基于弱特征增强的轻量化小目标检测方法
CSTR:
作者:
作者单位:

1. 江南大学 物联网工程学院,江苏 无锡 214122;2. 江南大学 物联网技术应用教育部工程研究中心,江苏 无锡 214122

作者简介:

通讯作者:

E-mail: wuzhihai@jiangnan.edu.cn.

中图分类号:

TP183

基金项目:

国家自然科学基金项目(61876073).


Lightweight small target detection method based on weak feature enhancement
Author:
Affiliation:

1. School of Internet of Things Engineering,Jiangnan University,Wuxi 214122,China;2. Engineering Research Center of Internet of Things Technology Applications of MOE,Jiangnan University,Wuxi 214122,China

Fund Project:

  • 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:

    针对复杂背景下小目标特征经多次卷积被背景噪声淹没导致的检测精度低的问题,提出一种增强弱特征表达的一阶段轻量级小目标检测算法SA-YOLO.首先,用改进的ShuffleNetv2网络构建骨干网络,通过嵌入SE注意力模块和Inception结构,提升网络在复杂背景下的特征提取能力,有效地抑制背景噪声,充分提取弱特征;其次,在颈部网络,采用新的特征融合模块,以含有弱特征较多的低层级特征块的空间位置信息对高层级特征进行权重调整,提高不同层级的特征融合利用率,减少小目标的特征损失;最后,在头部网络,用解耦的检测头替换原YOLO耦合的检测头,解耦分类任务和回归任务,提高弱特征的解码能力,增强小目标检测的性能.在公开数据集COCO2017上进行实验,结果表明,SA-YOLO参数量仅有1.14M,小目标平均检测召回率$\rm AR_S$达到31.6%.同时,将所提出算法与近几年主流算法进行对比,结果表明,所提出算法在小目标检测方面具有较强的竞争力.

    Abstract:

    Aiming at the problem of low detection accuracy caused by small target features in complex backgrounds, which are submerged by background noise after multiple convolutions, a one-stage lightweight small target detection algorithm SA-YOLO is proposed to enhance weak feature expression. First, the improved ShuffleNetv2 network is used to build the backbone network, and by embedding the SE attention module and the Inception structure, the feature extraction ability of the network in complex backgrounds is improved, background noise is effectively suppressed, and weak features are fully extracted. Second, in the neck network, a new feature fusion module is adopted to adjust the weights of high-level features based on the spatial location information of low-level feature blocks containing more weak features, so as to improve the utilization rate of feature fusion at different levels and reduce the feature loss of small targets. Finally, it replaces the original YOLO-coupled detection head with the decoupled detection head, decouples thei classification task and the regression task, improves the decoding ability of weak features, and enhances the performance of small target detection. Experiments are carried out on the public dataset COCO2017, and the results show that the parameter size of the SA-YOLO is only 1.14M, and the average detection recall rate $\rm AR_S$ of small targets reaches 31.6%. At the same time, the proposed algorithm is compared with the mainstream algorithms in recent years. The results show that the proposed algorithm has strong competitiveness in small target detection.

    参考文献
    相似文献
    引证文献
引用本文

周葳楠,吴治海,张正道,等.基于弱特征增强的轻量化小目标检测方法[J].控制与决策,2024,39(2):381-390

复制
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:
  • 最后修改日期:
  • 录用日期:
  • 在线发布日期: 2024-01-18
  • 出版日期: 2024-02-20
文章二维码