您的位置:山东大学 -> 科技期刊社 -> 《山东大学学报(工学版)》

山东大学学报 (工学版) ›› 2022, Vol. 52 ›› Issue (6): 115-122.doi: 10.6040/j.issn.1672-3961.0.2022.087

• 机器学习与数据挖掘 • 上一篇    

面向小样本目标检测任务的自适应特征重构算法

刘丁菠1,2,刘学艳2,3,于东然2,4,杨博2,3,李伟5*   

  1. 1.吉林大学软件学院, 吉林 长春 130012;2.吉林大学符号计算与知识工程教育部重点实验室, 吉林 长春 130012;3.吉林大学计算机科学与技术学院, 吉林 长春 130012;4.吉林大学人工智能学院, 吉林 长春 130012;5.吉林大学商学与管理学院, 吉林 长春 130012
  • 发布日期:2022-12-23
  • 作者简介:刘丁菠(1997— ),女,吉林珲春人,硕士研究生,主要研究方向为计算机视觉. E-mail:dbliujlu@126.com. *通信作者简介:李伟(1977— ),女,辽宁朝阳人,讲师,博士,主要研究方向为最优化理论. E-mail:W_li@jlu.edu.cn
  • 基金资助:
    国家自然科学基金项目(62172185、61876069);国家重点研发计划项目(2021ZD0112501、2021ZD0112502);吉林省重点科技研发项目(20180201067GX、20180201044GX);吉林省自然科学基金项目(20200201036JC)

Adaptive feature reconstruction algorithm for few-shot object detection

LIU Dingbo1,2, LIU Xueyan2,3, YU Dongran2,4, YANG Bo2,3, LI Wei5*   

  1. 1. College of Software, Jilin University, Changchun 130012, Jilin, China;
    2. Key Laboratory of Symbolic Computation and Knowledge Engineering of Ministry of Education, Jilin University, Changchun 130012, Jilin, China;
    3. College of Computer Science and Technology, Jilin University, Changchun 130012, Jilin, China;
    4. School of Artificial Intelligence, Jilin University, Changchun 130012, Jilin, China;
    5. School of Business and Management, Jilin University, Changchun 130012, Jilin, China
  • Published:2022-12-23

摘要: 针对目标检测任务中样本量不足时新类别检测性能变差的问题,提出面向小样本目标检测任务的自适应特征重构算法。该算法包含两个模块:基础类别特征偏移缓解模块,用于获取预训练阶段基础类别的特征方向;场景特征自适应约束模块,用于根据场景特征与各类别原型特征的相关性确定当前场景对于某些类别的偏好,从而自适应地调整基础类别偏移方向对实例特征的影响。试验结果表明,在PASCAL VOC和MS COCO数据集上,该模型对于小样本目标检测任务的检测能力均优于对比算法,在保证对于基础类别实例检测能力的基础上,对新类别的检测精度最高可分别提升12.4%与2.1%。本研究提出的模型可以保证对于基础类别相关实例的检测能力,并提升新类别实例检测性能。

关键词: 目标检测, 小样本学习, 自适应特征重构, 场景特征, 类别偏好

中图分类号: 

