Pub Date : 2025-08-01Epub Date: 2025-09-01DOI: 10.1016/j.ophoto.2025.100100
Miriam Jäger, Markus Hillemann, Boris Jutzi
3D Gaussian Splatting (3DGS) has emerged as a powerful approach for 3D scene reconstruction using 3D Gaussians. However, neither the centers nor surfaces of the Gaussians are accurately aligned to the object surface, complicating their direct use in point cloud and mesh reconstruction. Additionally, 3DGS typically produces floater artifacts, increasing the number of Gaussians and storage requirements. To address these issues, we present FeatureGS, which incorporates an additional geometric loss term based on an eigenvalue-derived 3D shape feature into the optimization process of 3DGS. The goal is to improve geometric accuracy and enhance properties of planar surfaces with reduced structural entropy in local 3D neighborhoods, typically given in man-made environments. We present four alternative formulations for the geometric loss term based on ‘planarity’ of Gaussians, as well as ‘planarity’, ‘omnivariance’, and ‘eigenentropy’ of Gaussian neighborhoods. On the small-scale DTU benchmark with man-made scenes, FeatureGS achieves a 20% improvement in geometric accuracy, suppresses floater artifacts by 90%, and reduces the number of Gaussians by 95%. FeatureGS proves to be a strong method for geometrically accurate, artifact-reduced and memory-efficient 3D scene reconstruction, enabling the direct use of Gaussian centers for geometric representation.
{"title":"FeatureGS: Eigenvalue-feature optimization in 3D Gaussian Splatting for geometrically accurate and artifact-reduced reconstruction","authors":"Miriam Jäger, Markus Hillemann, Boris Jutzi","doi":"10.1016/j.ophoto.2025.100100","DOIUrl":"10.1016/j.ophoto.2025.100100","url":null,"abstract":"<div><div>3D Gaussian Splatting (3DGS) has emerged as a powerful approach for 3D scene reconstruction using 3D Gaussians. However, neither the centers nor surfaces of the Gaussians are accurately aligned to the object surface, complicating their direct use in point cloud and mesh reconstruction. Additionally, 3DGS typically produces floater artifacts, increasing the number of Gaussians and storage requirements. To address these issues, we present FeatureGS, which incorporates an additional geometric loss term based on an eigenvalue-derived 3D shape feature into the optimization process of 3DGS. The goal is to improve geometric accuracy and enhance properties of planar surfaces with reduced structural entropy in local 3D neighborhoods, typically given in man-made environments. We present four alternative formulations for the geometric loss term based on ‘planarity’ of Gaussians, as well as ‘planarity’, ‘omnivariance’, and ‘eigenentropy’ of Gaussian neighborhoods. On the small-scale DTU benchmark with man-made scenes, FeatureGS achieves a 20% improvement in geometric accuracy, suppresses floater artifacts by 90%, and reduces the number of Gaussians by 95%. FeatureGS proves to be a strong method for geometrically accurate, artifact-reduced and memory-efficient 3D scene reconstruction, enabling the direct use of Gaussian centers for geometric representation.</div></div>","PeriodicalId":100730,"journal":{"name":"ISPRS Open Journal of Photogrammetry and Remote Sensing","volume":"17 ","pages":"Article 100100"},"PeriodicalIF":0.0,"publicationDate":"2025-08-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"144988574","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
Pub Date : 2025-04-01Epub Date: 2025-04-16DOI: 10.1016/j.ophoto.2025.100089
Ivana Petrovska, Boris Jutzi
Image-based 3D reconstruction offers realistic scene representation for applications that require accurate geometric information. Although the assumption that images are simultaneously captured, perfectly posed and noise-free simplifies the 3D reconstruction, this rarely holds in real-world settings. A real-world scene comprises multiple objects which obstruct each other and certain object parts are occluded, thus it can be challenging to generate a complete and accurate geometry. Being a part of our environment, we are particularly interested in vegetation that often obscures important structures, leading to incomplete reconstruction of the underlying features. In this contribution, we present a comparative analysis of the geometry behind vegetation occlusions reconstructed by traditional Multi-View Stereo (MVS) and radiance field methods, namely: Neural Radiance Fields (NeRFs), 3D Gaussian Splatting (3DGS) and 2D Gaussian Splatting (2DGS). Excluding certain image parts and investigating how different level of vegetation occlusions affect the geometric reconstruction, we consider Synthetic masks with different occlusion coverage of 10% (Very Sparse), 30% (Sparse), 50% (Medium), 70% (Dense) and 90% (Very Dense). To additionally demonstrate the impact of spatially consistent 3D occlusions, we use Natural masks (up to 35%) where the vegetation is stationary in the 3D scene, but relative to the view-point. Our investigations are based on real-world scenarios; one occlusion-free indoor scenario, on which we apply the Synthetic masks and one outdoor scenario, from which we derive the Natural masks. The qualitative and quantitative 3D evaluation is based on point cloud comparison against a ground truth mesh addressing accuracy and completeness. The conducted experiments and results demonstrate that although MVS shows lowest accuracy errors in both scenarios, the completeness manifests a sharp decline as the occlusion percentage increases, eventually failing under Very Dense masks. NeRFs manifest robustness in the reconstruction with highest completeness considering masks, although the accuracy proportionally decreases with increasing the occlusions. 2DGS achieves second best accuracy results outperforming NeRFs and 3DGS, indicating a consistent performance across different occlusion scenarios. Additionally, by using MVS for initialization, 3DGS and 2DGS completeness improves without significantly sacrificing the accuracy, due to the more densely reconstructed homogeneous areas. We demonstrate that radiance field methods can compete against traditional MVS, showing robust performance for a complete reconstruction under vegetation occlusions.
