Deepening the Accuracy of Tree Species Classification: A Deep Learning-Based Methodology DOI Open Access
Sungeun Cha, Joongbin Lim, Kyoung-Min Kim

et al.

Forests, Journal Year: 2023, Volume and Issue: 14(8), P. 1602 - 1602

Published: Aug. 8, 2023

The utilization of multi-temporally integrated imageries, combined with advanced techniques such as convolutional neural networks (CNNs), has shown significant potential in enhancing the accuracy and efficiency tree species classification models. In this study, we explore application CNNs for using imageries. By leveraging temporal variations captured our goal is to improve models’ discriminative power overall performance. results study reveal a notable improvement compared previous approaches. Specifically, when random forest model’s 84.5% Gwangneung region, CNN-based model achieved higher 90.5%, demonstrating 6% improvement. Furthermore, by extending same Chuncheon observed further enhancement accuracy, reaching 92.1%. While additional validation necessary, these findings suggest that proposed can be applied beyond single its broader applicability. Our experimental confirm effectiveness deep learning approach achieving high classification. integration imageries algorithm presents promising avenue advancing classification, contributing improved management, conservation, monitoring context climate change.

Language: Английский

Spatially autocorrelated training and validation samples inflate performance assessment of convolutional neural networks DOI Creative Commons
Teja Kattenborn, Felix Schiefer, Julian Frey

et al.

ISPRS Open Journal of Photogrammetry and Remote Sensing, Journal Year: 2022, Volume and Issue: 5, P. 100018 - 100018

Published: June 21, 2022

Deep learning and particularly Convolutional Neural Networks (CNN) in concert with remote sensing are becoming standard analytical tools the geosciences. A series of studies has presented seemingly outstanding performance CNN for predictive modelling. However, such models is commonly estimated using random cross-validation, which does not account spatial autocorrelation between training validation data. Independent method, dependence will inevitably inflate model performance. This problem ignored most CNN-related suggests a flaw their procedure. Here, we demonstrate how neglecting during cross-validation leads to an optimistic assessment, example tree species segmentation multiple, spatially distributed drone image acquisitions. We evaluated CNN-based predictions test data sampled from 1) randomly hold-outs 2) blocked hold-outs. Assuming that block provides realistic performance, holdouts overestimated by up 28%. Smaller sample size increased this optimism. Spatial among observations was significantly higher within than different Thus, should be tested strategies multiple independent Otherwise, any geospatial deep method likely overestimated.

Language: Английский

Citations

83
Status, advancements and prospects of deep learning methods applied in forest studies DOI Creative Commons
Ting Yun, Jian Li, Lingfei Ma

et al.

International Journal of Applied Earth Observation and Geoinformation, Journal Year: 2024, Volume and Issue: 131, P. 103938 - 103938

Published: June 4, 2024

Language: Английский

Citations

35
Deep learning-based individual tree crown delineation in mangrove forests using very-high-resolution satellite imagery DOI Creative Commons
Guillaume Lassalle, Matheus Pinheiro Ferreira, Laura Elena Cué La Rosa

et al.

ISPRS Journal of Photogrammetry and Remote Sensing, Journal Year: 2022, Volume and Issue: 189, P. 220 - 235

Published: May 25, 2022

Mangrove forests are vulnerable ecosystems that require broad-scale monitoring. Various solutions based on satellite imagery have emerged for this purpose but still suffer from the lack of methods to accurately delineate individual tree crowns (ITCs). Within-stand variability in crown size and shape, clumping fragmentation, understory vegetation hamper delineation these ecosystems. To cope with factors, proposed method combines a deep learning-based enhancement ITCs marker-controlled watershed segmentation algorithm. The MT-EDv3 neural network is employed compute normalized Euclidean distance pixels treetops Laplacian Gaussian filter applied resulting image enhance borders before segmentation. was WorldView over four mangrove sites worldwide compared previously published using standardized metrics. Accurate detection (Overall Accuracy ≥ 0.93 Kappa 0.87) area estimation (R2 0.66, NRMSE ≤ 12%) achieved all either panchromatic band or combination pan-sharpened visible-near-infrared bands. Based Precision, Recall, F1-score, outperformed previous software-based algorithms delineation, as well Mask R-CNN framework. viewing geometry images forest heterogeneity were identified important contributors accuracy. This study first achieve accurate across sites, opening perspectives applications satellite-based shows promising transferability other very-high-resolution sensors aerial unmanned vehicle could be improved by including more spectral information LiDAR-derived canopy height models.

Language: Английский

Citations

60
KOH activated carbons from Brazil nut shell: Preparation, characterization, and their application in phenol adsorption DOI

Maria C. F. Da Silva,

C Schnorr, Sabrina F. Lütke

et al.

Process Safety and Environmental Protection, Journal Year: 2022, Volume and Issue: 187, P. 387 - 396

Published: Sept. 9, 2022

Language: Английский

Citations

45
Large-Scale Date Palm Tree Segmentation from Multiscale UAV-Based and Aerial Images Using Deep Vision Transformers DOI Creative Commons
Mohamed Barakat A. Gibril, Helmi Zulhaidi Mohd Shafri, Rami Al‐Ruzouq

et al.

