Search Results (4,628)

Remote sensing research increasingly relies on heterogeneous satellite, UAV, hyperspectral, multispectral, SAR, and environmental monitoring data to support land, urban, hydrological, and environmental applications. However, these data are often affected by sensor differences, spatial and temporal heterogeneity, missing observations, irregular sampling, noise, and non-stationary environmental processes. This systematic review was conducted within the context of the Romanian Hub for Artificial Intelligence (HRIA) project, which supports the development of strategic artificial intelligence technologies. The review synthesizes current research on adaptive neural networks for remote sensing and Earth observation, with particular attention to Liquid Neural Networks and related continuous-time neural models. A systematic search was conducted across IEEE Xplore, Scopus, Web of Science, ScienceDirect, SpringerLink, Wiley Online Library, Google Scholar, and reference lists. After duplicate removal, screening, and full-text assessment, 61 studies published between 2018 and 2026 were included in the qualitative synthesis. The findings show that adaptive neural networks have gained increasing attention after 2022 and are mainly applied to image-centered remote sensing tasks, including classification, mapping, object detection, segmentation, enhancement, and change detection. Most studies adapt established deep learning architectures through multi-scale processing, adaptive feature fusion, attention mechanisms, graph relationships, or task-specific refinement. Continuous-time models are used less frequently but are relevant for irregular observations and dynamic environmental processes. Liquid Neural Networks remain emerging, and current evidence suggests only preliminary, task-specific relevance for irregular, noisy, multimodal, and dynamic remote sensing applications. Full article

Mycotoxins are widespread contaminants in animal feeds and continue to pose serious threats to animal welfare, production efficiency, food security, and sustainable economic growth worldwide. Chickens are highly susceptible to multiple mycotoxins, such as aflatoxins (AFLs), deoxynivalenol (DON), fumonisins (FBs), zearalenone (ZEA), ochratoxin A (OTA), and T-2 toxin. Exposure to these toxins can damage intestinal integrity, compromise immune functions, impair nutrient absorption, and ultimately reduce production efficiency even at subclinical concentrations. Conventional mycotoxin detection methods, such as enzyme-linked immunosorbent assays (ELISA) and chromatographic techniques, provide accurate quantification but remain expensive, labor-intensive, time-consuming, and inefficient for large-scale screening, particularly when masked mycotoxins are present. The frequent co-occurrence of multiple mycotoxins further complicates risk assessment and effective management. Emerging analytical technologies, including hyperspectral imaging, biosensors, Internet of Things (IoT)-based platforms, and machine-learning algorithms, offer promising advancements for rapid detection and predictive risk forecasting. This review summarizes the toxicological impacts of major mycotoxins on poultry, outlines critical challenges in detection and prevention, and evaluates current artificial intelligence (AI) and machine learning (ML) approaches for mycotoxin identification, prediction, and management. A systematic literature search conducted across PubMed, ScienceDirect and Google Scholar identified 176 unique studies published between 2010 and 2026. Key knowledge gaps include limited availability of high-quality datasets and source code, inconsistent model interpretability, and poor reproducibility across production environments. By integrating conventional toxicology with data-driven approaches, this review highlights how predictive modeling can strengthen mycotoxin surveillance and support proactive mitigation strategies in modern poultry production systems. Full article