{"title":"Seeing beyond vegetation: A comparative occlusion analysis between Multi-View Stereo, Neural Radiance Fields and Gaussian Splatting for 3D reconstruction","authors":"Ivana Petrovska, Boris Jutzi","doi":"10.1016/j.ophoto.2025.100089","DOIUrl":"10.1016/j.ophoto.2025.100089","url":null,"abstract":"<div><div>Image-based 3D reconstruction offers realistic scene representation for applications that require accurate geometric information. Although the assumption that images are simultaneously captured, perfectly posed and noise-free simplifies the 3D reconstruction, this rarely holds in real-world settings. A real-world scene comprises multiple objects which obstruct each other and certain object parts are occluded, thus it can be challenging to generate a complete and accurate geometry. Being a part of our environment, we are particularly interested in vegetation that often obscures important structures, leading to incomplete reconstruction of the underlying features. In this contribution, we present a comparative analysis of the geometry behind vegetation occlusions reconstructed by traditional Multi-View Stereo (MVS) and radiance field methods, namely: Neural Radiance Fields (NeRFs), 3D Gaussian Splatting (3DGS) and 2D Gaussian Splatting (2DGS). Excluding certain image parts and investigating how different level of vegetation occlusions affect the geometric reconstruction, we consider Synthetic masks with different occlusion coverage of 10% (Very Sparse), 30% (Sparse), 50% (Medium), 70% (Dense) and 90% (Very Dense). To additionally demonstrate the impact of spatially consistent 3D occlusions, we use Natural masks (up to 35%) where the vegetation is stationary in the 3D scene, but relative to the view-point. Our investigations are based on real-world scenarios; one occlusion-free indoor scenario, on which we apply the Synthetic masks and one outdoor scenario, from which we derive the Natural masks. The qualitative and quantitative 3D evaluation is based on point cloud comparison against a ground truth mesh addressing accuracy and completeness. The conducted experiments and results demonstrate that although MVS shows lowest accuracy errors in both scenarios, the completeness manifests a sharp decline as the occlusion percentage increases, eventually failing under Very Dense masks. NeRFs manifest robustness in the reconstruction with highest completeness considering masks, although the accuracy proportionally decreases with increasing the occlusions. 2DGS achieves second best accuracy results outperforming NeRFs and 3DGS, indicating a consistent performance across different occlusion scenarios. Additionally, by using MVS for initialization, 3DGS and 2DGS completeness improves without significantly sacrificing the accuracy, due to the more densely reconstructed homogeneous areas. We demonstrate that radiance field methods can compete against traditional MVS, showing robust performance for a complete reconstruction under vegetation occlusions.</div></div>","PeriodicalId":100730,"journal":{"name":"ISPRS Open Journal of Photogrammetry and Remote Sensing","volume":"16 ","pages":"Article 100089"},"PeriodicalIF":0.0,"publicationDate":"2025-04-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"144107473","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
Pub Date : 2025-04-01Epub Date: 2025-04-07DOI: 10.1016/j.ophoto.2025.100088
Eugeniu Vezeteu , Aimad El Issaoui , Heikki Hyyti , Teemu Hakala , Jesse Muhojoki , Eric Hyyppä , Antero Kukko , Harri Kaartinen , Ville Kyrki , Juha Hyyppä
This paper presents a novel real-time fusion pipeline for integrating georeferenced airborne laser scanning (ALS) and online mobile laser scanning (MLS) data to enable accurate localization and mapping in complex natural environments. To address sensor drift caused by relative Light Detection and Ranging (lidar) and inertial measurements, occlusion affecting the Global Navigation Satellite System (GNSS) signal quality, and differences in the fields of view of the sensors, we propose a tightly coupled lidar-inertial registration system with an adaptive, robust Iterated Error-State Extended Kalman Filter (RIEKF). By leveraging ALS-derived prior maps as a global reference, our system effectively refines the MLS registration, even in challenging environments like forests. A novel coarse-to-fine initialization technique is introduced to estimate the initial transformation between the local MLS and global ALS frames using online GNSS measurements. Experimental results in forest environments demonstrate significant improvements in both absolute and relative trajectory accuracy, with relative mean localization errors as low as 0.17 m for a prior map based on dense ALS data and 0.22 m for a prior map based on sparse ALS data. We found that while GNSS does not significantly improve registration accuracy, it is essential for providing the initial transformation between the ALS and MLS frames, enabling their direct and online fusion. The proposed system predicts poses at an inertial measurement unit (IMU) rate of 400 Hz and updates the pose at the lidar frame rate of 10 Hz.
