Landslide is one of the multitudinous serious geological hazards. The key to its control and reduction lies on dynamic monitoring and early warning. The article points out the insufficiency of traditional measuring me...Landslide is one of the multitudinous serious geological hazards. The key to its control and reduction lies on dynamic monitoring and early warning. The article points out the insufficiency of traditional measuring means applied for large-scale landslide monitoring and proposes the method for extensive landslide displacement field monitoring using high- resolution remote images. Matching of cognominal points is realized by using the invariant features of SIFT algorithm in image translation, rotation, zooming, and affine transformation, and through recognition and comparison of characteristics of high-resolution images in different landsliding periods. Following that, landslide displacement vector field can be made known by measuring the distances and directions between cognominal points. As evidenced by field application of the method for landslide monitoring at West Open Mine in Fushun city of China, the method has the attraction of being able to make areal measurement through satellite observation and capable of obtaining at the same time the information of large- area intensive displacement field, for facilitating automatic delimitation of extent of landslide displacement vector field and sliding mass. This can serve as a basis for making analysis of laws governing occurrence of landslide and adoption of countermeasures.展开更多
为提升传统算法对高分辨率遥感图像中地物目标的检测效果,将深度学习目标检测框架快速区域卷积神经网络(faster regions with convolutional neural network,Faster R-CNN)应用于高分辨率遥感图像目标检测任务中。以机场为检测场景、飞...为提升传统算法对高分辨率遥感图像中地物目标的检测效果,将深度学习目标检测框架快速区域卷积神经网络(faster regions with convolutional neural network,Faster R-CNN)应用于高分辨率遥感图像目标检测任务中。以机场为检测场景、飞机为检测目标进行实验,首先,利用高分辨率遥感图像数据集训练Faster R-CNN框架,得到相应的目标检测模型;然后,采用该模型对高分辨率遥感图像中的飞机目标进行检测;最后,对实验结果进行统计分析及评价。实验结果表明,Faster R-CNN模型能够全面而准确地检测飞机目标,最优 F1分数值为0.976 3,并且同一个模型可以对多种高分辨率遥感图像进行目标检测。展开更多
在高分辨率遥感图像目标检测中,受云雾、光照、复杂背景、噪声等因素影响,现有目标检测方法虚警率高、速度慢、精确度低.为此提出基于深度神经网络剪枝的两阶段目标检测(object detection based on deep pruning,ODDP)方法.首先,给出深...在高分辨率遥感图像目标检测中,受云雾、光照、复杂背景、噪声等因素影响,现有目标检测方法虚警率高、速度慢、精确度低.为此提出基于深度神经网络剪枝的两阶段目标检测(object detection based on deep pruning,ODDP)方法.首先,给出深度神经网络剪枝方法,基于深度神经网络剪枝分别提出自主学习区域提取网络算法与优化训练分类网络算法;然后,将上述两算法用于卷积神经网络,得到两阶段目标检测模型.实验结果表明,在NWPU VHR-10高分辨率遥感数据集上,相比现有目标检测方法,ODDP的检测速度和精度均有一定提升.展开更多
针对目前使用机器学习解决高分辨率遥感图像分类主要存在下采样导致的细节信息丢失问题,提出了一种基于DeepLabv3架构的小波域DeepLabv3-MRF(Markov random field,MRF)算法。选择当前较为普遍的DeepLabv3架构分类算法,能够获得更为精确...针对目前使用机器学习解决高分辨率遥感图像分类主要存在下采样导致的细节信息丢失问题,提出了一种基于DeepLabv3架构的小波域DeepLabv3-MRF(Markov random field,MRF)算法。选择当前较为普遍的DeepLabv3架构分类算法,能够获得更为精确的分类结果;采用小波域DeepLabv3-MRF算法,还能够获得更为清晰的边缘细节信息。选取南方某地区高分辨率无人机遥感图像进行分类实验,通过小波变换的方向性、非冗余性以及MRF变换像素空间的交互性这三个方面,将分类结果与原始DeepLabv3架构分类结果对比分析。结果表明,所提出的分类方法精度明显高于原始DeepLabv3架构分类算法的精度,总体精度可提升3%左右,并且可以充分表达高分辨率遥感图像细节信息。展开更多
文摘Landslide is one of the multitudinous serious geological hazards. The key to its control and reduction lies on dynamic monitoring and early warning. The article points out the insufficiency of traditional measuring means applied for large-scale landslide monitoring and proposes the method for extensive landslide displacement field monitoring using high- resolution remote images. Matching of cognominal points is realized by using the invariant features of SIFT algorithm in image translation, rotation, zooming, and affine transformation, and through recognition and comparison of characteristics of high-resolution images in different landsliding periods. Following that, landslide displacement vector field can be made known by measuring the distances and directions between cognominal points. As evidenced by field application of the method for landslide monitoring at West Open Mine in Fushun city of China, the method has the attraction of being able to make areal measurement through satellite observation and capable of obtaining at the same time the information of large- area intensive displacement field, for facilitating automatic delimitation of extent of landslide displacement vector field and sliding mass. This can serve as a basis for making analysis of laws governing occurrence of landslide and adoption of countermeasures.
文摘在高分辨率遥感图像目标检测中,受云雾、光照、复杂背景、噪声等因素影响,现有目标检测方法虚警率高、速度慢、精确度低.为此提出基于深度神经网络剪枝的两阶段目标检测(object detection based on deep pruning,ODDP)方法.首先,给出深度神经网络剪枝方法,基于深度神经网络剪枝分别提出自主学习区域提取网络算法与优化训练分类网络算法;然后,将上述两算法用于卷积神经网络,得到两阶段目标检测模型.实验结果表明,在NWPU VHR-10高分辨率遥感数据集上,相比现有目标检测方法,ODDP的检测速度和精度均有一定提升.
文摘针对目前使用机器学习解决高分辨率遥感图像分类主要存在下采样导致的细节信息丢失问题,提出了一种基于DeepLabv3架构的小波域DeepLabv3-MRF(Markov random field,MRF)算法。选择当前较为普遍的DeepLabv3架构分类算法,能够获得更为精确的分类结果;采用小波域DeepLabv3-MRF算法,还能够获得更为清晰的边缘细节信息。选取南方某地区高分辨率无人机遥感图像进行分类实验,通过小波变换的方向性、非冗余性以及MRF变换像素空间的交互性这三个方面,将分类结果与原始DeepLabv3架构分类结果对比分析。结果表明,所提出的分类方法精度明显高于原始DeepLabv3架构分类算法的精度,总体精度可提升3%左右,并且可以充分表达高分辨率遥感图像细节信息。