The eddy covariance (EC) technique is a key tool in environmental monitoring, enabling continuous and non-invasive measurement of carbon dioxide (CO₂) fluxes at the ecosystem–atmosphere interface. In environments characterized by high levels of airborne particulates, such as volcanic regions, the reliability of enclosed-path EC measurements can be compromised by frequent filter clogging, potentially affecting data continuity, and increasing maintenance requirements. This study investigates whether the chemical and mineralogical signatures of particulate matter accumulated on clogged Swagelok pre-Licor filters can be used to identify dominant particle sources and provide insights into filter clogging processes. A multi-analytical workflow combining scanning electron microscopy with energy-dispersive X-ray spectroscopy (SEM–EDS), portable Raman spectroscopy, and hyperspectral imaging (HSI) was applied to recovered filter residues. The combined approach provided complementary chemical, mineralogical, and morphological information, allowing discrimination among volcanogenic material (e.g., glass shards, crystals, and lithic fragments), aeolian lithogenic dust, including Saharan inputs, and biogenic particles such as plant fibers. The results revealed two dominant particulate groups, volcanogenic mineral phases and biogenic material, with a minor contribution from wind-transported lithogenic dust. Volcanogenic phases, enriched in Si, Al, and Fe, dominated the inorganic fraction, whereas O-, C-, and N-rich particles were mainly associated with local biogenic sources. No clear evidence of significant anthropogenic contributions was identified. These findings demonstrate that multi-analytical characterization of particles accumulated on EC pre-filters can provide qualitative source attribution and valuable information on the processes responsible for filter loading and clogging. By linking particle characteristics with meteorological and environmental conditions, this approach has the potential to support site-specific, predictive, and event-driven maintenance strategies, contributing to improved EC data quality and more efficient long-term monitoring in high-aerosol environments. Full article

Hyperspectral retrieval of nitrogen-related vegetation variables has undergone rapid methodological advances driven by the emergence of protein-sensitive radiative transfer models (RTMs), modern machine learning (ML), and operational imaging spectroscopy. This review synthesizes recent developments in hyperspectral retrieval of nitrogen-related vegetation variables across leaf and canopy scales, with particular emphasis on advances reported between 2020 and 2026. We examine the evolution from classical parametric regression and nonlinear ML approaches towards physically based RTM inversion and hybrid RTM–ML frameworks that integrate the complementary strengths of physical modeling and statistical learning. Particular attention is given to protein-sensitive RTMs, advanced ML approaches, and uncertainty-aware retrieval. Recent developments highlight the potential of hybrid RTM–ML frameworks to combine physical consistency with computationally efficient statistical learning, while probabilistic methods such as Gaussian Process Regression provide additional capabilities for uncertainty characterization. The review further discusses the transition from experimental studies to operational applications enabled by airborne and satellite imaging spectroscopy, including PRISMA, EnMAP, and forthcoming missions such as CHIME. Remaining challenges include the inherently ill-posed nature of nitrogen retrieval, limited and insufficiently representative calibration data, uncertainty characterization, and generalization across sensors, species, and ecosystems. Overall, the reviewed evidence points towards increasingly integrated retrieval frameworks, while emphasizing that robust transferability and operational implementation remain dependent on representative data, physical realism, and rigorous uncertainty assessment. Full article

Hyperspectral images contain abundant spectral information and provide fine-grained spatial representations. However, their high dimensionality, severe spectral redundancy, subtle inter-class differences, mixed boundary regions, and limited labeled samples pose significant challenges to accurate classification. Convolutional neural networks (CNNs) have limited capability in modeling non-Euclidean structural relationships, whereas graph convolutional networks (GCNs) are susceptible to the quality of superpixel segmentation and noise propagation over graph structures. To address these issues in hyperspectral image classification, this paper proposes a Reliability-Aware Cross-Branch Contrastive Graph Convolutional Network (RACB-CGCN). The proposed method employs a dual-branch CNN–GCN architecture to extract pixel-level local spectral–spatial features and superpixel-level structural features, respectively. A superpixel reliability estimation and propagation control mechanism is introduced to assess node reliability based on the discrepancy between pixel-level features and superpixel-reconstructed features. This mechanism effectively suppresses the propagation of noisy information caused by impure superpixels and mixed boundary regions. Meanwhile, a cross-branch supervised contrastive learning strategy is developed to enhance semantic consistency between the CNN and GCN branches, thereby improving intra-class compactness and inter-class separability. In addition, a class-adaptive fusion module is designed to dynamically adjust the contributions of the two branches according to the feature characteristics of different land-cover classes. Experimental results demonstrate that the proposed method effectively exploits the complementary information between pixel-level fine-grained features and superpixel-level structural features, leading to improved classification accuracy and robustness in hyperspectral image classification. Full article

