Multi-Size Voxel Cube (MSVC) Algorithm—A Novel Method for Terrain Filtering from Dense Point Clouds Using a Deep Neural Network

Remote Sensing, Journal Year: 2025, Volume and Issue: 17(4), P. 615 - 615

Published: Feb. 11, 2025

When filtering highly rugged terrain from dense point clouds (particularly in technical applications such as civil engineering), the most widely used approaches yield suboptimal results. Here, we proposed and tested a novel ground-filtering algorithm, multi-size voxel cube (MSVC), utilizing deep neural network. This is based on voxelization of cloud, classification individual voxels ground or non-ground using surrounding (a “voxel cube” 9 × voxels), gradual reduction size, allowing acquisition custom-level detail clouds. The MSVC performance two clouds, capturing areas with vegetation cover, was compared that cloth simulation filter (CSF) manually classified reference. consistently outperformed CSF terms correctly identified points, balanced accuracy, F-score. Another advantage this lay its easy adaptability to any type terrain, enabled by utilization machine learning. only disadvantage necessity prepare training data. On other hand, aim account for future producing networks trained landscape types, thus eliminating phase work.

Language: Английский

Assessment of the Solar Potential of Buildings Based on Photogrammetric Data

Energies, Journal Year: 2025, Volume and Issue: 18(4), P. 868 - 868

Published: Feb. 12, 2025

In recent years, a growing demand for alternative energy sources, including solar energy, has been observed. This article presents methodology assessing the potential of buildings using images from Unmanned Aerial Vehicles (UAVs) and point clouds airborne LIDAR. The proposed method includes following stages: DSM generation, extraction building footprints, determination roof parameters, map removing areas that are not suitable installation systems, calculation power per each building, conversion irradiance into mapping generation. paper describes also Detecting Photovoltaic Panels algorithm with use deep learning techniques. enabled efficiency photovoltaic panels comparing results maps buildings, as well identifying require optimization. analysis, which had conducted in test village on campus university, confirmed usefulness above methods. analysis provides UAV image data enable generation higher accuracy (MAE = 8.5 MWh) than LIDAR 10.5 MWh).

Language: Английский

Combining LiDAR, SAR, and DEM Data for Estimating Understory Terrain Using Machine Learning-Based Methods

Forests, Journal Year: 2024, Volume and Issue: 15(11), P. 1992 - 1992

Published: Nov. 11, 2024

Currently, precise estimation of understory terrain faces numerous technical obstacles and challenges that are difficult to overcome. To address this problem, paper combines LiDAR, SAR, DEM data estimate terrain. The high multivariable-precision spaceborne LiDAR ICESat-2 data, validated by the NEON, divided into training validation sets. dataset is used as a dependent variable, SRTM Sentinel-1 SAR regarded independent variables, total 13 feature parameters with contributions extracted construct Multiple Linear Regression model (MLR), BAGGING model, Random Forest (RF), Long Short-Term Memory (LSTM). results indicate RF exhibits highest accuracy among four models, R2 = 0.999, RMSE 0.701 m, MAE 0.249 m. Then, based on at regional scale generated, an assessment performed using dataset, yielding 0.847 0.517 Furthermore, quantitatively analyzes effects slope, vegetation coverage, canopy height show increase, for gradually decreases. estimated relatively stable not easily affected height. research holds significant practical implications forest resource management, ecological conservation, biodiversity protection, well natural disaster prevention.

Language: Английский