基于注意力机制的单发多框检测器算法

赵辉; 李志伟; 张天琪

doi:10.11999/JEIT200304

基于注意力机制的单发多框检测器算法

doi: 10.11999/JEIT200304

1.
重庆邮电大学通信与信息工程学院重庆 400065
2.
信号与信息处理重庆市重点实验室重庆 400065

基金项目: 国家自然科学基金(61671095)

详细信息

作者简介:
赵辉：女，1980年生，教授，博士生导师，研究方向为深空通信、信号理论与信息处理、图像信息处理

李志伟：男，1996年生，硕士，研究方向为计算机视觉、深度学习、目标检测

张天琪：男，1971年生，教授，博士生导师，研究方向为无线通信的智能信号处理、通信抗干扰和信息对抗

通讯作者:
赵辉　zhaohui@cqupt.edu.cn

中图分类号: TN911.73; TP391.41
计量
- 文章访问数: 1129
- HTML全文浏览量: 377
- PDF下载量: 142
- 被引次数: 0
出版历程
- 收稿日期: 2020-04-24
- 修回日期: 2021-02-15
- 网络出版日期: 2021-03-31
- 刊出日期: 2021-07-10

Attention Based Single Shot Multibox Detector

1.
School of Communication and Information Engineering, Chongqing University of Posts and Telecommunications, Chongqing 400065, China
2.
Chongqing Key Laboratory of Signal and Information Processing, Chongqing 400065, China

Funds: The National Natural Science Foundation of China (61671095)

摘要

摘要: 单发多框检测器SSD是一种在简单、快速和准确性之间有着较好平衡的目标检测器算法。SSD网络结构中检测层单一的利用方式使得特征信息利用不充分，将导致小目标检测不够鲁棒。该文提出一种基于注意力机制的单发多框检测器算法ASSD。ASSD算法首先利用提出的双向特征融合模块进行特征信息融合以获取包含丰富细节和语义信息的特征层，然后利用提出的联合注意力单元进一步挖掘重点特征信息进而指导模型优化。最后，公共数据集上进行的一系列相关实验表明ASSD算法有效提高了传统SSD算法的检测精度，尤其适用于小目标检测。
- 目标检测 /
- 注意力机制 /
- 特征融合 /
- 单发多框检测器
Abstract: Single Shot multibox Detector (SSD) is a object detection algorithm that provides the optimal trade-off among simplicity, speed and accuracy. The single use of detection layers in SSD network structure makes the feature information not fully utilized, which will lead to the small object detection are not robust enough. In this paper, an Attention based Single Shot multibox Detector (ASSD) is proposed. The ASSD algorithm first uses the proposed two-way feature fusion module to fuse the feature information to obtain the feature layer which containing rich details and semantic information. Then, the proposed joint attention unit is used to mine further the key feature information to guide the model optimization. Finally, a series of experiments on the common data set show that the ASSD algorithm effectively improves the detection accuracy of conventional SSD algorithm, especially for small object detection.
- Object detection /
- Attention mechanism /
- Feature fusion /
- Single Shot multibox Detector (SSD)

HTML全文

图 1 ASSD算法框架结构图

下载: 全尺寸图片幻灯片

图 2 模块网络结构图

下载: 全尺寸图片幻灯片

图 3 联合注意力单元的网络结构图与算法流程框图

下载: 全尺寸图片幻灯片

图 4 SSD和ASSD的P-R曲线对比图

下载: 全尺寸图片幻灯片

图 5 ASSD算法与SSD算法样例检测结果对比图

下载: 全尺寸图片幻灯片

表 1 Pascal VOC2007 test上的ASSD消融实验 (1 M=10⁶)

方法	输入尺寸	mAP	#参数量	fps
SSD300	300×300	77.5	50.14 M	60.2
SSD512	512×512	79.8	51.98 M	25.2
SSD300+TwFFM	300×300	78.8	64.18 M	47.1
SSD300+SDPA^[22]	300×300	78.2	56.38 M	53.1
SSD300+SEB^[23]	300×300	78.4	54.43 M	51.5
SSD300+JAU	300×300	78.6	60.67 M	48.1
ASSD300	300×300	79.1	74.71 M	39.6
ASSD512	512×512	80.9	76.55 M	20.8

