A study-based ranking of LiDAR data visualization schemes aided by georectified aerial images

Light Detection and Ranging (LiDAR) collects dense 3D topographic information in the form of points. LiDAR data can be displayed either through direct rendering of the point cloud or by generalizing features extracted through classification or segmentation. We are working in the domain of visualizing LiDAR data sets and have developed certain pipelines for visualization. These pipelines have been presented elsewhere. We present a technique for the evaluation of visualization schemes for LiDAR data, by conducting a visualization experience survey for 13 pre-processing and visualization schemes where 60 participants rated these schemes on a 10 point scale on a questionnaire. The paper establishes a ranking for the different visualization schemes described herein. Finally, this paper establishes that our heuristic-based algorithm (presented elsewhere) performs almost equal to a classification-based visualization pipeline made using professional software. We believe that the presented technique can be used to assess other geospatial visualization schemes.     

Assessing reference dataset representativeness through confidence metrics based on information density

Land cover maps obtained from classification of remotely sensed imagery provide valuable information in numerous environmental monitoring and modeling tasks. However, many uncertainties and errors can directly or indirectly affect the quality of derived maps. This work focuses on one key aspect of the supervised classification process of remotely sensed imagery: the quality of the reference dataset used to develop a classifier. More specifically, the representative power of the reference dataset is assessed by contrasting it with the full dataset (e.g. entire image) needing classification. Our method is applicable in several ways: training or testing datasets (extracted from the reference dataset) can be compared with the full dataset. The proposed method moves beyond spatial sampling schemes (e.g. grid, cluster) and operates in the multidimensional feature space (e.g. spectral bands) and uses spatial statistics to compare information density of data to be classified with data used in the reference process. The working hypothesis is that higher information density, not in general but with respect to the entire classified image, expresses higher confidence in obtained results. Presented experiments establish a close link between confidence metrics and classification accuracy for a variety of image classifiers namely maximum likelihood, decision tree, Backpropagation Neural Network and Support Vector Machine. A sensitivity analysis demonstrates that spatially-continuous reference datasets (e.g. a square window) have the potential to provide similar classification confidence as typically-used spatially-random datasets. This is an important finding considering the higher acquisition costs for randomly distributed datasets. Furthermore, the method produces confidence maps that allow spatially-explicit comparison of confidence metrics within a given image for identification of over- and under-represented image portions. The current method is presented for individual image classification but, with sufficient evaluation from the remote sensing community it has the potential to become a standard for reference dataset reporting and thus allowing users to assess representativeness of reference datasets in a consistent manner across different classification tasks.     

Selection of LiDAR geometric features with adaptive neighborhood size for urban land cover classification

LiDAR has been an effective technology for acquiring urban land cover data in recent decades. Previous studies indicate that geometric features have a strong impact on land cover classification. Here, we analyzed an urban LiDAR dataset to explore the optimal feature subset from 25 geometric features incorporating 25 scales under 6 definitions for urban land cover classification. We performed a feature selection strategy to remove irrelevant or redundant features based on the correlation coefficient between features and classification accuracy of each features. The neighborhood scales were divided into small (0.5–1.5 m), medium (1.5–6 m) and large (>6 m) scale. Combining features with lower correlation coefficient and better classification performance would improve classification accuracy. The feature depicting homogeneity or heterogeneity of points would be calculated at a small scale, and the features to smooth points at a medium scale and the features of height different at large scale. As to the neighborhood definition, cuboid and cylinder were recommended. This study can guide the selection of optimal geometric features with adaptive neighborhood scale for urban land cover classification.     

高分辨率遥感影像场景的多尺度神经网络分类法

高分辨率遥感影像场景分类是实现复杂场景快速自动识别的基础,在军事、救灾等领域有十分重要的意义。为了在有限的遥感数据集上获得高识别精度,本文提出了一种基于联合多尺度卷积神经网络模型的高分辨率遥感影像场景分类方法。不同于传统的卷积神经网络模型,JMCNN建立了一个具有3个不同尺度通道的端对端多尺度联合卷积网络模型,包括多通道特征提取器、多尺度特征联合和Softmax分类3个部分。首先,多通道特征提取器提取图像中、高层多尺度特征;然后,多尺度特征联合对多个通道的中、高层多尺度特征进行多次融合以增强特征表达;最后,Softmax对高层特征进行分类。本文在UC Merced和SIRI遥感数据集进行测试,试验表明JMCNN模型在特征表达和计算速度方面均有显著提高,在小样本数据量下分别达到89.3%和88.3%的识别精度。     

