Search Results (357)

The continued acidification of red soil reduces phosphorus availability and microbial activity, which restricts corn growth. Phosphorus tailings, a waste product from phosphate mining, can neutralize soil acidity and supply controlled-release phosphorus, but their effects on the red soil-corn system remain unclear. A field experiment in Qujing, Yunnan (2023–2024), tested four treatments: CK (standard fertilization), T1 (CK plus phosphorus tailings), T2 (80% of standard fertilizer plus phosphorus tailings), and T3 (80% of standard fertilizer plus phosphorus tailings and organic fertilizer, both applied at 6.0 t·ha⁻¹). Using high-throughput sequencing, redundancy analysis (RDA), and structural equation modeling (SEM), the study evaluated impacts on soil properties, microbial communities, and corn yield and quality. Results showed: (1) Phosphorus tailings reduced soil acidification; T3 raised soil pH in the top 0–10 cm by 0.54–0.9 units compared to CK and increased total, available, and soluble phosphorus in the 0–20 cm layer to 952.82, 28.46, and 2.04 mg/kg, respectively. (2) T3 exhibited the highest microbial diversity (Chao1 and Shannon indices increased by 177.57% and 37.80% versus CK) and a more complex bacterial co-occurrence network (114 edges versus 107 in CK), indicating enhanced breakdown of aromatic compounds. (3) Corn yield under T3 improved by 13.72% over CK, with increases in hundred-grain weight (+6.02%), protein content (+18.04%), and crude fiber (+9.00%). (4) Effective nitrogen, ammonium nitrogen, available phosphorus, and soil conductivity were key factors affecting gcd/phoD phosphorus-reducing bacteria. (5) Phosphorus tailings indirectly increased yield by modifying soil properties and pH, both positively linked to yield, while gcd-carrying bacteria had a modest positive influence. In summary, combining phosphorus tailings with a 20% reduction in chemical fertilizer reduces fertilizer use, recycles mining waste, and boosts corn production in acidic red soil, though further studies are needed to evaluate long-term environmental effects. Full article

The successful launch of Korea Multipurpose Satellite-3/Advanced Earth Imaging Sensor System (KOMPSAT-3/AEISS) on 18 May 2012 allowed the Republic of Korea to meet the growing demand for high-resolution satellite imagery. However, like all satellite sensors, KOMPSAT-3/AEISS experienced temporal changes post-launch and thus requires ongoing evaluation and calibration. Although more than a decade has passed since launch, the KOMPSAT-3/AEISS mission and its multi-year data archive remain widely used. This study focused on the cross-calibration of KOMPSAT-3/AEISS with Sentinel-2A/Multispectral Instrument (MSI) by comparing the radiometric responses of the two satellite sensors under similar observation conditions, leveraging the linear relationship between Digital Numbers (DN) and top-of-atmosphere (TOA) radiance. Cross-calibration was performed using near-simultaneous satellite images of the same region, and the Spectral Band Adjustment Factor (SBAF) was calculated and applied to account for differences in spectral response functions (SRF). Additionally, Bidirectional Reflectance Distribution Function (BRDF) correction was applied using MODIS-based kernel models to minimize angular reflectance effects caused by differences in viewing and illumination geometry. This study aims to evaluate the radiometric consistency of KOMPSAT-3/AEISS relative to Sentinel-2A/MSI over Baotou scenes acquired in 2022–2023, derive band-specific calibration coefficients and compare them with prior results, and conduct a side-by-side comparison of cross-calibration and vicarious calibration. Furthermore, the cross-calibration yielded band-specific gains of 0.0196 (Blue), 0.0237 (Green), 0.0214 (Red), and 0.0136 (NIR). These findings offer valuable implications for Earth observation, environmental monitoring, and the planning and execution of future satellite missions. Full article