Background/Objectives: Estimating the post-mortem interval (PMI) of human skeletal remains remains challenging because bone undergoes structural, molecular and optical changes that evolve differently over time and are strongly influenced by taphonomic conditions. Rather than assuming that a single multimodal classifier performs equally well across all PMI intervals, this study aimed to determine which imaging or spectroscopic modality is most informative for specific forensic decision tasks and to derive an exploratory task-specific diagnostic decision-support framework based on internally evaluated sample-level analyses. Methods: Human femoral bone samples were assigned to five PMI classes ranging from 0–2 weeks to >100 years and examined using micro-computed tomography, hyperspectral imaging, handheld and microscopic Raman spectroscopy, and NIR-ONE spectroscopy. Repeated acquisitions were aggregated at the physical-sample level. Four targeted diagnostic contrasts were defined for the present reanalysis: archaeological class 5 versus classes 1–4, early classes 1 + 2 versus later classes 4 + 5, classes 1 + 2 versus class 4 within the Raman-compatible forensic range, and class 1 versus class 2. In addition, an exploratory direct pairwise comparison of class 4 versus class 5 was performed to specifically assess the boundary between the latest forensic interval and archaeological material. Parameter-wise ROC analysis, bootstrap confidence intervals, exploratory operating points at approximately 95% specificity, and repeated stratified cross-validation were used. All preprocessing for multivariate modelling was performed within the respective cross-validation folds. Results: Diagnostic performance was strongly task-dependent. Archaeological class 5 was distinguished from classes 1–4 by micro-CT Mean2 (AUC 0.984), NIR-ONE reflectance at 1944 nm (AUC 0.980), and HSI-derived tissue water index (TWI; AUC 0.961). In the additional exploratory direct C4-versus-C5 analysis, micro-CT Mean2 showed complete separation of the available samples (AUC 1.000), while NIR-ONE reflectance at 1944 nm retained excellent discriminatory performance (AUC 0.963). In contrast, HSI-derived TWI showed only moderate direct C4-versus-C5 discrimination (AUC 0.742). For early classes 1 + 2 versus later classes 4 + 5, the device-derived HSI StO₂ index showed the highest univariate performance (AUC 0.940), followed by TWI (AUC 0.894) and the NIR spectral slope between 1550 and 1950 nm (AUC 0.860). Within the Raman-compatible forensic range, HSI remained highly informative, while Raman carbonate/phosphate and crystallinity parameters provided complementary molecular information. Class 1 versus class 2 discrimination remained moderate. Corrected five-class modelling achieved only moderate balanced accuracy and did not consistently improve upon NIR-ONE alone. Conclusions: Diagnostic performance was strongly task-dependent. In this internally evaluated cohort, micro-CT Mean2 and high-wavelength NIR-ONE features showed the strongest discrimination of archaeological-compatible C5 material, whereas HSI-derived optical indices were most informative for the separated early-versus-later contrast and Raman spectroscopy provided complementary molecular information within C1–C4. Because C5 comprised only six unique physical samples and archaeological context was confounded with chronological age, the very high C5-related AUC estimates should be regarded as exploratory, hypothesis-generating estimates rather than validated measures of chronological PMI. The proposed decision-support framework is likewise exploratory and requires independent external validation before forensic implementation. Full article