  • TP391
[1] 谢富, 朱定局.深度学习目标检测方法综述[J].计算机系统应用, 2022, 31(2): 1-12. XIE Fu, ZHU Dingju. Survey on deep learning object detection[J]. Computer Systems & Applications, 2022, 31(2):1-12.
[2] HU H, BAI S, LI A, et al. Dense relation distillation with context-aware aggregation for few-shot object detection[C] //Proceedings of CVPR-21. Nashville, USA: IEEE, 2021: 10185-10194.
[3] TAN M, PANG R, LE Q V. Efficientdet: scalable and efficient object detection[C] //Proceedings of CVPR-20. Seattle, USA: IEEE, 2020: 10781-10790.
[4] ZHANG Y, KANG B, HOOI B, et al. Deep long-tailed learning:a survey[EB/OL].(2021-10-09)[2022-02-25]. https://arxiv.org/pdf/2110.04596.pdf.
[5] KÖHLER M, EISENBACH M, GROSS H M. Few-shot object detection: a survey[EB/OL].(2021-12-22)[2022-02-25].https://arxiv.org/pdf/2112.11699.pdf.
[6] BUDA M, MAKI A, MAZUROWSKI M A. A systematic study of the class imbalance problem in convolutional neural networks[J]. Neural Networks, 2018, 106: 249-259.
[7] LIN T Y, GOYAL P, GIRSHICK R, et al. Focal loss for dense object detection[J]. IEEE Transactions on Pattern Analysis and Machine Intelligence, 2017, 42(2):318-327.
[8] KANG B, LIU Z, WANG X, et al. Few-shot object detection via feature reweighting[C] //Proceedings of ICCV-19. Seoul, Korea: IEEE, 2019: 8420-8429.
[9] WANG Y X, RAMANAN D, HEBERT M. Meta-learning to detect rare objects[C] //Proceedings of ICCV-19. Seoul, Korea: IEEE, 2019: 9925-9934.
[10] YAN X, CHEN Z, XU A, et al. Meta r-cnn: towards general solver for instance-level low-shot learning[C] // Proceedings of ICCV-19. Seoul, Korea: IEEE, 2019: 9577-9586.
[11] WANG X, HUANG T E, DARRELL T, et al. Frustratingly simple few-shot object detection[C] // Proceedings of ICML-20. Online: ACM, 2022: 9919-9928.
[12] WU J, LIU S, HUANG D, et al. Multi-scale positive sample refinement for few-shot object detection[C] // Proceedings of ECCV-20. Online: Springer, 2020: 456-472.
[13] SUN B, LI B, CAI S, et al. FSCE: few-shot object detection via contrastive proposal encoding[C] // Proceedings of CVPR-21. Nashville, USA: IEEE, 2021: 7352-7362.
[14] TANG K, HUANG J, ZHANG H. Long-tailed classification by keeping the good and removing the bad momentum causal effect[J]. Advances in Neural Information Processing Systems, 2020, 33: 1513-1524.
[15] XU H, JIANG C, LIANG X, et al. Reasoning-rcnn: Unifying adaptive global reasoning into large-scale object detection[C] //Proceedings of CVPR-19. Long Beach, USA: IEEE, 2019: 6419-6428.
[16] REN S, HE K, GIRSHICK R, et al. Faster R-CNN: towards real-time object detection with region proposal networks[J]. IEEE Transactions on Pattern Analysis and Machine Intelligence, 2017, 39(6): 1137-1149.
[17] EVERINGHAM M, ZISSERMAN A, WILLIAMS C K I, et al. The PASCAL visual object classes(VOC)challenge[J]. International Journal of Computer Vision, 2010, 88(2):303-338.
[18] LIN T Y, MAIRE M, BELONGIE S, et al. Microsoft COCO: common objects in context[C] //Proceedings of ECCV-14. Zurich, Switzerland: Springer, 2014: 740-755.
[19] CHEN K, WANG J, PANG J, et al. MMDetection: open mmlab detection toolbox and benchmark[EB/OL].(2019-06-17)[2022-02-25]. https://arxiv.org/pdf/1906.07155.pdf.
[20] HEK, ZHANG X, REN S, et al. Deep residual learning for image recognition[C] //Proceedings of CVPR-16. Las Vegas, USA: IEEE, 2016:770-778.
[21] LIN T Y, DOLLÁR P, GIRSHICK R, et al. Feature pyramid networks for object detection[C] //Proceedings of CVPR-17. Honolulu, USA: IEEE, 2017:2117-2125.
[1] 孟令灿,聂秀山,张雪. 基于遮挡目标去除的公交车拥挤度分类算法[J]. 山东大学学报 (工学版), 2022, 52(4): 83-88.
[2] 万鹏. 基于F-PointNet的3D点云数据目标检测[J]. 山东大学学报 (工学版), 2019, 49(5): 98-104.
[3] 惠开发,成科扬,詹永照. 基于改进ViBe算法的视频浓缩[J]. 山东大学学报(工学版), 2017, 47(3): 43-48.
[4] 刘英霞,王希常,唐晓丽,常发亮. 基于小波域特征和贝叶斯估计的目标检测算法[J]. 山东大学学报(工学版), 2017, 47(2): 63-70.
[5] 杨健梅1,黄添强1,2*,江伟坚1. 基于人脸色温的拼接图像篡改检测[J]. 山东大学学报(工学版), 2013, 43(5): 24-30.
[6] 王秀芬,王汇源,王松. 基于背景差分法和显著性图的海底目标检测方法[J]. 山东大学学报(工学版), 2011, 41(1): 12-16.
[7] 乔伟1,王汇源1,2,吴晓娟1,刘鹏威1. 基于混沌动力学模型的群体目标检测与分类[J]. 山东大学学报(工学版), 2010, 40(2): 19-23.
[8] 吕行,史忠科 . DirectShow框架下实时运动目标检测与跟踪方法的研究与应用[J]. 山东大学学报(工学版), 2007, 37(6): 5-9 .
Viewed
Full text


Abstract

Cited

  Shared   
  Discussed   
No Suggested Reading articles found!