{"title":"Direct integration of ALS and MLS for real-time localization and mapping","authors":"Eugeniu Vezeteu , Aimad El Issaoui , Heikki Hyyti , Teemu Hakala , Jesse Muhojoki , Eric Hyyppä , Antero Kukko , Harri Kaartinen , Ville Kyrki , Juha Hyyppä","doi":"10.1016/j.ophoto.2025.100088","DOIUrl":"10.1016/j.ophoto.2025.100088","url":null,"abstract":"<div><div>This paper presents a novel real-time fusion pipeline for integrating georeferenced airborne laser scanning (ALS) and online mobile laser scanning (MLS) data to enable accurate localization and mapping in complex natural environments. To address sensor drift caused by relative Light Detection and Ranging (lidar) and inertial measurements, occlusion affecting the Global Navigation Satellite System (GNSS) signal quality, and differences in the fields of view of the sensors, we propose a tightly coupled lidar-inertial registration system with an adaptive, robust Iterated Error-State Extended Kalman Filter (RIEKF). By leveraging ALS-derived prior maps as a global reference, our system effectively refines the MLS registration, even in challenging environments like forests. A novel coarse-to-fine initialization technique is introduced to estimate the initial transformation between the local MLS and global ALS frames using online GNSS measurements. Experimental results in forest environments demonstrate significant improvements in both absolute and relative trajectory accuracy, with relative mean localization errors as low as 0.17 m for a prior map based on dense ALS data and 0.22 m for a prior map based on sparse ALS data. We found that while GNSS does not significantly improve registration accuracy, it is essential for providing the initial transformation between the ALS and MLS frames, enabling their direct and online fusion. The proposed system predicts poses at an inertial measurement unit (IMU) rate of 400 Hz and updates the pose at the lidar frame rate of 10 Hz.</div></div>","PeriodicalId":100730,"journal":{"name":"ISPRS Open Journal of Photogrammetry and Remote Sensing","volume":"16 ","pages":"Article 100088"},"PeriodicalIF":0.0,"publicationDate":"2025-04-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"143816764","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
This study addresses the challenge of oil spill detection using Synthetic Aperture Radar (SAR) satellite imagery, employing deep learning techniques to improve accuracy and efficiency. We investigated the effectiveness of various neural network architectures and encoders for this task, focusing on scenarios with limited training data. The research problem centered on enhancing feature extraction from single-channel SAR data to improve oil spill detection performance.
Our methodology involved developing a novel preprocessing pipeline that converts single-channel SAR data into a three-channel RGB representation. The preprocessing technique normalizes SAR intensity values and encodes extracted features into RGB channels.
Through an experiment, we have shown that a combination of the LinkNet with an EfficientNet-B4 is superior to pairs of other well-known architectures and encoders.
Quantitative evaluation revealed a significant improvement in F1-score of 0.064 compared to traditional dB-scale preprocessing methods. Qualitative assessment on independent SAR scenes from the Mediterranean Sea demonstrated better detection capabilities, albeit with increased sensitivity to look-alike.
We conclude that our proposed preprocessing technique shows promise for enhancing automatic oil spill segmentation from SAR imagery. The study contributes to advancing oil spill detection methods, with potential implications for environmental monitoring and marine ecosystem protection.
