基于改进<b>YOLO</b>的双网络桥梁表观病害快速检测算法

彭雨诺; 刘敏; 万智; 蒋文博; 何文轩; 王耀南

doi:10.16383/j.aas.c210807

基于改进YOLO的双网络桥梁表观病害快速检测算法

doi: 10.16383/j.aas.c210807

彭雨诺^{1, 2,},
刘敏^{1, 2,},
万智^3,,
蒋文博^{1, 2,},
何文轩^{1, 2,},
王耀南^{1, 2,}

1.
湖南大学电气与信息工程学院长沙 410082
2.
湖南大学机器人视觉感知与控制技术国家工程研究中心长沙 410082
3.
湖南桥康智能科技有限公司长沙 410021

基金项目: 国家自然科学基金(61771189, 62073126, 62027810), 湖南省自然科学基金杰出青年基金(2020JJ2008), 湖南省交通运输厅科技进步与创新计划项目(201734, 202138)资助

详细信息

作者简介:
彭雨诺：湖南大学电气与信息工程学院硕士研究生. 2019年获得湖南大学学士学位. 主要研究方向为深度学习, 图像处理. E-mail: pengyunuo@hnu.edu.cn

刘敏：湖南大学电气与信息工程学院教授. 2012年获美国加州大学河滨分校博士学位. 主要研究方向为模式识别和机器学习. 本文通信作者. E-mail: liu_min@hnu.edu.cn

万智：湖南桥康智能科技有限公司总工程师. 1999年获湖南大学学士学位, 2010年获中南大学博士学位. 主要研究方向为公路交通和环保监测. E-mail: xbh0n3@163.com

蒋文博：湖南大学电气与信息工程学院硕士研究生. 2018年获桂林电子科技大学学士学位. 主要研究方向为图像处理和模式识别. E-mail: jiang_wenbo@hnu.edu.cn

何文轩：湖南大学电气与信息工程学院硕士研究生. 2020年获昆明理工大学学士学位. 主要研究方向为图像处理和模式识别. E-mail: hwx@hnu.edu.cn

王耀南：中国工程院院士, 湖南大学电气与信息工程学院教授. 1995年获湖南大学博士学位. 主要研究方向为机器人学, 智能控制和图像处理. E-mail: yaonan@hnu.edu.cn

计量
- 文章访问数: 1844
- HTML全文浏览量: 855
- PDF下载量: 649
- 被引次数: 0
出版历程
- 收稿日期: 2021-03-06
- 录用日期: 2021-11-17
- 网络出版日期: 2021-12-20
- 刊出日期: 2022-04-13

A Dual Deep Network Based on the Improved YOLO for Fast Bridge Surface Defect Detection

PENG Yu-Nuo^{1, 2
,},
LIU Min^{1, 2
,},
WAN Zhi^3
,,
JIANG Wen-Bo^{1, 2
,},
HE Wen-Xuan^{1, 2
,},
WANG Yao-Nan^{1, 2
,}

1.
College of Electrical and Information Engineering, Hunan University, Changsha 410082
2.
National Engineering Research Center for Robot Visual Perception and Control Technology, Hunan University, Changsha 410082
3.
Hunan Qiaokang Intelligent Technology Company Limited, Changsha 410021

Funds: Supported by National Natural Science Foundation of China (61771189, 62073126, 62027810), Natural Science Foundation for Distinguished Young Scholars of Hunan Province (2020JJ2008), Science and Technology Progress and Innovation Plan of Department of Transportation of Hunan Province (201734, 202138)

More Information

Author Bio:
PENG Yu-Nuo　Master student at the College of Electrical and Information Engineering, Hunan University. He received his bachelor degree from the Hunan University in 2019. His research interest covers deep learning and image processing

LIU Min　Professor at the College of Electrical and Information Engineering, Hunan University. He received his Ph.D. degree from University of California, Reverside in 2012. His research interest covers pattern recognition and machine learning. Corresponding author of this paper

