Multi-source cross-domain hyperspectral image (HSI) classification is challenged by heterogeneous sensor configurations, scene-dependent distribution shifts, and limited labeled target data, which hinder effective knowledge transfer across multiple scenes. Motivated by progressive feature correction, we propose a three-stage Residual Feature Discrepancy Refinement (RFD) framework for collaborative representation learning across heterogeneous HSI domains. RFD formulates this correction as a deterministic, discrepancy-conditioned residual refinement process. First, domain-specific encoders project four source domains and the target domain, which may have unequal spectral dimensions and label spaces, into a common-dimensional feature space. Adaptive severity and domain weighting uses first- and second-order feature discrepancies to estimate source-specific conditioning coordinates and collaborative contribution weights. A shared discrepancy-conditioned residual refiner then performs multi-step feature refinement to reduce domain-dependent statistical deviations. Finally, an exponential-moving-average historical prototype memory stabilizes target adaptation, followed by cosine 1-nearest-neighbor classification. Across ten randomized runs, RFD achieves mean overall accuracies of 94.65%, 94.87%, and 96.95% on NC12, Salinas, and WHU-Hi-LongKou, respectively, and obtains the highest mean overall accuracy, average accuracy, and κ among the evaluated unified-protocol methods.
Small objects in UAV imagery are easily obscured by sensor noise, adverse illumination, and weather-like appearance degradation. To address this problem, this paper proposes YOLO-ROSS, a lightweight detector built on YOLOv11. Its two architectural contributions are a C3k2_DTAB feature-extraction block, which combines grouped channel self-attention and masked-window self-attention to protect weak local evidence while adding non-local context, and an AFPN-P2 neck, which preserves high-resolution geometry and progressively aligns shallow detail with deep semantics. For end-to-end deployment, the detector also adopts the rank-consistent one-to-many/one-to-one assignment used by YOLOv10; this adopted component is evaluated separately but is not claimed as a new label-assignment algorithm. On the mixed-corruption VisDrone benchmark, YOLO-ROSS obtains 0.309 mean mAP@0.5 over eight runs, an absolute improvement of 0.042 over YOLOv11n. Its all-class peak F1 is approximately 0.40, compared with 0.36 for YOLOv11n, although the optimal confidence threshold shifts from 0.146 to 0.186. A direct-transfer evaluation on SODA-D-Robustness provides an additional check under a combined driving-domain and appearance-corruption shift. These results support improved accuracy under the specified controlled RGB corruptions, while not establishing universal real-weather or cross-modal robustness.
UAV-based 3D object detection is important for remote-sensing applications such as autonomous inspection, search and rescue, aerial mapping, flight cooperation, and scene-level environmental understanding, where a detector must localize diverse objects in large and sparsely observed point clouds. Existing 3D object detectors provide a strong foundation, but most of them are developed around autonomous-driving scenarios and are not fully adapted to UAV scenes. Compared with road scenes, UAV point clouds usually cover larger areas, contain more diverse object categories, and include many small, sparse, and structurally varied targets. Many small targets therefore occupy only a few BEV cells and contain limited point returns; subsequent feature aggregation and downsampling can further smooth these sparse local responses, making object boundaries and center-related responses less distinguishable from the background. We propose CenterPoint-UAV, an end-to-end voxel-based detector that refines BEV features for UAV-based 3D object detection. CenterPoint-UAV introduces Context-Detail BEV Enhance (CDBE), which uses a Context Enhancement Branch (CEB) and a Detail Enhancement Branch (DEB) to produce complementary BEV feature maps and fuses them using Adaptive Residual Fusion (ARF). It then uses Cross-Level BEV Fusion (CLBF) to combine early BEV details with deep semantic features, followed by a Fine Center Head (FCH) for denser center prediction. Experiments on WiSAR3D, a large-scale real-world UAV point-cloud dataset for multi-category object detection, show that CenterPoint-UAV achieves state-of-the-art mAP among existing methods and maintains a low parameter budget, demonstrating its effectiveness for UAV-based 3D remote sensing.
