Objective and accurate classification model or method of cloud image is a prerequisite for accurate weather monitoring and forecast.Thus safety of aircraft taking off and landing and air flight can be guaranteed.Thres...Objective and accurate classification model or method of cloud image is a prerequisite for accurate weather monitoring and forecast.Thus safety of aircraft taking off and landing and air flight can be guaranteed.Thresholding is a kind of simple and effective method of cloud classification.It can realize automated ground-based cloud detection and cloudage observation.The existing segmentation methods based on fixed threshold and single threshold cannot achieve good segmentation effect.Thus it is difficult to obtain the accurate result of cloud detection and cloudage observation.In view of the above-mentioned problems,multi-thresholding methods of ground-based cloud based on exponential entropy/exponential gray entropy and uniform searching particle swarm optimization(UPSO)are proposed.Exponential entropy and exponential gray entropy make up for the defects of undefined value and zero value in Shannon entropy.In addition,exponential gray entropy reflects the relative uniformity of gray levels within the cloud cluster and background cluster.Cloud regions and background regions of different gray level ranges can be distinguished more precisely using the multi-thresholding strategy.In order to reduce computational complexity of original exhaustive algorithm for multi-threshold selection,the UPSO algorithm is adopted.It can find the optimal thresholds quickly and accurately.As a result,the real-time processing of segmentation of groundbased cloud image can be realized.The experimental results show that,in comparison with the existing groundbased cloud image segmentation methods and multi-thresholding method based on maximum Shannon entropy,the proposed methods can extract the boundary shape,textures and details feature of cloud more clearly.Therefore,the accuracies of cloudage detection and morphology classification for ground-based cloud are both improved.展开更多
Ground-based cloud classification is challenging due to extreme variations in the appearance of clouds under different atmospheric conditions. Texture classification techniques have recently been introduced to deal wi...Ground-based cloud classification is challenging due to extreme variations in the appearance of clouds under different atmospheric conditions. Texture classification techniques have recently been introduced to deal with this issue. A novel texture descriptor, the salient local binary pattern （SLBP）, is proposed for ground-based cloud classification. The SLBP takes advantage of the most frequently occurring patterns （the salient patterns） to capture descriptive information. This feature makes the SLBP robust to noise. Experimental results using ground-based cloud images demonstrate that the proposed method can achieve better results than current state-of-the-art methods.展开更多
The automatic definition of the ground from 3D point clouds has been a common process for the last two decades,with many different approaches and applications that can be found in a vast literature.This paper presents...The automatic definition of the ground from 3D point clouds has been a common process for the last two decades,with many different approaches and applications that can be found in a vast literature.This paper presents a comparison of three different methodological concepts for ground classification,in order to establish the advantages and drawbacks of each method.First,a heuristic method,based on previous knowledge of the geometry and context of the 3D data.Secondly,a Deep Convolutional Network based on SegNet that classifies 2D images generated from the 3D point cloud.Finally,the third method applies a Deep Learning classification based on PointNet,which takes 3D points directly as inputs.To validate each method and compare them,public and labelled point clouds from the Actueel Hoogtebestand Nederland dataset are employed.Furthermore,the three methods are validated against the ISPRS 3D Semantic Labeling Contest benchmark.The results obtained show that the deep learning-based approaches outperform the heuristic method,with F-scores above 96%.The best results were obtained using a shallower version of SegNet,with F-score above 97%.展开更多
Light Detection And Ranging (LiDAR) is a well-established active remote sensing technology that can provide accurate digital elevation measurements for the terrain and non-ground objects such as vegetations and buildi...Light Detection And Ranging (LiDAR) is a well-established active remote sensing technology that can provide accurate digital elevation measurements for the terrain and non-ground objects such as vegetations and buildings, etc. Non-ground objects need to be removed for creation of a Digital Terrain Model (DTM) which is a continuous surface representing only ground surface points. This study aimed at comparative analysis of three main filtering approaches for stripping off non-ground objects namely;Gaussian low pass filter, focal analysis mean filter and DTM slope-based filter of varying window sizes in creation of a reliable DTM from airborne LiDAR point clouds. A sample of LiDAR data provided by the ISPRS WG III/4 captured at Vaihingen in Germany over a pure residential area has been used in the analysis. Visual analysis has indicated that Gaussian low pass filter has given blurred DTMs of attenuated high-frequency objects and emphasized low-frequency objects while