A Comparison of Deep Learning Methods for Airborne Lidar Point Clouds Classification

Li, Nan; Kähler, Olaf; Pfeifer, Norbert

doi:10.1109/jstars.2021.3091389

Cited by 22 publications

(13 citation statements)

References 64 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…Unsurprisingly, the overall best results were achieved by the semantic segmentation methods, whereby the 3D network was better still than the 2D version and closely followed by random forest. With F1 scores over 97%, we reached the limits of ground truth quality [86]. Similar results can be expected from other point cloud classification networks [84,86,119].…”

Section: Stem Precision and Recallsupporting

confidence: 86%

“…With F1 scores over 97%, we reached the limits of ground truth quality [86]. Similar results can be expected from other point cloud classification networks [84,86,119].…”

Section: Stem Precision and Recallsupporting

confidence: 86%

“…Prevalent examples are PointNet++ [76], sparse convolutional networks [77][78][79], PointCNN [80], and KPConv [81], just to name a few. Such neural networks proved to be accurate and fast tools for semantic segmentation of large ALS point clouds [82][83][84][85][86]. Yet, since deep learning tends to require large amounts of training data, generating semantic point labels to train such a deep neural network is a tedious task.…”

Section: Three-dimensional Deep Learningmentioning

confidence: 99%

“…This is particularly useful to build convolutional neural networks for sparse 3D point cloud data. SCN architectures proved to be successful for ALS point cloud classification [82,86]. Furthermore, in contrast to more specialized point cloud architectures such as PointNet++ or KPConv, this kind of network allows us to conduct ablations with a direct 2D CNN counterpart (see Sections 4.3 and 4.4.6).…”

Section: Architecturementioning

confidence: 99%

See 3 more Smart Citations

Individual Tree Detection in Urban ALS Point Clouds with 3D Convolutional Networks

2022

View full text Add to dashboard Cite

Since trees are a vital part of urban green infrastructure, automatic mapping of individual urban trees is becoming increasingly important for city management and planning. Although deep-learning-based object detection networks are the state-of-the-art in computer vision, their adaptation to individual tree detection in urban areas has scarcely been studied. Some existing works have employed 2D object detection networks for this purpose. However, these have used three-dimensional information only in the form of projected feature maps. In contrast, we exploited the full 3D potential of airborne laser scanning (ALS) point clouds by using a 3D neural network for individual tree detection. Specifically, a sparse convolutional network was used for 3D feature extraction, feeding both semantic segmentation and circular object detection outputs, which were combined for further increased accuracy. We demonstrate the capability of our approach on an urban topographic ALS point cloud with 10,864 hand-labeled ground truth trees. Our method achieved an average precision of 83% regarding the common 0.5 intersection over union criterion. 85% percent of the stems were found correctly with a precision of 88%, while tree area was covered by the individual tree detections with an F1 accuracy of 92%. Thereby, we outperformed traditional delineation baselines and recent detection networks.

show abstract

Section: Stem Precision and Recallsupporting

confidence: 86%

“…With F1 scores over 97%, we reached the limits of ground truth quality [86]. Similar results can be expected from other point cloud classification networks [84,86,119].…”

Section: Stem Precision and Recallsupporting

confidence: 86%

Section: Three-dimensional Deep Learningmentioning

confidence: 99%

Section: Architecturementioning

confidence: 99%

See 2 more Smart Citations

Individual Tree Detection in Urban ALS Point Clouds with 3D Convolutional Networks

2022

View full text Add to dashboard Cite

show abstract

“…R. M. Asiyabi is with the Center for Spatial Information (CEOSpaceTech) of the University POLITEHNICA of Bucharest (UPB), Romania (e-mail: reza.mohammadi@upb.ro). information of LIDAR data [7], [8], and other EO data, using many different algorithms.…”

Section: Introductionmentioning

confidence: 99%

Earth Observation Semantic Data Mining: Latent Dirichlet Allocation-Based Approach

Asiyabi

Datcu

2022

IEEE J. Sel. Top. Appl. Earth Observations Remote Sensing

View full text Add to dashboard Cite

Recent advances in remote sensing technology have provided (very) high spatial resolution Earth Observation (EO) data with abundant latent semantic information. Conventional data processing algorithms are not capable of extracting the latent semantic information form these data and harness their full potential. As a result, semantic information discovery methods, based on data mining techniques, such as Latent Dirichlet Allocation (LDA) and Bag of Visual Words (BOVW) models, can discover the latent information. Despite their crucial rule, there are only a few studies in the field of semantic data mining for remote sensing applications. This study is focused on this shortage. Three different scenarios are used to evaluate the semantic information discovery in various remote sensing applications, including both optical and Synthetic Aperture Radar (SAR) data with different spatial resolutions. In the first scenario, semantic discovery method correlated the semantic perception of the user and machine to correct and enhance the user defined Ground Truth (GT) map in very high-resolution RGB data. The potential of the semantic discovery is evaluated for wildfire affected area detection in Sentinel-2 data in the second scenario. Finally, in the third scenario, the semantic discovery method is utilized to detect the misclassifications as well as the patches with ambiguous or multiple semantic labels in a Sentinel-1 SAR patch-based benchmark dataset, to enhance the robustness and accuracy of the annotation in the dataset. Our results in these three scenarios demonstrated the capability of the data mining-based semantic information discovery methods for various remote sensing applications.

show abstract

Human‐in‐the‐loop development of spatially adaptive ground point filtering pipelines—An archaeological case study

Doneus

Höfle

Kempf

et al. 2022

Archaeological Prospection

View full text Add to dashboard Cite

LiDAR data have become indispensable for research in archaeology and a variety of other topographic applications. To derive products (e.g. digital terrain or feature models, individual trees, buildings), the 3D LiDAR points representing the desired objects of interest within the acquired and georeferenced point cloud need to be identified. This process is known as classification, where each individual point is assigned to an object class. In archaeological prospection, classification focuses on identifying the object class 'ground points'. These are used to interpolate digital terrain models exposing the microtopography of a terrain to be able to identify and map archaeological and palaeoenvironmental features. Setting up such classification workflows can be time-consuming and prone to information loss, especially in geographically heterogeneous landscapes. In such landscapes, one classification setting can lead to qualitatively very different results, depending on varying terrain parameters such as steepness or vegetation density. In this paper, we are focussing on a special workflow for optimal classification results in these heterogeneous environments, which integrates expert knowledge. We present a novel Pythonbased open-source software solution, which helps to optimize this process and creates a single digital terrain model by an adaptive classification based on spatial segments. The advantage of this approach for archaeology is to produce coherent digital terrain models even in geomorphologically heterogenous areas or areas with patchy vegetation. The software is also useful to study the effects of different algorithm and parameter combinations on digital terrain modelling with a focus on a practical and time-saving implementation. As the developed pipelines and all meta-information are made available with the resulting data set, classification is white boxed and consequently scientifically comprehensible and repeatable.Together with the software's ability to simplify classification workflows significantly, it will be of interest for many applications also beyond the examples shown from archaeology.

show abstract

A Comparison of Deep Learning Methods for Airborne Lidar Point Clouds Classification

Cited by 22 publications

References 64 publications

Individual Tree Detection in Urban ALS Point Clouds with 3D Convolutional Networks

Individual Tree Detection in Urban ALS Point Clouds with 3D Convolutional Networks

Earth Observation Semantic Data Mining: Latent Dirichlet Allocation-Based Approach

Human‐in‐the‐loop development of spatially adaptive ground point filtering pipelines—An archaeological case study

Contact Info

Product

Resources

About