Canopy urban heat island (CUHI), defined as the air temperature difference between the urban near-surface atmosphere and surrounding rural areas, exhibits diurnal variability and affects urban thermal environments and well-being. However, the nonlinear effects of urban morphology on daytime and nighttime CUHI remain insufficiently understood. Taking the Yangtze River Delta (YRD) as the study area, this study integrates remote sensing, building morphology, and ground-based meteorological data to characterize urban morphology and canopy urban heat island intensity (CUHII). Extreme Gradient Boosting (XGBoost), SHapley Additive exPlanations (SHAP), and dependence plots were used to quantify and interpret nonlinear morphology–CUHII relationships. The models showed moderate explanatory ability, indicating that the selected morphology indicators explained only part of CUHII variability. Nighttime CUHII was stronger than daytime CUHII, with average values of 0.848 °C and 0.431 °C, respectively, and high-value areas concentrated in Shanghai, northern Zhejiang, and southern Jiangsu. During the daytime, the Aggregation Index (AI) was the most important selected morphology variable and was positively associated with CUHII, suggesting that compact built-up patterns may enhance heat accumulation by increasing heat absorption and limiting ventilation. At night, the Splitting Index (SPLIT) ranked first, with higher values generally associated with weaker CUHII, possibly reflecting greater spatial openness and reduced continuity of built-up surfaces. The nonlinear transition ranges of AI and SPLIT further reveal diurnal asymmetry in morphology–CUHII relationships. These findings support time-specific urban heat mitigation while avoiding attribution of overall CUHII variability solely to urban morphology.
Automated screening of geomorphologically defined landslide potential-hazard candidates from high-resolution topographic data remains challenging in mountainous regions where comprehensive field inventories are unavailable. This study proposes a two-stage framework for extracting rule-defined boot-shaped terrain candidates from airborne LiDAR digital elevation model (DEM) data. First, an expert-informed screening rule formalizes a steep-upper–gentle-lower terrain morphology using representative longitudinal profiles of slope units, and the screened units are converted into rule-derived reference masks. Second, semantic segmentation models are trained to approximate these reference patterns directly from DEM-derived raster inputs. Four architectures—U-Net, U-Net++, DeepLabV3+, and SegFormer-B0—were evaluated using 406 patches of 256 × 256 pixels at 2 m resolution from four LiDAR-covered subregions in Zhenxiong County, China. Under spatially grouped three-fold cross-validation, DeepLabV3+ with DEM + slope-gradient input and Dice + Focal loss achieved a mean pixel-level F1-score of 0.351, mIoU of 0.557, and Patch-F1 of 0.814. Input-feature experiments showed that slope-gradient information was particularly informative, whereas the incremental contribution of aspect was configuration-dependent; DEM + slope was retained as a parsimonious two-channel input. Sensitivity analysis showed that the rule-derived candidate definition changed materially with the screening parameters. Leave-one-subregion-out evaluation yielded a macro-averaged F1 of 0.321, indicating measurable within-county cross-subregion transfer. However, whole-area evaluation under natural candidate prevalence reduced the macro-average F1 to 0.072 at a fixed threshold and 0.095 using validation-derived operating thresholds. These results indicate that the proposed model is best interpreted as a raster-based surrogate for rule-derived geomorphological screening rather than as an independently validated landslide detector.
This study investigates how dust and sulfate aerosols modulate cloud properties and rainfall over Sudan, a key part of the Sahara–Sahel dust belt. Satellite and reanalysis products (MODIS, CHIRPS, MERRA 2, EAC4) are combined with four CMIP6 models to analyze rainy season (JJAS) aerosol optical depth (AOD), cloud water path (CWP), cloud effective radius (Reff), and precipitation for 2003–2014, and to assess future changes under SSP1 2.6, SSP2 4.5, and SSP5 8.5 during 2041–2100. Reanalysis data show that natural mineral dust dominates aerosol loading over Sudan, accounting for approximately 70–85% of total annual mean AOD, with substantial spatial variability across the domain and the highest contributions occurring over the Sahara–Sahel transition zone, whereas sulfate AOD peaks over urban and agricultural regions in central and eastern Sudan. Observations reveal that dust AOD is negatively correlated with CWP and precipitation in northern and central Sudan, while sulfate AOD shows positive correlations with CWP and rainfall in the southeast. All datasets exhibit negative AOD–Reff relationships that are consistent with a Twomey-like signature. However, because AOD is a column-integrated measure that does not directly represent cloud-based cloud condensation nuclei (CCN), these relationships should not be interpreted as direct evidence of the Twomey effect. The models also overestimate the positive AOD–CWP and AOD–precipitation correlations, suggesting that they may simulate stronger aerosol-related cloud persistence and precipitation responses than indicated by the observations. Multi-model projections indicate substantial twenty first century declines in sulfate and total AOD under all SSPs, driven by emission controls, whereas dust AOD shows weaker, climate- and land-use-controlled changes. Together, these results suggest that CMIP6 likely overestimates the sensitivity of Sudan’s hydrological cycle to aerosol perturbations and highlight the need for improved dust parameterizations and high-resolution regional modeling to constrain future water resource risks.