{"title":"Transfer learning and single-polarized SAR image preprocessing for oil spill detection","authors":"Nataliia Kussul , Yevhenii Salii , Volodymyr Kuzin , Bohdan Yailymov , Andrii Shelestov","doi":"10.1016/j.ophoto.2024.100081","DOIUrl":"10.1016/j.ophoto.2024.100081","url":null,"abstract":"<div><div>This study addresses the challenge of oil spill detection using Synthetic Aperture Radar (SAR) satellite imagery, employing deep learning techniques to improve accuracy and efficiency. We investigated the effectiveness of various neural network architectures and encoders for this task, focusing on scenarios with limited training data. The research problem centered on enhancing feature extraction from single-channel SAR data to improve oil spill detection performance.</div><div>Our methodology involved developing a novel preprocessing pipeline that converts single-channel SAR data into a three-channel RGB representation. The preprocessing technique normalizes SAR intensity values and encodes extracted features into RGB channels.</div><div>Through an experiment, we have shown that a combination of the LinkNet with an EfficientNet-B4 is superior to pairs of other well-known architectures and encoders.</div><div>Quantitative evaluation revealed a significant improvement in F1-score of 0.064 compared to traditional dB-scale preprocessing methods. Qualitative assessment on independent SAR scenes from the Mediterranean Sea demonstrated better detection capabilities, albeit with increased sensitivity to look-alike.</div><div>We conclude that our proposed preprocessing technique shows promise for enhancing automatic oil spill segmentation from SAR imagery. The study contributes to advancing oil spill detection methods, with potential implications for environmental monitoring and marine ecosystem protection.</div></div>","PeriodicalId":100730,"journal":{"name":"ISPRS Open Journal of Photogrammetry and Remote Sensing","volume":"15 ","pages":"Article 100081"},"PeriodicalIF":0.0,"publicationDate":"2025-01-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"143137194","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"OA","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
Pub Date : 2025-01-01Epub Date: 2025-01-15DOI: 10.1016/j.ophoto.2025.100083
Zhouxin Xi, Dani Degenhardt
Accurately defining and isolating 3D tree space is critical for extracting and analyzing tree inventory attributes, yet it remains a challenge due to the structural complexity and heterogeneity within natural forests. This study introduces TreeisoNet, a suite of supervised deep neural networks tailored for robust 3D tree segmentation across natural forest environments. These networks are specifically designed to identify tree locations, stem components (if available), and crown clusters, making them adaptable to varying scales of laser scanning from airborne laser scannner (ALS), terrestrial laser scanner (TLS), and unmanned aerial vehicle (UAV). Our evaluation used three benchmark datasets with manually isolated tree references, achieving mean intersection-over-union (mIoU) accuracies of 0.81 for UAV, 0.76 for TLS, and 0.59 for ALS, which are competitive with contemporary algorithms such as ForAINet, Treeiso, Mask R-CNN, and AMS3D. Noise from stem point delineation minimally impacted stem location detection but significantly affected crown clustering. Moderate manual refinement of stem points or tree centers significantly improved tree segmentation accuracies, achieving 0.85 for UAV, 0.86 for TLS, and 0.80 for ALS. The study confirms SegFormer as an effective 3D point-level classifier and an offset-based UNet as a superior segmenter, with the latter outperforming unsupervised solutions like watershed and shortest-path methods. TreeisoNet demonstrates strong adaptability in capturing invariant tree geometry features, ensuring transferability across different resolutions, sites, and sensors with minimal accuracy loss.
{"title":"A new unified framework for supervised 3D crown segmentation (TreeisoNet) using deep neural networks across airborne, UAV-borne, and terrestrial laser scans","authors":"Zhouxin Xi, Dani Degenhardt","doi":"10.1016/j.ophoto.2025.100083","DOIUrl":"10.1016/j.ophoto.2025.100083","url":null,"abstract":"<div><div>Accurately defining and isolating 3D tree space is critical for extracting and analyzing tree inventory attributes, yet it remains a challenge due to the structural complexity and heterogeneity within natural forests. This study introduces TreeisoNet, a suite of supervised deep neural networks tailored for robust 3D tree segmentation across natural forest environments. These networks are specifically designed to identify tree locations, stem components (if available), and crown clusters, making them adaptable to varying scales of laser scanning from airborne laser scannner (ALS), terrestrial laser scanner (TLS), and unmanned aerial vehicle (UAV). Our evaluation used three benchmark datasets with manually isolated tree references, achieving mean intersection-over-union (mIoU) accuracies of 0.81 for UAV, 0.76 for TLS, and 0.59 for ALS, which are competitive with contemporary algorithms such as ForAINet, Treeiso, Mask R-CNN, and AMS3D. Noise from stem point delineation minimally impacted stem location detection but significantly affected crown clustering. Moderate manual refinement of stem points or tree centers significantly improved tree segmentation accuracies, achieving 0.85 for UAV, 0.86 for TLS, and 0.80 for ALS. The study confirms SegFormer as an effective 3D point-level classifier and an offset-based UNet as a superior segmenter, with the latter outperforming unsupervised solutions like watershed and shortest-path methods. TreeisoNet demonstrates strong adaptability in capturing invariant tree geometry features, ensuring transferability across different resolutions, sites, and sensors with minimal accuracy loss.</div></div>","PeriodicalId":100730,"journal":{"name":"ISPRS Open Journal of Photogrammetry and Remote Sensing","volume":"15 ","pages":"Article 100083"},"PeriodicalIF":0.0,"publicationDate":"2025-01-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"143137151","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"OA","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
Pub Date : 2025-01-01Epub Date: 2025-01-14DOI: 10.1016/j.ophoto.2025.100082
David Pedley, Justin Morgenroth
Urban trees provide a multitude of environmental and amenity benefits for city occupants yet face ongoing risk of removal due to urban pressures and the preferences of landowners. Understanding the extent and location of canopy loss is critical for the effective management of urban forests. Although city-scale assessments of urban forest canopy cover are common, the accurate identification of fine-scale canopy loss remains challenging. Evaluating change at the property scale is of particular importance given the localised benefits of urban trees and the scale at which tree removal decisions are made.
