Plant Species Diversity Assessment in the Temperate Grassland Region of China Using UAV Hyperspectral Remote Sensing DOI Creative Commons
Hong Wang,

Chunyong Feng,

Xiaobing Li

et al.

Diversity, Journal Year: 2024, Volume and Issue: 16(12), P. 775 - 775

Published: Dec. 20, 2024

Biodiversity conservation is a critical environmental challenge, with accurate assessment being essential for efforts. This study addresses the limitations of current plant diversity methods, particularly in recognizing mixed and stunted grass species, by developing an enhanced species recognition approach using unmanned aerial vehicle (UAV) hyperspectral data deep learning models steppe region Xilinhot, Inner Mongolia. We compared five models—support vector machine (SVM), two-dimensional convolutional neural network (2D-CNN), three-dimensional (3D-CNN), hybrid spectral CNN (HybridSN), improved HybridSN+—for identification. The results show that SVM 2D-CNN have relatively poor effects on distribution individuals, while HybridSN HybridSN+ can effectively identify important region, accuracy model reach 96.45 (p < 0.05). Notably, 3D-CNN model’s performance was inferior to model, especially densely populated smaller species. optimized from demonstrated individuals under equivalent conditions, leading discernible enhancement overall (OA). Diversity indices (Shannon–Wiener diversity, Simpson Pielou evenness) were calculated identification spatial maps generated each index. A comparative analysis derived ground survey revealed strong correlation consistency, minimal differences between two methods. provides feasible technical efficient meticulous biodiversity assessment, offering crucial scientific references regional conservation, management, restoration.

Language: Английский

Classification Model of Grassland Desertification Based on Deep Learning DOI Open Access
Huilin Jiang,

Rigeng Wu,

Yongan Zhang

et al.

Sustainability, Journal Year: 2024, Volume and Issue: 16(19), P. 8307 - 8307

Published: Sept. 24, 2024

Grasslands are one of the most important ecosystems on earth, and impact grassland desertification earth’s environment ecosystem cannot be ignored. Accurately distinguishing types has application value. The appropriate grazing strategies can implemented based these distinctions. Grassland conservation measures tailored accordingly. This contributes to further protecting restoring vegetation. project takes color images labeled with grasslands as research object, uses currently popular deep learning model classification tool, then establishes a image-based feature extraction network, Vision Transformer model, by comparing various image models. experimental results show that, despite complex structure large number parameters obtained in this project, test accuracy rate reaches 88.72% training loss is only 0.0319. Compared models such VGG16, ResNet50, ResNet101, DenseNet101, DenseNet169, DenseNet201, so on, demonstrates clear advantages accuracy, fitting ability, generalization capacity. By integrating technology, applied management ecological restoration. Mobile devices used conveniently capture data, information processed quickly. provides efficient tools for managers, environmental scientists, organizations. These assist quickly assessing extent desertification, optimizing decisions. Furthermore, strong technical support offered restoration sustainable grasslands.

Language: Английский

Citations

0
Plant Species Diversity Assessment in the Temperate Grassland Region of China Using UAV Hyperspectral Remote Sensing DOI Creative Commons
Hong Wang,

Chunyong Feng,

Xiaobing Li

et al.

Diversity, Journal Year: 2024, Volume and Issue: 16(12), P. 775 - 775

Published: Dec. 20, 2024

Biodiversity conservation is a critical environmental challenge, with accurate assessment being essential for efforts. This study addresses the limitations of current plant diversity methods, particularly in recognizing mixed and stunted grass species, by developing an enhanced species recognition approach using unmanned aerial vehicle (UAV) hyperspectral data deep learning models steppe region Xilinhot, Inner Mongolia. We compared five models—support vector machine (SVM), two-dimensional convolutional neural network (2D-CNN), three-dimensional (3D-CNN), hybrid spectral CNN (HybridSN), improved HybridSN+—for identification. The results show that SVM 2D-CNN have relatively poor effects on distribution individuals, while HybridSN HybridSN+ can effectively identify important region, accuracy model reach 96.45 (p < 0.05). Notably, 3D-CNN model’s performance was inferior to model, especially densely populated smaller species. optimized from demonstrated individuals under equivalent conditions, leading discernible enhancement overall (OA). Diversity indices (Shannon–Wiener diversity, Simpson Pielou evenness) were calculated identification spatial maps generated each index. A comparative analysis derived ground survey revealed strong correlation consistency, minimal differences between two methods. provides feasible technical efficient meticulous biodiversity assessment, offering crucial scientific references regional conservation, management, restoration.

Language: Английский

Citations

0