《计算机应用》唯一官方网站 ›› 2022, Vol. 42 ›› Issue (11): 3617-3624.DOI: 10.11772/j.issn.1001-9081.2021091683

• 人工智能 • 上一篇    

基于负边距损失的小样本目标检测

杜芸彦1,2, 李鸿1,2, 杨锦辉1,2, 江彧1,2, 毛耀1,2()   

  1. 1.中国科学院大学,北京 100049
    2.中国科学院光束控制重点实验室(中国科学院光电技术研究所),成都 610207
  • 收稿日期:2021-09-27 修回日期:2022-05-25 接受日期:2022-05-26 发布日期:2022-11-14 出版日期:2022-11-10
  • 通讯作者: 毛耀
  • 作者简介:杜芸彦(1997—),女,四川成都人,硕士研究生,主要研究方向:目标检测、小样本学习
    李鸿(1996—),男,贵州毕节人,硕士研究生,主要研究方向:深度学习、轻量化目标检测
    杨锦辉(1996—),男,甘肃平凉人,硕士研究生,主要研究方向:轻量化目标检测
    江彧(1977—),女,安徽黄山人,副研究员,硕士,主要研究方向:人机交互系统
    毛耀(1978—),男,四川眉山人,研究员,博士,CCF会员,主要研究方向:机器视觉、强化学习。maoyao@ioe.ac.cn
  • 基金资助:
    国家重点研发计划项目(2017YFB1103002)

Few‑shot target detection based on negative‑margin loss

Yunyan DU1,2, Hong LI1,2, Jinhui YANG1,2, Yu JIANG1,2, Yao MAO1,2()   

  1. 1.University of Chinese Academy of Sciences,Beijing 100049,China
    2.Key Laboratory of Optical Engineering,Chinese Academy of Sciences (Institute of Optics and Electronics),Chengdu Sichuan 610207,China
  • Received:2021-09-27 Revised:2022-05-25 Accepted:2022-05-26 Online:2022-11-14 Published:2022-11-10
  • Contact: Yao MAO
  • About author:DU Yunyan, born in 1997, M. S. candidate. Her research interests include target detection, few-shot learning.
    LI Hong, born in 1996, M. S. candidate. His research interests include deep learning, lightweight target detection.
    YANG Jinhui, born in 1996, M. S. candidate. His research interests include lightweight target detection.
    JIANG Yu, born in 1977, M. S., associate researcher. Her research interests include computer application technology, human computer interaction system.
    MAO Yao, born in 1978, Ph. D., researcher. His research interests include computer technology, machine vision, reinforcement learning.
  • Supported by:
    The National Key Research and Development Program of China(2017YFB1103002)

摘要:

现有的大部分目标检测算法都依赖于大规模的标注数据集来保证检测的正确率,但某些场景往往很难获得大量标注数据,且耗费大量人力、物力。针对这一问题,提出了基于负边距损失的小样本目标检测方法(NM?FSTD),将小样本学习(FSL)中属于度量学习的负边距损失方法引入目标检测,负边距损失可以避免将同一新类的样本错误地映射到多个峰值或簇,有助于小样本目标检测中新类的分类。首先采用大量训练样本和基于负边距损失的目标检测框架训练得到具有良好泛化性能的模型,之后通过少量具有标签的目标类别的样本对模型进行微调,并采用微调后的模型对目标类别的新样本进行目标检测。为了验证NM?FSTD的检测效果,使用MS COCO进行训练和评估。实验结果表明,所提方法AP50达到了22.8%,与Meta R?CNN和MPSR相比,准确率分别提高了3.7和4.9个百分点。NM?FSTD能有效提高在小样本情况下对目标类别的检测性能,解决目前目标检测领域中数据不足的问题。

关键词: 目标检测, 小样本学习, 负边距损失, 度量学习

Abstract:

Most of the existing target detection algorithms rely on large?scale annotation datasets to ensure the accuracy of detection, however, it is difficult for some scenes to obtain a large number of annotation data and it consums a lot of human and material resources. In order to resolve this problem, a Few?Shot Target Detection method based on Negative Margin loss (NM?FSTD) was proposed. The negative margin loss method belonging to metric learning in Few?Shot Learning (FSL) was introduced into target detection, which could avoid mistakenly mapping the samples of the same novel classes to multiple peaks or clusters and helping to the classification of novel classes in few?shot target detection. Firstly, a large number of training samples and the target detection framework based on negative margin loss were used to train the model with good generalization performance. Then, the model was finetuned through a small number of labeled target category samples. Finally, the finetuned model was used to detect the new sample of target category. To verify the detection effect of NM?FSTD, MS COCO was used for training and evaluation. Experimental results show that the AP50 of NM?FSTD reaches 22.8%; compared with Meta R?CNN (Meta Regions with CNN features) and MPSR (Multi?Scale Positive Sample Refinement), the accuracies are improved by 3.7 and 4.9 percentage points, respectively. NM?FSTD can effectively improve the detection performance of target categories in the case of few?shot, and solve the problem of insufficient data in the field of target detection.

Key words: target detection, Few?Shot Learning (FSL), negative?margin loss, metric learning

中图分类号: