Search Results (132)

Plant secondary metabolites play important roles in plant defense, environmental adaptation, and ecosystem functioning, yet large-scale monitoring of foliar phenolics remains limited because of the high cost and restricted spatial coverage of airborne imaging spectroscopy and the limited spectral resolution of multispectral satellites. This study explored a cross-scale remote sensing framework to map foliar phenolics through the synergy of airborne imaging spectroscopy and Sentinel-2 multispectral imagery. Foliar samples were collected from 634 plots across seven National Ecological Observatory Network (NEON) ecological domains in the United States, representing six plant functional types. Community-weighted mean foliar phenolic concentrations were linked with NEON Airborne Observation Platform (AOP) imaging spectroscopy to develop phenolic retrieval models using partial least squares regression (PLSR) and Gaussian process regression (GPR). The optimized airborne-derived phenolics were subsequently aggregated across multiple spatial windows and used as reference data to train Sentinel-2 models using PLSR, random forest regression (RFR), and GPR. Both airborne hyperspectral models achieved strong predictive performance, with comparable accuracy between PLSR (R² = 0.770, RMSE = 16.11 mg·g⁻¹) and GPR (R² = 0.771, RMSE = 16.16 mg·g⁻¹). However, PLSR showed substantially lower predictive uncertainty (4.62 mg·g⁻¹) than GPR (12.58 mg·g⁻¹), indicating more stable predictions across NEON samples. Spectral importance analysis identified consistent phenolic-sensitive wavelength regions in the visible and shortwave infrared domains, particularly near previously reported absorption features. For Sentinel-2 upscaling, prediction accuracy increased consistently with larger spatial aggregation windows, indicating improved agreement between Sentinel-2 observations and airborne-derived phenolics through reduced spatial scale mismatch and geolocation misalignment. Among the evaluated approaches, RFR achieved the best performance, improving from R² = 0.479 at the 10-pixel window to R² = 0.776 (NRMSE = 7.0%) at the 100-pixel window. Feature importance analysis showed increasing contributions of red-edge and shortwave infrared information at larger aggregation scales. Spatial comparisons demonstrated that Sentinel-2 successfully reproduced major phenolic distribution patterns observed by airborne imaging spectroscopy. These results demonstrate that airborne imaging spectroscopy can effectively bridge field observations and satellite multispectral imagery for foliar phenolics estimation and highlight the potential of Sentinel-2 as a scalable approach for monitoring vegetation chemical traits across heterogeneous ecosystems. Full article

The chlorophyll fluorescence parameter Fv/Fm, representing the maximum photochemical efficiency of photosystem II, is an important indicator for evaluating plant photosynthetic performance and stress responses. However, rapid and non-destructive monitoring of Fv/Fm at the leaf scale remains challenging because conventional fluorescence measurements are time-consuming and require specialized equipment. Although spectral techniques provide an efficient alternative, existing spectral-based models mainly rely on single-source information and often lack sufficient integration of environmental conditions and physiological interpretability. Therefore, this study aimed to develop an explainable multi-source information fusion framework by integrating leaf spectral characteristics and environmental variables for accurate and interpretable estimation of grape leaf Fv/Fm. The grape cultivar ‘Queen Nina’ grown under protected cultivation was used as the experimental subject in this study. Visible–near-infrared reflectance spectra, measured Fv/Fm values, and environmental variables were synchronously collected under different water-stress conditions. Sensitive wavelengths were extracted by multiplicative scatter correction (MSC), competitive adaptive reweighted sampling (CARS), and the successive projections algorithm (SPA), and an XGBoost model incorporating both spectral and environmental features was established. The results demonstrated that: (1) the proposed multi-source fusion strategy effectively integrated spectral and environmental information for Fv/Fm prediction, with 12 sensitive wavelengths identified by MSC-CARS-SPA; (2) the XGBoost-EF model achieved R², RMSE, and MAE values of 0.906, 0.0432, and 0.0352, respectively, under vine-level five-fold cross-validation, outperforming the spectral-only XGBoost model; and (3) SHAP analysis provided an interpretable explanation of model predictions by quantifying the contributions of key spectral and environmental features, highlighting the importance of leaf temperature and red-edge wavelengths. It is concluded that the accuracy, robustness, and interpretability of non-destructive Fv/Fm monitoring in grape leaves can be substantially improved through multi-source information fusion. Full article

The objective of this research is the development of a copper-based optical sensor for the detection of ovarian cancer marker CA 125 synthesized using low-temperature plasma. Optical materials produced with metals show unique advantages due to their ability to interact with light. There are different methods currently used for the synthesis of optical materials that can be associated with longer processing times and low material yield. The novelty of this study is the development of copper-based optical material (CuPy) using low-temperature plasma and subsequent modification for the detection of CA 125. Introduction: Plasma consists of a mixture of fully and partially ionized gas. It comprises diverse, highly energized species of atoms, ions, electrons, excited molecules, and charged species. These energized species are used to create new materials, for surface modifications, and in medical applications. Plasma can create a controlled environment for the creation of novel materials. Using low-temperature plasma, it will be possible to have precise control of the chemical composition and structure due to the creation of excited molecules, ions, and free radicals. Method: The CuPy material was synthesized using radio-frequency-assisted low-temperature plasma. Prior to synthesis, the plasma chamber was cleaned using radio frequency (RF) plasma without any reagents or gases. RF plasma was used for the synthesis of CuPy for 10 min and subsequent hydrogen plasma (50 sccm) for another 10 min. Two types of products were extracted from the chamber (one in water and another in methanol). These two products were analyzed using UV–visible absorbance spectroscopy, fluorescence spectroscopy, X-ray photoelectron spectroscopy (XPS), and Fourier transform infrared spectroscopy (FTIR). The methanol extracted samples were further modified with CA 125 antibody. Zeta potential measurements were performed to confirm the binding of the CA 125 antibody to the sensor. The sensing efficacy of the sensor towards CA 125 antigen was monitored using fluorescence spectroscopy. Results: The absorbance spectrum of methanol extracted CuPy shows absorbances around 251 nm, 282 nm, and 339 nm. The extracted product exhibited a red edge excitation emission in the visible region. The elemental composition and oxidation state of the sample were evaluated using XPS. CA 125 antibody conjugation with CuPy was confirmed using UV–visible absorbance spectroscopy, fluorescence spectroscopy, and FTIR spectroscopy. The antibody binding resulted in the fluorescence shifts towards higher wavelengths with an increase in the emission intensity compared with CuPy. Zeta potential measurements also confirmed the binding of the CA 125 antibody to the sensor. Different concentrations of CA 125 antigen resulted in the quenching of fluorescence. This change in the fluorescence intensity was used for the detection of CA 125. Conclusions: A copper-based optical material was developed using low-temperature plasma, and it was found to be effective for the detection of CA 125 ovarian cancer marker. Full article

Selective energy conversion in confined catalytic nanocavities is examined through a coupled reactive–photonic framework. The practical target is a combustor-integrated selective emitter for thermophotovoltaic (TPV) conversion and cascaded thermoelectric (TEG) recovery, in which Pt-coated anodic porous alumina (APA) functions simultaneously as a catalytic reactor, a cavity-modified electromagnetic environment and a heat-routing structure. Visible/near-infrared spectra (380–780 nm) show that Pt-coated APA exhibits a substantially stronger non-grey red-edge depression than a smooth zirconia reference. This observation establishes a spectral contrast in the measured window but is not used to identify an experimental cutoff wavelength, because a finite, open, lossy and array-coupled pore does not exhibit the abrupt edge predicted for an ideal cylindrical waveguide. For the mid-infrared, analytical scaling shows that the principal H₂O and CO₂ bands at 2.7, 4.3, 6.3 and 15.0 µm all lie deep in the evanescent regime relative to the ideal TE₁₁ cutoff wavelength λc ≈ 0.513 µm for a 300 nm pore. A converged finite-difference time-domain benchmark at the CO₂ 4.3 µm band yields a source-local Purcell factor F_p ≈ 0.38, indicating suppression of the total local density of optical states, while aperture flux is more than six orders of magnitude smaller than the near-field power budget. The specific contribution is therefore not the established fact of below-cutoff attenuation, but the co-design and separate quantification of a catalytic nanocavity as a reactive compartment, photonic environment and energy-branching element. The results provide a bounded mechanistic basis for combustor-integrated TPV and hybrid TPV/TEG architectures. Full article

