深度残差网络的无人机多目标识别
|
摘要
传统目标识别算法中,经典的区域建议网络(RPN)在提取目标候选区域时计算量大,时间复杂度较高,因此提出一种级联区域建议网络(CRPN)的搜索模式对其进行改善。此外,深层次的卷积神经网络训练中易产生退化现象,而引入残差学习的深度残差网络(ResNet),能够有效抑制该现象。对多种不同深度以及不同参数的网络模型进行研究,将两层残差学习模块与三层残差学习模块结合使用,设计出一种占用内存更小、时间复杂度更低的新型多捷联式残差网络模型(Mu-ResNet)。采用Mu-ResNet与CRPN结合的网络模型在无人机目标数据集以及PASCAL VOC数据集上进行多目标识别测试,较使用ResNet与RPN结合的网络模型,识别准确率提升了近2个百分点。
In traditional target recognition algorithms, the classical region proposal net(RPN) has large amount of computation and high complexity of time at extracting the target candidate region.Cascade region proposal network(CRPN) is proposed as a new search method for improving the performance of RPN, in which residual learning based deep residual network(ResNet) is also used effectively to suppress the degradation phenomenon in deep-level convolution neural networks.Aimed at the network models with different depths and parameters, a novel multi-strapdown residual network(Mu-ResNet) model, which is of less memory and lower time complexity, is designed by combining two-layer and three-layer residual learning modules. The combination model of Mu-ResNet and CRPN is used for multi-target recognition test by using the unmanned aerial vehicle(UAV) target data and PASCAL VOC data. The results have shown that nearly 2% of recognition accuracy is increased compared with the combination model of ResNet and RPN.
In traditional target recognition algorithms, the classical region proposal net(RPN) has large amount of computation and high complexity of time at extracting the target candidate region.Cascade region proposal network(CRPN) is proposed as a new search method for improving the performance of RPN, in which residual learning based deep residual network(ResNet) is also used effectively to suppress the degradation phenomenon in deep-level convolution neural networks.Aimed at the network models with different depths and parameters, a novel multi-strapdown residual network(Mu-ResNet) model, which is of less memory and lower time complexity, is designed by combining two-layer and three-layer residual learning modules. The combination model of Mu-ResNet and CRPN is used for multi-target recognition test by using the unmanned aerial vehicle(UAV) target data and PASCAL VOC data. The results have shown that nearly 2% of recognition accuracy is increased compared with the combination model of ResNet and RPN.
引文
[1]李伟,张旭东.基于卷积神经网络的深度图像超分辨率重建方法[J].电子测量与仪器学报,2017,31(12):1918-1928.
[2]蒋帅.基于卷积神经网络的图像识别[D].长春:吉林大学,2017.
[3]CHANG L,DUARTE M M,SUCAR L E,et al.ABayesian approach for object classification based on clusters of SIFT local features[J].Expert Systems with Applications,2012,39(2):1679-1686.
[4]KRIZHEVSKY A,SUTSKEVER I,HINTON G E.Imagenet classification with deep convolutional neural networks[C]//Advances in Neural Information Processing Systems.Nevada:MIT Press,2012:1097-1105.
[5]LECUN Y,BOSER B,DENKER J S,et al.Backpropagation applied to handwritten zip code recognition[J].Neural Computation,1989,1(4):541-551.
[6]SERMANET P,EIGEN D,ZHANG X,et al.Overfeat:Integrated recognition,localization and detection using convolutional networks[EB/OL].[2018-03-23].https://arxiv.org/abs/1312.6229.
[7]ZEILER M D,FERGUS R.Visualizing and understanding convolutional networks[C]//European Conference on Computer Vision.Cham:Springer International Publishing,2014:818-833.
[8]SRIVASTAVA R K,GREFF K,SCHMIDHUBER J.Highway networks[EB/OL].[2018-03-25].https://arxiv.org/abs/1505.00387.
[9]SAXE A M,MCCLELLAND J L,GANGULI S.Exact solutions to the nonlinear dynamics of learning in deep linear neural networks[EB/OL].[2018-03-23].https://arxiv.org/abs/1312.6120.
[10]HE K,ZHANG X,REN S,et al.Delving deep into rectifiers:Surpassing human-level performance on imagenet classification[C]//Proceedings of the IEEEInternational Conference on Computer Vision.New York:IEEE Press,2015:1026-1034.
[11]GLOROT X,BENGIO Y.Understanding the difficulty of training deep feed forward neural networks[EB/OL].[2018-04-01].http://www.doc88.com/p-7738903804120.html.
[12]HE K,ZHANG X,REN S,et al.Deep residual learning for image recognition[C]//Proceedings of the IEEEConference on Computer Vision and Pattern Recognition.New York:IEEE Press,2016:770-778.
[13]张珂,高策,郭丽茹,等.非受限条件下多级残差网络人脸图像年龄估计[J].计算机辅助设计与图形学学报,2018,30(2):346-353.
[14]陆永帅,李元祥,刘波,等.基于深度残差网络的高光谱遥感数据霾监测[J].光学学报,2017,37(11):314-324.
[15]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,2015,39(6):1137-1149.
[2]蒋帅.基于卷积神经网络的图像识别[D].长春:吉林大学,2017.
[3]CHANG L,DUARTE M M,SUCAR L E,et al.ABayesian approach for object classification based on clusters of SIFT local features[J].Expert Systems with Applications,2012,39(2):1679-1686.
[4]KRIZHEVSKY A,SUTSKEVER I,HINTON G E.Imagenet classification with deep convolutional neural networks[C]//Advances in Neural Information Processing Systems.Nevada:MIT Press,2012:1097-1105.
[5]LECUN Y,BOSER B,DENKER J S,et al.Backpropagation applied to handwritten zip code recognition[J].Neural Computation,1989,1(4):541-551.
[6]SERMANET P,EIGEN D,ZHANG X,et al.Overfeat:Integrated recognition,localization and detection using convolutional networks[EB/OL].[2018-03-23].https://arxiv.org/abs/1312.6229.
[7]ZEILER M D,FERGUS R.Visualizing and understanding convolutional networks[C]//European Conference on Computer Vision.Cham:Springer International Publishing,2014:818-833.
[8]SRIVASTAVA R K,GREFF K,SCHMIDHUBER J.Highway networks[EB/OL].[2018-03-25].https://arxiv.org/abs/1505.00387.
[9]SAXE A M,MCCLELLAND J L,GANGULI S.Exact solutions to the nonlinear dynamics of learning in deep linear neural networks[EB/OL].[2018-03-23].https://arxiv.org/abs/1312.6120.
[10]HE K,ZHANG X,REN S,et al.Delving deep into rectifiers:Surpassing human-level performance on imagenet classification[C]//Proceedings of the IEEEInternational Conference on Computer Vision.New York:IEEE Press,2015:1026-1034.
[11]GLOROT X,BENGIO Y.Understanding the difficulty of training deep feed forward neural networks[EB/OL].[2018-04-01].http://www.doc88.com/p-7738903804120.html.
[12]HE K,ZHANG X,REN S,et al.Deep residual learning for image recognition[C]//Proceedings of the IEEEConference on Computer Vision and Pattern Recognition.New York:IEEE Press,2016:770-778.
[13]张珂,高策,郭丽茹,等.非受限条件下多级残差网络人脸图像年龄估计[J].计算机辅助设计与图形学学报,2018,30(2):346-353.
[14]陆永帅,李元祥,刘波,等.基于深度残差网络的高光谱遥感数据霾监测[J].光学学报,2017,37(11):314-324.
[15]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,2015,39(6):1137-1149.