A Space-Time GIS Approach to Exploring Large Individual-based Spatiotemporal Datasets

The increasing number of large individual-based spatiotemporal datasets in various research fields has challenged the GIS community to develop analysis tools that can efficiently help researchers explore the datasets in order to uncover useful information. Rooted in Hägerstrand's time geography, this study presents a generalized space-time path (GSTP) approach to facilitating visualization and exploration of spatiotemporal changes among individuals in a large dataset. The fundamental idea of this approach is to derive a small number of representative space-time paths (i.e. GSTPs) from the raw dataset by identifying spatial cluster centers of observed individuals at different time periods and connecting them according to their temporal sequence. A space-time GIS environment is developed to implement the GSTP concept. Different methods of handling temporal data aggregation and the creation of GSTPs are discussed in this article. Using a large individual-based migration history dataset, this study successfully develops an operational space-time GIS prototype in ESRI's ArcScene and ArcMap to provide a proof-of-concept study of this approach. This space-time GIS system demonstrates that the proposed GSTP approach can provide a useful exploratory analysis and geovisualization environment to help researchers effectively search for hidden patterns and trends in such datasets.     

Crop type mapping using LiDAR,Sentinel-2 and aerial imagery with machine learning algorithms

LiDAR data are becoming increasingly available, which has opened up many new applications. One such application is crop type mapping. Accurate crop type maps are critical for monitoring water use, estimating harvests and in precision agriculture. The traditional approach to obtaining maps of cultivated fields is by manually digitizing the fields from satellite or aerial imagery and then assigning crop type labels to each field - often informed by data collected during ground and aerial surveys. However, manual digitizing and labeling is time-consuming, expensive and subject to human error. Automated remote sensing methods is a cost-effective alternative, with machine learning gaining popularity for classifying crop types. This study evaluated the use of LiDAR data, Sentinel-2 imagery, aerial imagery and machine learning for differentiating five crop types in an intensively cultivated area. Different combinations of the three datasets were evaluated along with ten machine learning. The classification results were interpreted by comparing overall accuracies, kappa, standard deviation and f-score. It was found that LiDAR data successfully differentiated between different crop types, with XGBoost providing the highest overall accuracy of 87.8％. Furthermore, the crop type maps produced using the LiDAR data were in general agreement with those obtained by using Sentinel-2 data, with LiDAR obtaining a mean overall accuracy of 84.3％ and Sentinel-2 a mean overall accuracy of 83.6％. However, the combination of all three datasets proved to be the most effective at differentiating between the crop types, with RF providing the highest overall accuracy of 94.4％. These findings provide a foundation for selecting the appropriate combination of remotely sensed data sources and machine learning algorithms for operational crop type mapping.     

Assessing the potential for leaf-off LiDAR data to model canopy closure in temperate deciduous forests

Estimates of canopy closure have many important uses in forest management and ecological research. Field measurements, however, are typically not practical to acquire over expansive areas or for large numbers of locations. This problem has been addressed, in recent years, through the use of airborne light detection and ranging (LiDAR) technology which has proven effective in modeling canopy closure remotely. The techniques developed to use LiDAR for this purpose have been designed and evaluated for datasets acquired during leaf-on conditions. However, a large number of LiDAR datasets are acquired during leaf-off conditions since their primary purpose is to generate bare-earth Digital Elevation Models. In this paper, we develop and evaluate techniques for leveraging small-footprint leaf-off LiDAR data to model leaf-on canopy closure in temperate deciduous forests.We evaluate three techniques for modeling canopy closure: (1) the canopy-to-total-return-ratio (CTRR), (2) the canopy-to-total-pixel-ratio (CTPR), and (3) the hemispherical-viewshed (HV). The first technique has been used widely, in various forms, and has been shown to be effective with leaf-on LiDAR datasets. The CTRR technique that we tested uses the first-return LiDAR data only. The latter two techniques are new contributions that we develop and present in this paper. These techniques use Canopy Height Models (CHM) to detect significant gaps in the forest canopy which are of primary importance in estimating closure.The techniques we tested each showed good promise for predicting canopy closure using leaf-off LiDAR data with the CTPR and HV models having particularly high correlations with closure estimates from hemispherical photographs. The CTRR model had performance on par with results from previous studies that used leaf-on LiDAR, although, with leaf-off data the model tended to be negatively biased with respect to species having simple and compound leaf types and positively biased for coniferous species. The CTPR and HV models also showed some slight negative biases for compound-leaf species. The biases for the CTPR and HV models were mitigated when the CHM data were smoothed to fill in small gaps. The CHM-based models were robust to changes in the CHM model resolution which suggests that these methods may be applicable to a variety of small-footprint LiDAR datasets. In this research, the new CTPR and HV methods showed a strong ability to predict canopy closure using leaf-off data, however, future work will be needed to test the applicability of the models to variations in LiDAR datasets, forest types, and topography.     

