ALS Point Cloud Classification With Small Training Data Set Based on Transfer Learning

Zhao, Chuan; Guo, Hao; Lu, Jun; Yu, Donghang; Li, Daoji; Chen, Xiaowei

doi:10.1109/lgrs.2019.2947608

Cited by 21 publications

(11 citation statements)

References 18 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…Once the parameters are determined, automatic interpretation can be performed on large-scale ALS point clouds. In addition, it takes only about 0.5 h to train a CRF model on the Vaihingen Dataset in our work, while the training time in a deep learning framework takes about three to six days [54].…”

Section: Discussionmentioning

confidence: 99%

Large-Scale ALS Data Semantic Classification Integrating Location-Context-Semantics Cues by Higher-Order CRF

Han

Wang

Huang

et al. 2020

Sensors

View full text Add to dashboard Cite

We designed a location-context-semantics-based conditional random field (LCS-CRF) framework for the semantic classification of airborne laser scanning (ALS) point clouds. For ALS datasets of high spatial resolution but with severe noise pollutions, more contexture and semantics cues, besides location information, can be exploited to surmount the decrease of discrimination of features for classification. This paper mainly focuses on the semantic classification of ALS data using mixed location-context-semantics cues, which are integrated into a higher-order CRF framework by modeling the probabilistic potentials. The location cues modeled by the unary potentials can provide basic information for discriminating the various classes. The pairwise potentials consider the spatial contextual information by establishing the neighboring interactions between points to favor spatial smoothing. The semantics cues are explicitly encoded in the higher-order potentials. The higher-order potential operates at the clusters level with similar geometric and radiometric properties, guaranteeing the classification accuracy based on semantic rules. To demonstrate the performance of our approach, two standard benchmark datasets were utilized. Experiments show that our method achieves superior classification results with an overall accuracy of 83.1% on the Vaihingen Dataset and an overall accuracy of 94.3% on the Graphics and Media Lab (GML) Dataset A compared with other classification algorithms in the literature. 2 of 28 from different equipment. Therefore, we incorporate location, spatial contextual, and semantics cues within a higher-order conditional random field (CRF) framework to provide complementary information from varying perspectives, so that it can address the common misjudgment of semantic classes in ALS point clouds, from the perspectives of the accuracy of each class and the overall accuracy. Related Works for ALS Point Cloud ClassificationAccording to the type of entity used for classification, existing methods can be categorized as point-based and cluster-based (or segment-based) [2,3]. Point-based methods classify each point of the ALS data by using features as the inputs for supervised or unsupervised classifiers [4], while cluster-based methods segment the ALS data into clusters, then class labels are assigned to the clusters in which all points share the same class label [5,6]. We briefly review the aforementioned methods, and demonstrate the rationale for our method in what follows.Point-based methods generally extract point-wise features locally from the neighborhood defined by a sphere or cylinder. Therefore, such methods usually focus on the selection of discriminative features and effective classifiers. For instance, Reference [7] worked on 3D scene analysis, including geometric features extraction and optimal neighbors selection. Then an optimal eigenentropy-based scale selection method was proposed. Reference [2] combined airborne LiDAR with images to extract more discriminative features. Then, based on these ...

show abstract

Section: Discussionmentioning

confidence: 99%

Large-Scale ALS Data Semantic Classification Integrating Location-Context-Semantics Cues by Higher-Order CRF

Han

Wang

Huang

et al. 2020

Sensors

View full text Add to dashboard Cite

show abstract

“…As an example, a recent work employed a high-level multi-level feature selection strategy based on the intensity and normalized height features [36]. Although their proposed method achieved satisfactory results in reducing training time and the number of training samples, densely structured and deeper CNNs would obtain more accurate classification results.…”

Section: Deep Learning-based Methodsmentioning

confidence: 99%

High-Resolution Terrain Modeling Using Airborne LiDAR Data with Transfer Learning