it has achieved improved removal of non-ground object at larger window sizes. Focal analysis mean filter has shown better removal of nonground objects compared to Gaussian low pass filter especially at large window sizes where details of non-ground objects almost have diminished in the DTMs from window sizes of 25 × 25 and greater. DTM slope-based filter has created bare earth models that have been full of gabs at the positions of the non-ground objects where the sizes and numbers of that gabs have increased with increasing the window sizes of filter. Those gaps have been closed through exploitation of the spline interpolation method in order to get continuous surface representing bare earth landscape. Comparative analysis has shown that the minimum elevations of the DTMs increase with increasing the filter widow sizes till 21 × 21 and 31 × 31 for the Gaussian low pass filter and the focal analysis mean filter respectively. On the other hand, the DTM slope-based filter has kept the minimum elevation of the original data, that could be due to noise in the LiDAR data unchanged. Alternatively, the three approaches have produced DTMs of decreasing maximum elevation values and consequently decreasing ranges of elevations due to increases in the filter window sizes. Moreover, the standard deviations of the created DTMs from the three filters have decreased with increasing the filter window sizes however, the decreases have been continuous and steady in the cases of the Gaussian low pass filter and the focal analysis mean filters while in the case of the DTM slope-based filter the standard deviations of the created DTMs have decreased with high rates till window size of 31 × 31 then they have kept unchanged due to more increases in the filter window sizes.展开更多
因高度向分辨能力缺失,地基干涉雷达应用于建筑成像时会发生严重的高度向叠掩现象。层析合成孔径雷达(Tomographic Synthetic Aperture Radar,TomoSAR)技术具备高度向分辨能力,能够实现建筑三维成像。地基层析圆弧扫描合成孔径雷达(Grou...因高度向分辨能力缺失,地基干涉雷达应用于建筑成像时会发生严重的高度向叠掩现象。层析合成孔径雷达(Tomographic Synthetic Aperture Radar,TomoSAR)技术具备高度向分辨能力,能够实现建筑三维成像。地基层析圆弧扫描合成孔径雷达(Ground-based Tomographic Arc-scanning Synthetic Aperture Radar,GB-TomoArcSAR)通过双轴转台控制天线在不同俯仰角度的水平面内进行圆周扫描来获取高度向合成孔径,实现三维层析成像。本文提出了GB-TomoArcSAR的三维点云生成方法,首先构建了适用于高度向弧形采样条件的层析成像几何模型。其次利用基于巴特沃斯滤波器的奇异值分解(Singular Value Decomposition,SVD)方法进行谱估计,找出层析谱中的峰值及其对应的峰值位置,构成层析向目标候选集。随后利用自对消顺序广义似然比(Sequential Generalized Likelihood Ratio Test with Cancellation,SGLRTC)检测器估计散射体的数目与位置,通过设置检测门限将真实目标的峰值及对应的峰值位置从候选集中筛选出来。最后采用基于空间几何分布的点云优化方法剔除误差点,生成点云图像。文章通过点目标和面目标的仿真实验,验证了所提方法适用于GB-TomoArcSAR,能够有效解决高度向多散射体目标的叠掩问题;进一步开展了实测数据验证,基于所提方法获取了北京市一处建筑基坑的层析点云,其与实际场景几何特征一致。展开更多
基金Supported by the National Natural Science Foundation of China(60872065)the Open Foundation of Key Laboratory of Meteorological Disaster of Ministry of Education at Nanjing University of Information Science & Technology(KLME1108)the Priority Academic Program Development of Jiangsu Higher Education Institutions
文摘Objective and accurate classification model or method of cloud image is a prerequisite for accurate weather monitoring and forecast.Thus safety of aircraft taking off and landing and air flight can be guaranteed.Thresholding is a kind of simple and effective method of cloud classification.It can realize automated ground-based cloud detection and cloudage observation.The existing segmentation methods based on fixed threshold and single threshold cannot achieve good segmentation effect.Thus it is difficult to obtain the accurate result of cloud detection and cloudage observation.In view of the above-mentioned problems,multi-thresholding methods of ground-based cloud based on exponential entropy/exponential gray entropy and uniform searching particle swarm optimization(UPSO)are proposed.Exponential entropy and exponential gray entropy make up for the defects of undefined value and zero value in Shannon entropy.In addition,exponential gray entropy reflects the relative uniformity of gray levels within the cloud cluster and background cluster.Cloud regions and background regions of different gray level ranges can be distinguished more precisely using the multi-thresholding strategy.In order to reduce computational complexity of original exhaustive algorithm for multi-threshold selection,the UPSO algorithm is adopted.It can find the optimal thresholds quickly and accurately.As a result,the real-time processing of segmentation of groundbased cloud image can be realized.The experimental results show that,in comparison with the existing groundbased cloud image segmentation methods and multi-thresholding method based on maximum Shannon entropy,the proposed methods can extract the boundary shape,textures and details feature of cloud more clearly.Therefore,the accuracies of cloudage detection and morphology classification for ground-based cloud are both improved.
基金Supported by the National Natural Science Foundation of China (61172103, 60933010, and 60835001)
文摘Ground-based cloud classification is challenging due to extreme variations in the appearance of clouds under different atmospheric conditions. Texture classification techniques have recently been introduced to deal with this issue. A novel texture descriptor, the salient local binary pattern （SLBP）, is proposed for ground-based cloud classification. The SLBP takes advantage of the most frequently occurring patterns （the salient patterns） to capture descriptive information. This feature makes the SLBP robust to noise. Experimental results using ground-based cloud images demonstrate that the proposed method can achieve better results than current state-of-the-art methods.