As the residual products of severe chemical weathering, bauxite deposits serve both as essential economic Al-Fe resources and geochemical archives that reveal information about the parent rocks’ composition, paleoenvironments and paleoclimates, and the tectonic settings responsible for their genesis. The well-developed Early Paleocene bauxite deposits of the Salt Range, Pakistan, provide an opportunity for deciphering their ore genesis and parental affinities. The deposits occur as lenticular bodies and are typically composed of three consecutive stratigraphic facies from base to top: (1) massive dark-red facies (L-1), (2) composite conglomeratic–pisolitic facies (L-2), and (3) Kaolinite-rich clayey facies (L-3). Results from optical microscopy, X-ray powder diffraction (XRPD), and scanning electron microscopy with Energy-Dispersive X-Ray Spectroscopy (SEM-EDS) reveal that facies L-1 contains kaolinite, hematite, and goethite as major minerals, with minor amounts of muscovite, quartz, anatase, and rutile. In contrast, facies L-2 primarily consists of kaolinite, boehmite, hematite, gibbsite, goethite, alunite/natroalunite, and zaherite, with anatase, rutile, and quartz as minor constituents. L-3 is dominated by kaolinite, quartz, and anatase, while hematite and goethite exist in minor concentrations. Geochemical analysis reveals elevated concentrations of ${\mathrm{Al}}_2$${\mathrm{O}}_3$, ${\mathrm{Fe}}_2$${\mathrm{O}}_3$, ${\mathrm{SiO}}_2$, and ${\mathrm{TiO}}_2$. Trace elements, including Th, U, Ga, Y, Zr, Nb, Hf, V, and Cr, exhibit a positive trend across all sections when normalized to Upper Continental Crust (UCC) values. Field observations and analytical data suggest a polygenetic origin of these deposits. L-1 suggests in situ lateritization of some sort of precursor materials, with enrichment in stable and ultra-stable heavy minerals such as zircon, tourmaline, rutile, and monazite. This facies is mineralogically mature with bauxitic components, but lacks the typical bauxitic textures. In contrast, L-2 is texturally and mineralogically mature, characterized by various-sized pisoids and ooids within a microgranular-to-microclastic matrix. The L-3 mineralogy and texture suggest that the conditions were still favorable for bauxite formation. However, the ongoing tectonic activities and wet–dry climate cycles post-depositionally disrupted the bauxitization process. The accumulation of highly stable detrital minerals, such as zircon, rutile, tourmaline, and monazite, indicates prolonged weathering and multiple cycles of sedimentary reworking. These deposits have parental affinity with acidic-to-intermediate/-argillaceous rocks, resulting from the weathering of sediments derived from UCC sources, including cratonic sandstone and shale. Full article

Seismic attribute analysis, guided by well data, reveals widespread stratigraphic anomalies caused by erosion or karstification in the late Ordovician-early Devonian Hunton Group in the Arkoma Basin, eastern Oklahoma, USA. This study shows that these strata are more extensive than previously known. Well data and seismic mapping in the Red Oak petroleum field identify approximately 40 m thick Hunton lenses, averaging 3 km in diameter, surrounded by karsted rock. These lenses may be remnants of incomplete erosion during the Middle Devonian period (pre-Woodford unconformity) or result from Hunton rocks sagging into sinkholes caused by karstification and collapse of underlying Viola or Bromide carbonates. Using formation tops from wells, correlated with attribute and structure maps from a 3D seismic volume, we identify (1) areas lacking Hunton seismic markers, indicating complete removal; (2) areas with Hunton contacts, showing where Hunton remains; and (3) zones without Hunton but with a thin layer underlying carbonate strata, interpreted as an incipient karst zone, often near areas with Hunton. We also observe that the thickness of the overlying Woodford Shale, a key hydrocarbon target, correlates with karstic and erosional thinning of Hunton carbonates. Interpretation of 3D seismic data reveals a strong connection between thinned Hunton and thickened Woodford Shale. Full article