WAN Zhi　Chief engineer at Hunan Qiaokang Intelligent Technology Company Limited. He received his bachelor’s degree from Hunan University in 1999 and Ph.D. degree from the Central South University in 2010. His research interest covers highway traffic and environmental protection monitoring

JIANG Wen-Bo　Master student at the College of Electrical and Information Engineering, Hunan University. He received his bachelor degree from the Guilin University of Electronic Technology in 2018. His research interest covers image processing and pattern recognition

HE Wen-Xuan　Master student at the College of Electrical and Information Engineering, Hunan University. He received his bachelor degree from Kunming University of Science and Technology in 2020. His research interest covers image processing and pattern recognition

WANG Yao-Nan　Academician at Chinese Academy of Engineering, professor at College of Electrical and Information Engineering, Hunan University. He received his Ph.D. degree from the Hunan University in 1995. His research interest covers robotics, intelligent control and image processing

摘要

摘要: 桥梁表观病害检测是确保桥梁安全的关键步骤. 然而, 桥梁表观病害类型多样, 不同病害间外观差异显著且病害之间可能发生重叠, 现有算法无法实现快速且准确的桥梁多病害检测. 针对这一问题, 对YOLO (You only look once) 进行了改进, 提出了YOLO-lump和YOLO-crack以提高网络检测多病害的能力, 进而形成基于双网络的桥梁表观病害快速检测算法. 一方面, YOLO-lump在较大的滑动窗口图像上实现块状病害的检测. 在YOLO-lump中, 提出了混合空洞金字塔模块, 其结合了混合空洞卷积与空间金字塔池化, 用于提取稀疏表达的多尺度特征, 同时可以避免空洞卷积造成的局部信息丢失; 另一方面, YOLO-crack在较小的滑动窗口图像上实现裂缝病害的检测. 在YOLO-crack中, 提出了下采样注意力模块, 利用1×1卷积和3×3分组卷积分别解耦特征的通道相关性和空间相关性, 可以增强裂缝在下采样阶段的前景响应, 减少空间信息的损失. 实验结果表明, 该算法能够提高桥梁表观病害检测的精度, 同时可实现病害的实时检测.
- 桥梁表观病害检测 /
- 深度卷积神经网络 /
- 空间金字塔模块 /
- 注意力机制
Abstract: Surface defect detection is a critical step to ensure bridge safety. However, there are various types of bridge surface defects, different defects have a wide range of variation in appearance and generally overlap with each other. The existing algorithms cannot efficiently and precisely detect such defects. To solve this problem, we improve the YOLO (You only look once) to enhance the performance of the network to detect multiple defects, YOLO-lump and YOLO-crack are proposed to form a dual deep network for fast bridge surface defect detection. On the one hand, the YOLO-lump can realize the detection of the lump defects on larger sub-images, by employing a hybrid dilated pyramid module based on the hybrid dilated convolution and the spatial pyramid pooling to extract multi-scale features and to avoid losing local information caused by the dilated convolution. On the other hand, the YOLO-crack can realize the detection of the crack defects on smaller sub-images, by proposing a downsampling attention module which uses the 1×1 convolution and the 3×3 group convolution to respectively map cross-channel correlation and spatial correlation of features, enhancing the foreground response of the crack in the downsampling stage and reducing the loss of spatial information. Experimental results show that the proposed algorithm can improve the detection accuracy of the bridge surface defects and realize real-time detection.
- Bridge surface defect detection /
- deep convolutional neural network /
- spatial pyramid module /
- attention mechanism
注释:

1) 收稿日期 2021-03-06 录用日期 2021-11-17 Manuscript received March 6, 2021; accepted November 17,2021 国家自然科学基金 (61771189, 62073126, 62027810), 湖南省自然科学基金杰出青年基金 (2020JJ2008), 湖南省交通运输厅科技进步与创新计划项目 (201734, 202138) 资助 Supported by National Natural Science Foundation of China (61771189, 62073126, 62027810), Natural Science Foundation for Distinguished Young Scholars of Hunan Province (2020JJ2008), Science and Technology Progress and Innovation Plan of Department of Transportation of Hunan Province (201734, 202138)