Estimates of gross primary production (GPP) derived from satellite remote sensing are crucial for assessing the terrestrial carbon dynamics. While model comparison is important, existing GPP products rely on a limited number of satellite sensors. In this study, the contemporary fraction of absorbed photosynthetically active radiation product derived from PROBA-V and Sentinel-3 was used to develop a new GPP product (EU-GPP) based on a light use efficiency (LUE) model for the European continent. EU-GPP accounted for the distinct responses of biomes to temperature and water stresses by incorporating biome-specific environmental scalars. Evaluation against eddy covariance GPP and model comparison against other LUE-based GPP products demonstrated the high model accuracy of EU-GPP within Europe. The model also responded well to the drought-induced stress, indicating its ability to capture interannual variability and the impact of extreme weather events. EU-GPP highlighted increasing GPP trends in croplands during 2014–2023, underlining the recent advancements in agricultural management practices. The Mediterranean forest ecosystems exhibited weakening GPP, suggesting the strong adverse impact of summer droughts on these ecosystems. This study presents an LUE-based GPP product based on novel remote sensing data and provides an independent perspective for the monitoring of GPP across the European continent.
Background: Pathogenesis-related (PR) proteins, primarily thaumatin-like proteins (TLPs) and chitinases, are the principal cause of protein haze formation in white wines, increasing bentonite requirements and affecting winemaking efficiency. However, evaluating their spatial variability before harvest remains challenging because conventional analytical methods are destructive, labor-intensive, and spatially limited. Methods: This study developed a non-destructive framework to predict the accumulation of PR proteins in Vitis vinifera cv. Chardonnay and Sauvignon Blanc by integrating multitemporal UAV-derived multispectral imagery with stem water potential (Ψstem). Vegetation indices and physiological measurements were acquired throughout berry development over two growing seasons. Elastic Net, XGBoost, and CatBoost models were developed using the 2023 growing season as the calibration dataset through 20 independent jackknife training iterations, with five-fold cross-validation for hyperparameter optimization and an internal 80/20 split used exclusively for early stopping. The models were subsequently externally validated using the independent 2024 growing season. Model interpretation was performed using SHAP to identify influential predictors of model predictions. Results: CatBoost provided the most consistent predictive performance across response variables and was therefore selected for spatial prediction. SHAP analysis revealed cultivar-specific predictor hierarchies, with GNDVI at harvest dominating predictions in Chardonnay, whereas multitemporal NDVI variables were the most influential predictors in Sauvignon Blanc. Stem water potential acquired during Berry Filling II and pre-harvest consistently contributed to model performance in both cultivars, highlighting the importance of late-season physiological conditions. Spatial prediction maps revealed marked intra-vineyard heterogeneity in PR protein accumulation, identifying vineyard sectors with contrasting predicted protein concentrations. Conclusions: Integrating multitemporal UAV multispectral imagery, stem water potential, and explainable machine learning provides an accurate and interpretable framework for predicting the accumulation of pathogenesis-related proteins before harvest. This approach expands the application of remote sensing from conventional assessments of vine vigor to the prediction of biochemical traits directly associated with wine protein stability, supporting targeted sampling, selective harvesting, and more efficient bentonite management in precision viticulture.