Predicting canopy traits non-destructively is important for understanding crop growth and improving phenotyping efficiency. Hyperspectral reflectance provides detailed spectral information, but the role of band selection in regression-based trait prediction at the canopy scale remains unclear. In this study, we evaluated the effects of different band-selection algorithms on the prediction accuracy of aboveground biomass (AGB), leaf area index (LAI), and canopy cover (CC) in soybeans using canopy hyperspectral reflectance in the visible to near-infrared (VNIR) range from 501 to 801 nm. The dataset included multiple sites, years, cultivars, and irrigation treatments. We compared a full-band partial least squares regression (PLS) model with three band-selection methods (PLS-Variable Importance in Projection (VIP), Bootstrapped least absolute shrinkage and selection operator (LASSO) (BoLASSO), and an ensemble approach). Model performance was assessed using Kennard–Stone validation and leave-one-year-out cross-validation. The results showed that the effectiveness of band selection depended on the target trait. Full-band PLS performed well for AGB under Kennard–Stone validation, whereas BoLASSO achieved comparable accuracy to PLS for LAI and CC using a reduced number of selected bands. Leave-one-year-out cross-validation showed that year-to-year transferability was more difficult for AGB than for LAI and CC. The selected wavelengths were located mainly in the visible, red-edge, and near-infrared regions. These results indicate that band-selection strategies should be tailored to the target trait and that selected VNIR bands can provide candidate spectral regions for simplified sensing of soybean canopy traits. Full article

Urban freshwater ecosystems affected by eutrophication and recurrent algal blooms require monitoring approaches capable of representing optical complexity and spatial heterogeneity. This study evaluated an integrated workflow combining proximal in situ hyperspectral sensing, radiometric calibration, spectral filtering, predictor-band selection, data transformation, and machine-learning regression to estimate chlorophyll-a (chl-a) in a tropical eutrophic urban reservoir. Monthly field campaigns were conducted from September 2022 to February 2023, with simultaneous chl-a measurements and hyperspectral image acquisition. After preprocessing, noise removal, and exclusion of anomalous spectra, 82 matched hyperspectral–chl-a observations were retained for model development. Predictor bands were selected using Pearson correlation and F-test analysis, identifying five relevant wavelengths: 530, 535, 682, 687, and 732 nm. Multiple Linear Regression, Random Forest Regressor, Support Vector Regressor, and XGBoost Regressor were tested under different data transformations. The Support Vector Regressor with logarithmic transformation achieved the best performance, with R² = 0.86 and RMSE = 6.89 µg L⁻¹. The selected wavelengths correspond to spectral regions associated with green reflectance, red chl-a absorption, and red-edge/NIR responses in productive waters. The results indicate that proximal hyperspectral sensing combined with machine learning can support chl-a estimation in optically complex urban reservoirs and provide complementary information for eutrophication monitoring and bloom-management strategies. Full article

Canopy SPAD index is a practical indicator for evaluating the photosynthetic status and health of Malus sieversii, an endangered wild apple resource in Xinjiang. To develop a rapid and non-destructive monitoring approach, 255 canopy samples were collected across the flower fading stage, fruit stage, and fruit mature stage using synchronized UAV hyperspectral imaging and ground SPAD measurements. Spectral preprocessing, feature-band selection, regression modeling, and SHAP interpretation were evaluated using training-set optimization and independent test-set validation. SG-FD produced the strongest preprocessing response, with a maximum absolute correlation coefficient of 0.70. SiPLS reduced 220 effective bands to 84 wavelengths; subsequent CARS, GA, and SPA screening retained 28, 8, and 12 wavelengths, respectively. The SiPLS-CARS-based Transformer-LSTM model achieved the best performance, with R² = 0.91 and RMSE = 2.12 in training and R² = 0.86 and RMSE = 2.47 in testing. SHAP results indicated that red-edge wavelengths and visible sensitive bands contributed most to prediction. The proposed UAV hyperspectral and Transformer-LSTM framework provides an interpretable proof-of-concept method for canopy SPAD index estimation in Malus sieversii and supports non-destructive monitoring of wild fruit forest health. Full article