The objective of this study was to develop a method to accurately detect and quantify the city-wide loss of urban tree canopy (UTC) at the scale of individual properties using publicly available remote sensing data. The study area was the city of Christchurch, New Zealand, with the study focussed on UTC loss that occurred between 2016 and 2021. To accurately delineate the 2016 UTC, a semantic segmentation deep learning model (DeepLabv3+) was pretrained using existing UTC data and fine-tuned using high resolution aerial imagery. The output of this model was then segmented into polygons representing individual trees using the Segment Anything Model. To overcome poor alignment of aerial imagery, LiDAR point cloud data was utilised to identify changes in height between 2016 and 2021, which was overlaid across the 2016 UTC to map areas of UTC loss. The accuracy of UTC loss predictions was validated using a visual comparison of aerial imagery and LiDAR data, with UTC loss quantified for each property within the study area.
The loss detection method achieved accurate results for the property-scale identification of UTC loss, including a mean F1 score of 0.934 and a mean IOU of 0.883. Precision values were higher than recall values (0.941 compared to 0.811), which reflected a deliberately conservative approach to avoid false positive detections. Approximately 14.5% of 2016 UTC was lost by 2021, with 74.9% of the UTC loss occurring on residential land. This research provides a novel geospatial method for evaluating fine-scale city-wide tree dynamics using remote sensing data of varying type and quality with imperfect alignment. This creates the opportunity for detailed evaluation of the drivers of UTC loss on individual properties to enable better management of existing urban forests.
{"title":"Detecting and measuring fine-scale urban tree canopy loss with deep learning and remote sensing","authors":"David Pedley, Justin Morgenroth","doi":"10.1016/j.ophoto.2025.100082","DOIUrl":"10.1016/j.ophoto.2025.100082","url":null,"abstract":"<div><div>Urban trees provide a multitude of environmental and amenity benefits for city occupants yet face ongoing risk of removal due to urban pressures and the preferences of landowners. Understanding the extent and location of canopy loss is critical for the effective management of urban forests. Although city-scale assessments of urban forest canopy cover are common, the accurate identification of fine-scale canopy loss remains challenging. Evaluating change at the property scale is of particular importance given the localised benefits of urban trees and the scale at which tree removal decisions are made.</div><div>The objective of this study was to develop a method to accurately detect and quantify the city-wide loss of urban tree canopy (UTC) at the scale of individual properties using publicly available remote sensing data. The study area was the city of Christchurch, New Zealand, with the study focussed on UTC loss that occurred between 2016 and 2021. To accurately delineate the 2016 UTC, a semantic segmentation deep learning model (DeepLabv3+) was pretrained using existing UTC data and fine-tuned using high resolution aerial imagery. The output of this model was then segmented into polygons representing individual trees using the Segment Anything Model. To overcome poor alignment of aerial imagery, LiDAR point cloud data was utilised to identify changes in height between 2016 and 2021, which was overlaid across the 2016 UTC to map areas of UTC loss. The accuracy of UTC loss predictions was validated using a visual comparison of aerial imagery and LiDAR data, with UTC loss quantified for each property within the study area.</div><div>The loss detection method achieved accurate results for the property-scale identification of UTC loss, including a mean F1 score of 0.934 and a mean IOU of 0.883. Precision values were higher than recall values (0.941 compared to 0.811), which reflected a deliberately conservative approach to avoid false positive detections. Approximately 14.5% of 2016 UTC was lost by 2021, with 74.9% of the UTC loss occurring on residential land. This research provides a novel geospatial method for evaluating fine-scale city-wide tree dynamics using remote sensing data of varying type and quality with imperfect alignment. This creates the opportunity for detailed evaluation of the drivers of UTC loss on individual properties to enable better management of existing urban forests.</div></div>","PeriodicalId":100730,"journal":{"name":"ISPRS Open Journal of Photogrammetry and Remote Sensing","volume":"15 ","pages":"Article 100082"},"PeriodicalIF":0.0,"publicationDate":"2025-01-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"143137195","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"OA","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