Object-oriented semantic labelling of spectral–spatial LiDAR point cloud for urban land cover classification and buildings detection

The urban land cover mapping and automated extraction of building boundaries is a crucial step in generating three-dimensional city models. This study proposes an object-based point cloud labelling technique to semantically label light detection and ranging (LiDAR) data captured over an urban scene. Spectral data from multispectral images are also used to complement the geometrical information from LiDAR data. Initial object primitives are created using a modified colour-based region growing technique. Multiple classifier system is then applied on the features extracted from the segments for classification and also for reducing the subjectivity involved in the selection of classifier and improving the precision of the results. The proposed methodology produces two outputs: (i) urban land cover classes and (ii) buildings masks which are further reconstructed and vectorized into three-dimensional buildings footprints. Experiments carried out on three airborne LiDAR datasets show that the proposed technique successfully discriminates urban land covers and detect urban buildings.     

通过训练样本采样处理改善小宗作物遥感识别精度

训练样本质量是决定农作物遥感识别精度的关键因素,虽然高空间分辨率卫星的发展有效地解决了农作物遥感识别过程中的混合像元问题,但是当区域内不同作物种植面积差异较大时,训练集中不同类别样本数量往往相差较大,这样的不均衡数据集影响分类器的训练,导致少数类别的识别精度不理想。为研究作物遥感识别过程中的不均衡样本问题,本文基于GF-2号卫星数据,首先挖掘了地物的光谱信息、纹理信息,用特征递归消除RFE (Recursive Feature Elimination)方法进行特征优选,然后从数据处理的角度采用了5种采样算法对不均衡训练集进行处理,最后使用采样后的均衡数据集训练分类器,对比数据采样前后决策树与Adaboost(Adaptive Boosting)两种分类器的识别结果,发现:(1)经过采样处理后两种分类算法明显提升了小宗作物的分类精度;(2)经过ADASYS (Adaptive synthetic sampling)采样处理后,分类器性能提升最多,决策树的Kappa系数提高了14.32%,Adaboost的Kappa系数提高了10.23%,达到最高值0.9336;(3)过采样的处理效果优于欠采样,过采样对分类器的性能提升更多。综上所述,选择合适的采样方法和分类方法是提高不均衡数据集遥感分类精度的有效途径。     

Feature fusion to improve road network extraction in high-resolution SAR images

This letter aims at the extraction of roads and road networks from high-resolution synthetic aperture radar data. Classical methods based on line detection do not use all the information available; indeed, in high-resolution data, roads are large enough to be considered as regions and can be characterized also by their statistics. This property can be used in a classification scheme. Therefore, this letter presents a road extraction method which is based on the fusion of classification (statistical information) and line detection (structural information). This fusion is done at the feature level, which helps to improve both the level of likelihood and the number of the extracted roads. The proposed approach is tested with two classification methods and one line extractor. Results on two different datasets are discussed.     

A multiresolution hierarchical classification algorithm for filtering airborne LiDAR data

We presented a multiresolution hierarchical classification (MHC) algorithm for differentiating ground from non-ground LiDAR point cloud based on point residuals from the interpolated raster surface. MHC includes three levels of hierarchy, with the simultaneous increase of cell resolution and residual threshold from the low to the high level of the hierarchy. At each level, the surface is iteratively interpolated towards the ground using thin plate spline (TPS) until no ground points are classified, and the classified ground points are used to update the surface in the next iteration. 15 groups of benchmark dataset, provided by the International Society for Photogrammetry and Remote Sensing (ISPRS) commission, were used to compare the performance of MHC with those of the 17 other publicized filtering methods. Results indicated that MHC with the average total error and average Cohen’s kappa coefficient of 4.11% and 86.27% performs better than all other filtering methods.     