The economic impact of food fraud is difficult to quantify precisely, because fraud is structurally designed to evade detection; available estimates are indirect projections rather than direct forensic accounting and are commonly cited in the range of USD 10–15 billion annually. The integration of artificial intelligence (AI) with analytical instrumentation has generated a rapidly expanding body of research aimed at detecting adulteration, mislabeling, and substitution across food matrices. This systematic review examines the extent to which AI-assisted instrumental technologies contribute to food fraud prevention (as distinct from laboratory detection) and characterizes the structural factors that constrain real-world translation. A systematic search of the peer-reviewed literature published between 2021 and 2026 yielded 83 eligible records (80 primary studies and 3 review articles) after applying predefined inclusion criteria. Data were extracted into a structured seven-sheet workbook covering study characteristics, instrumental technologies, AI architectures, performance metrics, industrial-validation status, implementation evidence, and methodological quality. The corpus shows consistently high reported analytical accuracy under controlled laboratory conditions (median of extractable classification accuracies ≈ 99–100%; ≥95% in 86% of studies with an extractable value). At the same time, 68 of 83 studies (82%) reported no external validation, no study (0/83) achieved inter-laboratory validation, no study documented routine-monitoring application, and only one study reported testing in a genuine industrial environment. The most frequently featured platforms were NIR spectroscopy and electronic-nose arrays (each featuring in 30/83 studies, frequently in data-fusion combinations), followed by gas-chromatography-based systems (16/83) and hyperspectral imaging (13/83). Classical machine learning predominated (57/83 studies coded as classical ML, with a further 11 hybrid ML/DL designs and 12 deep-learning-only designs). A direct statistical comparison found no significant difference in reported accuracy between classical-ML and deep-learning studies (median 100% vs. 98.2%; Mann–Whitney U test, p = 0.16). A pre-specified test of the hypothesis that high reported accuracy is itself a marker of overfitting was not supported by the corpus: reported accuracy was not negatively associated with external-validation status (Fisher’s exact p = 0.51) or with methodological-quality score (Spearman ρ = 0.15, p = 0.23). Methodological quality was predominantly moderate (49/83 scored 3/5; 22 scored 2/5; 11 scored 4/5; one study scored 5/5), and 19/83 (23%) carried a high risk of bias. The review’s central observation—a measurable gap between demonstrated laboratory detection and evidenced real-world prevention—is well supported by the deployment, inter-laboratory, and routine-monitoring data. We deliberately separate this strongly evidenced conclusion from weaker inferences (e.g., the overfitting hypothesis) that the corpus cannot currently establish, and we outline a validation-driven, deployment-oriented research agenda. Full article

The Sadiola Hill gold deposit is located in Mali in the Kédougou–Kénieba Inlier (KKI) of the West African craton. The gold deposit is hosted in metamorphosed carbonaceous and clastic sedimentary rocks that are intruded by diorite dykes and granitoids. The deposit records a multistage history of alteration, including complex compositional variation and speciation of carbonate group minerals and white and dark micas. We present the results of spectral and petrographic studies of 36 drill core samples from six diamond boreholes from the Sadiola Hill gold mine to (1) refine paragenetic studies and (2) define alteration assemblages accompanying the main gold event. The methodologies used included high-resolution short-wave infrared (SWIR) hyperspectral imaging, electron microprobe analysis, and conventional petrography. The alteration assemblages that correspond with the gold mineralizing event include ferroan dolomite, dolomite, ankerite, Al-poor muscovite/illite to muscovite/muscovitic illite, and phlogopite. Pervasive and vein-type carbonate alteration includes ferroan dolomite–ankerite in greywacke and dolomite in the marble (with minor Fe²⁺ substitutions). The texture and spectral character of pervasive hydrothermal dolomite differ from metamorphic dolomite. In the greywacke, it is dominantly expressed as Au-stage Al-poor illite and Al-poor muscovite. In marble, it is expressed as moderately crystalline muscovitic illite and muscovite. In well-mineralized marble samples, white mica is notably more crystalline and aluminous. Pre-gold Mg-rich illite is fully consumed. Phengite, calcite, and chlorite occur late in the paragenetic sequence irrespective of lithology and post-date the main gold event. Results demonstrate that the integrated study of hyperspectral imaging with petrography and microprobe analysis characterizes the alteration mineralogy of the gold-mineralizing event and refines the paragenetic sequence of the gold deposit. Full article