Pub Date : 2025-01-01Epub Date: 2025-03-05DOI: 10.1016/j.ophoto.2025.100087
Zuoya Liu , Harri Kaartinen , Teemu Hakala , Heikki Hyyti , Juha Hyyppä , Antero Kukko , Ruizhi Chen
Accurate individual tree locations enable efficient forest inventory management and automation, support precise forest surveys, management decisions and future individual-tree harvesting plans. In this paper, we compared and analyzed in detail the performance of an ultra-wideband (UWB) data-driven method for mapping individual tree locations in boreal forest sample plots of varying complexity. Twelve forest sample plots selected from varying forest-stand conditions representing different developing stages, stem densities and abundance of sub canopy growth in boreal forests were tested. These plots were classified into three categories (“Easy”, “Medium” and “Difficult”) according to these varying stand conditions. The experimental results show that UWB data-driven method is able to map individual tree locations accurately with total root-mean-squared-errors (RMSEs) of 0.17 m, 0.2 m, and 0.26 m for “Easy”, “Medium” and “Difficult” forest plots, respectively, providing a strong reference for forest surveys.
{"title":"Performance analysis of ultra-wideband positioning for measuring tree positions in boreal forest plots","authors":"Zuoya Liu , Harri Kaartinen , Teemu Hakala , Heikki Hyyti , Juha Hyyppä , Antero Kukko , Ruizhi Chen","doi":"10.1016/j.ophoto.2025.100087","DOIUrl":"10.1016/j.ophoto.2025.100087","url":null,"abstract":"<div><div>Accurate individual tree locations enable efficient forest inventory management and automation, support precise forest surveys, management decisions and future individual-tree harvesting plans. In this paper, we compared and analyzed in detail the performance of an ultra-wideband (UWB) data-driven method for mapping individual tree locations in boreal forest sample plots of varying complexity. Twelve forest sample plots selected from varying forest-stand conditions representing different developing stages, stem densities and abundance of sub canopy growth in boreal forests were tested. These plots were classified into three categories (“Easy”, “Medium” and “Difficult”) according to these varying stand conditions. The experimental results show that UWB data-driven method is able to map individual tree locations accurately with total root-mean-squared-errors (RMSEs) of 0.17 m, 0.2 m, and 0.26 m for “Easy”, “Medium” and “Difficult” forest plots, respectively, providing a strong reference for forest surveys.</div></div>","PeriodicalId":100730,"journal":{"name":"ISPRS Open Journal of Photogrammetry and Remote Sensing","volume":"15 ","pages":"Article 100087"},"PeriodicalIF":0.0,"publicationDate":"2025-01-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"143580158","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"OA","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
In recent years, transformer-based deep learning networks have gained popularity in Hyperspectral (HS) unmixing applications due to their superior performance. Most of these networks use an Endmember Extraction Algorithm(EEA) for the initialization of their network. As EEAs performance depends on the environment, single initialization does not ensure optimum performance. Also, only a few networks utilize the spatial context in HS Images to solve the unmixing problem. In this paper, we propose Hyperspectral Unmixing with Spatial Context and Endmember Ensemble Learning with Attention Mechanism (SCEELA) to address these issues. The proposed method has three main components, Signature Predictor (SP), Pixel Contextualizer (PC) and Abundance Predictor (AP). SP uses an ensemble of EEAs for each endmember as the initialization and the attention mechanism within the transformer enables ensemble learning to predict accurate endmembers. The attention mechanism in the PC enables the network to capture the contextual data and provide a more refined pixel to the AP to predict the abundance of that pixel. SCEELA was compared with eight state-of-the-art HS unmixing algorithms for three widely used real datasets and one synthetic dataset. The results show that the proposed method shows impressive performance when compared with other state-of-the-art algorithms.