Evaluation of LiDAR and image segmentation based classification techniques for automatic building footprint extraction for a segment of Atlantic County,New Jersey

Extracting high-quality building footprints is a basic requirement in multiple sectors of town planning, disaster management, 3D visualization, etc. In the current study, we compare three different techniques for acquiring building footprints using (i) LiDAR, (ii) object-oriented classification (OOC) applied on high-resolution aerial photographs and (iii) digital surface models generated from interpolated LiDAR point cloud data. The three outputs were compared with a digitized sample of building polygons quantitatively by computing the errors of commission and omission, and qualitatively using statistical operations. These findings showed that building footprints derived from OOC gave highest regression and correlation values with least commission error. The R² and R values (0.86 and 0.92, respectively) imply that the footprint areas derived by OOC matched more closely with the actual area of buildings, while a low commission error of 24.7% represented a higher number of footprints as correctly classified.     

融合航空影像的LiDAR地物点云分类

点云滤波分类是LiDAR后续应用的基础工作,在点云滤波的基础上,以航空影像为辅助条件,结合点云高程信息,设计一套地物点云的分类方法。该方法首先融合航空影像与LiDAR数据,将对应RGB值赋予每个点,根据植被的光谱特征提取出部分植被点云;然后再根据文中定义的点云高程纹理,在剩余地物点云中提取出建筑物点,最后根据回波次数信息分离出剩余植被点,完成地物点云的分类。采用北京凤凰岭地区一组机载LiDAR数据进行实验。实验结果表明,该方法能够有效地将地物点云进行分类并且满足一定的精度要求,具有一定的实用价值。     

Automatic registration of optical imagery with 3D LiDAR data using statistical similarity

The development of robust and accurate methods for automatic registration of optical imagery and 3D LiDAR data continues to be a challenge for a variety of applications in photogrammetry, computer vision and remote sensing. This paper proposes a new approach for the registration of optical imagery with LiDAR data based on the theory of Mutual Information (MI), which exploits the statistical dependency between same- and multi-modal datasets to achieve accurate registration. The MI-based similarity measures quantify dependencies between aerial imagery, and both LiDAR intensity data and 3D point cloud data. The needs for specific physical feature correspondences, which are not always attainable in the registration of imagery with 3D point clouds, are avoided. Current methods for registering 2D imagery to 3D point clouds are first reviewed, after which the mutual MI approach is presented. Particular attention is given to adoption of the Normalised Combined Mutual Information (NCMI) approach as a means to produce a similarity measure that exploits the inherently registered LiDAR intensity and point cloud data so as to improve the robustness of registration between optical imagery and LiDAR data. The effectiveness of local versus global similarity measures is also investigated, as are the transformation models involved in the registration process. An experimental program conducted to evaluate MI-based methods for registering aerial imagery to LiDAR data is reported and the results obtained in two areas with differing terrain and land cover, and with aerial imagery of different resolution and LiDAR data with different point density are discussed. These results demonstrate the potential of the MI and especially the CMI methods for registration of imagery and 3D point clouds, and they highlight the feasibility and robustness of the presented MI-based approach to automated registration of multi-sensor, multi-temporal and multi-resolution remote sensing data for a wide range of applications.     

Feature reduction using a singular value decomposition for the iterative guided spectral class rejection hybrid classifier

Feature reduction in a remote sensing dataset is often desirable to decrease the processing time required to perform a classification and improve overall classification accuracy. This paper introduces a feature reduction method based on the singular value decomposition (SVD). This SVD-based feature reduction method reduces the storage and processing requirements of the SVD by utilizing a training dataset. This feature reduction technique was applied to training data from two multitemporal datasets of Landsat TM/ETM+ imagery acquired over a forested area in Virginia, USA and Rondônia, Brazil. Subsequent parallel iterative guided spectral class rejection (pIGSCR) forest/non-forest classifications were performed to determine the quality of the feature reduction. The classifications of the Virginia data were five times faster using SVD-based feature reduction without affecting the classification accuracy. Feature reduction using the SVD was also compared to feature reduction using principal components analysis (PCA). The highest average accuracies for the Virginia dataset (88.34%) and for the Rondônia dataset (93.31%) were achieved using the SVD. The results presented here indicate that SVD-based feature reduction can produce statistically significantly better classifications than PCA.     

An intelligent geospatial processing unit for image classification based on geographic vector agents (GVAs)

Spatial modeling methods usually use pixels and image objects as fundamental processing units to address real‐world objects, geo‐objects, in image space. To do this, both pixel‐based and object‐based approaches typically employ a linear two‐staged workflow of segmentation and classification. Pixel‐based methods segment a classified image to address geo‐objects in image space. In contrast, object‐based approaches classify a segmented image to identify geo‐objects from raster datasets. These methods lack the ability to simultaneously integrate the geometry and theme of geo‐objects in image space. This article explores Geographical Vector Agents (GVAs) as an automated and intelligent processing unit to directly address real‐world objects in the process of remote sensing image classification. The GVA is a distinct type of geographic automata characterized by elastic geometry, dynamic internal structure, neighborhoods and their respective rules. We test this concept by modeling a set of objects on a subset IKONOS image and LiDAR DSM datasets without the setting parameters (e.g. scale, shape information), usually applied in conventional Geographic Object‐Based Image Analysis (GEOBIA) approaches. The results show that the GVA approach achieves more than 3.5% improvement for correctness, 2% improvement for quality, although no significant improvement for completeness to GEOBIA, thus demonstrating the competitive performance of GVAs classification.     

