加权网络下的小目标检测算法

doi:10.3778/j.issn.1673-9418.2101040

计算机科学与探索 ›› 2022, Vol. 16 ›› Issue (9): 2143-2150.DOI: 10.3778/j.issn.1673-9418.2101040

加权网络下的小目标检测算法

陈灏然¹, 彭力¹^,⁺(), 李文涛¹, 戴菲菲²

1.物联网技术应用教育部工程研究中心（江南大学物联网工程学院）,江苏无锡 214122
2.台州市产品质量安全监测研究院,浙江台州 318000

收稿日期:2021-01-11 修回日期:2021-03-10 出版日期:2022-09-01 发布日期:2021-03-29
通讯作者: + E-mail: jnpengli@outlook.com
作者简介:陈灏然（1995—）,男,江苏盐城人,硕士研究生,主要研究方向为小目标检测、数据增强、深度学习。
彭力（1967—）,男,河北唐山人,博士,教授,博士生导师,CAAI会员,CCF会员,主要研究方向为视觉物联网、行为识别、深度学习。
李文涛（1996—）,男,安徽合肥人,硕士研究生,主要研究方向为深度学习、计算机视觉。
戴菲菲（1988—）,女,浙江临海人,硕士,工程师,主要研究方向为大数据、视觉物联网。
基金资助:
国家自然科学基金(61873112);教育部-中国移动科研基金项目(MCM20170204);国家重点研发计划(2018YFD0400902)

Small Object Detection Algorithm Based on Weighted Network

CHEN Haoran¹, PENG Li¹^,⁺(), LI Wentao¹, DAI Feifei²

1. Engineering Research Center of Internet of Things Technology Applications (School of Internet of Things Enginee-ring, Jiangnan University), Ministry of Education, Wuxi, Jiangsu 214122, China
2. Taizhou Product Quality and Safety Monitoring Institute, Taizhou, Zhejiang 318000, China

Received:2021-01-11 Revised:2021-03-10 Online:2022-09-01 Published:2021-03-29
About author:CHEN Haoran, born in 1995, M.S. candidate. His research interests include small object detection, data augmentation and deep learning.
PENG Li, born in 1967, Ph.D., professor, Ph.D. supervisor, member of CAAI and CCF. His research interests include visual Internet of things, action recognition and deep learning.
LI Wentao, born in 1996, M.S. candidate. His research interests include deep learning and computer vision.
DAI Feifei, born in 1988, M.S., engineer. Her research interests include big data and visual Internet of things.
Supported by:
National Natural Science Foundation of China(61873112);Research Fund Project of Ministry of Education-China Mobile(MCM20170204);National Key Research and Development Program of China(2018YFD0400902)

摘要/Abstract

摘要：

对于一幅图的观察,本能上会更多关注这幅图中相对更醒目的对象。通常这类对象会在这幅图中占据较大比重,从而导致小目标被忽视。由于小目标所在区域往往为弱测区域,检测器提取特征的过程中能够提取的特征较少,且在提取完特征后在特征信息传递的过程中容易丢失,使得针对小目标检测的效果并不是很好。因此,在单阶检测器的基础上,加入了跨信道交互的机制确保层间信息的完整,同时采取对训练样本进行目标增强并且设计了一个通用的损失函数,在此基础上改进样本加权网络预测样本的任务权重。提出的框架UWN在VOC公开数据集上的mAP为81.2%,在自制的小目标航拍数据集的mAP为82.3%。相对于FSSD算法,牺牲了部分速度,得到了精度方面的较大提升。

关键词: 小目标检测, 目标增强, 跨信道交互, 加权网络

Abstract:

For the observation of a picture, people may instinctly pay more attention to the eye-catching objects in the picture. Usually such objects tend to occupy a larger proportion in the picture, which leads to small targets being ignored. Because the area where the small target is located is often a weak detection area, and the features that can be extracted in the process of extracting features by the detector are few and are easily lost in the process of feature information transmission after the feature is extracted, the effect of small target detection is not good. Therefore, on the basis of the single-order detector, this paper adds a cross-channel interaction mechanism to ensure the integrity of the information between layers, adopts target enhancement of training samples and designs a general loss function. Apart from this, this paper improves the sample weighting on the basis of the loss function to predict weight of samples. The mAP of this paper framework UWN (unified weighted network) on the VOC public dataset is 81.2% and the mAP on the self-made small target aerial photography dataset is 82.3%. Compared with the FSSD algorithm, some speed is sacrificed, and the accuracy is greatly improved.

Key words: small target detection, target enhancement, cross-channel interaction, weighted network

中图分类号:

TP391.4

陈灏然, 彭力, 李文涛, 戴菲菲. 加权网络下的小目标检测算法[J]. 计算机科学与探索, 2022, 16(9): 2143-2150.

CHEN Haoran, PENG Li, LI Wentao, DAI Feifei. Small Object Detection Algorithm Based on Weighted Network[J]. Journal of Frontiers of Computer Science and Technology, 2022, 16(9): 2143-2150.