Droughts significantly impact agriculture, water resources, and ecosystems. Their timely detection is essential for implementing effective mitigation strategies. This study explores the use of multispectral Sentinel-2 remote sensing indices and machine learning techniques to detect drought conditions in three distinct regions of India, such as Jodhpur, Amravati, and Thanjavur, during the Rabi season (October–April). Twelve remote sensing indices were studied to assess different aspects of vegetation health, soil moisture, and water stress, and their possible joint use and influence as indicators of regional drought events. Reference data used to define drought conditions in each region were primarily sourced from official government drought declarations and regional and national news publications, which provide seasonal maps of drought conditions across the country. Based on this information, a district vs. year (3 × 10) ground truth is created, indicating the presence or absence of drought (Drought/No Drought) for each region across the ten-year period. Using this ground truth table, we extended the remote sensing dataset by adding a binary drought label for each observation: 1 for “Drought” and 0 for “No Drought”. The dataset is organized by year (2016–2025) in a two-dimensional format, with indices as columns and observations as rows. Each observation represents a single measurement of the remote sensing indices. This enriched dataset serves as the foundation for training and evaluating machine learning models aimed at classifying drought conditions based on spectral information. The resultant remote sensing dataset was used to predict drought events through various machine learning models, including Random Forest, XGBoost, Bagging Classifier, and Gradient Boosting. Among the models, XGBoost achieved the highest accuracy (84.80%), followed closely by the Bagging Classifier (83.98%) and Random Forest (82.98%). In terms of precision, Bagging Classifier and Random Forest performed comparably (82.31% and 81.45%, respectively), while XGBoost achieved a precision of 81.28%. We applied a seasonal majority voting strategy, assigning a final drought label for each region and Rabi season based on the majority of predicted monthly labels. Using this method, XGBoost and Bagging Classifier achieved $96.67\%$ accuracy, precision, and recall, while Random Forest and Gradient Boosting reached $90\%$ and $83.33\%$, respectively, across all metrics. Shapley Additive Explanation (SHAP) analysis revealed that Normalized Multi-band Drought Index (NMDI) and Day of Season (DOS) consistently emerged as the most influential features in determining model predictions. This finding is supported by the Borda Count and Weighted Sum analysis, which ranked NMDI, and DOS as the top feature across all models. Additionally, Red-edge Chlorophyll Index (RECI), Normalized Difference Water Index (NDWI), Normalized Difference Moisture Index (NDMI), and Ratio Drought Index (RDI) were identified as important features contributing to model performance. These features help reveal the underlying spatiotemporal dynamics of drought indicators, offering interpretable insights into model decisions. To evaluate the impact of feature selection, we further conducted a feature ablation study. We trained each model using different combinations of top features: Top 1, Top 2, Top 3, Top 4, and Top 5. The performance of each model was assessed based on accuracy, precision, and recall. XGBoost demonstrated the best overall performance, especially when using the Top 5 features. Full article

Extreme sea surface temperature (SST) events, such as marine heatwaves (MHWs) and marine cold spells (MCSs), severely affect warm water coral reefs. However, further study is required on their historical and future spatiotemporal patterns, driving mechanisms, and impacts in coral reef regions. This study analyzed the spatiotemporal patterns in MHWs/MCSs for the periods 1982–2022 and 2023–2070 using ten indices based on OISSTv2.1 and CMIP6 data, respectively, identified key MHW drivers via four machine learning methods (Random Forest, Extreme Gradient Boosting, Light Gradient Boosting Machine, and categorical boosting) and SHAP values (Shapley Additive Explanations), and then examined their relationship with coral coverage across ten global marine regions. Our results revealed that (1) MHWs are not only increasing in their average intensity but also becoming more extreme, while MCSs have declined. More MHW days are observed in regions like the Red Sea, the Persian Gulf, and the South Pacific Islands, with increases of up to 28 days per decade. (2) Higher-latitude coral reefs are experiencing more severe MHWs than equatorial regions, with up to 1.24 times more MHW days, emphasizing the urgent need to protect coral refuges. (3) MHWs are projected to occur nearly year-round by 2070 under scenario SSP5–8.5. The area ratio of MHWs to MCSs is expected to rise sharply from 2040 onward, reaching approximately 100-fold under the SSP2–4.5 scenario and 196-fold under the SSP5–8.5 scenario, particularly in the Marshall Islands and Caribbean Sea regions. (4) The coefficient of variation (CV) of annual temperature, annual ocean heat content, and monthly temperature were the top three factors driving MHW intensity. We emphasize that future MHW predictions should focus more on the CV of forecasting indicators rather than just the climate means. (5) Coral coverage exhibited post-mortality processes following MHWs, showing a strong negative correlation (r = −0.54, p < 0.01) with MHWs while demonstrating a significant positive correlation (r = 0.6, p < 0.01) with MCSs. Our research underscores the sustained efforts to protect and restore coral reefs amid escalating climate-induced stressors. Full article

