Search Results (141)

Peppers from the Capsicum genus are highly valued worldwide for their culinary, medicinal, and nutritional uses. However, accurately classifying and developing new varieties to enhance these traits remains a challenge due to the limitations of traditional methods, which often lack precision and are time-consuming. This study aimed to overcome these limitations by applying advanced multivariate statistical techniques and machine learning models (KNN, RF, XGBoost) to characterize and classify Capsicum genotypes based on genetic and phenotypic features. Sixteen Capsicum genotypes were analyzed using methods such as MANOVA, PCA, and cluster analysis to explore their variabilities and similarities. Cluster analysis revealed the formation of distinct groups, indicating phenotypic similarity patterns among specific varieties. The machine learning models were evaluated using Leave-One-Out cross-validation to address the challenges posed by small datasets. The results indicated that Random Forest outperformed the other models, exhibiting superior class discrimination with an AUC of 0.96, while KNN and XGBoost achieved AUC values of 0.95 and 0.85, respectively. Despite the slightly superior performance of Random Forest relative to KNN, both models demonstrated strong predictive performance, whereas XGBoost exhibited moderate performance. In addition, key agronomic traits such as pericarp thickness, fruit diameter, seeds per fruit, and corolla color were identified as the most relevant variables for classification. Principal component analysis indicated that the first components explained a substantial proportion of the total variance, supporting efficient dimensionality reduction and pattern recognition. Furthermore, the Random Forest model achieved high overall performance, with accuracy, precision, recall, and F1-score values close to 0.93, reinforcing its robustness in multiclass classification. This study highlights the effectiveness of machine learning in overcoming the constraints of traditional classification methods, providing a robust approach for the accurate identification and improvement of pepper varieties. Full article

Wearable inertial sensors produce multi-axis motion signals with rich spatial and temporal structure. Existing deep-learning pipelines for human activity recognition (HAR) rarely tackle three issues jointly: explicit modeling of the body-part grouping of multi-location inertial channels, bidirectional temporal modeling at linear-time cost, and dynamic, time-varying attention for non-stationary motion. We aim to close these three gaps within a single architecture. To this end we propose Dynamic-Attentive Selective Mamba (DASM), which combines three components: Group-Aware Convolutions (GroupConv) for body-part-aware local features, a Bidirectional Mamba (BiMamba) module for linear-time forward and backward temporal context, and a Dynamic CBAM (DCBAM) that produces per-timestep channel and spatial attention for non-stationary windows. On the UCI Daily and Sports Activities dataset (19 classes, 8 subjects), under stratified segment-level 5-fold cross-validation (3 seeds, 15 runs/model), DASM reaches 99.89% accuracy and F1, a 0.11% gain over CNN-BiGRU-CBAM and 0.50% over Multi-STMT; under leave-one-subject-out (LOSO), it reaches 89.34%, 1.69% above the strongest baseline. The 10.55% drop under LOSO shows that segment-level results overestimate cross-subject generalization. Ablations show small but statistically detectable gains (Cohen’s $d\in [0.4,0.7]$ per module, $d\ge 1.5$ full-vs-baseline). We therefore position the contribution as a structured architecture within a near-saturated benchmark; broader deployment claims require multi-dataset subject-independent validation. Full article