Land subsidence threatens the operational safety of railways in soft-soil plains. This study investigates the spatiotemporal evolution and mechanisms of subsidence along the Beijing–Shanghai Conventional Railway (BSR) and High-Speed Railway (HSR) in Shanghai using 2015–2025 Sentinel-1 imagery. To explicitly decouple macroscopic environmental background subsidence from localized engineering disturbances, we propose a novel framework integrating SBAS-InSAR monitoring, RF-SHAP multi-source attribution, and LightGBM baseline prediction. Results reveal significant deformation heterogeneity governed by foundation designs: the shallow-subgrade BSR experienced a mean subsidence rate of −2.03 mm/yr (with 4.69% extreme pixels), whereas the deep-anchored HSR remained highly stable at −0.81 mm/yr. Attribution analysis demonstrates that anthropogenic factors primarily drive regional deformation, contributing 70.9% to the variance, with distance to the BSR, groundwater levels, and building density identified as core nonlinear predictors. Furthermore, by analyzing dynamic prediction residuals, the framework accurately traced high-risk structural anomalies, successfully isolating −17.9 mm/yr of acute settlement induced by short-term construction and 100–120 mm of cumulative consolidation triggered by long-term static loads. This approach provides a robust, data-driven diagnostic tool to assist in targeted track-bed maintenance for railway safety management.
Precipitation nowcasting aims to predict short-term precipitation evolution over forecast lead times of 1–6 h and can support hydrological-risk and disaster-prevention applications when near-real-time observations are available. However, precipitation forecasting remains challenging because of rapid spatiotemporal evolution, spatial displacement, and the difficulty of representing localized high-intensity precipitation. To address these issues, this study proposes STAMP-GAN, a spatiotemporal attention-modulated generative adversarial network for regional precipitation sequence prediction. STAMP-GAN combines an AM-ConvLSTM temporal evolution module with spatial attention, efficient channel attention, large-receptive-field context modeling, and temporal-index-conditioned feature modulation. A spatially aligned two-dimensional digital elevation model (DEM) field is retained as static auxiliary geographical information. The STAMP-Net generator uses hierarchical multi-scale feature extraction to reconstruct precipitation structures at different spatial scales while a dual-branch temporal PatchGAN provides adversarial supervision for both the complete forecast sequence and the final three forecast frames. A hybrid objective combines regression, event-based, structural, temporal, and adversarial constraints. Experiments on the ERA5 and CMA-S datasets show that, compared with the best-performing baseline for each metric, STAMP-GAN achieves relative CSI improvements of approximately 6.5% and 9.5%, respectively. The proposed framework provides a data-driven approach for retrospective hourly regional precipitation sequence prediction under gridded meteorological-data conditions, rather than a fully validated operational real-time nowcasting system.
Windthrow is a major disturbance risk for radiata pine (Pinus radiata D. Don) plantations, but operational susceptibility models must transfer across regions and storm events. We developed a multi-regional framework combining airborne laser scanning (ALS), aerial imagery, mapped stand and site variables, climate, soils, and event-period weather. These data were used to detect storm damage and model windthrow susceptibility following major storm events in Gisborne, Hawke’s Bay and Tasman, New Zealand. Windthrow was mapped from repeat ALS canopy-height differencing in Gisborne and Hawke’s Bay, and from post-storm aerial imagery in Tasman, producing 29,244 balanced windthrow and no-windthrow plot observations. Random-forest models were evaluated using stand-grouped, spatially blocked and leave-one-region-out validation. Stand structure provided the strongest predictive signal, with windthrow concentrated in older, taller and higher-volume stands. Adding long-term climate produced the largest improvement beyond the Base stand/site formulation, giving a pooled ROC–AUC of 0.901 ± 0.013. Spatially blocked ROC–AUC for the selected model ranged across the three regions from 0.780 to 0.856, while leave-one-region-out ROC–AUC ranged from 0.649 to 0.802, demonstrating useful but region-dependent transfer. Adding soil and event-period weather did not consistently improve transferability. Prevalence-calibrated conditional scenario estimates increased with stand development under the mapped regional prevalence and the conditions represented by the reference events. These estimates provide a scalable basis for comparative windthrow-risk screening but should not be interpreted as independently validated absolute or annual windthrow probabilities.