近年来，基于变压器的深度学习网络由于其优越的性能在高光谱（HS）解混应用中得到了广泛的应用。这些网络大多使用端点提取算法（end - member Extraction Algorithm， EEA）来初始化它们的网络。由于EEAs的性能取决于环境，因此单个初始化并不能确保最佳性能。此外，只有少数网络利用HS图像中的空间上下文来解决解混问题。本文提出了基于空间上下文的高光谱解混和基于注意机制的端元集成学习（SCEELA）来解决这些问题。该方法由三个主要部分组成：特征预测器（SP）、像素上下文预测器（PC）和丰度预测器（AP）。SP对每个端成员使用eea集合作为初始化，转换器内的注意机制使集成学习能够预测准确的端成员。PC中的注意力机制使网络能够捕获上下文数据，并向AP提供更精细的像素，以预测该像素的丰度。在3个广泛使用的真实数据集和1个合成数据集上，对SCEELA与8种最先进的HS解混算法进行了比较。结果表明，与其他先进算法相比，该方法具有令人印象深刻的性能。
{"title":"Hyperspectral unmixing with spatial context and endmember ensemble learning with attention mechanism","authors":"R.M.K.L. Ratnayake, D.M.U.P. Sumanasekara, H.M.K.D. Wickramathilaka, G.M.R.I. Godaliyadda, H.M.V.R. Herath, M.P.B. Ekanayake","doi":"10.1016/j.ophoto.2025.100086","DOIUrl":"10.1016/j.ophoto.2025.100086","url":null,"abstract":"<div><div>In recent years, transformer-based deep learning networks have gained popularity in Hyperspectral (HS) unmixing applications due to their superior performance. Most of these networks use an Endmember Extraction Algorithm(EEA) for the initialization of their network. As EEAs performance depends on the environment, single initialization does not ensure optimum performance. Also, only a few networks utilize the spatial context in HS Images to solve the unmixing problem. In this paper, we propose Hyperspectral Unmixing with Spatial Context and Endmember Ensemble Learning with Attention Mechanism (SCEELA) to address these issues. The proposed method has three main components, Signature Predictor (SP), Pixel Contextualizer (PC) and Abundance Predictor (AP). SP uses an ensemble of EEAs for each endmember as the initialization and the attention mechanism within the transformer enables ensemble learning to predict accurate endmembers. The attention mechanism in the PC enables the network to capture the contextual data and provide a more refined pixel to the AP to predict the abundance of that pixel. SCEELA was compared with eight state-of-the-art HS unmixing algorithms for three widely used real datasets and one synthetic dataset. The results show that the proposed method shows impressive performance when compared with other state-of-the-art algorithms.</div></div>","PeriodicalId":100730,"journal":{"name":"ISPRS Open Journal of Photogrammetry and Remote Sensing","volume":"15 ","pages":"Article 100086"},"PeriodicalIF":0.0,"publicationDate":"2025-01-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"143420353","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"OA","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
Pub Date : 2025-01-01Epub Date: 2024-12-09DOI: 10.1016/j.ophoto.2024.100080
Daniel Mederer , Hannes Feilhauer , Eya Cherif , Katja Berger , Tobias B. Hank , Kyle R. Kovach , Phuong D. Dao , Bing Lu , Philip A. Townsend , Teja Kattenborn
Plant traits play a pivotal role in steering ecosystem dynamics. As plant canopies have evolved to interact with light, spectral data convey information on a variety of plant traits. Machine learning techniques have been used successfully to retrieve diverse traits from hyperspectral data. Nonetheless, the efficacy of machine learning is restricted by limited access to high-quality reference data for training. Previous studies showed that aggregating data across domains, sensors, or growth forms provided by collaborative efforts of the scientific community enables the creation of transferable models. However, even such curated databases are still sparse for several traits. To address these challenges, we investigated the potential of filling such data gaps with simulated hyperspectral data generated through the most widely-used radiative transfer model (RTM) PROSAIL. We coupled trait information from the TRY plant trait database with information on plant communities from the sPlot database, to build a realistic input trait dataset for the RTM-based simulation of canopy spectra. Our findings indicate that simulated data can alleviate the effects of data scarcity for highly underrepresented traits. In most other cases, however, the effects of including simulated data from RTMs are negligible or even negative. While more complex RTM models promise further improvements, their parameterization remains challenging. This highlights two key observations: firstly, RTM models, such as PROSAIL, exhibit limitations in producing realistic spectra across diverse ecosystems; secondly, real-world data repurposed from various sources exhibit superior retrieval success compared to simulated data. As a result, we advocate to emphasize the importance of active data sharing over secrecy and overreliance on modeling to address data limitations.
