第16卷第2期 智能系统学报 Vol.16 No.2 2021年3月 CAAI Transactions on Intelligent Systems Mar.2021 D0:10.11992tis.201910004 网络出版地址：https:/kns.cnki.net/kcms/detail/23.1538.TP.20200717.1727.008.html 基于Faster R-CNN的多任务增强裂缝图像检测方法 毛莺池，唐江红，王静，平萍，王龙宝 (河海大学计算机与信息学院，江苏南京211100) 摘要：针对Faster R-CNN算法对多目标、小目标检测精度不高的问题，本文提出一种基于Faster R-CNN的多 任务增强裂缝图像检测(Multitask Enhanced Dam Crack Image Detection Based on Faster R-CNN,ME-Faster R- CNN)方法。同时提出一种基于K-means的多源自适应平衡TrAdaBoost的迁移学习方法(multi-source adaptive balance TrAdaBoost based on K-neans,K-MABtrA)轴助网络训练，解决样本不足问题。ME-Faster R-CNN将图片 输入ResNet--50网络提取特征：然后将所得特征图输入多任务增强RPN模型，同时改善RPN模型的锚盒尺寸和 大小以提高检测识别精度，生成候选区域：最后将特征图和候选区域发送到检测处理网络。K-MABA方法利 用K-means聚类删除与目标源差别较大的图像，再在多元自适应平衡TrAdaBoost迁移学习方法下训练模型。 实验结果表明：将ME-Faster R-CNN在K-MABtrA迁移学习的条件下应用于小数据集大坝裂缝图像集的平均 IoU为82.52%，平均精度mAP值为80.02%.与相同参数设置下的Faster R-CNN检测算法相比，平均IoU和 mAP值分别提高了1.06%和1.56%。 关键词：裂缝图像检测；Faster R-CNN;多任务检测；小目标检测；迁移学习；大坝安全；区域建议网络；小样本 中图分类号：TP391文献标志码：A文章编号：1673-4785(2021)02-0286-08 中文引用格式：毛莺池，唐江红，王静，等.基于Faster R-CNN的多任务增强裂缝图像检测方法.智能系统学报，2021， 16(2):286-293. 英文引用格式：MAO Yingchi,,TANG Jianghong,.WANG Jing,etal.Mlulti-.-task enhanced dam crack image detection based on Faster R-CNNIJI.CAAI transactions on intelligent systems,2021,16(2):286-293. Multi-task enhanced dam crack image detection based on Faster R-CNN MAO Yingchi,TANG Jianghong,WANG Jing,PING Ping,WANG Longbao (College of Computer and Information,Hohai University,Nanjing 211100,China) Abstract:To improve the accuracy of the detection of multiple small targets using the faster R-CNN model,we pro- pose a multi-task enhanced dam-crack image detection method based on faster R-CNN(ME-Faster R-CNN).In addition, to solve the problem of insufficient dam-crack samples,we propose a transfer learning method,multi-source adaptive balance TrAdaBoost based on K-means(K-MABtrA),to assist with network training.In the ME-Faster R-CNN,the ResNet-50 network is adopted to extract features from original images,obtain the feature map,and input a multi-task en- hanced region-proposal-network module to generate candidate regions by adopting the appropriate size and dimensions of the anchor box.Lastly,the features map and candidate regions are processed to detect dam cracks.The K-MABtrA method first uses K-means clustering to delete unsuitable images.Then,models are trained using the multi-source adapt- ive balance TrAdaBoost method.Our experimental results show that the proposed ME Faster R-CNN with the K-MAB- trA method can obtain an 82.52%average intersection over union (IoU)and 80.02%mean average precision(mAP). Compared with Faster R-CNN detection method using the same parameters,the average IoU and mAP values was in- creased by 1.06%and 1.56%,respectively. Keywords:crack image detection;Faster R-CNN;Multi-task detection;small targets detection;transfer learning;dam safety;RPN;small sample 收稿日期：2019-09-15.网络出版日期：2020-07-18. 我国是世界上拥有水库大坝最多的国家山， 基金项目：国家重点研发课题(2018Y℉C0407105):国家自然科 学基金重点项目(61832005)：国网新源科技项目 但随着时间的推移和坝龄的增长，大坝表面和内 (SGTYHT/19-JS-217):华能集团重点研发课题 (HNKJ19-H12). 部发生形变，出险几率增加，威胁人民生命财产 通信作者：唐江红.E-mail:15195897810@163.com. 安全。裂缝是大坝的主要危害之一。