改进YOLOv3的快速遥感机场区域目标检测

doi:10.19665/j.issn1001-2400.2021.05.019

摘要/Abstract

摘要：

遥感机场区域目标的检测有很大的军事意义和民用意义。为了取得快速且精确的检测效果,自主构建了更加符合具体任务的数据集;以一步回归全局检测为基础框架,针对数据集中类别分布不均衡问题,提出使用生成的方法用生成对抗网络进行有针对性的数据扩充,以获得具有领域变换特性、类数据分布更为均衡的数据集。同时,使用改进的双权重特征金字塔网络检测部件,来融合得到深层次可区分性的更加鲁棒的特征。实验结果表明,相比原网络,改进网络带来了4.98%的多类目标平均检测精确度以及8.33%的平均交并比的提升,分别达到了89.07%的多类目标平均检测精度以及61.97%的平均交并比。此外,改进网络的平均检测时间为0.062 5 s,相比类似检测率的RetinaNet-101网络速度约快5.3倍,体现了该网络的有效性以及对具体任务的实用性。

关键词: 目标检测, 图像处理, 神经网络, 机场区域, 遥感

Abstract:

The detection of remote sensing airport regional objects is of great military and civilian significance.In order to achieve fast and accurate detection results,a data set that is more mission-specific is independently constructed for the detection task.We use the representative network YOLOv3 of the one-step regression global detection method as the basic framework.For the problem of uneven distribution of categories in the data set,the use of generated data isproposed.The method uses generative adversarial networks to perform targeted data expansion to obtain data sets with domain transformation characteristics and more balanced distribution of different types of data.At the same time,the improved DWFPN detection component is used to fusion deeper distinguishable and more robust features.Experiments show that,compared with the original network,the improved network brings 4.98% increase in mean average precision (mAP) and 8.33% increase in average IOU,which reaches 89.07% mAP and 61.97% average IOU,respectively.At the same time,the average detection time of the improved network is 0.0625s,which is 7 times faster than the RetinaNet-101 network with a similar detection rate.Experiments prove the effectiveness of the network and its practicality for specific tasks.

Key words: object detection, image processing, neural network, airport area, remote sensing

中图分类号:

TP183

韩永赛,马时平,何林远,李承昊,朱明明,张飞. 改进YOLOv3的快速遥感机场区域目标检测[J]. 西安电子科技大学学报, 2021, 48(5): 156-166.

HAN Yongsai,MA Shiping,HE Linyuan,LI Chenghao,ZHU Mingming,ZHANG Fei. Detection of the object in the fast remote sensing airport area on the improved YOLOv3[J]. Journal of Xidian University, 2021, 48(5): 156-166.