Lipid metabolites in maize kernels determine grain quality by influencing nutritional value, oxidative stability, and post-harvest deterioration, making their profiling essential for quality improvement and breeding. This study applies hyperspectral imaging (HSI) with a spectral range of 900–1700 nm to detect maize lipid metabolites. Six lipid metabolites—9-Octadecynoic acid (stearolic acid), pinolenic acid (Δ5,9,12 18:3), N-Acylethanolamine (16:0), N-Acylethanolamine (18:0), N-Acylethanolamine (18:1), and propionic acid—were selected due to their strong relevance to maize kernel quality and favorable spectral response. First, two-trace two-dimensional (2T2D) correlation spectroscopy with heterogeneous preprocessing is employed to capture both synchronous and asynchronous correlations across different preprocessing spectra. A convolutional autoencoder (CAE) was subsequently used to extract low-dimensional latent features from heterogeneous 2T2D-COS representations, followed by regression modeling using random forest (RF), support vector regression (SVR), gradient boosting (GB), and partial least squares regression (PLSR). A total of 82 maize seed varieties were employed for experimental validation. Compared with one-dimensional spectral, homogeneous preprocessing, and PCA-based feature extraction, the proposed approach provided improved predictive performance across the six lipid metabolites, with the optimal CAE-based models achieving R_P² values of 0.629–0.887, RMSEP values of 0.241–0.565, and RPD values of 1.656–2.069. Overall, this approach provides a rough screening solution for metabolite prediction in maize crop. Full article

Hyperspectral image (HSI) classification benefits from rich spectral information; however, high dimensionality of HSI data increases computational cost, noise sensitivity, and the risk of overfitting when labeled samples are limited. Most pretrained computer vision networks are designed for three-channel inputs, making direct application to hyperspectral cubes difficult. Conventional principal component analysis (PCA) ranks components by total variance without distinguishing useful signal variance from noise-related variance, which can reduce the reliability of the resulting representation when only a few components are retained. This paper proposes a data-augmented Noise-Adjusted Principal Component Analysis (DA-NAPCA) framework for deep learning-based HSI classification. By accounting for estimated noise covariance, NAPCA orders the transformed components by signal-to-noise ratio rather than total variance, while data augmentation mitigates the overfitting risk when labeled samples are limited. Unlike typical NAPCA/MNF applications, which select the number of retained components empirically, DA-NAPCA deliberately retains three noise-adjusted components to form a compact three-channel representation, enabling pretrained models designed for three-channel inputs to be fine-tuned without modifying their input layers. The framework is evaluated using a 3D convolutional neural network (3D-CNN) for spatial–spectral feature learning and a pretrained EfficientNet-B0 model for lightweight transfer learning. Although this paper uses 3D-CNN and EfficientNet-B0 as illustrative examples, the proposed DA-NAPCA framework is a representation-level preprocessing approach and does not require architecture-specific modification. Experiments conducted on the Indian Pines, University of Pavia, and Salinas datasets compare DA-NAPCA with RGB, band selection, PCA-based dimensionality reduction, and ablation variants. Across the three datasets, DA-NAPCA achieved mean overall accuracies of 93.11–94.71% with 3D-CNN and 95.93–97.44% with EfficientNet-B0. Compared with the second-best baseline method, DA-NAPCA improved overall accuracy by 2.75–7.58 percentage points with 3D-CNN and 1.28–2.12 percentage points with EfficientNet-B0. These results demonstrate that combining a compact noise-adjusted representation with spatial augmentation provides an effective input representation for deep learning-based HSI classification. Full article