Drones, Journal Year: 2023, Volume and Issue: 7(2), P. 93 - 93

Published: Jan. 29, 2023

The reliable and efficient large-scale mapping of date palm trees from remotely sensed data is crucial for developing tree inventories, continuous monitoring, vulnerability assessments, environmental control, long-term management. Given the increasing availability UAV images with limited spectral information, high intra-class variance trees, variations in spatial resolutions data, differences image contexts backgrounds, accurate very-high resolution (VHSR) can be challenging. This study aimed to investigate reliability efficiency various deep vision transformers extracting multiscale multisource VHSR images. Numerous transformers, including Segformer, Segmenter, UperNet-Swin transformer, dense prediction levels model complexity, were evaluated. models developed evaluated using a set comprehensive UAV-based aerial generalizability transferability compared convolutional neural network-based (CNN) semantic segmentation (including DeepLabV3+, PSPNet, FCN-ResNet-50, DANet). results examined generally comparable several CNN-based models. investigated achieved satisfactory images, an mIoU ranging 85% 86.3% mF-score 91.62% 92.44%. Among models, Segformer generated highest on testing datasets. model, followed by outperformed all dataset additional unseen dataset. In addition delivering remarkable versatile was among those small number parameters relatively low computing costs. Collectively, could used efficiently updating inventories palms other species.

Language: Английский

Citations

29
More than 10,000 pre-Columbian earthworks are still hidden throughout Amazonia DOI Open Access
Vinícius Peripato, Carolina Levis, Guido A. Moreira

et al.

Science, Journal Year: 2023, Volume and Issue: 382(6666), P. 103 - 109

Published: Oct. 6, 2023

Indigenous societies are known to have occupied the Amazon basin for more than 12,000 years, but scale of their influence on Amazonian forests remains uncertain. We report discovery, using LIDAR (light detection and ranging) information from across basin, 24 previously undetected pre-Columbian earthworks beneath forest canopy. Modeled distribution abundance large-scale archaeological sites Amazonia suggest that between 10,272 23,648 remain be discovered most will found in southwest. also identified 53 domesticated tree species significantly associated with earthwork occurrence probability, likely suggesting past management practices. Closed-canopy contain thousands undiscovered around which actively modified forests, a discovery opens opportunities better understanding magnitude ancient human its current state.

Language: Английский

Citations

24
Early detection of red palm weevil infestations using deep learning classification of acoustic signals DOI
Wadii Boulila, Ayyub Alzahem, Anis Koubâa

et al.

Computers and Electronics in Agriculture, Journal Year: 2023, Volume and Issue: 212, P. 108154 - 108154

Published: Aug. 28, 2023

Language: Английский

Citations

20
Multi-task fully convolutional network for tree species mapping in dense forests using small training hyperspectral data DOI
Laura Elena Cué La Rosa, Camile Sothe, Raul Queiroz Feitosa

et al.

ISPRS Journal of Photogrammetry and Remote Sensing, Journal Year: 2021, Volume and Issue: 179, P. 35 - 49

Published: July 28, 2021

Language: Английский

Citations

41
Individual Tree Species Classification Based on Convolutional Neural Networks and Multitemporal High-Resolution Remote Sensing Images DOI Creative Commons
Xianfei Guo, Hui Li,

Linhai Jing

et al.

Sensors, Journal Year: 2022, Volume and Issue: 22(9), P. 3157 - 3157

Published: April 20, 2022

The classification of individual tree species (ITS) is beneficial to forest management and protection. Previous studies in ITS that are primarily based on airborne LiDAR aerial photographs have achieved the highest accuracies. However, because complex high cost data acquisition, it difficult apply large-area forests. High-resolution, satellite remote sensing abundant sources significant application potential classification. Based Worldview-3 Google Earth images, convolutional neural network (CNN) models were employed improve accuracy by fully utilizing feature information contained different seasonal images. Among three CNN models, DenseNet yielded better performances than ResNet GoogLeNet. It offered an OA 75.1% for seven using only WorldView-3 image 78.1% combinations autumn results indicated images with suitable temporal detail could be as auxiliary accuracy.

Language: Английский

Citations

24
Early-Stage Pine Wilt Disease Detection via Multi-Feature Fusion in UAV Imagery DOI Open Access
Wanying Xie, Han Wang, Wenping Liu

et al.

Forests, Journal Year: 2024, Volume and Issue: 15(1), P. 171 - 171

Published: Jan. 14, 2024

Pine wilt disease (PWD) is a highly contagious and devastating forest disease. The timely detection of pine trees infected with PWD in the early stage great significance to effectively control spread protect resources. However, spatial domain, features early-stage are not distinctly evident, leading numerous missed detections false positives when directly using spatial-domain images. we found that frequency domain information can more clearly express characteristics PWD. In this paper, propose method based on deep learning for by comprehensively utilizing domain. An attention mechanism introduced further enhance features. Employing two deformable convolutions fuse both domains, aim fully capture semantic information. To substantiate proposed method, study employs UAVs images at Dahuofang Experimental Forest Fushun, Liaoning Province. A dataset affected curated facilitate future research infestations trees. results indicate that, compared Faster R-CNN, DETR YOLOv5, best-performing improves average precision (AP) 17.7%, 6.2% 6.0%, F1 scores 14.6%, 3.9% 5.0%, respectively. provides technical support tree counting localization field areas lays foundation wood nematode

Language: Английский

Citations

5