River systems and their transitional zones near estuaries are characterized by strong spatial and temporal variability in both hydro-chemical and optical conditions. These dynamics make the monitoring of key water quality indicators such as chlorophyll-a (Chl-a) particularly demanding. Unmanned aerial systems (UASs) equipped with multispectral sensors have increasingly been used to address these challenges, providing high spatial resolution observations in environments where satellite imagery is often constrained by narrow channel widths and complex optical conditions. This systematic review examines the use of multispectral sensors for the detection, estimation, and mapping of chlorophyll-a in riverine, estuarine and transitional environments. Following the PRISMA 2020 framework, sixteen peer-reviewed studies published between 2016 and 2025 were identified and analyzed, focusing on the observation platforms employed, spectral band configurations, radiometric processing procedures, and the modeling approaches used to retrieve chlorophyll-a concentrations. Across the reviewed literature, most applications rely on empirical spectral indices based on red, red-edge, and near-infrared wavelengths, usually calibrated with concurrent in situ measurements. Machine learning methods appear mainly in more recent publications, yet their performance remains strongly tied to site-specific calibration datasets. Notable differences in radiometric correction workflows, validation protocols, and documentation of results complicate direct comparison among studies. This review highlights the strong potential of UAS multispectral observations for resolving small-scale spatial patterns of chlorophyll-a in dynamic river systems, while underscoring the need for greater methodological consistency in future research. Full article

Wild carrot (Daucus carota subsp. carota), the wild relative of cultivated carrot, is globally identified as an invasive weed that threatens hybrid carrot seed production through natural cross-pollination, resulting in compromised genetic purity. Manual identification across the large areas required to ensure genetic purity in carrot seed crops is impractical. Remote sensing offers an alternative; however, morphological similarities among wild carrot, cultivated carrot, and common weeds hinder reliable detection. Early identification, however, remains essential for preventing genetic contamination. This study evaluated leaf-level hyperspectral reflectance spectroscopy (400–2450 nm) with machine learning to discriminate wild carrot from hybrid carrots, parental lines, and 19 co-occurring weed species. Spectral data from 266 wild carrot plants across three New Zealand sites and six weeks (5–10 weeks after emergence) showed negligible spatial effects (R² = 0.034–0.055, pseudo-F = 1.46–2.39, p > 0.05) and moderate temporal variation (R² = 0.136–0.151, pseudo-F = 5.48–6.17, p < 0.001), indicating broadly stable spectral signatures suitable for model generalization. Savitzky–Golay filtering, with min–max normalization outperformed SNV, yielding high full-spectrum accuracies for wild carrot vs. other species (90.35%, κ = 0.80), wild carrot vs. weeds (96.03%, κ = 0.92), and a multi-class model (90.79%, κ = 0.88). After removing atmospheric water-absorption bands to follow airborne sensing, reduced-band models based on airborne-compatible wavelengths maintained strong performance, including 89.40% accuracy (κ = 0.79) for wild carrot vs. weeds using a 20-band Subspace Discriminant model (400–402, 527, 705–720 nm). These findings demonstrate that stable wild carrot spectra and carefully selected visible and red-edge bands can underpin cost-effective UAV/UGV-mounted hyperspectral or multispectral sensors for site-specific wild carrot management. Full article

Acaryochloris marina is a distinctive cyanobacterium that uses chlorophyll d as its primary photosynthetic pigment and possesses two major light-harvesting systems: membrane-integral chlorophyll-binding Pcb/CBP complexes and water-soluble phycobiliproteins. How these antenna systems respond at the transcriptome level to contrasting light environments remains incompletely characterized. Here, we re-analyzed a publicly available RNA-seq dataset for A. marina MBIC11017 (NCBI BioProject PRJNA1130970), comparing cells grown under low-irradiance far-red light (LL-FR; 1.5–2 µmol photons m⁻² s⁻¹, 710-nm peak) and high-irradiance white light (HL-WL; 30–35 µmol photons m⁻² s⁻¹). Because light quality and irradiance both differ in this experimental design, the two effects cannot be separated; all transcriptional changes are therefore interpreted as responses to the combined LL-FR versus HL-WL contrast rather than to far-red wavelength alone. Of 8439 expressed genes, 1810 (21.4%) were significantly differentially expressed (adjusted p < 0.05). Using GFF-verified locus tags which corrected mis-annotations propagated in earlier analyses, the PS-I core gene set showed a mean log₂ fold-change of +1.96 (3.9-fold; 11/11 loci significant), whereas the PS-II core gene set showed a mean log₂ fold-change of +1.10 (2.1-fold; 12/20 loci significant). Light-harvesting genes showed the strongest response: 17/18 phycobiliprotein-pathway genes in KEGG amr00196 were upregulated, together with multiple putative Pcb/CBP loci (mean antenna log₂FC = +3.51; 11.4-fold). Weighted gene co-expression network analysis placed the antenna-associate genes examined here within a module positively correlated with the LL-FR condition (r = 0.802, p = 0.017), and STRING analysis supported an enriched network of predicted or known protein associations (1115 nodes, 4763 edges; PPI enrichment p < 1.0 × 10⁻¹⁶). Recent matched-irradiance experiments indicate that, at equal photon flux, far-red wavelengths reduce phycobilisome content relative to white light. The transcriptional pattern reported here is therefore most parsimoniously interpreted as predominantly a low-irradiance response, with possible wavelength-associated CA5 contributions that cannot be isolated in the present design. Overall, the analysis reveals coordinated transcript-level changes across plasmid-encoded reacquired phycobiliprotein genes, chromosomal Pcb/CBP loci, chlorophyll biosynthesis genes, and photosystem core genes, consistent with coordinated regulation of light-harvesting components in A. marina. Full article

