Article
Forestry
Dong-Hyeon Kim, Chi-Ung Ko, Dong-Geun Kim, Jin-Taek Kang, Jeong-Mook Park, Hyung-Ju Cho
Summary: This study proposes a new approach using the PointNet++ model to segment the canopy, trunk, and branches of trees and identifies an optimal learning environment for the model. The approach is validated using LiDAR point cloud data from 435 tree samples, and the best performance is achieved with 4096 representative points.
Article
Environmental Sciences
Yifan Huang, Yan He, Xiaolei Zhu, Jiayong Yu, Yongqiang Chen
Summary: Airborne LIDAR is an active remote sensing technology that uses specific wavelengths of laser light to penetrate seawater. By using a depth extraction method based on the PointConv deep learning model, faint seafloor echoes can be successfully extracted and distinguished from background noise.
Article
Environmental Sciences
Kai Xiao, Jia Qian, Teng Li, Yuanxi Peng
Summary: In this study, a deep learning network is proposed to complete semantic segmentation of large-scale point clouds by leveraging contextual semantic information. By fusing local geometry and feature content, as well as implementing data preprocessing with principal component extraction, the network's processing capability is improved and satisfactory results are achieved.
Article
Environmental Sciences
Reza Mahmoudi Kouhi, Sylvie Daniel, Philippe Giguere
Summary: Currently, 3D point clouds are widely used for presenting 3D objects and accurately localizing them. However, the lack of semantic information in raw point clouds has led to the development of deep neural networks for semantic segmentation. Few prior works have studied the impact of data preparation on network performance. Therefore, this study proposes novel data preparation methods that improve the performance of deep neural networks for point cloud semantic segmentation.
Article
Remote Sensing
Xin Xu, Federico Iuricich, Kim Calders, John Armston, Leila De Floriani
Summary: This paper introduces an automated tree segmentation method based on terrestrial laser scanning technology, which utilizes a new topological algorithm to segment point clouds into individual trees without user interactions. The algorithm identifies tree bottoms and tops and reconstructs single trees using relevant topological features. Experimental results demonstrate its higher segmentation accuracy and potential for wide applications in the forest ecology community.
INTERNATIONAL JOURNAL OF APPLIED EARTH OBSERVATION AND GEOINFORMATION
(2023)
Article
Computer Science, Artificial Intelligence
Fan Lu, Guang Chen, Yinlong Liu, Lijun Zhang, Sanqing Qu, Shu Liu, Rongqi Gu, Changjun Jiang
Summary: In this paper, a hierarchical network called HRegNet is proposed for efficient registration of large-scale outdoor LiDAR point clouds. By performing registration on hierarchically extracted keypoints and descriptors, the network combines reliable features from deeper layers with precise position information from shallower layers to achieve robust and accurate registration. Experimental results demonstrate the high accuracy and efficiency of HRegNet.
IEEE TRANSACTIONS ON PATTERN ANALYSIS AND MACHINE INTELLIGENCE
(2023)
Article
Geochemistry & Geophysics
Mohammad Pashaei, Michael J. Starek, Craig L. Glennie, Jacob Berryhill
Summary: Information derived from full-waveform (FW) lidar data is relevant for point cloud analysis. Traditionally, waveform attributes are obtained through fitting the echo waveform with a parametric function. However, it is challenging for some systems to describe the system response using a simple parametric function. This study explores the direct exploitation of multireturn waveform signals for point cloud classification in a built environment. The classification performance of calibrated waveform attributes and deep learning-based FW data classification technique is compared, showing that the latter contains higher information content.
IEEE TRANSACTIONS ON GEOSCIENCE AND REMOTE SENSING
(2023)
Review
Computer Science, Artificial Intelligence
Alok Jhaldiyal, Navendu Chaudhary
Summary: This paper focuses on the importance of LiDAR sensors in real-time decision-making applications and the application of deep learning methods based on point clouds. Among them, projection-based methods are widely used in various applications but have been less studied. This paper examines the recent progress of projection-based methods in detail and summarizes important interventions.