下载: 导出CSV

表 2 各种算法在Pascal VOC2007测试集上的性能对比

方法	基础骨干网	输入尺寸	mAP	fps
YOLOv2^[11]	Darkent-19	416×416	76.8	67
YOLOv2+^[11]	Darkent-19	544×544	78.6	40
Faster RCNN^[7]	ResNet-101	～1000×600	76.4	2.4
SSD300	VGG-16	300×300	77.5	60.2
SSD512	VGG-16	512×512	79.8	25.2
DSSD321^[16]	ResNet-101	300×300	78.6	9.5
DSSD513^[16]	ResNet-101	513×513	81.5	5.5
RSSD300^[20]	VGG-16	300×300	78.5	35.0
RSSD512^[20]	VGG-16	300×300	80.8	16.6
FSSD300^[21]	VGG-16	300×300	78.8	65.8
FSSD512^[21]	VGG-16	512×512	80.9	35.7
ASSD300	VGG-16	300×300	79.1	39.6
ASSD512	VGG-16	512×512	81.0	20.8

下载: 导出CSV

表 3 各种算法在Pascal VOC2007测试集中20个类别的性能对比

方法	mAP	aero	bike	bird	boat	bottle	bus	car	cat	chair	cow
SSD300	77.5	79.5	83.9	76.0	69.6	50.5	87.0	85.7	88.1	60.3	81.5
SSD512	79.8	85.8	85.6	79.5	74.1	59.9	86.8	88	89.1	63.8	86.3
ASSD300	79.1	85.4	84.1	78.7	71.8	54.0	86.2	85.3	89.5	60.4	87.4
ASSD512	81.0	86.8	85.2	84.1	75.2	60.5	88.3	88.4	89.3	63.5	87.6

方法	mAP	table	dog	horse	mbike	person	plant	sheep	sofa	train	tv
SSD300	77.5	77.0	86.1	87.5	84.0	79.4	52.3	77.9	79.5	87.6	76.8
SSD512	79.8	75.8	87.4	87.5	83.2	83.5	56.3	81.6	77.7	87	77.4
ASSD300	79.1	77.1	87.4	86.8	84.8	79.5	57.8	81.5	80.1	87.4	76.9
ASSD512	81.0	76.6	88.2	86.7	85.7	82.8	59.2	83.6	80.5	87.5	80.8

下载: 导出CSV

参考文献(26)