图/表 14

图1 two-stage算法流程图

Fig.1 Network structure of two-stage algorithm

图2 one-stage算法流程图

Fig.2 Network structure of one-stage algorithm

图3 VOC数据集样本

Fig.3 VOC dataset sample

图4 自制航拍数据集样本

Fig.4 Self-made aerial dataset sample

图5 SENet信道交互图

Fig.5 SENet channel interaction

图6 k邻域跨信道交互图

Fig.6 k-neighbour cross-channel interaction

图7 信道交互图

Fig.7 Channel interaction

图8 整体算法结构图

Fig.8 Block diagram of algorithm

表1 不同 k值在VOC2007+2012下检测结果

Table 1 Test results of different k under VOC2007+2012

Method	Fusion-layer	Enhancement?	k	mAP/%	FPS（1080Ti）
ele-sum	all	√	3	78.2	23
concat	all	√	3	77.3	28
ele-sum	Conv4-fc7-Conv7	√	3	81.2	79
ele-sum	Conv4-fc7-Conv7	×	3	80.3	77
ele-sum	Conv4-fc7-Conv7	√	5	79.8	67
ele-sum	Conv4-fc7-Conv7	×	5	77.5	60
ele-sum	Conv4-fc7-Conv7	√	7	79.8	62
ele-sum	Conv4-fc7-Conv7	×	7	77.5	57

表2 不同损失函数在VOC2007+2012下检测结果

Table 2 Test results of different methods under VOC2007+2012

Algorithm	Change focal loss?	mAP/%
UWN	√	81.2
UWN	×	80.9

表3 不同检测算法在VOC2007+2012下检测结果

Table 3 Test results of different methods under VOC2007+2012

Method	Backbone	Train	mAP/%	FPS(1080Ti)
Faster R-CNN	ResNet-101	VOC2007+2012	73.2	7
YOLO	VGG-16	VOC2007+2012	66.4	96
YOLOv2	DarkNet-19	VOC2007+2012	78.6	80
SSD	VGG-16	VOC2007+2012	77.2	120
DSSD	ResNet-101	VOC2007+2012	78.6	10
RFB-Net	VGG-16	VOC2007+2012	80.5	75
UWN	VGG-16	VOC2007+2012	81.2	68
UWN	VGG-16	Aerial photography	82.3	66

表4 样本数据在不同增强方式下的检测结果

Table 4 Test results of sample data under different enhancement

Data	Method	mAP/%
VOC2007+2012	扭曲长宽	78.2
	翻转	78.4
	扭曲色域	78.8
	扭曲长宽+翻转	79.2
	扭曲长宽+扭曲色域	79.6
	扭曲色域+翻转	80.3
	本文方式	81.2

