基于迁移学习与图像增强的夜间航拍车辆识别方法
|
摘要
为了对夜间航拍图片中的车辆进行有效识别,提出基于二次迁移学习和Retinex算法的图像处理方法,仅利用小规模的数据集训练网络,采用基于Faster R-CNN的深度学习算法即可实现车辆的快速检测.首先在ImageNet大规模数据集和中国科学院日间航拍中规模数据集之间应用一次迁移学习;然后在日间中规模数据集与夜间航拍小规模数据集之间应用二次迁移学习;最后利用Retinex迭代算法对夜间图片进行处理以增强其与日间图片的相似性,使二次迁移学习有效进行.实验结果表明,在深度学习平台上,该方法利用小规模航拍数据集训练出有效的识别网络,检测结果优于传统的机器学习方法,在军事侦察及交通管控等方面具有一定的应用价值.
In order to identify vehicles in night-time aerial images effectively, this paper proposed an image processing technique based on two-time transfer learning and the Retinex algorithm. It only used a small-scale data set to train the network and then employed a deep learning algorithm based on Faster R-CNN to achieve quick detection of vehicles. Firstly, a transfer learning process was applied between the large-scale ImageNet data set and the mid-scale Chinese Academy of Sciences daytime aerial data set, and then a second transfer learning algorithm was utilized from the day-time mid-scale data set to the night-time small-scale data set. At the same time, the Retinex iterative algorithm was used to process the night-time pictures to enhance their similarity with the day-time pictures, so that the second transfer learning can be effectively performed. The experimental results show that this method can train an effective recognition network on deep learning platforms with small-scale data sets, and its detection performance is superior to the traditional machine learning methods. It also has certain application values in military reconnaissance and traffic control field.
In order to identify vehicles in night-time aerial images effectively, this paper proposed an image processing technique based on two-time transfer learning and the Retinex algorithm. It only used a small-scale data set to train the network and then employed a deep learning algorithm based on Faster R-CNN to achieve quick detection of vehicles. Firstly, a transfer learning process was applied between the large-scale ImageNet data set and the mid-scale Chinese Academy of Sciences daytime aerial data set, and then a second transfer learning algorithm was utilized from the day-time mid-scale data set to the night-time small-scale data set. At the same time, the Retinex iterative algorithm was used to process the night-time pictures to enhance their similarity with the day-time pictures, so that the second transfer learning can be effectively performed. The experimental results show that this method can train an effective recognition network on deep learning platforms with small-scale data sets, and its detection performance is superior to the traditional machine learning methods. It also has certain application values in military reconnaissance and traffic control field.
引文
[1]Jian Mingquan. Drone’s unique advantages in traffic management[J].ChinaPublicSecurity,2016(12):40-44(inChinese)(简明全.无人机在交通管理中的独有优势[J].中国公共安全, 2016(12):40-44)
[2]Cheng H Y, Weng C C, Chen Y Y. Vehicle detection in aerial surveillanceusingdynamicBayesiannetworks[J].IEEE Transactions on Image Processing, 2012, 21(4):2152-2159
[3]Freund Y, Schapire R E. Experiments with a new boosting algorithm[C]//Proceedings of the 13th International Conference onInternationalConferenceonMachineLearning.SanFrancisco:Morgan Kaufmann Publishers Inc, 1996:148-156
[4]Kim D, Lee D, Myung H, et al. Object detection and tracking forautonomousunderwaterrobotsusingweightedtemplate matching[C]//ProceedingsofOceans.LosAlamitos:IEEE Computer Society Press, 2012:1-5
[5]Hinz S, Baumgartner A. Vehicle detection in aerial images using generic features, grouping, and context[C]//Proceedings of the23rdSymposiumoftheGermanAssociationforPattern Recognition. Heidelberg:Springer, 2001, 2191:45-52
[6]Dalal N, Triggs B. Histograms of oriented gradients for human detection[C]//ProceedingsoftheIEEEComputerSociety Conference on Computer Vision and Pattern Recognition. Los Alamitos:IEEE Computer Society Press, 2005:886-893
[7]Ergun H, Sert M. Fusing deep convolutional networks for large scale visual concept classification[C]//Proceedings of the 2nd IEEEInternationalConferenceonMultimediaBigData.Los Alamitos:IEEE Computer Society Press, 2016:210-213
