The developed system, named EcoVision, establishes a practical foundation for scalable, high-resolution salt marsh monitoring, demonstrating how AI-driven workflows can translate pixel-level predictions into ecologically interpretable metrics.
Abstract
High-resolution RGB imagery acquired from low-altitude UAV surveys was processed through a modular pipeline incorporating transformer-based semantic segmentation, connected-component vegetation extraction, fine-grained species classification using a ConvNeXt architecture, and grid-based dominance scoring at 2x2m resolution. The framework targeted two ecologically significant halophytic grasses, Spartina maritima and Puccinellia maritima, and was trained using a curated and manually annotated UAV imagery, along with biodiversity imagery sourced from publicly accessible datasets. In order to identify these plants from the imagery, our segmentation yielded reliable species masks (mean IoU = 0.56; pixel-level accuracy = 0.96), while object-level classification achieved very good discrimination (F1 = 0.99). Dominance estimates closely matched quadrat-based field surveys, with mean absolute differences below 8%, preserving fine-scale spatial structure under realistic survey conditions. The developed system, named EcoVision, establishes a practical foundation for scalable, high-resolution salt marsh monitoring, demonstrating how AI-driven workflows can translate pixel-level predictions into ecologically interpretable metrics.
Fine-grained mapping of riparian vegetation is important for ecological monitoring, invasive species control, and ecosystem restoration. However, riparian plant communities often exhibit fragmented patches, broad transition zones and high visual similarity among classes, making stable species-level segmentation difficult from conventional satellite imagery or lower-resolution UAV imagery. To address this gap, we present WetVeg-2mm, an ultra-high-resolution UAV dataset for fine-grained riparian vegetation semantic segmentation. Built from UAV surveys over a representative riparian section of the Jiuzhou River in Guangxi, China, the dataset provides 2054 image chips (1024 × 1024) with pixel-level annotations at 2 mm ground sampling distance. It contains 17 semantic classes in total, including 14 representative wetland plant classes, such as Colocasia, Eichhornia and Phragmites, together with water, bareland and background. Five baseline models, namely U-Net, Attention U-Net, DeepLabV3+, PSPNet and SegFormer, were evaluated using per-class IoU, mIoU, mDice, PA, Precision and Recall. Across all evaluated baseline settings, SegFormer with ImageNet pretraining achieved the best overall performance, with 76.23% mIoU, 86.01% mDice, 85.17% PA, 87.30% Recall and 85.42% Precision on the test set. Overall, WetVeg-2mm provides a reproducible and challenging benchmark for fine-grained riparian vegetation semantic segmentation.
Accurate characterization of riverbed substrate from remote sensing imagery is essential for applications in fluvial geomorphology, habitat modeling, and river management. While recent advances in computer vision, particularly deep learning, have improved sediment mapping capabilities, their reliance on large annotated datasets and computational resources limits their broader applicability. This study presents a scalable workflow for categorical substrate classification using ultra-high-resolution aerial RGB orthoimagery in clear-water river environments. The approach integrates spectral information with statistical and structural texture descriptors derived from Gray-Level Co-occurrence Matrices (GLCM) and Local Binary Patterns (LBP), combined within a Random Forest classification framework. The methodology is structured as a semi-automated, five-stage workflow: (1) expert-based ground-truth substrate annotation; (2) feature set generation; (3) spatially aware model optimization; (4) full-domain classification; and (5) design-based validation for independent accuracy assessment. Model performance is evaluated using spatially aware cross-validation and design-based probability sampling to account for spatial autocorrelation and provide unbiased accuracy estimates. The method was applied in four geomorphologically distinct alpine river reaches, achieving design-based overall accuracy ranging from 70% to 88%. These results demonstrate that RGB-based approaches can achieve reliable reach-scale categorical substrate classification when combined with appropriate feature representation and rigorous validation strategies. However, limitations remain for visually similar or transitional substrate classes, particularly fine sediments such as sand and clay, which are difficult to distinguish consistently even during manual annotation. The workflow is implemented using open-source tools and is applicable to clear-water conditions where the riverbed remains optically visible.
Tulio Soto Parra, D. Farò, G. Zolezzi· Remote Sensing· 0 citations
Accurate and up-to-date information is essential for environmental monitoring, particularly in regions characterized by complex topography and heterogeneous landscapes. This study presents a multisource remote sensing–based approach for forest vegetation classification on La Palma Island (Canary Islands, Spain), which was further used to illustrate its potential for monitoring the temporal dynamics of different forest habitat types. Very high-resolution multispectral data from the WorldView-2/3 satellites were used, complemented by multispectral and LiDAR data acquired by an unmanned aerial vehicle (UAV). Four target forest vegetation types were mapped within a six-class classification scheme that also included “Other vegetation” and “Soil/Others” as non-target/background classes. The performance of ten supervised classification algorithms was evaluated, including Minimum Distance, Mahalanobis Distance, Parallelepiped, Spectral Angle Mapper, Maximum Likelihood, Naïve Bayes, K-Nearest Neighbors, Random Forest, Support Vector Machine, and the transformer-based deep learning model SegFormer. The results indicate that Random Forest achieved the highest overall accuracy, while Support Vector Machine and SegFormer also showed competitive performance, particularly when spectral information was integrated with vegetation indices and topographic variables. The study provides practical evidence on the selection of input data and classifiers for detailed forest vegetation mapping in a large and topographically complex island.