图9 不同算法在VOC数据集上的对比结果

Fig.9 Comparison results of different algorithms on VOC dataset

图10 不同算法在航拍数据集上的对比结果

Fig.10 Comparison results of different algorithms on Aerial photography dataset

参考文献 23

[1]	REN S Q, HE K M, GIRSHICK R B, 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 URL
[2]	LIU W, ANGUELOV D, ERHAN D, et al. SSD: single shot multibox detector[C]// LNCS 9905: Proceedings of the 14th European Conference on Computer Vision, Amsterdam, Oct 11-14, 2016. Cham: Springer, 2016: 21-37.
[3]	REDMON J, DIVVALA S, GIRSHICK R, et al. You only look once: unified, real-time object detection[C]// Proceedings of the 2016 IEEE Conference on Computer Vision and Pattern Recognition, Las Vegas, Jun 27-30, 2016. Washington: IEEE Computer Society, 2016: 779-788.
[4]	FU C Y, LIU W, RANGA A, et al. DSSD: deconvolutional single shot detector[J]. arXiv.1701.06659, 2017.
[5]	SZEGEDY C, IOFFE S, VANHOUCKE V, et al. Inception-v4, Inception-ResNet and the impact of residual connections on learning[C]// Proceedings of the 31st AAAI Conference on Artificial Intelligence, San Francisco, Feb 4-9, 2017. Menlo Park: AAAI, 2017: 4278-4284.
[6]	LI Z, ZHOU F. FSSD: feature fusion single shot multibox detector[J]. arXiv:1712.00960, 2017.
[7]	ZHANG N, DONAHUE J, GIRSHICK R B, et al. Part-based R-CNNs for fine-grained category detection[C]// LNCS 8689: Proceedings of the 13th European Conference on Computer Vision, Zurich, Sep 6-12, 2014. Cham: Springer, 2014: 834-849.
[8]	HU J, SHEN L, SUN G, et al. Squeeze-and-excitation networks[J]. IEEE Transactions on Pattern Analysis and Machine Intelligence, 2020, 42(8): 2011-2023. DOI URL
[9]	CHENG J L, SCHARENBROICH L, BALDI P, et al. Sigmoid: a software infrastructure for pathway bioinformatics and systems biology[J]. IEEE Intelligent Systems, 2005, 20(3): 68-75.
[10]	GAO F, YANG C G, GE Y S, et al. Dense receptive field network: a backbone network for object detection[C]// LNCS 11729: Proceedings of the 28th International Conference on Artificial Neural Networks, Munich, Sep 17-19, 2019. Cham: Springer, 2019: 105-118.
[11]	江雯, 陈更生, 杨帆, 等. 基于Sobel算子的自适应图像缩放算法[J]. 计算机工程, 2010, 36(7): 214-216.
	JIANG W, CHEN G S, YANG F, et al. Self-adaptive image scaling algorithm based on Sobel operator[J]. Computer Engineering, 2010, 36(7): 214-216.
[12]	罗海波, 许凌云, 惠斌, 等. 基于深度学习的目标跟踪方法研究现状与展望[J]. 红外与激光工程, 2017, 46(5): 6-12.
	LUO H B, XU L Y, HUI B, et al. Status and prospect of target tracking based on deep learning[J]. Infrared and Laser Engineering, 2017, 46(5): 6-12.
[13]	路雪, 刘坤, 程永翔. 一种深度学习的非机动车辆目标检测算法[J]. 计算机工程与应用, 2019, 55(8): 182-188. DOI
	LU X, LIU K, CHENG Y X. Non-motor vehicle target detection based on deep learning[J]. Computer Engineering and Applications, 2019, 55(8): 182-188.
[14]	陈幻杰, 王琦琦, 杨国威, 等. 多尺度卷积特征融合的SSD目标检测算法[J]. 计算机科学与探索, 2019, 13(6): 1049-1061. DOI
	CHEN H J, WANG Q Q, YANG G W, et al. SSD object detection algorithm with multi-scale convolution feature fusion[J]. Journal of Frontiers of Computer Science and Technology, 2019, 13(6): 1049-1061.
[15]	LIU S T, HUANG D, WANG Y H. Receptive field block net for accurate and fast object detection[C]// LNCS 11215: Proceedings of the 15th European Conference on Computer Vision, Munich, Sep 8-14, 2018. Cham: Springer, 2018: 404-419.
[16]	ALY H A, DUBOIS E. Image up-sampling using total-variation regularization with a new observation model[J]. IEEE Transactions on Image Processing, 2005, 14(10): 1647-1659. DOI URL
[17]	奚琦, 张正道, 彭力. 基于改进MDSSD的小目标实时检测算法[J]. 激光与光电子学进展, 2020, 57(20): 89-97.
	XI Q, ZHANG Z D, PENG L. Small target real-time detection algorithm based on improved MDSSD[J]. Laser & Optoelectronics Progress, 2020, 57(20): 89-97.
[18]	ZHANG Y L, YUAN Y, FENG Y C, et al. Hierarchical and robust convolutional neural network for very high-resolution remote sensing object detection[J]. IEEE Transactions on Geoscience and Remote Sensing, 2019, 57(8): 5535-5548. DOI URL
[19]	LOWE D G. Distinctive image features from scale-invariant keypoints[J]. International Journal of Computer Vision, 2004, 60(2): 91-110. DOI URL
[20]	LIN T Y, GOYAL P, GIRSHICK R B, et al. Focal loss for dense object detection[J]. IEEE Transactions on Pattern Analysis and Machine Intelligence, 2020, 42(2): 318-327. DOI URL
[21]	WANG Z, DEDO M I, GUO K, et al. Efficient recognition of the propagated orbital angular momentum modes in turbulences with the convolutional neural network[J]. IEEE Photonics Journal, 2019, 11(3): 1-14.
[22]	DOLLAR P, APPEL R, BELONGIE , et al. Fast feature pyramids for object detection[J]. IEEE Transactions on Pattern Analysis and Machine Intelligence, 2014, 36(8): 1532-1545. DOI URL
[23]	陈璠, 彭力. 多层级重叠条纹特征融合的行人重识别[J]. 计算机科学与探索, 2021, 15(9): 1753-1761. DOI
	CHEN F, PENG L. Person re-identification based on multi-level feature fusion with overlapping stripes[J]. Journal of Frontiers of Computer Science and Technology, 2021, 15(9): 1753-1761. DOI

加权网络下的小目标检测算法

Small Object Detection Algorithm Based on Weighted Network

RichHTML

PDF

可视化

摘要/Abstract

引用本文

使用本文

图/表 14

参考文献 23

相关文章 3

编辑推荐

Metrics

[1]	赵鹏飞, 谢林柏, 彭力. 融合注意力机制的深层次小目标检测算法[J]. 计算机科学与探索, 2022, 16(4): 927-937.
[2]	李科岑, 王晓强, 林浩, 李雷孝, 杨艳艳, 孟闯, 高静. 深度学习中的单阶段小目标检测方法综述[J]. 计算机科学与探索, 2022, 16(1): 41-58.
[3]	李玉华, 贺人贵, 钟开, 李瑞轩. 动态加权网络中节点重要度评估[J]. 计算机科学与探索, 2012, 6(2): 134-143.