The Use of LiDAR in Digital Surface Modelling: Issues and Errors

Modelling and extracting 3D geographical data presents numerous challenges that require continual research to attempt to evolve an efficient, reliable and accurate solution. LiDAR data capture and analysis has become a preferred acquisition choice for elevation data because the resulting quality and level of detail far exceeds traditional methods for large survey areas. As with any data collection system, LiDAR is prone to errors. Analysing these errors, ascertaining causes and producing error correction strategies is vital if accurate and confident results are to be obtained. Eight years of LiDAR datasets (from 1998 to 2005) have been closely analysed for a large coastal area of South Wales. This article provides a detailed and accurate summary of the identified LiDAR data issues and subsequent errors which affect the accuracy of end products such as Digital Surface Models (DSMs).     

基于深度残差网络的机载LiDAR点云分类

机载LiDAR点云的分类是利用其进行城市场景三维重建的关键步骤之一。为充分利用现有的图像领域性能较好的深度学习网络模型,提高点云分类精度,并降低训练时间和对训练样本数量的要求,本文提出一种基于深度残差网络的机载LiDAR点云分类方法。首先提取归一化高程、表面变化率、强度和归一化植被指数4种具有较高区分度的点云低层次特征;然后通过设置不同的邻域大小和视角,利用所提出的点云特征图生成策略,得到多尺度和多视角点云特征图;再将点云特征图输入到预训练的深度残差网络,提取多尺度和多视角深层次特征;最后构建并训练神经网络分类器,利用训练的模型对待分类点云进行预测,经后处理得到分类结果。利用ISPRS三维语义标记竞赛的公开标准数据集进行试验,结果表明,本文方法可有效区分建筑物、地面、车辆等8类地物,分类结果的总体精度为87.1%,可为城市场景三维重建提供可靠的信息。     

Demonstration of a virtual active hyperspectral LiDAR in automated point cloud classification

In this paper, a measurement system for the acquisition of a virtual hyperspectral LiDAR dataset is presented. As commercial hyperspectral LiDARs are not yet available, the system provides a novel type of data for the testing and developing of future hyperspectral LiDAR algorithms. The measurement system consists of two parts: first, backscattered reflectance spectra are collected using a spectrometer and a cutting-edge technology, white-light supercontinuum laser source; second, a commercial monochromatic LiDAR system is used for ranging. A virtual hyperspectral LiDAR dataset is produced by data fusion. Such a dataset was collected on a Norway spruce (Picea abies) sample. The performance of classification was tested using an experimental hyperspectral algorithm based on a novel combination of the Spectral Correlation Mapper and a region growing algorithm. The classifier was able to automatically distinguish between needles, branches and background, in other words, perform a difficult task using only traditional TLS data.     

Radiometric normalization of overlapping LiDAR intensity data for reduction of striping noise

Airborne LiDAR data are usually collected with partially overlapping strips in order to serve a seamless and fine resolution mapping purpose. One of the factors limiting the use of intensity data is the presence of striping noise found in the overlapping region. Though recent researches have proposed physical and empirical approaches for intensity data correction, the effect of striping noise has not yet been resolved. This paper presents a radiometric normalization technique to normalize the intensity data from one data strip to another one with partial overlap. The normalization technique is built based on a second-order polynomial function fitted on the joint histogram plot, which is generated with a set of pairwise closest data points identified within the overlapping region. The proposed method was tested with two individual LiDAR datasets collected by Teledyne Optech's Gemini (1064?nm) and Orion (1550?nm) sensors. The experimental results showed that radiometric correction and normalization can significantly reduce the striping noise found in the overlapping LiDAR intensity data and improve its capability in land cover classification. The coefficient of variation of five selected land cover features was reduced by 19–65%, where a 9–18% accuracy improvement was achieved in different classification scenarios. With the proven capability of the proposed method, both radiometric correction and normalization should be applied as a pre-processing step before performing any surface classification and object recognition.