{"title":"Plant trait retrieval from hyperspectral data: Collective efforts in scientific data curation outperform simulated data derived from the PROSAIL model","authors":"Daniel Mederer , Hannes Feilhauer , Eya Cherif , Katja Berger , Tobias B. Hank , Kyle R. Kovach , Phuong D. Dao , Bing Lu , Philip A. Townsend , Teja Kattenborn","doi":"10.1016/j.ophoto.2024.100080","DOIUrl":"10.1016/j.ophoto.2024.100080","url":null,"abstract":"<div><div>Plant traits play a pivotal role in steering ecosystem dynamics. As plant canopies have evolved to interact with light, spectral data convey information on a variety of plant traits. Machine learning techniques have been used successfully to retrieve diverse traits from hyperspectral data. Nonetheless, the efficacy of machine learning is restricted by limited access to high-quality reference data for training. Previous studies showed that aggregating data across domains, sensors, or growth forms provided by collaborative efforts of the scientific community enables the creation of transferable models. However, even such curated databases are still sparse for several traits. To address these challenges, we investigated the potential of filling such data gaps with simulated hyperspectral data generated through the most widely-used radiative transfer model (RTM) PROSAIL. We coupled trait information from the TRY plant trait database with information on plant communities from the sPlot database, to build a realistic input trait dataset for the RTM-based simulation of canopy spectra. Our findings indicate that simulated data can alleviate the effects of data scarcity for highly underrepresented traits. In most other cases, however, the effects of including simulated data from RTMs are negligible or even negative. While more complex RTM models promise further improvements, their parameterization remains challenging. This highlights two key observations: firstly, RTM models, such as PROSAIL, exhibit limitations in producing realistic spectra across diverse ecosystems; secondly, real-world data repurposed from various sources exhibit superior retrieval success compared to simulated data. As a result, we advocate to emphasize the importance of active data sharing over secrecy and overreliance on modeling to address data limitations.</div></div>","PeriodicalId":100730,"journal":{"name":"ISPRS Open Journal of Photogrammetry and Remote Sensing","volume":"15 ","pages":"Article 100080"},"PeriodicalIF":0.0,"publicationDate":"2025-01-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"143137153","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"OA","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
Pub Date : 2025-01-01Epub Date: 2024-12-11DOI: 10.1016/j.ophoto.2024.100079
Florian Schill , Christoph Holst , Daniel Wujanz , Jens Hartmann , Jens-André Paffenholz
After more than twenty years of commercial use, laser scanners have reached technical maturity and consequently became a standard tool for 3D-data acquisition across various fields of application. Yet, meaningful stochastic information regarding the achieved metric quality of recorded points remains an open research question. Recent research demonstrated that raw intensity values can be deployed to derive stochastic models for reflectorless rangefinders. Yet, all existing studies focused on single instances of particular laser scanners while the derivation of the stochastic models required significant efforts.
Motivated by the aforementioned shortcomings, the focus of this study is set on the comparison of stochastic models for a series of eight identical phase-based scanners that differ in age, working hours and date of last calibration. In order to achieve this, a standardised methodological workflow is suggested to derive the unknown parameters of the individual stochastic models. Based on the generated outcome, a comparison is conducted which clarifies if a universally applicable stochastic model (type calibration) can be used for a particular scanner model or if individual parameter sets are still required for every scanner (instance calibration) to validate the practical benefit and usability of those models. The generated results successfully demonstrate that the computed stochastic model is transferable to all individual scanners of the series.
{"title":"Intensity-based stochastic model of terrestrial laser scanners: Methodological workflow, empirical derivation and practical benefit","authors":"Florian Schill , Christoph Holst , Daniel Wujanz , Jens Hartmann , Jens-André Paffenholz","doi":"10.1016/j.ophoto.2024.100079","DOIUrl":"10.1016/j.ophoto.2024.100079","url":null,"abstract":"<div><div>After more than twenty years of commercial use, laser scanners have reached technical maturity and consequently became a standard tool for 3D-data acquisition across various fields of application. Yet, meaningful stochastic information regarding the achieved metric quality of recorded points remains an open research question. Recent research demonstrated that raw intensity values can be deployed to derive stochastic models for reflectorless rangefinders. Yet, all existing studies focused on single instances of particular laser scanners while the derivation of the stochastic models required significant efforts.</div><div>Motivated by the aforementioned shortcomings, the focus of this study is set on the comparison of stochastic models for a series of eight identical phase-based scanners that differ in age, working hours and date of last calibration. In order to achieve this, a standardised methodological workflow is suggested to derive the unknown parameters of the individual stochastic models. Based on the generated outcome, a comparison is conducted which clarifies if a universally applicable stochastic model (type calibration) can be used for a particular scanner model or if individual parameter sets are still required for every scanner (instance calibration) to validate the practical benefit and usability of those models. The generated results successfully demonstrate that the computed stochastic model is transferable to all individual scanners of the series.</div></div>","PeriodicalId":100730,"journal":{"name":"ISPRS Open Journal of Photogrammetry and Remote Sensing","volume":"15 ","pages":"Article 100079"},"PeriodicalIF":0.0,"publicationDate":"2025-01-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"143137152","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"OA","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
扫码关注我们
求助内容：
应助结果提醒方式：
京公网安备 11010802042870号