基金the Spanish Ministry of Economy and Competitiveness through the Human Resources program FPI[grant number BES-2014-067736]Xunta de Galicia through grant number ED431C2016-038This project has received funding from the European Union’s Horizon 2020 research and innovation programme under grant agreement No.769255.
文摘The automatic definition of the ground from 3D point clouds has been a common process for the last two decades,with many different approaches and applications that can be found in a vast literature.This paper presents a comparison of three different methodological concepts for ground classification,in order to establish the advantages and drawbacks of each method.First,a heuristic method,based on previous knowledge of the geometry and context of the 3D data.Secondly,a Deep Convolutional Network based on SegNet that classifies 2D images generated from the 3D point cloud.Finally,the third method applies a Deep Learning classification based on PointNet,which takes 3D points directly as inputs.To validate each method and compare them,public and labelled point clouds from the Actueel Hoogtebestand Nederland dataset are employed.Furthermore,the three methods are validated against the ISPRS 3D Semantic Labeling Contest benchmark.The results obtained show that the deep learning-based approaches outperform the heuristic method,with F-scores above 96%.The best results were obtained using a shallower version of SegNet,with F-score above 97%.
文摘Light Detection And Ranging (LiDAR) is a well-established active remote sensing technology that can provide accurate digital elevation measurements for the terrain and non-ground objects such as vegetations and buildings, etc. Non-ground objects need to be removed for creation of a Digital Terrain Model (DTM) which is a continuous surface representing only ground surface points. This study aimed at comparative analysis of three main filtering approaches for stripping off non-ground objects namely;Gaussian low pass filter, focal analysis mean filter and DTM slope-based filter of varying window sizes in creation of a reliable DTM from airborne LiDAR point clouds. A sample of LiDAR data provided by the ISPRS WG III/4 captured at Vaihingen in Germany over a pure residential area has been used in the analysis. Visual analysis has indicated that Gaussian low pass filter has given blurred DTMs of attenuated high-frequency objects and emphasized low-frequency objects while it has achieved improved removal of non-ground object at larger window sizes. Focal analysis mean filter has shown better removal of nonground objects compared to Gaussian low pass filter especially at large window sizes where details of non-ground objects almost have diminished in the DTMs from window sizes of 25 × 25 and greater. DTM slope-based filter has created bare earth models that have been full of gabs at the positions of the non-ground objects where the sizes and numbers of that gabs have increased with increasing the window sizes of filter. Those gaps have been closed through exploitation of the spline interpolation method in order to get continuous surface representing bare earth landscape. Comparative analysis has shown that the minimum elevations of the DTMs increase with increasing the filter widow sizes till 21 × 21 and 31 × 31 for the Gaussian low pass filter and the focal analysis mean filter respectively. On the other hand, the DTM slope-based filter has kept the minimum elevation of the original data, that could be due to noise in the LiDAR data unchanged. Alternatively, the three approaches have produced DTMs of decreasing maximum elevation values and consequently decreasing ranges of elevations due to increases in the filter window sizes. Moreover, the standard deviations of the created DTMs from the three filters have decreased with increasing the filter window sizes however, the decreases have been continuous and steady in the cases of the Gaussian low pass filter and the focal analysis mean filters while in the case of the DTM slope-based filter the standard deviations of the created DTMs have decreased with high rates till window size of 31 × 31 then they have kept unchanged due to more increases in the filter window sizes.
文摘因高度向分辨能力缺失,地基干涉雷达应用于建筑成像时会发生严重的高度向叠掩现象。层析合成孔径雷达(Tomographic Synthetic Aperture Radar,TomoSAR)技术具备高度向分辨能力,能够实现建筑三维成像。地基层析圆弧扫描合成孔径雷达(Ground-based Tomographic Arc-scanning Synthetic Aperture Radar,GB-TomoArcSAR)通过双轴转台控制天线在不同俯仰角度的水平面内进行圆周扫描来获取高度向合成孔径,实现三维层析成像。本文提出了GB-TomoArcSAR的三维点云生成方法,首先构建了适用于高度向弧形采样条件的层析成像几何模型。其次利用基于巴特沃斯滤波器的奇异值分解(Singular Value Decomposition,SVD)方法进行谱估计,找出层析谱中的峰值及其对应的峰值位置,构成层析向目标候选集。随后利用自对消顺序广义似然比(Sequential Generalized Likelihood Ratio Test with Cancellation,SGLRTC)检测器估计散射体的数目与位置,通过设置检测门限将真实目标的峰值及对应的峰值位置从候选集中筛选出来。最后采用基于空间几何分布的点云优化方法剔除误差点,生成点云图像。文章通过点目标和面目标的仿真实验,验证了所提方法适用于GB-TomoArcSAR,能够有效解决高度向多散射体目标的叠掩问题;进一步开展了实测数据验证,基于所提方法获取了北京市一处建筑基坑的层析点云,其与实际场景几何特征一致。