2) 本文责任编委胡清华 Recommended by Associate Editor HU Qing-Hua 1. 湖南大学电气与信息工程学院长沙 410082 2. 湖南大学机器人视觉感知与控制技术国家工程研究中心长沙 410082 3. 湖南桥康智能科技有限公司长沙 410021 1. College of Electrical and Information Engineering, HunanUniversity, Changsha 410082 2. National Engineering Research Center for Robot Visual Perception and Control Technology, Hunan University, Changsha 410082 3. Hunan Qiaokang Intelligent Technology Company Limited, Changsha 410021

HTML全文

图 1 双网络桥梁表观病害快速检测算法整体框架

Fig. 1 Overview of the dual deep network for fast bridge surface defect detection

下载: 全尺寸图片幻灯片

图 2 GAN网络生成的桥梁表观病害图像

Fig. 2 Bridge surface defect images generated by GAN network

下载: 全尺寸图片幻灯片

图 3 混合空洞金字塔模块

Fig. 3 The hybrid dilated pyramid module

下载: 全尺寸图片幻灯片

图 4 下采样注意力模块

Fig. 4 The downsampling attention module

下载: 全尺寸图片幻灯片

图 5 BIR-X-LITE机器人数据采集过程

Fig. 5 The process of data acquisition by the BIR-X-LITE robot

下载: 全尺寸图片幻灯片

图 6 不同病害之间大小比较

Fig. 6 Comparison of defects with different sizes

下载: 全尺寸图片幻灯片

图 7 训练数据示例

Fig. 7 Examples of the training dataset

下载: 全尺寸图片幻灯片

图 8 不同块状病害检测网络的PR曲线

Fig. 8 Precision-recall curves of different detectors on the lump dataset

下载: 全尺寸图片幻灯片

图 9 本文方法和其他方法在不同桥梁表观图像上的测试结果

Fig. 9 Results of the proposed method and other methods on various bridge surface images

下载: 全尺寸图片幻灯片

图 10 不同裂缝病害检测网络的PR曲线

Fig. 10 Precision-Recall curves of different detectors on the crack dataset

下载: 全尺寸图片幻灯片

图 11 Grad-CAM++可视化结果

Fig. 11 Grad-CAM++ visualization results

下载: 全尺寸图片幻灯片

表 1 桥梁表观图像数据库

Table 1 Dataset of the bridge surface images

采集时间	桥梁名称	图像数目	数据大小
2018-09	东临路大桥	2126张	6.4 GB
	红旗二号桥	10470张	31.4 GB
	荒唐亭大桥	6090张	18.2 GB
	马家河大桥	1402张	4.2 GB
2018-10	南川河大桥	4055张	13.6 GB
	宁家冲大桥	17119张	48.4 GB
	天马大桥	3614张	11.9 GB
2018-11	铜陵长江大桥	19961张	116.0 GB
2019-07	新庆大桥	25784张	225.5 GB
2019-04	广东潮汕大桥	79000张	317.1 GB
总计	10座大桥	169621张	792.7 GB

下载: 导出CSV

表 2 训练/验证/测试数据集

Table 2 Training/validation/testing datasets

类型	训练集 (正/负样本)	验证集 (正/负样本)	测试集 (正/负样本)
块状病害	7668张 (2611/5057)	2924张 (978/1946)	51231张 (345/50886)
裂缝病害	5643张 (3283/2360)	1453张 (873/580)	51079张 (193/50886)

下载: 导出CSV

表 3 不同输入大小下块状病害检测结果对比

Table 3 Results of lump defect detection with different input sizes

输入大小	召回率	准确率	F1	mAP	检测时间
704 × 704	80.6%	79.3%	79.9%	85.6%	37.8 ms
608 × 608	86.5%	83.9%	85.2%	89.2%	24.7 ms
512 × 512	85.8%	84.5%	85.1%	88.6%	18.8 ms
416 × 416	80.7%	77.4%	79.0%	82.1%	16.6 ms