The objective of the study was to develop and validate a ground-based method using a depth image sensor equipped with depth, visible red, green, blue (RGB), and near-infrared bands to measure the leaf area index (LAI) based on the relative illuminance of foliage only. The method was applied in a Itajii chinkapin (Castanopsis sieboldii (Makino) Hatus. ex T.Yamaz. & Mashiba )forest in Aichi Prefecture, Japan, and validated by comparing estimates with conventional methods (LAI-2200 and fisheye photography). To apply the 5-band sensor to actual forests, a methodology is proposed for matching the color camera and near-infrared camera in units of pixels, along with a method for widening the exposure range through multi-step camera exposure. Based on these advancements, the RGB color band, near-infrared band, and depth band are converted into several physical properties. Employing these properties, each pixel of the canopy image is classified into upper foliage, lower foliage, sky, and non-assimilated parts (stems and branches). Subsequently, the LAI is calculated using the gap-fraction method, which is based on the relative illuminance of the foliage. In comparison with existing indirect LAI estimations, this technique enabled the distinction between upper and lower canopy layers and the exclusion of non-assimilated parts. The findings indicate that the plant area index (PAI) ranged from 2.23 to 3.68 m² m⁻², representing an increase from 33% to 34% compared to the LAI calculated after excluding non-assimilating parts. The findings of this study underscore the necessity of distinguishing non-assimilated components in the estimation of LAI. The PAI estimates derived from the depth image sensor exhibited moderate to strong agreement with the LAI-2200, contingent upon canopy rings (R² = 0.48–0.98), thereby substantiating the reliability of the system’s performance. The developed approaches also permit the evaluation of the distributions of leaves and branches at various heights from the ground surface to the top of the canopy. The novel LAI measurement method developed in this study has the potential to provide precise, reliable foundational data to support research in ecology and hydrology related to complex tree structures. Full article

Accurate semantic segmentation of indoor architectural elements, such as staircases and elevators, is critical for safe and efficient robotic navigation, particularly in complex multi-floor environments. Traditional fusion methods struggle with occlusions, reflections, and low-contrast regions. In this paper, we propose a novel feature fusion module, Skip-Connection Fusion (SCF), that dynamically integrates RGB (Red, Green, Blue) and depth features through an adaptive weighting mechanism and skip-connection integration. This approach enables the model to selectively emphasize informative regions while suppressing noise, effectively addressing challenging conditions such as partially blocked staircases, glossy elevator doors, and dimly lit stair edges, which improves obstacle detection and supports reliable human–robot interaction in complex environments. Extensive experiments on a newly collected dataset demonstrate that SCF consistently outperforms state-of-the-art methods, including PSPNet and DeepLabv3, in both overall mIoU (mean Intersection over Union) and challenging-case performance. Specifically, our SCF module improves segmentation accuracy by 5.23% in the top 10% of challenging samples, highlighting its robustness in real-world conditions. Furthermore, we conduct a sensitivity analysis on the learnable weights, demonstrating their impact on segmentation quality across varying scene complexities. Our work provides a strong foundation for real-world applications in autonomous navigation, assistive robotics, and smart surveillance. Full article

Accurate body condition score (BCS) monitoring in dairy cows is essential for optimizing health, productivity, and welfare. Traditional manual scoring methods are labor-intensive and subjective, driving interest in automated imaging-based systems. This study evaluated the effectiveness of 2D imaging and deep learning for BCS classification using three camera perspectives—front, back, and top-down—to identify the most reliable viewpoint. The research involved 56 Norwegian Red milking cows at the Center for Livestock Experiments (SHF) of Norges Miljo-og Biovitenskaplige Universitet (NMBU) in Norway. Images were classified into BCS categories of 2.5, 3.0, and 3.5 using a YOLOv8 model. The back view achieved the highest classification precision (mAP@0.5 = 0.439), confirming that key morphological features for BCS assessment are best captured from this angle. Challenges included misclassification due to overlapping features, especially in Class 2.5 and background data. The study recommends improvements in algorithmic feature extraction, dataset expansion, and multi-view integration to enhance accuracy. Integration with precision farming tools enables continuous monitoring and early detection of health issues. This research highlights the potential of 2D imaging as a cost-effective alternative to 3D systems, particularly for small and medium-sized farms, supporting more effective herd management and improved animal welfare. Full article

This study addresses the challenges of redundant crop identification features and low computational efficiency in complex agricultural environments, particularly in arid regions. Focusing on the Hexi region of Gansu Province, we utilized the Google Earth Engine (GEE) to integrate Sentinel-2 optical imagery (10 bands) and Sentinel-1 radar data (VV/VH polarization), constructing a 96-feature set that comprises spectral, vegetation index, red-edge, and texture variables. The recursive feature elimination random forest (RF-RFE) algorithm was employed for feature selection and model optimization. Key findings include: (1) Variables driven by spatiotemporal differentiation were effectively selected, with red-edge bands (B5–B7) during the grain-filling stage in August accounting for 56.7% of the top 30 features, which were closely correlated with canopy chlorophyll content (p < 0.01). (2) A breakthrough in lightweight modeling was achieved, reducing the number of features by 69%, enhancing computational efficiency by 62.5% (from 8 h to 3 h), and decreasing memory usage by 66.7% (from 12 GB to 4 GB), while maintaining classification accuracy (PA: 97.69%, UA: 97.20%, Kappa: 0.89). (3) Multi-source data fusion improved accuracy by 11.54% compared to optical-only schemes, demonstrating the compensatory role of radar in arid, cloudy regions. This study offers an interpretable and transferable lightweight framework for precision crop monitoring in arid zones. Full article