DOI: 10.11992/tis.201910004 网络出版地址: https://kns.cnki.net/kcms/detail/23.1538.TP.20200717.1727.008.html 基于 Faster R-CNN 的多任务增强裂缝图像检测方法 毛莺池，唐江红，王静，平萍，王龙宝 （河海大学 计算机与信息学院，江苏 南京 211100） 摘 要：针对 Faster R-CNN 算法对多目标、小目标检测精度不高的问题，本文提出一种基于 Faster R-CNN 的多 任务增强裂缝图像检测 (Multitask Enhanced Dam Crack Image Detection Based on Faster R-CNN, ME-Faster R￾CNN) 方法。同时提出一种基于 K-means 的多源自适应平衡 TrAdaBoost 的迁移学习方法 (multi-source adaptive balance TrAdaBoost based on K-means, K-MABtrA) 辅助网络训练，解决样本不足问题。ME-Faster R-CNN 将图片 输入 ResNet-50 网络提取特征；然后将所得特征图输入多任务增强 RPN 模型，同时改善 RPN 模型的锚盒尺寸和 大小以提高检测识别精度，生成候选区域；最后将特征图和候选区域发送到检测处理网络。K-MABtrA 方法利 用 K-means 聚类删除与目标源差别较大的图像，再在多元自适应平衡 TrAdaBoost 迁移学习方法下训练模型。 实验结果表明：将 ME-Faster R-CNN 在 K-MABtrA 迁移学习的条件下应用于小数据集大坝裂缝图像集的平均 IoU 为 82.52%，平均精度 mAP 值为 80.02%，与相同参数设置下的 Faster R-CNN 检测算法相比，平均 IoU 和 mAP 值分别提高了 1.06% 和 1.56%。 关键词：裂缝图像检测；Faster R-CNN；多任务检测；小目标检测；迁移学习；大坝安全；区域建议网络；小样本 中图分类号：TP391 文献标志码：A 文章编号：1673−4785(2021)02−0286−08 中文引用格式：毛莺池, 唐江红, 王静, 等. 基于 Faster R-CNN 的多任务增强裂缝图像检测方法 [J]. 智能系统学报, 2021, 16(2): 286–293. 英文引用格式：MAO Yingchi, TANG Jianghong, WANG Jing, et al. Multi-task enhanced dam crack image detection based on Faster R-CNN[J]. CAAI transactions on intelligent systems, 2021, 16(2): 286–293. Multi-task enhanced dam crack image detection based on Faster R-CNN MAO Yingchi，TANG Jianghong，WANG Jing，PING Ping，WANG Longbao (College of Computer and Information, Hohai University, Nanjing 211100, China) Abstract: To improve the accuracy of the detection of multiple small targets using the faster R-CNN model, we pro￾pose a multi-task enhanced dam-crack image detection method based on faster R-CNN (ME-Faster R-CNN). In addition, to solve the problem of insufficient dam-crack samples, we propose a transfer learning method, multi-source adaptive balance TrAdaBoost based on K-means (K-MABtrA), to assist with network training. In the ME-Faster R-CNN, the ResNet-50 network is adopted to extract features from original images, obtain the feature map, and input a multi-task en￾hanced region-proposal-network module to generate candidate regions by adopting the appropriate size and dimensions of the anchor box. Lastly, the features map and candidate regions are processed to detect dam cracks. The K-MABtrA method first uses K-means clustering to delete unsuitable images. Then, models are trained using the multi-source adapt￾ive balance TrAdaBoost method. Our experimental results show that the proposed ME Faster R-CNN with the K-MAB￾trA method can obtain an 82.52% average intersection over union (IoU) and 80.02% mean average precision (mAP). Compared with Faster R-CNN detection method using the same parameters, the average IoU and mAP values was in￾creased by 1.06% and 1.56%, respectively. Keywords: crack image detection; Faster R-CNN; Multi-task detection; small targets detection; transfer learning; dam safety; RPN; small sample 我国是世界上拥有水库大坝最多的国家[1] ， 但随着时间的推移和坝龄的增长，大坝表面和内 部发生形变，出险几率增加，威胁人民生命财产 安全。裂缝是大坝的主要危害之一。 收稿日期：2019−09−15. 网络出版日期：2020−07−18. 基金项目：国家重点研发课题 (2018YFC0407105)；国家自然科 学基金重点项目 (61832005)；国网新源科技项目 (SGTYHT/19-JS-217)；华能集团重点研发课 题 (HNKJ19-H12). 通信作者：唐江红. E-mail：15195897810@163.com. 第 16 卷第 2 期 智 能 系 统 学 报 Vol.16 No.2 2021 年 3 月 CAAI Transactions on Intelligent Systems Mar. 2021