第 53 卷 第 6 期
2020 年 6 月
天津大学学报(自然科学与工程技术版)
Journal of Tianjin University(Science and Technology)
Vol. 53 No. 6
Jun. 2020
收稿日期:2019-04-04;修回日期:2019-08-26.
作者简介:杨爱萍(1977— ),女,博士,副教授.
通信作者:杨爱萍,yangaiping@tju.edu.cn.
基金项目:国家自然科学基金资助项目(61771329,61632018).
Supported by the National Natural Science Foundation of China(No. 61771329,No. 61632018).
DOI:10.11784/tdxbz201904007
多层特征图堆叠网络及其目标检测方法
杨爱萍,鲁立宇,冀 中
(天津大学电气自动化与信息工程学院,天津 300072)
摘 要:随着深度卷积神经网络的快速发展,基于深度学习的目标检测方法由于具有良好的特征表达能力及优良的
检测精度,成为当前目标检测算法的主流.为了解决目标检测中小目标漏检问题,往往使用多尺度处理方法.现有
的多尺度目标检测方法可以分为基于图像金字塔的方法和基于特征金字塔的方法.相比于基于图像金字塔的方法,
基于特征金字塔的方法速度更快,更能充分利用不同卷积层的特征信息.现有的基于特征金字塔的方法采用对应元
素相加的方式融合不同尺度的特征图,在特征融合过程中易丢失低层细节特征信息.针对该问题,本文基于特征金
字塔网络(feature pyramid network,FPN),提出一种多层特征图堆叠网络(multi-feature concatenation network,
MFCN)及其目标检测方法.该网络以 FPN 为基础,设计多层特征图堆叠结构,通过不同特征层之间的特征图堆叠
融合高层语义特征和低层细节特征,并且在每个层上进行目标检测,保证每层可包含该层及其之上所有层的特征信
息,可有效克服低层细节信息丢失.同时,为了能够充分利用 ResNet101 中的高层特征,在其后添加新的卷积层,
并联合其低层特征图,提取多尺度特征.在 PASCAL VOC 2007 数据集上的检测精度为 80.1%mAP,同时在
PASCAL VOC 2012 和 MS COCO 数据集上的表现都优于 FPN 算法.相比于 FPN 算法,MFCN 的检测性能更加
优秀.
关键词:特征金字塔网络;目标检测;特征图堆叠;语义信息
中图分类号:TP391 文献标志码:A 文章编号:0493-2137(2020)06-0647-06
Multi-Feature Concatenation Network for Object Detection
Yang Aiping,Lu Liyu,Ji Zhong
(School of Electrical and Information Engineering,Tianjin University,Tianjin 300072,China)
Abstract:
With the rapid development of deep the convolutional neural network,mainstream methods for objec
detection have been based on deep learning owing to its superior feature representation and excellent detection accu-
racy.To omit small objects in object detection,a multi-scale algorithm is usually adopted.Existing multi-scale
object detection methods can be categorized as image pyramid-based or feature pyramid-based.Compared with the
image pyramid-based method,the feature pyramid-based method is faster and better able to take full advantage of the
feature information of different convolution layers. The existing feature pyramid-based method fuses feature maps
from different scales by adding corresponding elements,which often results in loss of some detailed low-level feature
information.To tackle this problem,this paper proposes a multi-feature concatenation network(MFCN)based on a
feature pyramid network(FPN). A structure-performing,multi-layer feature map concatenation was designed. Se-
mantic high-level features and detailed low-level features were fused by concatenating feature maps from differen
feature layers.Objects on each layer were detected to ensure that each layer could contain the feature information o
the layer and all layers above it,effectively overcoming the loss of detailed low-level information.To make full use
of the high-level features in ResNet101,a new convolutional layer was added and combined with the low-level fea-
ture map to extract multi-scale features.Results of the new design showed that detection accuracy on the PASCA
万方数据