图/表 17

图1

表1

表2

图2

图3

图4

表3

图5

图6

表4

表5

表6

表7

图7

图8

表8

图9

参考文献 22

[1]	吴帅, 徐勇, 赵东宁. 基于深度卷积网络的目标检测综述[J]. 模式识别与人工智能, 2018, 31(04):335-346.
	WU Shuai, XU Yong, ZHAO Dongning. Overview of Target Detection Based on Deep Convolutional Networks[J]. Pattern Recognition and Artificial Intelligence, 2018, 31(04):335-346.
[2]	VON GIOI R G, JAKUBOWICZ J, MOREL J M, et al. LSD:A Fast Line Segment Detector with a False Detection Control[J]. IEEE Transactions on Pattern Analysis and Machine Intelligence, 2010, 32(4):722-732. doi: 10.1109/TPAMI.2008.300
[3]	DUDA R O, HART P E. Use of the Hough Transformation to Detect Lines and Curves in Pictures[J]. Communications of the ACM, 1972, 15(1):11-15. doi: 10.1145/361237.361242
[4]	LI W, XIANG S, WANG H, et al. Robust Airplane Detection in Satellite Images[C]// IEEE International Conference on Image Processing.Piscataway:IEEE, 2011:2821-2824.
[5]	LIU L, SHI Z W. Aircraft Detection Based on Rotation Invariant and Sparse Coding in Remote Sensing Images[J]. Optik-International Journal and Electron Optics, 2014, 125(18):5327-5333.
[6]	李旭冬, 叶茂, 李涛. 基于卷积神经网络的目标检测研究综述[J]. 计算机应用研究, 2017, 34(10):2881-2886.
	LI Xudong, YE Mao, LI Tao. A Survey of Target Detection Research Based on Convolutional Neural Networks[J]. Journal of Computer Applications, 2017, 34(10):2881-2886.
[7]	ZHANG P, NIU X, DOU Y, et al. Airport Detection on Optical Satellite Images Using Deep Convolutional Neural Networks[J]. IEEE Geoscience and Remote Sensing Letters, 2017, 14(8):1183-1187. doi: 10.1109/LGRS.2017.2673118
[8]	ZHU T H, LI Y H, YE Q K, et al. Integrating Saliency and ResNet for Airport Detection in Large-Size Remote Sensing Images[C]// International Conference on Image Vision and Computing.Piscataway:IEEE, 2017:20-25.
[9]	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 and Machine Intelligence, 2017, 39(6):1137-1149. doi: 10.1109/TPAMI.2016.2577031
[10]	REDMON J, DIVVALA S, GIRSHICK R, et al. You Only Look Once:Unified,Real-Time Object Detection[C]// Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition.Piscataway:IEEE, 2016:779-788.
[11]	许悦雷, 朱明明, 马时平, 等. 迁移学习结合难分样本挖掘的机场目标检测[J]. 西安电子科技大学学报, 2018, 45(05):190-196.
	XU Yuelei, ZHU Mingming, MA Shiping, et al. Airport Target Detection Based on Migration Learning and Mining of Difficult Samples[J]. Journal of Xidian University, 2018, 45(5):190-196.
[12]	余东行, 张宁, 张保明, 等. 结合卷积神经网络与显著性特征的机场检测[J]. 测绘通报, 2019(7):44-49.
	YU Dongxing, ZHANG Ning, ZHANG Baoming, et al. Airport Detection Combining Convolutional Neural Networks and Salient Features[J]. Bulletin of Surveying and Mapping, 2019(7):44-49.
[13]	易翔, 王炳健. 视觉显著性指导的红外与可见光图像融合算法[J]. 西安电子科技大学学报, 2019, 46(1):27-32.
	YI Xiang, WANG Bingjian. Infrared and Visible Image Fusion Algorithm Guided by Visual Saliency[J]. Journal of Xidian University, 2019, 46(1):27-32.
[14]	REDMON J, FARHADI A. YOLOv3:An Incremental Improvement[EB/OL]. [2019-07-18]. https://arxiv.org/abs/1804.02767.
[15]	CHEN F, REN R, DE VOORDE T V, et al. Fast Automatic Airport Detection in Remote Sensing Images Using Convolutional Neural Networks[J]. Remote Sensing, 2018, 10(3):443. doi: 10.3390/rs10030443
[16]	EVERINGHAM M, GOOL L V, 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
[17]	RADFORD A, METZ L, CHINTALA S. Unsupervised Representation Learning with Deep Convolutional Generative Adversarial Networks (2015)[EB/OL]. [2020-01-23]. https://arxiv.org/abs/1511.06434.
[18]	庄福振, 罗平, 何清, 等. 迁移学习研究进展[J]. 软件学报, 2015, 26(1):26-39.
	ZHUANG Fuzhen, LUO Ping, HE Qing, et al. Research Progress of Transfer Learning[J]. Journal of Software, 2015, 26(1):26-39.
[19]	IANDOLA F, MOSKEWICZ M, KARAYEV S, et al. DenseNet:Implementing Efficient ConvNet Descriptor Pyramids (2014)[EB/OL]. [2020-01-23]. https://arxiv.org/abs/1404.1869.
[20]	LIN T Y, GOYAL P, GIRSHICK R, et al. Focal Loss for Dense Object Detection[C]// Proceedings of the IEEE International Conference on Computer Vision.Piscataway:IEEE, 2017:2980-2988.
[21]	DAI J, LI Y, HE K, et al. R-FCN:Object Detection Via Region-based Fully Convolutional Networks[C]// Neural Information Processing Systems,Piscataway:IEEE, 2016:379-387.
[22]	FU C Y, LIU W, RANGA A, et al. DSSD:Deconvolutional Single Shot Detector (2014)[EB/OL]. [2020-01-23]. https://arxiv.org/abs/1701.06659.

组块	层级	参数
输入端	Inputs layer	(416,416,3)
卷积块1	Conv 32×3×3&Conv 64×3×3_s2	(416,416,32)&(208,208,64)
残差块1	Residual Block 1×64& Conv 128×3×3_s2	(208,208,64)& (104,104,128)
残差块2	Residual Block 2×128& Conv 256×3×3_s2	(104,104,128)& (52,52,256)
残差块3	Residual Block 8×256& Conv 512×3×3_s2	(52,52,256)& (26,26,512)
残差块4	Residual Block 8×512& Conv 1 024×3×3_s2	(26,26,512)& (13,13,1 024)
残差块5	Residual Block 4×1 024& Conv 3×(512×1×1+1 024×3×3)	(13,13,1 024)& (13,13,1 024)

算法	机场	民航飞机	战斗机	直升机	运输机	油罐	平均检测精确率	平均交并比
YOLOv3	0.899 1	0.893 5	0.785 7	0.839 1	0.813 1	0.814 8	0.840 9	0.536 4
YOLOv3+T1+T2	0.909 3	0.913 6	0.878 8	0.928 8	0.878 2	0.835 6	0.890 7	0.619 7

算法	机场	民航飞机	战斗机	直升机	运输机	油罐	平均漏警概率
YOLOv3	0.051 1	0.057 2	0.164 1	0.111 3	0.137 2	0.134 9	0.109 3
YOLOv3+T1+T2	0.040 9	0.037 1	0.070 8	0.020 9	0.072 3	0.114 1	0.059 4
增量	-0.010 2	-0.020 1	-0.093 3	-0.090 4	-0.064 9	-0.020 8	-0.049 9

算法	机场	民航飞机	战斗机	直升机	运输机	油罐	mFA
YOLOv3	0.111 2	0.116 9	0.224 1	0.170 8	0.197 2	0.194 8	0.169 2
YOLOv3+T1+T2	0.101 3	0.096 9	0.131 2	0.081 3	0.132 1	0.173 9	0.119 5
增量	-0.009 9	-0.020 0	-0.092 9	-0.089 5	-0.065 1	-0.020 9	-0.049 7

检测网络	Faster R-CNN	R-FCN	DSSD	YOLOv3	RetinaNet-101	文中网络
平均检测精确率	0.773 4	0.830 2	0.812 5	0.840 9	0.901 2	0.890 7
时间/s	0.215 2	0.167 3	0.163 1	0.058 2	0.396 0	0.062 5