下载: 导出CSV

表 4 YOLO-lump网络消融实验

Table 4 Ablation experiment on the YOLO-lump

网络模型	召回率	准确率	F1	mAP	检测时间
YOLOv4	85.8%	84.5%	85.1%	88.6%	18.8 ms
YOLO-lump-A	86.1%	84.8%	85.4%	89.3%	18.8 ms
YOLO-lump-B	87.2%	84.4%	85.8%	90.7%	18.8 ms
YOLO-lump-C	79.3%	68.1%	73.3%	74.9%	26.7 ms
YOLO-lump-D	84.4%	83.3%	83.8%	87.7%	20.1 ms
YOLO-lump	86.4%	89.7%	88.0%	92.7%	20.4 ms
YOLO-lump-E	88.7%	89.5%	89.1%	93.5%	24.3 ms

下载: 导出CSV

表 5 块状病害检测网络对比实验

Table 5 Comparison of different detectors on the lump dataset

网络模型	特征提取网络	mAP	检测时间
SSD	VGG-16	85.1%	30.3 ms
Faster-RCNN	ResNet-101	86.9%	34.9 ms
RetinaNet	ResNet-101	89.5%	41.5 ms
FCOS	ResNet-101	87.9%	28.8 ms
EfficientDet	EfficientNet	89.6%	22.3 ms
YOLOv3	Darknet-53	87.6%	15.4 ms
Improved-YOLOv3	Darknet-53	89.3%	15.4 ms
YOLOv4	CSPDarknet-53	88.6%	18.8 ms
YOLO-lump	CSPDarknet-53	92.7%	20.4 ms

下载: 导出CSV

表 6 YOLO-crack网络消融实验

Table 6 Ablation experiment on the YOLO-crack

网络模型	召回率	准确率	F1	mAP	检测时间
YOLOv4	80.8%	79.4%	80.2%	84.5%	29.7 ms
YOLO-crack-A	77.5%	82.6%	80.0%	85.0%	29.7 ms
YOLO-crack-B	85.6%	76.7%	81.0%	85.7%	29.7 ms
YOLO-crack-C	78.7%	79.1%	78.9%	83.8%	17.1 ms
YOLO-crack-D	79.0%	79.5%	79.2%	84.6%	16.5 ms
YOLO-crack	80.2%	81.2%	80.7%	86.2%	17.6 ms
YOLO-crack-E	77.9%	80.9%	79.4%	82.5%	17.7 ms

下载: 导出CSV

表 7 注意力模块对比实验

Table 7 Comparison of different attention modules

网络模型	mAP	检测时间
YOLO-crack-D	84.6%	16.5 ms
YOLO-crack-D+SE注意力模块	84.9%	16.9 ms
YOLO-crack-D+CBAM注意力模块	85.7%	17.4 ms
YOLO-crack-D+下采样注意力模块	86.2%	17.6 ms

下载: 导出CSV

表 8 裂缝病害检测网络对比实验

Table 8 Comparison of different detectors on the crack dataset

网络模型	特征提取网络	mAP	检测时间
SSD	VGG-16	79.8%	45.2 ms
Faster-RCNN	ResNet-101	81.2%	54.7 ms
RetinaNet	ResNet-101	82.9%	58.4 ms
FCOS	ResNet-101	83.4%	42.9 ms
EfficientDet	EfficientNet	83.5%	27.4 ms
YOLOv3	Darknet-53	82.3%	23.8 ms
Improved-YOLOv3	Darknet-53	84.1%	23.8 ms
YOLOv4	CSPDarknet-53	84.5%	29.7 ms
YOLO-crack	CSPDarknet-39	86.2%	17.6 ms