APPLIED INTELLIGENCE
(2023)
Article
Environmental Sciences
Huxiong Li, Weiya Ye, Jun Liu, Weikai Tan, Saied Pirasteh, Sarah Narges Fatholahi, Jonathan Li
Summary: This study introduces a novel workflow for automated Digital Terrain Model (DTM) extraction from Airborne LiDAR point clouds based on a convolutional neural network and transfer learning. The results demonstrate that the proposed workflow establishes a superior DTM extraction accuracy with a root mean square error of only 7.3 cm for the interpolated DTM at 1 m resolution.
Article
Environmental Sciences
Genping Zhao, Weiguang Zhang, Yeping Peng, Heng Wu, Zhuowei Wang, Lianglun Cheng
Summary: This paper introduces an efficient deep neural network for 3D LiDAR point cloud classification, achieving effective feature learning and relationship encoding through Point Expanded Grouping and Spatial Embedding units. Compared to other methods, the accuracy improved by 2% while achieving a 26% increase in efficiency on public datasets.
Article
Geochemistry & Geophysics
Yuanzhi Cai, Lei Fan, Peter M. Atkinson, Cheng Zhang
Summary: This research proposes a novel image enhancement method to reveal the local geometric characteristics of point cloud data in images. The method explores various feature channel combinations and achieves improved semantic segmentation accuracy. Experimental results on the Semantic3D benchmark demonstrate the superiority of this image-based approach.
IEEE TRANSACTIONS ON GEOSCIENCE AND REMOTE SENSING
(2022)
Article
Geochemistry & Geophysics
Likun Chen, Yanfeng Gu, Xian Li, Xiangrong Zhang, Baisen Liu
Summary: An article proposes a normalized spatial-spectral supervoxel segmentation method for multispectral point cloud (MPC) data, which can segment MPC without the need for any manual annotation and achieves better performance compared to other methods, as demonstrated in experiments.
IEEE TRANSACTIONS ON GEOSCIENCE AND REMOTE SENSING
(2023)
Article
Engineering, Multidisciplinary
Maohua Liu, Ziwei Han, Yiming Chen, Zhengjun Liu, Yanshun Han
Summary: Accurate tree species identification is crucial for ecological evaluation and other forest applications. The proposed LayerNet, a point-based deep neural network, shows significant advantages in tree species classification tasks, with the highest classification accuracy reaching 92.5% through 3D data processing and classification.
Article
Environmental Sciences
Eray Sevgen, Saygin Abdikan
Summary: Automatic point cloud classification (PCC) is a challenging task in large-scale urban point clouds. This study utilizes a traditional machine learning framework with neighborhood definition, multi-scale feature extraction, and classification steps. The framework adopts fast feature calculation with multi-scale radius neighborhood and a state-of-the-art GBM classifier, LightGBM. Results show that the framework outperforms traditional machine learning models and competes with DL-based methods.
Article
Environmental Sciences
Ming Wei, Ming Zhu, Yaoyuan Zhang, Jiaqi Sun, Jiarong Wang
Summary: The application of 3D scenes has been expanding in recent years, but the reliability of 3D point clouds acquired using sensors is limited, causing difficulties in their utilization. To address this issue, point cloud completion techniques can reconstruct and restore sparse and incomplete point clouds to enhance their realism. In this study, we propose a cyclic global guiding network structure that considers both local details and overall characteristics of the whole cloud for point cloud completion tasks. We introduce fitting planes and layered folding attention modules based on global guidance to strengthen the local effect. Experimental results demonstrate the effectiveness of our method on diverse datasets and its superiority over other networks.
Article
Computer Science, Software Engineering
Duosheng Yu, Takashi Kanai
Article
Computer Science, Software Engineering
Yuhang Huang, Yonghang Yu, Takashi Kanai
COMPUTER ANIMATION AND VIRTUAL WORLDS
(2019)
Article
Computer Science, Software Engineering
Tianxing Li, Rui Shi, Takashi Kanai
Summary: This paper proposes a graph-learning-based method for automatically generating nonlinear deformation for characters with any number of vertices, which encodes deformed meshes by constructing graphs and designs a multi-resolution graph network for better feature extraction. Experimental results show better performance in deformation approximation for unseen characters and poses.
COMPUTER GRAPHICS FORUM
(2021)