基于特征融合的深度学习目标检测算法研究

黄剑航; 王振友

doi:10.12052/gdutxb.200147

欢迎来到能源电力期刊网！

English Version 期刊登录

您当前的位置：

首页 >

文章列表页 >

基于特征融合的深度学习目标检测算法研究

更新时间：2026-05-27

- 基于特征融合的深度学习目标检测算法研究
- A Research on Deep Learning Object Detection Algorithm Based on Feature Fusion
- 广东工业大学学报 2021年38卷第4期页码：52-58
- 作者机构：
  
  广东工业大学应用数学学院,广东,广州,510520
- 作者简介：
  
  黄剑航(1996–)，男，硕士研究生，主要研究方向为目标检测
  王振友(1979–)，男，教授，主要研究方向为优化计算、医学统计分析，E-mail：zywang@gdut.edu.cn
- 基金信息：
  
  广东省基础与应用基础研究基金资助项目(2020B1515310001)
- DOI：10.12052/gdutxb.200147
  中图分类号： TP242.6+2
- 网络首发：2021-07-10，
  
  纸质出版：2021
- 稿件说明：
移动端阅览
黄剑航, 王振友. 基于特征融合的深度学习目标检测算法研究[J]. 广东工业大学学报, 2021, 38(4): 52-58.

Huang Jian-hang, Wang Zhen-you. A Research on Deep Learning Object Detection Algorithm Based on Feature Fusion[J]. 2021, 38(4): 52-58.
黄剑航, 王振友. 基于特征融合的深度学习目标检测算法研究[J]. 广东工业大学学报, 2021, 38(4): 52-58. DOI： 10.12052/gdutxb.200147.

Huang Jian-hang, Wang Zhen-you. A Research on Deep Learning Object Detection Algorithm Based on Feature Fusion[J]. 2021, 38(4): 52-58. DOI： 10.12052/gdutxb.200147.

摘要

通过研究卷积神经网络中的特征层级

发现高层特征图的分辨率低、语义信息强

低层特征图的分辨率强、语义信息较弱等问题。针对上述问题提出一种二次特征融合的目标检测算法

该算法在特征金字塔网络(Feature Pyramid Networks

FPN)的基础上对过渡特征重复使用并进行二次特征融合

使丰富的低层特征信息补充到高层。最终在COCO2014的数据集上平均精度AP(Average Precision)、AP

、AP

分别达到了35.3%

57.5%

36.6%

与未使用特征融合方法以及使用传统特征融合的方法相比

分别提升了2.4%

3.7%

2.4%

能改善漏检情况和有利于小目标的检测。

Abstract

Through the study of feature levels in convolutional neural networks

this paper found that high-level feature have stronger semantic information and low resolution

and low-level features have strong resolution and weaker semantic information. Aiming at these problems

a object detection algorithm based on secondary feature fusion is proposed. The algorithm reuses transitional features and performs secondary feature fusion on the basis of Feature Pyramid Networks to supplement the rich low-level feature information to the top. Finally

the average accuracy of AP

and AP

on the COCO2014 data set reach 35.3%

57.5%

and 36.6%

respectively. Compared with the unused feature fusion method and the traditional feature fusion method

the average accuracy is

increased by 2.4%

3.7% and 2.4%

which significantly improves the missed detection and the detection of small targets.

关键词

Keywords

references

HE K, GKIOXARI G, DOLLÁR P, et al . Mask R-CNN [J ] . IEEE transactions on pattern analysis machine intelligence, 2020, 42(2): 386-397.

REDMON J, DIVVALA S, GIRSHICK R, et al. You only look once: unified, real-time object detection[C]//IEEE Conference on Computer Vision and Pattern Recognition (CVPR). Las Vegas: IEEE, 2016: 779-788.

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 Machine Intelligence, 2017, 39(6): 1137-1149.

REDMON J, FARHADI A. YOLO9000: better, faster, stronger[C]//IEEE Conference on Computer Vision and Pattern Recognition(CVPR). Honolulu: IEEE, 2017: 6517-6525.

LI Y, CHEN Y, WANG N, et al. Scale-aware trident networks for object detection[C]//IEEE/CVF International Conference on Computer Vision (ICCV). Seoul: IEEE, 2019: 6053-6062.

BHARAT S, MAHYAR N, LARRY S D. SNIPER: efficient multi-scale training[J]. arXiv preprint arXiv: 1805.09300, 2018.

LIU W, ANUELOVG D, ERHAN D, et al. SSD: Single shot multibox detector[C]//European Conference on Computer Vision. Berlin: Springer, Cham, 2016: 21-37.

KONG T, YAO A, CHEN Y, et al. Hypernet: Towards accurate region proposal generation and joint object detection[C]//IEEE Conference on Computer Vision and Pattern Recognition(CVPR). Las Vegas: IEEE, 2016: 845-853.

LIU S T, HUANG D, WANG Y H. Receptive field block net for accurate and fast object detection[J]. arXiv preprint arXiv: 1711.07767, 2017.

LIN TY, DOLLÁR P, GIRSHICK R, et al. Feature pyramid networks for object detection[C]//IEEE Conference on Computer Vision and Pattern Recognition (CVPR). Las Vegas: IEEE, 2016: 2117-2125.

LI Z, ZHOU F Q. FSSD: feature fusion single shot multibox detector[J]. arXiv preprint arXiv: 1712.00960, 2017.

FU C Y, LIU W, RANGA A, et al. Dssd: deconvolutional single shot detector[J]. arXiv preprint arXiv: 1701.06659. 2017.

WEN J W, ZHANM W, LI C H, et al . Design of Atrous filter to strengthen small object detection capability of SSD [J ] . Application Research of Computers, 2019, 36(3): 861-865, 872.

GAO J Y, LIU W Y, YANG Z G. Object tracking combined with attention and feature fusion [J]. Journal of Guangdong University of Technology, 2019, 36(4): 18-23.

HE K, ZHANG X Y, REN S Q, et al. Deep residual learning for image recognition[C]//IEEE Conference on Computer Vision and Pattern Recognition(CVPR). Las Vegas: IEEE, 2016: 770-778.

REDMON J, FARHADI A. YOLOV3: an incremental improvement[J]. arXiv preprint arXiv: 1804.02767, 2018.

ABADI M, AGARWAL A, BARHAM P, et al. Tensorflow: large-scale machine learning on heterogeneous distributed systems[J]. arXiv preprint arXiv: 1603.04467, 2016.

RUSSAKOVSKY O, DENG J, SU H, et al . ImageNet large scale visual recognition challenge [J ] . International journal of computer vision, 2015, 115(3): 211-252.

DAI J F, LI Y, HE K. R-FCN: object detection via region-based fully convolutional networks[J]. arXiv preprint arXiv: 1605.06409, 2016.

浏览量

下载量

CSCD

文章被引用时，请邮件提醒。

提交

工具集

关联资源

基于YOLOv4-MP的绝缘子爆裂缺陷检测方法

一种基于多尺度的多层卷积稀疏编码网络

面向电力场景图像的动态自适应增量目标检测DA-IOD算法

用于三维点云分割和分类的高分辨率特征网络

SCA-YOLO: 基于可变形卷积与上下文感知注意力的实时目标检测