[8]Zhuang Fuzhen, Luo Ping, He Qing, et al. Survey on transfer learning research[J]. Journal of Software, 2015, 26(1):26-39(in Chinese)(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)
[9]Ren S Q, He K M, Girshick R, et al. Faster R-CNN:towards real-timeobjectdetectionwithregionproposalnetworks[J].IEEETransactionsonPatternAnalysisandMachineIntelligence, 2017, 39(6):1137-1149
[10]Russakovsky O, Deng J, Su H, et al. ImageNet large scale visual recognition challenge[J]. International Journal of Computer Vision, 2015, 115(3):211-252
[11]Girshick R. Fast R-CNN[C]//Proceedings of the IEEE InternationalConferenceonComputerVision.LosAlamitos:IEEE Computer Society Press, 2015:1440-1448
[12]Long J, Shelhamer E, Darrell T. Fully convolutional networks for semantic segmentation[C]//Proceedings of the Conference onComputerVisionandPatternRecognition.LosAlamitos:IEEE Computer Society Press, 2015:3431-3440
[13]Uijlings J R R, van de Sande K E A, Gevers T, et al. Selective search for object recognition[J]. International Journal of Computer Vision, 2013, 104(2):154-171
[14]LawrenceZitnickC,DollarP.Edgeboxes:locatingobject proposalsfromedges[C]//Proceedingsofthe13thEuropean ConferenceonComputerVision.Heidelberg:Springer,2014,8693:391-405
[15]Dai J F, He K M, Sun J. Instance-aware semantic segmentation via multi-task network cascades[C]//Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition. Los Alamitos:IEEE Computer Society Press, 2016:3150-3158
[2]Cheng H Y, Weng C C, Chen Y Y. Vehicle detection in aerial surveillanceusingdynamicBayesiannetworks[J].IEEE Transactions on Image Processing, 2012, 21(4):2152-2159
[3]Freund Y, Schapire R E. Experiments with a new boosting algorithm[C]//Proceedings of the 13th International Conference onInternationalConferenceonMachineLearning.SanFrancisco:Morgan Kaufmann Publishers Inc, 1996:148-156
[4]Kim D, Lee D, Myung H, et al. Object detection and tracking forautonomousunderwaterrobotsusingweightedtemplate matching[C]//ProceedingsofOceans.LosAlamitos:IEEE Computer Society Press, 2012:1-5
[5]Hinz S, Baumgartner A. Vehicle detection in aerial images using generic features, grouping, and context[C]//Proceedings of the23rdSymposiumoftheGermanAssociationforPattern Recognition. Heidelberg:Springer, 2001, 2191:45-52
[6]Dalal N, Triggs B. Histograms of oriented gradients for human detection[C]//ProceedingsoftheIEEEComputerSociety Conference on Computer Vision and Pattern Recognition. Los Alamitos:IEEE Computer Society Press, 2005:886-893
[7]Ergun H, Sert M. Fusing deep convolutional networks for large scale visual concept classification[C]//Proceedings of the 2nd IEEEInternationalConferenceonMultimediaBigData.Los Alamitos:IEEE Computer Society Press, 2016:210-213
[8]Zhuang Fuzhen, Luo Ping, He Qing, et al. Survey on transfer learning research[J]. Journal of Software, 2015, 26(1):26-39(in Chinese)(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)
[9]Ren S Q, He K M, Girshick R, et al. Faster R-CNN:towards real-timeobjectdetectionwithregionproposalnetworks[J].IEEETransactionsonPatternAnalysisandMachineIntelligence, 2017, 39(6):1137-1149
[10]Russakovsky O, Deng J, Su H, et al. ImageNet large scale visual recognition challenge[J]. International Journal of Computer Vision, 2015, 115(3):211-252
[11]Girshick R. Fast R-CNN[C]//Proceedings of the IEEE InternationalConferenceonComputerVision.LosAlamitos:IEEE Computer Society Press, 2015:1440-1448
[12]Long J, Shelhamer E, Darrell T. Fully convolutional networks for semantic segmentation[C]//Proceedings of the Conference onComputerVisionandPatternRecognition.LosAlamitos:IEEE Computer Society Press, 2015:3431-3440
[13]Uijlings J R R, van de Sande K E A, Gevers T, et al. Selective search for object recognition[J]. International Journal of Computer Vision, 2013, 104(2):154-171
[14]LawrenceZitnickC,DollarP.Edgeboxes:locatingobject proposalsfromedges[C]//Proceedingsofthe13thEuropean ConferenceonComputerVision.Heidelberg:Springer,2014,8693:391-405
[15]Dai J F, He K M, Sun J. Instance-aware semantic segmentation via multi-task network cascades[C]//Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition. Los Alamitos:IEEE Computer Society Press, 2016:3150-3158