This paper presents an enhanced hub location model tailored to port–hinterland logistics planning, grounded in the Capacitated Multiple-Allocation Hub Location Problem (CMAHLP). The formulation incorporates nonlinear cost structures, hub-specific operating costs, adaptive capacity constraints, and a feasibility condition based on the Social Net Present Value (NPVsocial) to support the design of intermodal freight networks under asymmetric spatial and socio-environmental conditions. The empirical case focuses on Spain, leveraging its strategic position between Asia, North Africa, and Europe. The model includes four major ports—Barcelona, Valencia, Málaga, and Algeciras—as intermodal gateways connected to the 47 provinces of peninsular Spain through calibrated cost matrices based on real distances and mode-specific road and rail costs. A Genetic Algorithm is applied to evaluate 120 scenarios, varying the number of active hubs (4, 6, 8, 10, 12), transshipment discounts (α = 0.2 and 1.0), and internal parameters. The most efficient configuration involved 300 generations, 150 individuals, a crossover rate of 0.85, and a mutation rate of 0.40. The algorithm integrates guided mutation, elitist reinsertion, and local search on the top 15% of individuals. Results confirm the central role of Madrid, Valencia, and Barcelona, frequently accompanied by high-performance inland hubs such as Málaga, Córdoba, Jaén, Palencia, León, and Zaragoza. Cities with active ports such as Cartagena, Seville, and Alicante appear in several of the most efficient network configurations. Their recurring presence underscores the strategic role of inland hubs located near seaports in supporting logistical cohesion and operational resilience across the system. The COVID-19 crisis, the Suez Canal incident, and the persistent tensions in the Red Sea have made clear the fragility of traditional freight corridors linking Asia and Europe. These shocks have brought renewed strategic attention to southern Spain—particularly the Mediterranean and Andalusian axes—as viable alternatives that offer both geographic and intermodal advantages. In this evolving context, the contribution of southern hubs gains further support through strong system-wide performance indicators such as entropy, cluster diversity, and Pareto efficiency, which allow for the assessment of spatial balance, structural robustness, and optimal trade-offs in intermodal freight planning. Southern hubs, particularly in coordination with North African partners, are poised to gain prominence in an emerging Euro–Maghreb logistics interface that demands a territorial balance and resilient port–hinterland integration. Full article

Background: Gastric cancer (GC) remains a major global health challenge, with rising incidence among patients post-Helicobacter pylori (H. pylori) eradication, particularly those with persistent intestinal metaplasia (IM). Current risk stratification tools are limited in this high-risk population. Aim: To develop, validate, and externally test a machine learning-based prediction model—termed the Early Gastric Cancer Model (EGCM)—for identifying early gastric cancer (EGC) risk in H. pylori-eradicated patients with IM, and to implement it as a web-based clinical tool. Methods: This retrospective, dual-center study enrolled 214 H. pylori-eradicated patients with histologically confirmed IM from 900 Hospital and Fujian Provincial People’s Hospital. The dataset was split into a training cohort (70%) and an internal validation cohort (30%), with an external test cohort from the second center. A total of 21 machine learning algorithms were screened using cross-validation and hyperparameter optimization. Boruta and SHAP analyses were employed for feature selection, and the final EGCM was constructed using the top five predictors: atrophy range, xanthoma, map-like redness (MLR), MLR range, and age. Model performance was evaluated via ROC curves, precision–recall curves, calibration plots, and decision curve analysis (DCA), and compared against conventional inflammatory biomarkers such as NLR and PLR. Results: The CatBoost algorithm demonstrated the best overall performance, achieving an AUC of 0.743 (95% CI: 0.70–0.80) in internal validation and 0.905 in the external test set. The EGCM exhibited superior discrimination compared to individual inflammatory markers (p < 0.01). Calibration analysis confirmed strong agreement between predicted and observed outcomes. DCA showed the EGCM yielded greater net clinical benefit. A web calculator was developed to facilitate clinical application. Conclusions: The EGCM is a validated, interpretable, and practical tool for stratifying EGC risk in H. pylori-eradicated IM patients across multiple centers. Its integration into clinical practice could improve surveillance precision and early cancer detection. Full article