下载: 导出CSV

表 9 实际应用测试结果

Table 9 Results of the practical application

测试数据集	图像数量	块状病害检测					裂缝病害检测					检测时间
测试数据集	图像数量	GT	TP	FN	FP	召回率	GT	TP	FN	FP	召回率	检测时间
东临路大桥	872	8	8	0	907	100%	6	5	1	1478	83.3%	995 ms/张
红旗二号桥	3265	26	25	1	2132	96.2%	17	17	0	13582	100%	995 ms/张
荒唐亭大桥	2929	22	19	3	3295	86.4%	26	25	1	10115	96.2%	994 ms/张
马家河大桥	836	11	9	2	1041	81.8%	7	7	0	3569	100%	993 ms/张
南川河大桥	2617	20	20	0	4238	100%	23	21	2	5331	91.3%	996 ms/张
宁家冲大桥	2453	28	27	1	3145	96.4%	28	26	2	9504	92.9%	997 ms/张
天马大桥	7107	65	62	3	7294	95.4%	90	86	4	22383	95.6%	996 ms/张
铜陵长江大桥	5962	46	45	1	8505	97.8%	57	55	2	26237	96.5%	996 ms/张
新庆大桥	6194	63	61	2	5598	96.8%	46	45	1	15394	97.8%	995 ms/张
广东潮汕大桥	19189	130	124	6	19869	95.4%	186	179	7	41908	96.2%	995 ms/张
总计	51424	419	400	19	56024	95.5%	486	466	20	149501	95.9%	995 ms/张

下载: 导出CSV

表 10 双网络算法与单网络性能比较

Table 10 Comparison of performance between the dual deep network and the single network

检测策略	蜂窝病害				漏筋病害				孔洞病害				裂缝病害				检测时间
检测策略	召回率	准确率	F1	mAP	召回率	准确率	F1	mAP	召回率	准确率	F1	mAP	召回率	准确率	F1	mAP	检测时间
双网络	85.2%	84.6%	84.9%	86.7%	87.1%	86.5%	86.8%	89.8%	87.3%	85.2%	86.2%	89.3%	80.8%	79.4%	80.1%	84.5%	34.8 ms
单网络	78.5%	77.2%	77.8%	80.6%	83.8%	84.3%	84.0%	84.4%	84.4%	83.1%	83.7%	85.0%	78.7%	76.8%	77.7%	80.1%	30.3 ms

下载: 导出CSV

参考文献(49)