Early detection of tree-seedling establishment is essential for monitoring regeneration success in coastal-dune plantations, where conventional field assessments remain labour-intensive and spatially limited. This study presents a deep-learning workflow for detecting early-stage Pinus pinaster seedlings using multispectral UAS-derived point clouds. Field surveys in the Quiaios National Forest, Portugal, mapped approximately 1500 seedlings using RTK GNSS positioning, biometric measurements, and field photographs. Multispectral imagery acquired with a DJI Mavic 3 Multispectral platform was processed through Structure-from-Motion to generate calibrated orthomosaics, terrain products, and dense point clouds. Training-data preparation combined pine-centred buffers, spectral conditioning, manual refinement and point-cloud class assignment. Point Transformer V3 models were trained in ArcGIS Pro and evaluated using field-mapped buffers withheld from model training within plantation-line areas. The Baseline high-recall model achieved 88% object-level recall at the operational threshold of at least three classified Pine-Seedling points per buffer. The refined hard-negative model retained 84% recall while reducing off-buffer detections from 243 to 41. False-negative analysis showed that omissions were associated with reduced crown diameter and limited branch development under the adopted buffer-based retrieval framework. These results support transformer-based multispectral point-cloud classification for scalable monitoring of early-stage pine regeneration in heterogeneous coastal environments.
Accurate estimation of regional sea-level trends and accelerations requires adequate quantification of their uncertainties. However, measurement and processing uncertainties in regional estimates remain poorly characterized. We constructed an error variance–covariance matrix incorporating major satellite-altimetry errors, including inter-mission offsets, wet tropospheric correction, glacial isostatic adjustment, orbit, reference-frame, and early-TOPEX errors, with spatially dependent uncertainty parameters evaluated separately for each region. The method was applied to regional sea-level records constructed from TOPEX/Poseidon, Jason-1, Jason-2, Jason-3, and Sentinel-6A along-track observations over the China Seas and their Adjacent Oceans (CSA) from January 1993 to May 2026. The CSA sea-level trend is 4.032 ± 0.323 mm yr−1, while the acceleration is −0.013 ± 0.033 mm yr−2 (covariance-propagated 90% uncertainties), indicating a persistent rise but no acceleration distinguishable from zero. Across the Bohai, Yellow, East China, and South China Seas, trends range from 3.603 to 4.231 mm yr−1, whereas accelerations range from −0.020 to 0.056 mm yr−2, revealing spatial differences in both quantities and greater regional variability in acceleration. Covariance-propagated and residual-based intervals differ, demonstrating that they characterize different aspects of parameter uncertainty. Explicit propagation of identifiable measurement and processing error covariance provides a traceable basis for quantifying uncertainties in regional sea-level trend and acceleration estimates.
High-resolution remote sensing semantic segmentation is essential for land-cover mapping, urban monitoring, and object-level geospatial analysis, but accurate prediction remains difficult because remote sensing images often contain complex backgrounds, shadows, weak object contrast, and complex texture variations. Moreover, spatial details lost during feature downsampling cannot be fully recovered by subsequent decoding. As a result, coarse predictions usually contain two coupled residual problems: spatial boundary displacement and local semantic inconsistency. To address these problems, we propose DGSRef (Decoupled Geometric-Semantic Refinement Network), a lightweight attachable refiner for improving coarse predictions from existing segmentation models. DGSRef treats coarse logits as semantic priors and refines them through two decoupled stages. In the geometric alignment stage, a displacement field is predicted to warp coarse logits in the output space, modeling boundary correction as spatial transport rather than direct reclassification. A Multi-Scale Semantic-Guided Structural Difference (MSGSD) module further provides semantic-guided structural cues for displacement estimation. In the semantic residual stage, gated residual logits are predicted to correct remaining local semantic inconsistencies without globally overwriting the aligned prediction. Experiments on ISPRS Vaihingen, ISPRS Potsdam, and LoveDA show that DGSRef improves diverse segmentation architectures with limited additional computation and parameters, confirming its effectiveness as a lightweight decoupled refinement framework.