Marine tunicates are a very attractive and abundant source of secondary metabolites with chemical diversity and biological activity. Fractionation and purification of the organic extract of the Red Sea tunicate Didemnum species resulted in the isolation and identification of three new compounds, didemnosides A and B (1 and 2) and 1,1′,3,3′-bisuracil (3), together with thymidine (4), 2′-deoxyuridine (5), homarine (6), and acetamide (7). Planar structures of the compounds were explained through analyses of their 1D (¹H and ¹³C) and 2D (¹H–¹H COSY, HSQC, and HMBC) NMR spectra and high-resolution mass spectral determinations. Compound 1 exhibited the highest growth inhibition toward the MCF-7 cancer cell line with IC₅₀ values of 0.597 μM, while other compounds were inactive (≥50 μM) against this cell line. On the other hand, compounds 1, 2, and 4–7 moderately inhibited SW-1222 and PC-3 cells with IC₅₀ values ranging between 5.25 and 9.36 μM. Molecular docking analyses of the top three active compounds on each tested cell line exposed stable interactions into the active pockets of estrogen receptor alpha (ESR1), human topoisomerase II alpha (TOP2A), and cyclin-dependent kinase 5 (CDK5) which are contemplated as essential targets in cancer treatments. Thus, compound 1 represents a scaffold for the development of more effective anticancer drugs. Full article

Italy is a global top wine producer, with emphasis on high-quality wines. This study investigates the Carbon Footprint (CF), Water Footprint (WF), and Ecological Footprint (EF) of twelve red wine producers in Piedmont, Northern Italy. The analysis was based on a 0.75 L wine bottle as functional unit (FU). Twelve producers were interviewed and given questionnaires, which made it possible to gather primary data for the environmental evaluation that described vineyard and agricultural operations and wine production. The average CF was 0.88 ± 0.3 kg ${CO}_{2eq},$ with 44% of CF associated with the glass bottle, 20% to the diesel fuel fed to the agricultural machines, 32% to electricity consumption, and 4% to other contributions. The average WF was 881 ± 252.4 L, with 98% Green WF due to evapotranspiration, and 2% Blue and Grey WF. The average EF was 81.3 ± 57.2 global ha, 73% ascribed to the vineyard area and 27% to CO₂ assimilation. The obtained CF and WF values align with existing literature, while no comparison is possible for the EF data, which are previously unknown. To reduce the environmental impacts of wine production, actions like using recycled glass bottles, electric agricultural machines and renewable energy can help. However, high-quality wine production in Piedmont is deeply rooted in tradition and mostly managed by small producers. Further research should investigate the social acceptance of such actions, and policies supporting economic incentives could be key enablers. Full article

In recent years, significant advancements in artificial intelligence, particularly in the field of deep learning, have increasingly been integrated into agricultural applications, including critical processes such as disease detection. Tomato, being one of the most widely consumed agricultural products globally and highly susceptible to a variety of fungal, bacterial, and viral pathogens, remains a prominent focus in disease detection research. In this study, we propose a deep learning-based approach for the detection of tomato diseases, a critical challenge in agriculture due to the crop’s vulnerability to fungal, bacterial, and viral pathogens. We constructed an original dataset comprising 6414 images captured under real production conditions, categorized into three image types: leaves, green tomatoes, and red tomatoes. The dataset includes five classes: healthy samples, late blight, early blight, gray mold, and bacterial cancer. Twenty-one deep learning models were evaluated, and the top five performers (EfficientNet-b0, NasNet-Large, ResNet-50, DenseNet-201, and Places365-GoogLeNet) were selected for feature extraction. From each model, 1000 deep features were extracted, and feature selection was conducted using MRMR, Chi-Square (Chi²), and ReliefF methods. The top 100 features from each selection technique were then used for reclassification with traditional machine learning classifiers under five-fold cross-validation. The highest test accuracy of 92.0% was achieved with EfficientNet-b0 features, Chi² selection, and the Fine KNN classifier. EfficientNet-b0 consistently outperformed other models, while the combination of NasNet-Large and Wide Neural Network yielded the lowest performance. These results demonstrate the effectiveness of combining deep learning-based feature extraction with traditional classifiers and feature selection techniques for robust detection of tomato diseases in real-world agricultural environments. Full article