[1]	马建, 孙守增, 杨琦. 中国桥梁工程学术研究综述: 2014. 中国公路学报, 2014, 27(5): 1-96 doi: 10.3969/j.issn.1001-7372.2014.05.001 Ma Jian, Sun Shou-Zeng, Yang Qi. Review on china's bridge engineering research: 2014. China Journal of High-way and Transport, 2014, 27(5): 1-96 doi: 10.3969/j.issn.1001-7372.2014.05.001
[2]	陈榕峰, 徐群丽, 秦凯强. 桥梁裂缝智能检测系统的研究新进展. 公路, 2019, 64(05): 101-105 Chen Rong-Feng, Xu Kai-Li, Qin Kai-Qiang. Research progress of intelligent bridge crack detection system. Highway, 2019, 64(05): 101-105
[3]	钟新谷, 彭雄, 沈明燕. 基于无人飞机成像的桥梁裂缝宽度识别可行性研究. 土木工程学报, 2019, 52(4): 52-61 Zhong Xin-Gu, Peng Xiong, Shen Ming-Yan. Study on the feasibility of identifying concrete crack width with images acquired by unmanned aerial vehicles. China Civil Engineering Journal, 2019, 52(4): 52-61
[4]	Lin W G, Sun Y C, Yang Q N. Real-time comprehensive image processing system for detecting concrete bridges crack. Computers and Concrete, 2019, 23(6): 445-457
[5]	Sutter B, Lelevé A, Pham M T, Gouin O, Jupille N, Kuhn M. A semi-autonomous mobile robot for bridge inspection. Automation Construction, 2018, 91(JUL.): 111-119
[6]	Hirai H, Ishii K. Development of dam inspection underwater robot. Journal of Robotics, Networking and Artificial Life, 2019, 6(1): 18-22 doi: 10.2991/jrnal.k.190531.004
[7]	钟钒, 周激流, 郎方年, 何坤, 黄梅. 边缘检测滤波尺度自适应选择算法. 自动化学报, 2007, 33(8): 867-870 Zhong Fan, Zhou Ji-Liu, He Kun, Huang Mei. Adaptive scale filtering for edge detection. Acta Automatica Sinica, 2007, 33(8): 867-870
[8]	Win M, Bushroa A R, Hassan M A. A contrast adjustment thresholding method for surface defect detection based on mesoscopy. IEEE Transactions on Industrial Informatics, 2017, 11(3): 642-649
[9]	Kamaliardakani M, Sun L, Ardakani M K. Sealed-crack detection algorithm using heuristic thresholding approach. Journal of Computing in Civil Engineering, 2016, 30(1): 04014110 doi: 10.1061/(ASCE)CP.1943-5487.0000447
[10]	German S, Brilakis I, Desroches R. Rapid entropy-based detection and properties measurement of concrete spalling with machine vision for post-earthquake safety assessments. Advanced Engineering Informatics, 2012, 26(4): 846-858 doi: 10.1016/j.aei.2012.06.005
[11]	张慧, 王坤峰, 王飞跃. 深度学习在目标视觉检测中的应用进展与展望. 自动化学报, 2017, 43(8): 1289−1305 Zhang Hui, Wang Kun-Feng, Wang Fei-Yue. Advances and perspectives on applications of deep learning in visual object detection. Acta Automatica Sinica, 2017, 43(8): 1289-1305
[12]	Shi Y, Cui L, Qi Z. Automatic road crack detection using random structured forests. IEEE Transactions on Intelligent Transportation Systems, 2016, 17(12): 3434-3445 doi: 10.1109/TITS.2016.2552248
[13]	勾红叶, 杨彪, 华辉, 谢蕊. 桥梁信息化及智能桥梁2019年度研究进展. 土木与环境工程学报, 2020, 42(5): 14-27 Gou Ye-Hong, Yang Biao, Hua Hui, Xie Rui. Research progress of bridge informatization and intelligent bridge in 2019. Journal of Civil and Environmental Engineering, 2020, 42(5): 14-27
[14]	Zou Q, Zhang Z, Li Q, Qi X, Wang Q, Wang S. Deepcrack: Learning hierarchical convolutional features for crack detection. IEEE Transactions on Image Processing, 2019, 28(3): 1498-1512 doi: 10.1109/TIP.2018.2878966
[15]	李良福, 马卫飞, 李丽, 陆铖. 基于深度学习的桥梁裂缝检测算法研究. 自动化学报, 2019, 45(9): 1727−1742 Li Liang-Fu, Ma Wei-Fei, Li Li, Lu Cheng. Research on detection algorithm for bridge cracks based on deep learning. Acta Automatica Sinica, 2019, 45(9): 1727-1742
[16]	Kim I H, Jeon H, Baek S C, Hong W H. Application of crack identification techniques for an aging concrete bridge inspection using an unmanned aerial vehicle. Sensors, 2018, 18(6): 1881 doi: 10.3390/s18061881