To address the limitation of traditional cellular automata models in effectively integrating temporal and spatial information, this study extends the previously developed Land use Simulation and Decision-Support system (LandSDS). By incorporating a graph attention network (GAT), a transformer, and an agent-based model (ABM) into a cellular automata framework informed by remote sensing time series, GT-LandSDS is constructed. Specifically, GAT dynamically captures higher-order spatial dependencies among land parcels; the self-attention mechanism of the transformer extracts land use change characteristics from multi-period observations; and ABM captures human behavioral decisions of three types, namely traffic, resident, and government. Based on this framework, GT-LandSDS derives CA transition rules from multiple dimensions and enhances the dynamic exploration of land use change across space and time. Using Guangxi Zhuang Autonomous Region as a case study, the model was validated with remote sensing land use data from six periods (2000, 2005, 2010, 2015, 2020, and 2023), and scenario-based future predictions were generated. The results show that: (1) The overall accuracy reaches 0.926, while the Kappa coefficient is 0.820, and the figure of merit (FoM) for change simulation is 0.034, indicating a relative advantage over ANN-CA, LSTM-CA, and UESP in overall pattern simulation, although fine-scale change reproduction remains limited; (2) Three development scenarios were then assessed: continuing historical trends, theoretical high-intensity urban expansion, and karst landform conservation under a green transformation development policy. The land use pattern of the area from 2023 to 2035 was predicted. The findings reveal that accelerating urbanization leads to rapid expansion of construction land, increasing by more than 88% compared with 2023, and causes substantial cropland loss. In contrast, intervention through the green transformation development policy limits construction land growth to 26.5%, effectively curbing urban sprawl while protecting forest, grassland, and cropland resources in the karst landscape. This study offers new insights into land use change simulation in ecologically fragile regions subject to strong policy interventions. It provides a scientific basis for coordinating ecological conservation and high-quality development in karst areas.
Wildfire impact assessment requires information on both burned areas and the biomass exposed within burned landscapes. We developed a field-calibrated, multi-source remote-sensing framework to estimate above-ground biomass (AGB) and quantify annual wildfire-related potential AGB exposure across the boreal forests of Quebec and Ontario, Canada, during 2018–2024. The dataset comprised 3725 plot-year AGB observations linked to optical, Sentinel-1 C-band, ALOS L-band synthetic aperture radar, environmental, and geographic predictors. Product-wise screening reduced the 91 candidate predictors to 28. An optimized extreme gradient boosting (XGBoost) model was evaluated using five-fold grouped cross-validation, with repeated observations from each plot assigned to a single fold. The model achieved an RMSE of 25.08 ± 0.36 t ha−1, an MAE of 20.89 ± 0.39 t ha−1, and an R2 of 0.53 ± 0.02. The full multi-source configuration outperformed all reduced-source and source-only configurations, while removing ALOS L-band SAR or environmental/geographic predictors produced among the largest performance declines. The model was applied to 9937 land-cover-stratified points within wildfire polygons using predictors from the year preceding each fire. Under the complete-loss assumption, cumulative potential AGB exposure was 269.20 Mt across 6.66 Mha of effective burned area, with a 95% bootstrap interval of 254.19–284.06 Mt reflecting finite-point sampling uncertainty and an area-weighted mean exposure intensity of 40.43 t ha−1. The 2023 fire season accounted for 206.44 Mt, representing 76.7% of cumulative exposure and 73.2% of effective burned area. Effective burned area and total potential exposure were strongly correlated (r = 0.99), whereas exposure intensity followed a distinct pattern and peaked in 2022 at 46.86 t ha−1. Thus, burned area was the primary correlate of regional potential biomass exposure, whereas exposure intensity reflected variation in pre-fire biomass among burned landscapes. These estimates represent potential exposure rather than measured combustion, mortality, or carbon emissions and demonstrate the value of integrating spatially explicit pre-fire AGB with wildfire perimeters.
Multispectral pan-sharpening aims to fuse high-resolution panchromatic and low-resolution multispectral imagery. However, this process introduces spatial artifacts and spectral distortions. Assessing the quality of fused images remains a fundamental challenge due to the absence of full-resolution ground-truth data. This paper provides a comprehensive review of Image Quality Assessment (IQA) frameworks tailored for pan-sharpened imagery. After overviewing major fusion approaches, including Component Substitution (CS), Multi-Resolution Analysis (MRA), Variational Optimization (VO), and Deep Learning (DL), the review analyzes the evaluation techniques used to benchmark them. It then systematically examines the evolution of evaluation protocols, from classical reference-based metrics relying on Wald’s protocol to full-resolution consistency models and recent no-reference (NR) algorithms. The analysis highlights critical methodological bottlenecks within the field, including unrealistic scale-invariance assumptions in consistency-based metrics, dependence on arbitrary parameters, and severe cross-sensor overfitting in deep learning approaches. Furthermore, the review addresses the mismatch between mathematical fidelity, human visual perception, and practical applicability. Finally, it outlines future research directions, focusing on spatial quality mapping and task-driven assessment protocols that validate fusion efficacy based on its impact on automated remote sensing applications.