[1]	赵凤, 孙文静, 刘汉强, 等. 基于近邻搜索花授粉优化的直觉模糊聚类图像分割[J]. 电子与信息学报, 2020, 42(4): 1005–1012. doi: 10.11999/JEIT190428 ZHAO Feng, SUN Wenjing, LIU Hanqiang, et al. Intuitionistic fuzzy clustering image segmentation based on flower pollination optimization with nearest neighbor searching[J]. Journal of Electronics &Information Technology, 2020, 42(4): 1005–1012. doi: 10.11999/JEIT190428
[2]	孙彦景, 石韫开, 云霄, 等. 基于多层卷积特征的自适应决策融合目标跟踪算法[J]. 电子与信息学报, 2019, 41(10): 2464–2470. doi: 10.11999/JEIT180971 SUN Yanjing, SHI Yunkai, YUN Xiao, et al. Adaptive strategy fusion target tracking based on multi-layer convolutional features[J]. Journal of Electronics &Information Technology, 2019, 41(10): 2464–2470. doi: 10.11999/JEIT180971
[3]	GIRSHICK R, DONAHUE J, DARRELL T, et al. Rich feature hierarchies for accurate object detection and semantic segmentation[C]. Proceedings of 2014 IEEE Conference on Computer Vision and Pattern Recognition, Columbus, USA, 2014: 580–587. doi: 10.1109/CVPR.2014.81.
[4]	HE Kaiming, ZHANG Xiangyu, REN Shaoqing, et al. Spatial pyramid pooling in deep convolutional networks for visual recognition[J]. IEEE Transactions on Pattern Analysis and Machine Intelligence, 2015, 37(9): 1904–1916. doi: 10.1109/TPAMI.2015.2389824
[5]	GIRSHICK R. Fast R-CNN[C]. Proceedings of 2015 IEEE International Conference on Computer Vision, Santiago, Chile, 2015: 1440–1448. doi: 10.1109/ICCV.2015.169.
[6]	SIMONYAN K and ZISSERMAN A. Very deep convolutional networks for large scale image recognition[C]. Proceedings of the 3rd International Conference on Learning Representations, San Diego, USA, 2015.
[7]	REN Shaoqing, HE Kaiming, GIRSHICK R, et al. Faster R-CNN: Towards real-time object detection with region proposal networks[J]. IEEE Transactions on Pattern Analysis and Machine Intelligence, 2017, 39(6): 1137–1149. doi: 10.1109/TPAMI.2016.2577031
[8]	REDMON J, DIVVALA S, GIRSHICK R, et al. You only look once: Unified, real-time object detection[C]. Proceedings of 2016 IEEE Conference on Computer Vision and Pattern Recognition, Las Vegas, USA, 2016: 779–788. doi: 10.1109/CVPR.2016.91.
[9]	LIU Wei, ANGUELOV D, ERHAN D, et al. SSD: Single shot MultiBox detector[C]. Proceedings of the 14th European Conference on Computer Vision, Amsterdam, The Netherlands, 2016: 21–37. doi: 10.1007/978-3-319-46448-0_2.
[10]	SZEGEDY C, LIU Wei, JIA Yangqing, et al. Going deeper with convolutions[C]. Proceedings of 2015 IEEE Conference on Computer Vision and Pattern Recognition, Boston, USA, 2015: 1–9. doi: 10.1109/CVPR.2015.7298594.
[11]	REDMON J and FARHADI A. YOLO9000: Better, faster, stronger[C]. Proceedings of 2017 IEEE Conference on Computer Vision and Pattern Recognition, Honolulu, USA, 2017: 7263–7271. doi: 10.1109/CVPR.2017.690.
[12]	ZEILER M D and FERGUS R. Visualizing and understanding convolutional networks[C]. Proceedings of the 13th European Conference on Computer Vision, Zurich, Switzerland, 2014: 818–833. doi: 10.1007/978-3-319-10590-1_53.
[13]	CHEN Chenyi, LIU Mingyu, TUZEL O, et al. R-CNN for small object detection[C]. Proceedings of the 13th Asian Conference on Computer Vision, Taipei, China, 2016: 214–230. doi: 10.1007/978-3-319-54193-8_14.
[14]	HE Kaiming, ZHANG Xiangyu, REN Shaoqing, et al. Deep residual learning for image recognition[C]. Proceedings of 2016 IEEE Conference on Computer Vision and Pattern Recognition, Las Vegas, USA, 2016: 770–778. doi: 10.1109/CVPR.2016.90.
[15]	HUANG Gao, LIU Zhuang, VAN DER MAATEN L, et al. Densely connected convolutional networks[C]. Proceedings of 2017 IEEE Conference on Computer Vision and Pattern Recognition, Honolulu, USA, 2017: 4700–4708. doi: 10.1109/CVPR.2017.243.
[16]	FU Chengyang, LIU Wei, RANGA A, et al. DSSD: Deconvolutional single shot detector[J]. arXiv: 1701.06659, 2017.
[17]	SHEN Zhiqiang, LIU Zhuang, LI Jianguo, et al. Dsod: Learning deeply supervised object detectors from scratch[C]. Proceedings of 2017 IEEE International Conference on Computer Vision, Venice, Italy, 2017: 1919–1927. doi: 10.1109/ICCV.2017.212.
[18]	ZEILER M D and FERGUS R. Visualizing and understanding convolutional networks[C]. Proceedings of the 13th European Conference on Computer Vision, Zurich, Switzerland, 2014: 818–833.
[19]	LIN T Y, DOLLÁR P, GIRSHICK R, et al. Feature pyramid networks for object detection[C]. Proceedings of 2017 IEEE Conference on Computer Vision and Pattern Recognition, Honolulu, USA, 2017: 2117–2125. doi: 10.1109/CVPR.2017.106.
[20]	JEONG J, PARK H, and KWAK N. Enhancement of SSD by concatenating feature maps for object detection[C]. Proceedings of 2017 British Machine Vision Conference, London, UK, 2017.
[21]	LI Zuoxin and ZHOU Fuqiang. FSSD: Feature fusion single shot Multibox detector[J]. arXiv: 1712.00960, 2017.
[22]	MNIH V, HEESS N, GRAVES A, et al. Recurrent models of visual attention[C]. Proceedings of the 27th International Conference on Neural Information Processing Systems, Montreal, Canada, 2014: 2204–2212.
[23]	BAHDANAU D, CHO K, BENGIO Y, et al. Neural machine translation by jointly learning to align and translate[C]. Proceedings of the 3rd International Conference on Learning Representations, San Diego, USA, 2015.
[24]	VASWANI A, SHAZEER N, PARMAR N, et al. Attention is all you need[C]. Proceedings of the 31st International Conference on Neural Information Processing Systems, Long Beach, USA, 2017: 5998–6008.
[25]	HU Jie, SHEN Li, and SUN Gang. Squeeze-and-excitation networks[C]. Proceedings of 2018 IEEE/CVF Conference on Computer Vision and Pattern Recognition, Salt Lake City, USA, 2018: 7132–7141. doi: 10.1109/CVPR.2018.00745.
[26]	EVERINGHAM M, VAN GOOL L, WILLIAMS C K I, et al. The pascal visual object classes (VOC) challenge[J]. International Journal of Computer Vision, 2010, 88(2): 303–338. doi: 10.1007/s11263-009-0275-4