Fog-dependent Tillandsia landbeckii in the hyper-arid Atacama Desert lacks the red-edge reflectance pattern that supports vegetation monitoring, motivating shortwave infrared (SWIR) approaches. We evaluated a newly developed UAV-borne multispectral SWIR camera system for estimating plant water status and additional plant functional traits (fresh and dry biomass, and N uptake) from four spectral bands (1100, 1200, 1510, and 1650 nm) across 20 destructively sampled plots. Of five traits tested, only canopy water content (CWC) retained statistically robust spectral associations after multiple-testing correction, with most significant predictors concentrated in the 1200–1510 nm wavelength region. A physically interpretable predictor, the mean spectral slope between 1200 and 1510 nm, yielded conditional cross-validated $R_{\mathrm{cv}}^2=0.51$ (${\mathrm{RMSE}}_{\mathrm{cv}}\approx 170$ g m⁻²), though fully selection-corrected estimates were substantially lower ($R_{\mathrm{cv}}^2=0.10$–$0.20$), reflecting feature-selection instability at the given sample size. The absence of robust biomass- and nitrogen-related signals is physically interpretable given the species’ atypical surface optics. While expanded sampling and independent validation remain necessary to establish transferable performance estimates, these results demonstrate that SWIR-based water-status retrieval is feasible for this spectrally challenging species, opening a pathway toward functional monitoring of fog-dependent desert ecosystems. Full article

Weed interference during early growth stages is a major constraint on sugarcane productivity, yet effective tools for species-specific detection remain limited in tropical agricultural systems. This study evaluated the spectral separability between Sugarcane (Saccharum officinarum) and a dominant weed species, Rottboellia cochinchinensis, to develop an accessible framework for early-stage weed mapping. Multispectral data acquired from an Unmanned Aerial Vehicle (UAV) and hyperspectral data obtained from a field spectrometer were utilized. Hyperspectral data were synthesized to reconstruct multispectral bands (UAV image bands) using a regularized linear synthesis model, thereby generating spectral signatures. Spectral separability between sugarcane and Rottboellia cochinchinensis was assessed visually and statistically (Jeffries–Matusita distance). Blue and Green bands provided the strongest differentiation between species, while RedEdge enhanced separability when paired with pigment-sensitive wavelengths. When using vegetation indices based on the near-infrared (NIR) band, the visual appearance of class separation was poor due to the NIR band’s sensitivity to variation in leaf internal structure, canopy architecture, water content, and spectral mixing with the soil background at the early stage of sugarcane. These results were used to differentiate weed coverage from sugarcane. Object-based image analysis (OBIA) outperformed the pixel-based method, achieving higher overall accuracy (0.9038) and a more spatially coherent weed delineation (Kappa = 0.8499). These findings suggest that synthesized spectral signatures of Rottboellia cochinchinensis and sugarcane, combined with targeted spectral indices and OBIA techniques, offer a practical and transferable approach for early detection of Rottboellia cochinchinensis at the farm level. Full article