[17]	Zhang C B, Chang C, Maziar J. Concrete bridge surface damage detection using a single-stage detector. Computer‐Aided Civil and Infrastructure Engineering, 2020, 35(4): 389-409 doi: 10.1111/mice.12500
[18]	Li S Y, Zhao X F, Zhou G Y. Automatic pixel‐level multiple damage detection of concrete structure using fully convolutional network. Computer‐Aided Civil and Infrastructure Engineering, 2019, 34(7): 616-634 doi: 10.1111/mice.12433
[19]	Yang L, Li B, Li W. Deep concrete inspection using unmanned aerial vehicle towards CSSC database. In: Proceeding of the 2017 IEEE/RSJ International Conference on Intelligent Robots and Systems. Vancouver, Canada: IEEE, 2017. 24−28
[20]	Mundt M, Majumder S, Murali S. Meta-learning convolutional neural architectures for multi-target concrete defect classification with the concrete defect bridge image dataset. In: Proceedings of the 2019 IEEE Conference on Computer Vision and Pattern Recognition. Long Beach, USA: IEEE, 2019. 11188−11197
[21]	Hüthwohla P, Lu R D, Brilakisa I. Multi-classifier for reinforced concrete bridge defect. Automation in Construction, 2019, 105(SEP.): 102824.1-102826.15
[22]	Redmon J, Divvala S, Girshick R, Farhadi A. You only look once: unified, real-time object detection. In: Proceedings of the 2016 IEEE Conference on Computer Vision and Pattern Recognition. Las Vegas, USA: IEEE, 2016. 779−788
[23]	Bochkovskiy A, Wang C Y. Yolov4: Optimal speed and accuracy of object detection [Online], available: https://arxiv.org/abs/2004.10934, April 23, 2020
[24]	He K M, Zhang X, Ren S. Spatial pyramid pooling in deep convolutional networks for visual recognition. IEEE Transactions on Pattern Analysis and Machine Intelligence, 2014, 37(9): 1904-1916
[25]	Wang P, Chen P, Yuan Y. Understanding convolution for semantic segmentation. In: Proceedings of the 2018 IEEE Winter Conference on Applications of Computer Vision. Lake Tahoe, USA: IEEE, 2018. 1451−1460
[26]	Hu J, Shen L, Surr G. Squeeze-and-excitation networks. In: Proceedings of the 2018 IEEE Conference on Computer Vision and Pattern Recognition. Salt Lake City, USA: IEEE, 2018. 7132−7141
[27]	Woo S, Park J, Lee J Y. CBAM: Convolutional block attention module. In: Proceedings of the 2018 European Conference on Computer Vision. Munich, Germany: Springer, 2018. 3−19
[28]	Chollet F. Xception: Deep learning with depth-wise separable convolutions. In: Proceedings of the 2017 IEEE Conference on Computer Vision and Pattern Recognition. Hawaii, USA: IEEE, 2017
[29]	Tan M X, Le Q V. EfficientNet: Rethinking model scaling for convolutional neural networks. In: Proceedings of the 2019 International Conference on Machine Learning. Los Angeles, USA: IEEE, 2019. 97: 6105−6114
[30]	Wang C Y, Mark-Liao H Y, Wu Y H, Chen P Y. CSPNet: A new backbone that can enhance learning capability of CNN. In: Proceedings of the 2020 IEEE Conference on Computer Vision and Pattern Recognition Workshop. Seattle, USA: IEEE, 2020. 390−391
[31]	Wang T H, Liu M Y, Zhu J Y, Tao A, Catanzaro B. High-resolution image synthesis and semantic manipulation with conditional GANs. In: Proceedings of the 2018 IEEE Conference on Computer Vision and Pattern Recognition. Salt Lake City, USA: IEEE, 2018. 8798−8807
[32]	Lin T Y, Goyal P, Girshick R. Focal loss for dense object detection. In: Proceedings of the 2017 IEEE International Conference on Computer Vision. Venice, Italy: IEEE, 2017. 2999−3007
[33]	Jiang W B, Liu M, Peng Y N, Wu L H, Wang Y N. HDCB-Net: A neural network with the hybrid dilated convolution for pixel-level crack detection on concrete bridges. IEEE Transactions on Industrial Informatics, 2020
