Street tree segmentation from mobile laser scanning data using deep learning-based image instance segmentation

Urban forestry & urban greening, Год журнала: 2024, Номер 92, С. 128200 - 128200

Опубликована: Янв. 20, 2024

Язык: Английский

---

Online segmentation of street trees from mobile laser scanning data via deep learning image instance segmentation

Optics & Laser Technology, Год журнала: 2025, Номер 184, С. 112603 - 112603

Опубликована: Фев. 13, 2025

Язык: Английский

---

YOLO-SegNet: A Method for Individual Street Tree Segmentation Based on the Improved YOLOv8 and the SegFormer Network

Agriculture, Год журнала: 2024, Номер 14(9), С. 1620 - 1620

Опубликована: Сен. 15, 2024

In urban forest management, individual street tree segmentation is a fundamental method to obtain phenotypes, which especially critical. Most existing image models have been evaluated on smaller datasets and lack experimental verification larger, publicly available datasets. Therefore, this paper, based large, dataset, proposes YOLO-SegNet for segmentation. the first stage of object detection task, BiFormer attention mechanism was introduced into YOLOv8 network increase contextual information extraction improve ability detect multiscale multishaped targets. second-stage SegFormer proposed edge more efficiently. The results indicate that our method, combines YOLOv8+BiFormer SegFormer, achieved 92.0% mean intersection over union (mIoU), 95.9% pixel accuracy (mPA), 97.4% dataset. Compared with those fully convolutional neural (FCN), lite-reduced atrous spatial pyramid pooling (LR-ASPP), scene parsing (PSPNet), UNet, DeepLabv3+, HRNet, mIoUs increased by 10.5, 9.7, 5.0, 6.8, 4.5, 2.7 percentage points, respectively. can effectively support smart agroforestry development.

Язык: Английский

---

A method for automatic extraction and individual segmentation of urban street trees from laser point clouds

Optics & Laser Technology, Год журнала: 2024, Номер 180, С. 111431 - 111431

Опубликована: Июль 19, 2024

Язык: Английский

---

Three-Dimensional Quantification and Visualization of Leaf Chlorophyll Content in Poplar Saplings under Drought Using SFM-MVS

Forests, Год журнала: 2023, Номер 15(1), С. 20 - 20

Опубликована: Дек. 20, 2023

As global temperatures warm, drought reduces plant yields and is one of the most serious abiotic stresses causing losses. The early identification great significance for making improvement decisions in advance. Chlorophyll closely related to photosynthesis nutritional status. By tracking changes chlorophyll between strains, we can identify impact on a plant’s physiological status, efficiently adjust ecosystem adaptability, achieve optimization planting management strategies resource utilization efficiency. Plant three-dimensional reconstruction character description are current research hot spots development phenomics, which three-dimensionally reveal structure phenotypes. This article obtains visible light multi-view images four poplar varieties before after drought. Machine learning algorithms were used establish regression models color vegetation indices content. model, based partial least squares (PLSR), reached best performance, with an R2 0.711. SFM-MVS algorithm was reconstruct point cloud perform correction, noise reduction, morphological calibration. trained PLSR prediction model combined information, re-rendered digitization Experimental found that under natural growth conditions, content trees showed gradient distribution state gradually increasing values from top bottom; being given short period mild stress, accumulated. Compared value it has improved, but no longer presents state. At same time, severe decreased as whole, lower leaves began turn yellow, wilt fall off; when stress intensity consistent duration, effect 895 < SY-1 110 3804. provides effective tool in-depth understanding mechanisms responses plants environmental stress. It improving agricultural forestry production protecting ecological environment. also decision-making solving problems caused by climate change.

Язык: Английский

---

Tree Instance Segmentation in Urban 3D Point Clouds Using a Coarse-to-Fine Algorithm Based on Semantic Segmentation

ISPRS annals of the photogrammetry, remote sensing and spatial information sciences, Год журнала: 2024, Номер X-4/W5-2024, С. 79 - 86

Опубликована: Июнь 27, 2024

Abstract. 3D point clouds acquired with terrestrial or mobile LiDAR sensors are increasingly used to map urban forests. The segmentation of separate tree instances, i.e., subsets points representing individual trees, is a relevant step in automatically extracting inventory data from clouds. Various algorithms have been proposed for instance segmentation, offering different trade-offs between accuracy, runtime, and robustness against incompleteness noise. In this work, we propose coarse-to-fine algorithm segmenting instances scanning that combines two existing techniques: (1) the computationally efficient marker-controlled Watershed (2) more accurate region growing algorithm. Initially, generates coarse which further improved by Voronoi segmentation-based error removal. Subsequently, refined areas overlapping crowns sufficient quality. both steps, our uses results prior semantic select suitable markers seed points. We evaluated an ablation study using datasets one dataset three German cities. Our show outperforms standard terms panoptic quality 3.7, 25.5, 29.6 percentage points, respectively, while being than approach purely based on growing.

Язык: Английский

---

Detecting natural gas storage microleakage based on K-means clustering under constraint of Jeffries-Matusita distance criterion using mobile LiDAR data

Journal of Environmental Management, Год журнала: 2024, Номер 370, С. 122539 - 122539

Опубликована: Сен. 21, 2024

Язык: Английский

---

A point cloud segmentation algorithm based on multi-feature training and weighted random forest

Measurement Science and Technology, Год журнала: 2024, Номер 36(1), С. 015407 - 015407

Опубликована: Окт. 16, 2024

Abstract Point cloud segmentation is the process of dividing point data into a series coherent subsets according to its attributes. It has been widely used in target recognition, digital protection cultural relics, medical research and other fields. To improve classification accuracy achieve accurate objects or scenes, algorithm based on multi–features training weighted random forest (RF) proposed. Firstly, feature vector composed 3D coordinate value, RGB echo intensity, density, normal direction average curvature train SVM classifier, ‘one–to–one’ strategy adopted initial multivariate rough cloud. Then, maximum information coefficient sample correlation (SCC) are evaluate decision tree, tree accordingly build weak RF, so as further The experiment verifies effectiveness proposed by segmenting outdoor scene model. results show that RF can segmentation, an effective method.

Язык: Английский

---

Estimation of Single Tree Height Based on Improved K-Means Method for Unmanned Aerial Vehicle Point Cloud Data

IGARSS 2022 - 2022 IEEE International Geoscience and Remote Sensing Symposium, Год журнала: 2024, Номер unknown, С. 8569 - 8572

Опубликована: Июль 7, 2024

Язык: Английский

---

Efficient Semantic Segmentation for Large-Scale Agricultural Nursery Managements via Point Cloud-Based Neural Network

Remote Sensing, Год журнала: 2024, Номер 16(21), С. 4011 - 4011

Опубликована: Окт. 29, 2024

Remote sensing technology has found extensive application in agriculture, providing critical data for analysis. The advancement of semantic segmentation models significantly enhances the utilization point cloud data, offering innovative technical support modern horticulture nursery environments, particularly area plant cultivation. Semantic results aid obtaining tree components, like canopies and trunks, detailed on growth environments. However, precise from large-scale areas can be challenging due to vast number points involved. Therefore, this paper introduces an improved model aimed at achieving superior performance points. incorporates direction angles between improve local feature extraction ensure rotational invariance. It also uses geometric relative distance information better adjustment different neighboring features. An external attention module extracts global spatial features, upsampling strategy integrates features encoder decoder. A specialized dataset was created real environments experiments. Results show that surpasses several point-based models, a Mean Intersection over Union (mIoU) 87.18%. This precision environment analysis supports autonomous managements.

Язык: Английский

---