We present the vicarious calibration and validation of atmospheric correction for the airborne Watersat Imaging Spectrometer Experiment (WISE) sensor, deployed over the south-west shores of Anticosti Island (Gulf of Saint Lawrence). Coincident in situ water-leaving reflectance was acquired from the research vessel Coriolis II (HyperSAS, above-water radiometry) and from a jet-ski (underwater radiometry). The HyperSAS Hyperspectral Surface Acquisition System dataset was used for calibration, while the jet-ski dataset was split into 30% calibration and 70% validation. We evaluated Ratio and Empirical Line (EL) calibration approaches and investigated the influence of residual sun glint. Modeled vs. measured at-sensor reflectance residuals show a clear dependence on angular distance to the specular direction, demonstrating that geometry-dependent glint bias affects vicarious gain estimation. The Ratio approach performs adequately with a limited calibration range (∼50% accuracy at 500 nm) but degrades when applied to a broader matchup dataset. In contrast, the EL method provides more stable gains by accommodating both additive and multiplicative discrepancies (∼20% accuracy at 500 nm). These results highlight the importance of accounting for residual sun glint in airborne aquatic imaging spectroscopy (for calibration or other purposes) and support the EL with Standard Major Axis regression as the most robust calibration strategy for quantitative retrieval of water-leaving reflectance from WISE. Full article

Alzheimer’s disease (AD) pathology is increasingly recognized to manifest in the retina, offering a non-invasive window for early biomarker discovery. This proof-of-concept study investigated whether multimodal retinal imaging—hyperspectral imaging (HSI), optical coherence tomography (OCT), and color fundus photography (CFP)—can differentiate individuals with and without cerebral amyloid-beta (Aβ) pathology in 40 participants with PET-confirmed Aβ status (17 Aβ+, cognitively normal or with mild cognitive impairment; 23 Aβ− cognitively normal controls). HSI-derived gray-level co-occurrence matrix (GLCM) texture, OCT-derived ganglion cell–inner plexiform layer (GC-IPL) thickness, and CFP-derived vascular biomarkers (VBMs) were extracted, and logistic regression with leave-one-out cross-validation assessed classification performance per modality, alone and combined; the cohort was supplemented with AD dementia patients for an exploratory cross-sectional comparison across disease-stage groups. HSI showed nominally lower GLCM correlation at 466 nm in Aβ+ participants, most pronounced in the inferior macula (AUC = 0.72). GC-IPL thickness showed a similar inferior-predominant regional pattern. Combining HSI and GC-IPL features yielded the best performance (AUC = 0.84; sensitivity = 0.82; specificity = 0.78), although this improvement over the unimodal models did not reach statistical significance, whereas vascular biomarkers contributed minimally. In an exploratory cross-sectional comparison across AD stage groups drawn from two cohorts, HSI features showed a non-monotonic pattern, decreasing in early Aβ+ stages and rising again in dementia. These findings provide preliminary evidence of complementary information between HSI and OCT for detecting retinal biomarkers of early-stage AD, supporting multimodal retinal imaging as a scalable screening approach warranting validation in larger, longitudinal cohorts. Full article

Hyperspectral technology has become an important method for identifying wild and cultivated Ophiocordyceps sinensis, but existing studies mainly focus on intact samples. In the actual circulation process, Ophiocordyceps sinensis often breaks, and samples missing the stroma part but retaining the larva part still possess practical identification value. However, accommodating the identification of both intact samples and stroma-missing samples requires clarifying the difference in spectroscopic identification information contribution between the larva part and the stroma part, thereby optimizing the data acquisition strategy for specialized Ophiocordyceps sinensis detection instruments. Based on the segmentation of hyperspectral images of intact Ophiocordyceps sinensis, this study constructed simulated single and mixed part datasets for machine learning model training to analyze the spectral feature differences of different parts. Furthermore, real stroma-missing samples were used to verify the identification capability of the models in the practical scene. Finally, the optimal detecting part was determined by comprehensively considering the highest identification accuracy and the maximum application scope. Both the larva part and the stroma part exhibit effective spectroscopic identification information contribution for identifying wild and cultivated Ophiocordyceps sinensis. However, the spectroscopic identification information contribution of the larva part is greater than that of the stroma part. Multiple comparative experimental results show that the model trained with intact samples can be directly used for the identification of stroma-missing Ophiocordyceps sinensis with the highest accuracy of 98.52%. This conclusion provides a basis for the practical application of hyperspectral technology in Ophiocordyceps sinensis. Full article

