全局与局部图像特征自适应融合的小目标检测算法
DOI:
作者:
作者单位:

西安建筑科技大学,信息与控制工程学院

作者简介:

通讯作者:

中图分类号:

TP183

基金项目:

国家自然科学基金项目(51209167;12002251);陕西省自然科学基金(2019JM-474);西安市科技计划项目(2020KJRC0055);陕西省岩土与地下空间工程重点实验室开放基金项目(YT202004)


Small object detection algorithm based on adaptive fusion of global and local image features
Author:
Affiliation:

1.School of Information and Control Engineering, Xi '2.'3.an University of Architecture and Technology

Fund Project:

National Natural Science Foundation of China(51209167; 12002251); Natural Science Foundation of Shaanxi Province(2019JM-474); Science and technology project of Xi’an(2020KJRC0055); Funding of Shaanxi Key Laboratory of Geotechnical and Underground Space Engineering(YT202004)

  • 摘要
  • |
  • 图/表
  • |
  • 访问统计
  • |
  • 参考文献
  • |
  • 相似文献
  • |
  • 引证文献
  • |
  • 资源附件
  • |
  • 文章评论
    摘要:

    针对现有目标检测算法对于小目标检测精度低的问题,本文提出一种全局与局部图像特征自适应融合的一阶段小目标检测算法 SODet。首先将 Transformer 与卷积神经网络相结合构建主干网络,分别提取图像全局与局部信息,并利用自适应特征选择模块 AFS 对二者输出进行融合;其次在特征融合网络中利用额外尺度特征图进行特征融合,同时利用大目标抑制单元约束大目标特征表达、转移小目标特征,输出四个尺度的特征图送入预测网络;最后在损失函数部分针对小目标检测利用 EIOU 和 Focal loss 进行优化。实验结果表明,SODet 算法在 MS COCO 验证集上 APS 达到 31.5%,相比于其他算法具有较强的竞争力,同时具有较高的推理速度。

    Abstract:

    Aiming at the problem that the existing object detectors have low accuracy for small objects. This paper proposes a one-stage small object detector SODet, which adaptively fuse the global and local image features. First, Transformer and convolutional neural network (CNN) are combined to construct a backbone network to extract global and local information of the image respectively. Then the adaptive feature selection module AFS is used to fuse the outputs of Transformer and CNN. Secondly, extra-scale feature maps are used in the feature fusion network. At the same time, the large object restraint unit is used to constrain the expression of large object features and transfer small object features. The feature maps of four scales are sent to the prediction network. Finally, in the loss function, EIOU and Focal loss are used to optimize small object detection. The experimental results show that the SODet has 31.5% in terms of APS on the MS COCO verification set, which is more competitive than other algorithms and has a higher inference speed.

    参考文献
    相似文献
    引证文献
引用本文
分享
文章指标
  • 点击次数:
  • 下载次数:
  • HTML阅读次数:
  • 引用次数:
历史
  • 收稿日期:2021-10-20
  • 最后修改日期:2022-03-08
  • 录用日期:2022-03-15
  • 在线发布日期:
  • 出版日期: