdetection AT jinwenma multiscalefeatureselectivematchingnetworkforobjectdetection |