Hyperspectral (HS) sensing provides detailed high-dimensional spectral data to identify subtle anomalies and material variations in complex scenes. HS anomaly detection (HS-AD) aims to identify small, spectrally distinct objects or materials in HS imagery without prior target information. However, most HS-AD methods assume that the training and test data are drawn from the same distribution (single-domain). This assumption is often violated in operational sensing because of differences in sensor characteristics, scene composition, atmospheric conditions, and acquisition geometry. To address this challenge, we propose Generalized MoE-AD (GMoE-AD), a neural Mixture-of-Experts (MoE) architecture for robust cross-domain hyperspectral anomaly detection. The framework fuses the outputs of six unsupervised base detectors with representations from a pretrained HS foundation model, combines four neural experts through learned top-2 routing, and applies gradient-reversal-based domain-adversarial training to improve robustness under distribution shifts. A drift-aware test-time adaptation (DTA) variant is evaluated separately. We evaluate GMoE-AD on six public real-world HS benchmark datasets—San-Diego, Salinas, HYDICE-Urban, ABU-Airport, ABU-Beach, and ABU-Urban—and one private Arizona dataset comprising 22 images acquired by four different sensors. Under all-domain training, the model uses the training portions of all seven datasets and is evaluated without access to domain identity or dataset-specific information. GMoE-AD achieves an average ROC-AUC of 0.943, PR-AUC of 0.623, and F1-macro of 0.821 in this setting. In a leave-one-dataset-out (LODO) evaluation, where the target dataset is completely unseen during training, the model maintains an average ROC-AUC of 0.910 and F1-macro of 0.773. The optional DTA variant increases mean ROC-AUC from 0.938 to 0.953 but reduces F1-macro from 0.822 to 0.815, indicating a metric- and dataset-dependent adaptation trade-off. These results suggest that combining neural expert routing, transfer learning, and domain-adversarial representation learning improves robustness across heterogeneous HS datasets and provides a unified approach to anomaly detection in high-dimensional sensor data, supporting deployment in real-world hyperspectral applications where training and deployment conditions differ. Full article

Pine wilt disease (PWD) is a devastating infectious disease of pine trees caused by the invasion of Bursaphelenchus xylophilus. Achieving rapid and accurate identification of pine trees in the early stages of infection is critical for preventing and controlling its spread, particularly for early warning and intervention before the plants exhibit obvious discoloration symptoms. The Plant Senescence Reflectance Index (PSRI) has shown significant potential for the early detection of PWD. However, existing studies often use its default band combinations for calculation, which may fail to fully exploit key wavelength information that is more sensitive to pre-symptomatic PWD stress. Based on unmanned aerial vehicle (UAV) hyperspectral imaging data, the present work systematically optimizes and reconstructs the three band parameters of the PSRI to explore an optimal wavelength combination more suitable for the early identification of PWD-infected trees. The results indicate that compared to the original settings of the PSRI, the wavelength combination of 490-666-700 nm performs better in pre-visually identifying infected pine trees in the early stages of infection, achieving a detection accuracy of 82.71%. Our work identifies an optimized PSRI wavelength combination more sensitive to the pre-visual early stage of PWD based on hyperspectral data. This method demonstrates promising potential for pre-visual detection of PWD before visible symptoms appear, which may provide an earlier opportunity for PWD monitoring and intervention. Full article