J. Marcello, F. Eugenio, A. Mederos-Barrera et al.· Remote Sensing· 0 citations
Abstract. Vegetation mapping in alpine environments is essential for monitoring ecosystem dynamics and climate change impacts, yet remains challenging when using very high-resolution UAV imagery under limited labeled data. This study proposes a data-centric, pixel- based classification framework for class-level vegetation mapping using multispectral UAV data acquired in an alpine study area. The approach prioritizes improving data representation rather than increasing model complexity. To address label scarcity, a feature-rich dataset was constructed by integrating spectral information, vegetation indices, and lightweight spatial descriptors to enhance class separability. Classification was performed using XGBoost, which is well suited for multispectral tabular data and robust under imbalanced conditions. The results show consistent classification performance across vegetation types and demonstrate the effectiveness of dataset enrichment under limited supervision, highlighting the importance of feature representation in data-scarce scenarios.
M. Elahi, Alessandra Spadaro, F. Matrone et al.· The International Archives o...· 0 citations
Accurate estimation of Above-Ground Biomass (AGB) in natural grasslands is critical for ecosystem management, yet traditional satellite and uncrewed aerial vehicle (UAV) remote sensing struggles to capture fine-scale vegetation details. To overcome these limitations, we developed “PraePercep-Y,” an integrated ground-based robotic perception platform tailored for typical grasslands. Within this system, we systematically established a theoretical model for fractional vegetation cover (FVC) estimation, determining that an optimal actual pixel area of 3–5 mm2 significantly enhances detection accuracy. Building upon this optimized data acquisition, we proposed the AGBT network for high-precision biomass inversion. The AGBT model is designed with a flexible dual-stream architecture that operates in two primary configurations: Mode 1 (RGB-based estimation), utilizing the visual stream to extract hierarchical RGB image features, and Mode 2 (Multisensor data fusion), which further integrates an attribute stream to encode vegetation height and FVC. To effectively fuse these modalities, we introduced a cross-modal attention mechanism for semantic alignment and an adaptive weight distribution strategy to dynamically adjust modality contributions. Validated through rigorous field experiments from 2022 to 2025 across 600 quadrats in Inner Mongolia, the system demonstrated superior performance. The AGBT model achieved strong predictive accuracy, with R2 values of 0.82 for Mode 1 (RGB-only) and 0.88 for Mode 2 (multisensor fusion). This study provides a highly accurate, automated methodological baseline for AGB assessment, significantly reducing manual labor and offering a scientific basis for grassland resource conservation.
Yuzhuo Zhang, Tianyi Wang, Decheng Wang et al.· IEEE Transactions on Geoscie...· 0 citations
Abstract. Posidonia oceanica is one of the most important seagrass species in the Mediterranean Sea, providing essential ecosystem services such as carbon sequestration, coastal protection and acting as a habitat and nursery ground for numerous marine species. These meadows have experienced significant decline in recent decades due to increasing anthropogenic pressures and environmental changes. Accurate and efficient mapping techniques are therefore essential for monitoring their spatial distribution and supporting conservation efforts. This study investigates the potential of very high-resolution Neo-Pléiades satellite imagery for mapping P. oceanica meadows along the northeastern coast of Sardinia (Italy). Two satellite acquisitions from 2021 and 2022 were orthorectified in PCI Catalyst (v.2023.0.0) using a Rational Polynomial Coefficient (RPC) model. Subsequently, a water column correction based on the Lyzenga depth-invariant index was applied to reduce depth-related spectral variability. The images were then classified using an object-based image analysis approach implemented in eCognition Developer (v.10.5), comparing three supervised algorithms: Nearest Neighbor (NN), Support Vector Machines (SVM), and Random Tree (RT). Accuracy assessment based on confusion matrices showed high classification performance, with overall accuracies up to 0.97 and Kappa values up to 0.96. Additional spatial validation using manually delineated reference areas confirmed classification reliability, although slightly lower agreement values were observed compared to confusion matrix estimates. The results highlight the strong potential of integrating high-resolution satellite imagery, water column correction, and object-based classification for mapping and monitoring P. oceanica habitats.
V. Baiocchi, F. Giannone, Chiara Magurano et al.· The International Archives o...· 0 citations