Liu

et al. 2021

Remote Sensing

View full text Add to dashboard Cite

This study presents a novel workflow for automated Digital Terrain Model (DTM) extraction from Airborne LiDAR point clouds based on a convolutional neural network (CNN), considering a transfer learning approach. The workflow consists of three parts: feature image generation, transfer learning using ResNet, and interpolation. First, each point is transformed into a featured image based on its elevation differences with neighboring points. Then, the feature images are classified into ground and non-ground using ImageNet pretrained ResNet models. The ground points are extracted by remapping each feature image to its corresponding points. Last, the extracted ground points are interpolated to generate a continuous elevation surface. We compared the proposed workflow with two traditional filters, namely the Progressive Morphological Filter (PMF) and the Progressive Triangulated Irregular Network Densification (PTD). Our results show that the proposed workflow establishes an advantageous DTM extraction accuracy with yields of only 0.52%, 4.84%, and 2.43% for Type I, Type II, and the total error, respectively. In comparison, Type I, Type II, and the total error for PMF are 7.82%, 11.60%, and 9.48% and for PTD 1.55%, 5.37%, and 3.22%, respectively. The root means square error (RMSE) for the 1 m resolution interpolated DTM is only 7.3 cm. Moreover, we conducted a qualitative analysis to investigate the reliability and limitations of the proposed workflow.

show abstract

“…The basic idea of subsurface growing is to incorporate the measured surface points of an ALS data set during the segmentation process and create and consider virtual points located below the real roof surface. The extension consists of the following two steps: creating virtual points in (1), and merging of neighboring segments in (2). Specifically, the creation of virtual points is generally not limited in number, but they must be located under real point measurements and must be close to a plane represented by the measured points of a segment.…”

Section: A Subsurface Growingmentioning

confidence: 99%

“…Airborne Light Detection and Ranging (LiDAR), also called Airborne Laser Scanning (ALS), is an active sensor used in remote sensing techniques for extracting various information about physical surfaces. ALS provides dense, discrete, detailed, and accurate three-dimensional (3D) point clouds that directly capture objects and ground surfaces [1].…”

Section: Introductionmentioning

confidence: 99%

Building Roof Superstructures Classification From Imbalanced and Low Density Airborne LiDAR Point Cloud

et al. 2021

View full text Add to dashboard Cite

Light Detection and Ranging (LiDAR), an active remote sensing technology, is becoming an essential tool for geoinformation extraction and urban planning. Airborne Laser Scanning (ALS) point clouds segmentation and accurate classification are challenging and crucial to produce different geoinformation products like three-dimensional (3D) city designs. This paper introduces an effective data-driven approach to build roof superstructures classification for airborne LiDAR point clouds with very low density and imbalanced classes, covering an urban area. Notably, it focuses on building roof superstructures (especially dormers and chimneys) and mitigating nonplanar objects' problems. Also, the imbalanced class problem of LiDAR data, to the best of our knowledge, is not yet addressed in the literature; it is considered in this study. The major advantage of the proposed approach is using only raw data without assumptions on the distribution underlying data. The main methodological novelties of this work are summarized in the following key elements. (i) At first, an adapted connected component analysis for 3D points cloud is proposed. (ii) Twelve geometry-based features are extracted for each component. (iii) A Support Vector Machine (SVM)-driven procedure is applied to classify the 3D components. (iv) Furthermore, a new component size-based sampling (CSBS) method is proposed to treat the imbalanced data problem and has been compared with several existing resampling strategies. In this study, components are classified into five classes: shed and gable dormers, chimneys, ground, and others. The results of this investigation show the satisfying classification performance of the proposed approach.Results also showed that the proposed approach outperformed machine learning methods, including SVM, Random Forest, Decision Tree, and Adaboost.

show abstract

ALS Point Cloud Classification With Small Training Data Set Based on Transfer Learning

Cited by 21 publications

References 18 publications

Large-Scale ALS Data Semantic Classification Integrating Location-Context-Semantics Cues by Higher-Order CRF

Large-Scale ALS Data Semantic Classification Integrating Location-Context-Semantics Cues by Higher-Order CRF

High-Resolution Terrain Modeling Using Airborne LiDAR Data with Transfer Learning

Building Roof Superstructures Classification From Imbalanced and Low Density Airborne LiDAR Point Cloud

Contact Info

Product

Resources

About