[34]	Goodfellow I, Mirza M, Xu B, Courville A, Bengio Y. Generative adversarial networks. In: Proceedings of the 2014 Conference and Workshop on Neural Information Processing Systems. Montreal, Canada: 2014.
[35]	刘建伟, 谢浩杰, 罗雄麟. 生成对抗网络在各领域应用研究进展. 自动化学报, 2020, 46(12): 2500−2536 Liu Jian-Wei, Xie Hao-Jie, Luo Xiong-Lin. Research progress on application of generative adversarial networks in various fields. Acta Automatica Sinica, 2020, 46(12): 2500−2536
[36]	林懿伦, 戴星原, 李力, 王晓, 王飞跃. 人工智能研究的新前线: 生成式对抗网络. 自动化学报, 2018, 44(5): 775-792 Lin Yi-Lun, Dai Xing-Yuan, Li Li, Wang Xiao, Wang Fei-Yue. The new frontier of AI research: generative adversarial networks. Acta Automatica Sinica, 2018, 44(5): 775-792
[37]	Zheng Z, Zheng L, Yang Y. Unlabeled samples generated by GAN improve the person reidentification baseline in vitro. In: Proceedings of the 2017 IEEE International Conference on Computer Vision. Venice, Italy: IEEE, 2017. 3774−3782
[38]	Yang Q, Yan P, Zhang Y. Low-dose CT image denoising using a generative adversarial network with wasserstein distance and perceptual loss. IEEE Transactions on Medical Imaging, 2018: 1348-1357
[39]	Nie D, Trullo R, Lian J, Wang L, Petitjean C. Medical image synthesis with deep convolutional adversarial networks. IEEE Transactions on Biomedical Engineering, 2018, 65(12): 2720-2730 doi: 10.1109/TBME.2018.2814538
[40]	Chen L, Papandreou G, Kokkinos I. DeepLab: Semantic image segmentation with deep convolutional nets, atrous convolution, and fully connected CRFs. IEEE Transactions on Pattern Analysis and Machine Intelligence, 2018, 40(4): 834-848 doi: 10.1109/TPAMI.2017.2699184
[41]	Misra D. Mish: A self regularized non-monotonic activation function. In: Proceedings of the 2020 British Machine Vision Virtual Conference. Virtual Event, UK: 2020.
[42]	Sandler M, Howard A, Zhu M, Zhmoginov A. MobileNetV2: Inverted residuals and linear bottlenecks. In: Proceedings of the 2018 IEEE Conference on Computer Vision and Pattern Recognition. Salt Lake City, USA: IEEE, 2018. 4510−4520
[43]	Liu S, Qi L, Qin H F, Shi J P, Jia J Y. Path aggregation network for instance segmentation. In: Proceedings of the 2018 IEEE Conference on Computer Vision and Pattern Recognition. Salt Lake City, USA: IEEE, 2018. 8759−8768
[44]	Lin T Y, Dollar P, Girshick R, He K M, Hariharan B, Belongie S. Feature pyramid networks for object detection. In: Proceedings of the 2017 IEEE Conference on Computer Vision and Pattern Recognition. Venice, Italy: IEEE, 2017. 2117–2125
[45]	Dollár P, Zitnick C L. Fast edge detection using structured forests. IEEE Transactions on Pattern Analysis and Machine Intelligence, 2014, 37(8): 1558-1570.
[46]	Liu W, Anguelov D, Erhan D, Szegedy C, Reed S. SSD: Single shot multi-box detector. In: Proceedings of the 2016 European Conference on Computer Vision. Amsterdam, Netherlands: Springer, 2016. 21−37
[47]	Ren S Q, He K M, Girshick R, Sun J. Faster R-CNN: Towards real-time object detection with region proposal networks. IEEE Transactions on Pattern Analysis and Machine Intelligence, 2017, 39(6): 1137-1149 doi: 10.1109/TPAMI.2016.2577031
[48]	Tian Z, Shen C H, Chen H, He T. FCOS: Fully convolutional one-stage object detection. In: Proceedings of the 2019 IEEE International Conference on Computer Vision. Seoul, Korea: IEEE, 2019. 9627–9636
[49]	Chattopadhyay A, Sarkar A, Howlader P. Grad-CAM++: Improved visual explanations for deep convolutional networks. In: Proceedings of the 2018 IEEE Winter Conference on Applications of Computer Vision. Lake Tahoe, USA: IEEE, 2018.