Phosphorus (P) deficiency severely limits maize growth and yield, yet early detection remains challenging, as visible symptoms appear only after prolonged starvation. This study evaluated the capability of hyperspectral imaging (HSI) combined with machine learning to detect P deficiency in maize seedlings at both symptomatic and pre-symptomatic stages. Two greenhouse experiments were conducted: a long-term pot system under high and low P conditions and a short-term hydroponic experiment with three P concentrations of 500, 100, and 0 μmol/L phosphate (Pi). After long-term P deficiency, significant reductions in shoot biomass and Pi content were observed, while root biomass increased and nutrient profiles were altered. Hyperspectral signatures revealed distinct wavelength-specific differences across visible, red-edge, and near-infrared (NIR) regions, with P-deficient leaves showing lower reflectance in green and NIR regions but higher reflectance in the red band. A multilayer perceptron machine learning model achieved 99.65% accuracy in discriminating between P treatments. In the short-term experiment, P deficiency significantly reduced tissue Pi content within one week without affecting pigment composition or photosynthetic parameters. Despite the absence of visible symptoms, hyperspectral measurements detected subtle spectral changes, particularly in older leaves, enabling classification accuracies of 80.71–84.56% in the first week and 85.88–90.98% in the second week of P treatment. Conventional vegetation indices showed weak correlations with Pi content and failed to detect early P deficiency. These findings demonstrate that HSI combined with machine learning can effectively detect P deficiency before visible symptoms emerge, offering a non-destructive, rapid diagnostic tool for precision nutrient management in maize production systems. Full article

Optimizing nitrogen (N) fertilization is essential for sustaining durum wheat yield and grain quality while reducing the environmental impacts associated with N over-application. Hyperspectral sensing provides a rapid and non-destructive approach for monitoring crop N status. However, high-dimensional data, phenology-dependent spectral responses, and spatial autocorrelation in field measurements limit robust nitrogen classification and interpretation. This study evaluated hyperspectral-based nitrogen status classification in durum wheat under Mediterranean field conditions and identified key spectral regions using explainable artificial intelligence. A field experiment was conducted in Southern Italy using ten N fertilization rates (0–180 kg N ha⁻¹). Canopy reflectance was acquired at the booting and heading stages from georeferenced sampling locations. Three nitrogen stratification strategies (binary Low–High, Extreme, and three-level) were evaluated using Random Forest, SVM-RBF, and XGBoost classifiers. Model performance was assessed using spatially independent Leave-One-Plot-Out cross-validation at both the sample and plot levels, with plot-level predictions derived through majority voting. Classification robustness was strongly influenced by the stratification strategy and phenological stage. The binary Low–High stratification achieved the highest sample-level accuracy, with a maximum of 0.78 at booting (SVM-RBF) and 0.75 at heading (SVM-RBF), whereas the Extreme stratification produced intermediate performance, with maximum accuracies of 0.73 at booting (SVM-RBF) and 0.63 at heading (XGBoost). Plot-level aggregation improved performance, reaching up to 0.90 at booting and 1.00 at heading. SHAP analysis highlighted red, red-edge, and near-infrared wavelengths as the dominant contributors, with increased reliance on longer wavelengths at the heading. Overall, explainable machine learning provides a robust framework for hyperspectral nitrogen monitoring in durum wheat. Full article

This study compared three deep learning architectures—one-dimensional convolutional neural network (1D-CNN), self-supervised learning (SSL), and Vision Transformer (ViT)—to evaluate their ability to predict carotenoid content from visible–near-infrared (VIS–NIR) spectral reflectance data (400–850 nm) acquired non-destructively from tea leaves. Model performance was evaluated using 10-fold cross-validation and analyzed through the mean SHapley Additive exPlanations values to identify key spectral features. The ViT model achieved the highest predictive accuracy (coefficient of determination [R²] = 0.81, root mean square error [RMSE] = 1.04, ratio of performance to deviation [RPD] = 2.32), followed by 1D-CNN (R² = 0.75, RMSE = 1.21, RPD = 1.99), whereas SSL showed substantially lower predictive performance (R² = 0.30, RMSE = 2.01, RPD = 1.20). Feature importance analysis revealed that ViT focused strongly on the red-edge region around 720 nm, which corresponds to spectral features associated with carotenoids and chlorophyll. The 1D-CNN relied mainly on blue (450–480 nm) and red (670–700 nm) regions, while SSL exhibited a broadly distributed importance pattern across wavelengths. These results indicate that ViT’s self-attention mechanism captures long-range spectral dependencies more effectively than conventional convolutional or self-supervised models. Overall, the study demonstrates that transformer-based architectures provide a powerful and interpretable framework for non-destructive estimation of carotenoid content from VIS–NIR reflectance spectroscopy. Full article