基于深度卷积神经网络的遥感图像场景分类
|
摘要
遥感图像场景分类在地理空间对象检测、自然灾害检测、地理图像检索、环境监测等方面具有广泛的应用前景,引起了人们的广泛关注.文章改进了传统的深度卷积神经网络(DCNN),将其应用于遥感图像场景分类研究,提出了一种改进后的7层网络结构,在激活函数的选择上,针对神经元通过Relu进行激活容易激活失败的情况,采用PRelu函数替代Relu;针对传统的深度学习方法不能融合多种细粒度深度学习特征的问题,采用分层特征融合的方法,通过实验对比,将第四个卷积层、池化层和最后一个全连接层提取到的特征进行串联融合,得到一种更加有效的深度特征.与传统深度学习方法相比,文章所提方法分类准确率提高了8.81%.实验结果表明,该方法在准确率、Kappa系数上均有良好表现,取得了良好的分类效果.
Scene classification of remote sensing image has a wide application prospect in geospatial object detection,natural disaster detection,Geographic Image retrieval,environmental monitoring and so on,which has attracted wide attention.This paper improves the traditional deep convolution neural network(DCNN)and applies it to remote sensing image scene classification research.An improved seven-layer network structure is proposed.In the selection of activation function,PRelu function is used to replace Relu when neurons are easily activated by Relu.For the problem of learning features,we use the method of hierarchical feature fusion.Through experimental comparison,the features extracted from the fourth convolution layer,the pooling layer and the last full connection layer are fused in series to obtain a more effective depth feature.Compared with the traditional in-depth learning method,the classification accuracy of the proposed method is improved by 8.81%.The experimental results show that the method performs well in accuracy and Kappa coefficient,and achieves good classification results.
Scene classification of remote sensing image has a wide application prospect in geospatial object detection,natural disaster detection,Geographic Image retrieval,environmental monitoring and so on,which has attracted wide attention.This paper improves the traditional deep convolution neural network(DCNN)and applies it to remote sensing image scene classification research.An improved seven-layer network structure is proposed.In the selection of activation function,PRelu function is used to replace Relu when neurons are easily activated by Relu.For the problem of learning features,we use the method of hierarchical feature fusion.Through experimental comparison,the features extracted from the fourth convolution layer,the pooling layer and the last full connection layer are fused in series to obtain a more effective depth feature.Compared with the traditional in-depth learning method,the classification accuracy of the proposed method is improved by 8.81%.The experimental results show that the method performs well in accuracy and Kappa coefficient,and achieves good classification results.
引文
[1]〗GUO Z,ZHANG L,ZHANG D.A completed modeling of local binary pattern operator for texture classification.[J].IEEETransactions on Image Processing,2010,19(6):1657-1663
[2]KRIZHEVSKY A,SUTSKEVER I,HINTON G E.ImageNet classification with deep convolutional neural networks[C]//International Conference on Neural Information Processing Systems.Curran Associates Inc.2012:1097-1105
[3]CHENG G,YANG C,YAO X,et al.When deep learning meets metric learning:remote sensing image scene classification via learning discriminative CNNs[J].IEEE Transactions on Geoscience&Remote Sensing,2018(99):1-11
[4]李昌俊.基于深度学习的农业遥感图像耕地提取技术研究[D].合肥:中国科学技术大学,2018
[5]曲景影,孙显,高鑫.基于CNN模型的高分辨率遥感图像目标识别[J].国外电子测量技术,2016(8):45-50
[6]党宇,张继贤,邓喀中,等.基于深度学习AlexNet的遥感影像地表覆盖分类评价研究[J].地球信息科学学报,2017,19(11):1530-1537
[7]HE K,ZHAN X,REN S,et al.Delving deep into rectifiers:surpassing human-level performance on image net classification[C].IEEE.International Conference on Compnter Vision,2015:1026-1034
[2]KRIZHEVSKY A,SUTSKEVER I,HINTON G E.ImageNet classification with deep convolutional neural networks[C]//International Conference on Neural Information Processing Systems.Curran Associates Inc.2012:1097-1105
[3]CHENG G,YANG C,YAO X,et al.When deep learning meets metric learning:remote sensing image scene classification via learning discriminative CNNs[J].IEEE Transactions on Geoscience&Remote Sensing,2018(99):1-11
[4]李昌俊.基于深度学习的农业遥感图像耕地提取技术研究[D].合肥:中国科学技术大学,2018
[5]曲景影,孙显,高鑫.基于CNN模型的高分辨率遥感图像目标识别[J].国外电子测量技术,2016(8):45-50
[6]党宇,张继贤,邓喀中,等.基于深度学习AlexNet的遥感影像地表覆盖分类评价研究[J].地球信息科学学报,2017,19(11):1530-1537
[7]HE K,ZHAN X,REN S,et al.Delving deep into rectifiers:surpassing human-level performance on image net classification[C].IEEE.International Conference on Compnter Vision,2015:1026-1034