Biodiversity monitoring is essential for evaluating mine rehabilitation success. Traditionally, assessments have relied on ground-based plot measurements, but advances in remote sensing offer opportunities to complement or replace plot-based surveys with spatially continuous monitoring approaches. We evaluated drone-derived multispectral imagery, the Soil Adjusted Vegetation Index (SAVI), and canopy height models (CHM) for mapping plant species used in mine rehabilitation in central Queensland, Australia. Ten classification models tested four combinations of spectral and structural data. Incorporating CHM improved overall accuracy by up to 10%, with notable gains for vegetation classes containing Eucalyptus and Acacia species. Species-level accuracy ranged from 79% to 100% for Eucalyptus and 74% to 100% for Acacia species. Misclassification was greatest among closely related red gums (Eucalyptus tereticornis Sm. and Eucalyptus camaldulensis Dehnh.) and Corymbia citriodora (Hook.) K.D.Hill & L.A.S.Johnson. These results demonstrate that structural information substantially improves species discrimination and has the potential to enhance biodiversity monitoring across plot (500 m²), block (1–100 ha), and landscape (100–1000 ha) scales.
Near-infrared imagery is essential for vegetation monitoring, precision agriculture, and environmental remote sensing, but multispectral UAV systems remain significantly more expensive and less accessible than conventional RGB imaging platforms. This study presents a lightweight artificial intelligence framework for reconstructing the NIR spectral band exclusively from the red spectral band acquired by a UAV. The proposed methodology formulates the reconstruction task as a pixel-wise nonlinear regression problem and employs a compact multilayer perceptron (MLP) containing only 609 trainable parameters, without exploiting spatial neighborhood information. The framework was developed and evaluated using 280 synchronized multispectral UAV image sets acquired with a DJI Phantom 4 Multispectral platform over a heterogeneous agricultural landscape in the Republic of Moldova. Of these, 252 image sets were used for model development, and 28 were reserved as a held-out within-mission test subset. Quantitative evaluation on a held-out test dataset from the same acquisition mission yielded a mean squared error of 0.010329, a root mean squared error of 0.101632, a mean absolute error of 0.079883, a coefficient of determination of 0.253383, and a Pearson correlation coefficient of 0.683637 between measured and reconstructed normalized NIR digital intensities. The results indicate that the model captures part of the red–NIR relationship under the evaluated acquisition conditions; however, the moderate coefficient of determination suggests that the reconstructed values are an approximation rather than a replacement for measured NIR observations. An illustrative NDVI-based assessment showed that broad spatial vegetation patterns remained identifiable. Rather than introducing a new neural network architecture, this work establishes a compact empirical baseline to investigate the practical performance and limitations of pixel-wise NIR reconstruction from a single red-band value with minimal model complexity.
Semantic segmentation of large-scale power-corridor LiDAR point clouds is essential for remote sensing-based transmission line inspection, vegetation encroachment monitoring, and intelligent grid maintenance. However, existing methods still struggle with massive data volumes, severe class imbalance, sparse power-related objects, and high computational cost in power corridor scenes. To address these challenges, this article proposes GSSP-KAN, an efficient semantic segmentation network that integrates Kansformer with grouped separable sparse convolution. The Kansformer module enhances nonlinear feature representation and contextual modeling, while the Grouped Separable Sparse Convolution Block (GSSP_Block) reduces redundant self-attention computation and preserves fine-grained local geometric structures. GSSP-KAN is evaluated on four large-scale datasets, including NW-3D, NeiMeng-3D, Nanning, and Toronto-3D. Experimental results show that GSSP-KAN achieves 98.20% OA/88.50% mIoU on NW-3D, 99.96%/98.58% on NeiMeng-3D, 98.20%/96.70% on Nanning, and 97.90%/83.80% on Toronto-3D. Compared with the baseline, the proposed model reduces the parameter count to 14.7 M and accelerates inference by 24.0% on NW-3D and 30.8% on Toronto-3D.