Electrolytic-plasma hardening of steel components requires reliable non-contact temperature monitoring, but traditional pyrometry is complicated by the variable emissivity of steel and the intense radiation of the plasma envelope. This work presents an approach that repurposes a compact multispectral AS7341 sensor into a virtual temperature sensor based on physically grounded spectral feature engineering and regularized machine learning. The use of logarithmic ratios of the near-infrared channel (940 nm) to the visible channels suppresses the plasma contribution and linearizes Wien’s radiation law. On a controlled dataset of 20 cycles, this increases the Pearson correlation with the peak temperature from r = 0.498 (raw NIR channel) to r = 0.781 for the log(NIR/Clear) feature. Current is identified as a confounding variable; normalizing the NIR/Clear ratio by the cycle-averaged current (r = 0.761) ensures correct signal interpretation under varying process conditions. Two narrow channels–NIR (940 nm) and F8 (680 nm)–provide accuracy equivalent to the broadband Clear channel (r = 0.778 vs. 0.781), thus simplifying hardware implementation. Ridge regression using three weakly correlated features (log(NIR/Clear), cycle duration, and initial temperature) achieves a mean absolute error of 91.4 °C under leave-one-out cross-validation (LOOCV) and 85.5 °C on an independent current-group test (R² = 0.536). Independent verification by scanning electron microscopy and Vickers microhardness on 30KhGSA steel confirms reliable separation of the three thermal regimes: underheating (<800 °C, 280–320 HV), optimal quenching (800–900 °C, 620–680 HV, fine-needle martensite), and overheating (>900 °C, 540–590 HV). The proposed set of spectral features provides a physically justified basis for a low-cost industrial temperature sensor for electrolytic-plasma processing. Full article

EEG-based driver fatigue detection faces three signal-level challenges: inter-subject spectral variability, coupled frequency–spatial–temporal dynamics that existing methods process independently, and dependence on a single labeling scheme. This paper presents DCAMNet, a lightweight CNN (12.3 K parameters) that addresses these challenges through three end-to-end blocks. An SE-driven dynamic convolution block adapts spectral sensitivity per sample via input-dependent kernel weighting—applied here for the first time to fatigue detection. A spatial convolution block encodes electrode-level cortical patterns, and a temporal attention block captures fatigue dynamics through windowed variance descriptors with group-wise attention scoring. DCAMNet was evaluated on SEED-VIG (PERCLOS labels) and MESD (reaction-time labels) under both subject-mixed and leave-one-subject-out (LOSO) protocols. Under LOSO cross-validation—the operationally relevant test that eliminates within-subject information leakage and simulates deployment on unseen drivers—DCAMNet achieved 85.43% accuracy on SEED-VIG with a 2.86-point advantage over the strongest baseline, and $79\pm 5$% accuracy on MESD with a 3-point advantage. As upper-bound estimates under the subject-mixed protocol, accuracy reached 97.47% (SEED-VIG) and 96.52% (MESD). With 1.35 ms inference latency on a standard GPU, the compact architecture suggests potential suitability for real-time embedded deployment, although on-device validation on representative automotive hardware remains necessary. Full article

Idiopathic pulmonary fibrosis (IPF) and autoimmune usual interstitial pneumonia (aUIP) share overlapping clinico-radiological features, complicating differential diagnosis. Electronic nose (eNose) technology characterizes exhaled breath profiles (“breathprints”) and may offer a non-invasive diagnostic approach in fibrotic interstitial lung diseases. To evaluate whether eNose breathprint analysis can discriminate between IPF and aUIP. In this cross-sectional study of 60 patients (34 IPF, 26 aUIP), breathprints were analyzed using principal component analysis (PCA, retaining eigenvalues > 1). Group differences were assessed via independent t-tests. Linear discriminant analysis (LDA) with leave-one-out cross-validation evaluated the discriminatory performance of PC combinations. PCA identified four principal components, with PC1 explaining 96% of the total variance. PC1 scores were significantly higher in aUIP compared to IPF (mean difference −0.53; 95% CI −1.04 to −0.02; p = 0.04); PC2-PC4 showed no significant differences (p > 0.3). LDA utilizing PC1 and PC3 achieved a cross-validated classification accuracy of 73.3% (95% CI 60.7–84.4, p < 0.05). eNose-derived breathprints showed preliminary discriminatory potential between IPF and autoimmune UIP, supporting further validation of this non-invasive adjunctive approach. Breathomics represents a promising non-invasive adjunctive tool for phenotyping fibrotic interstitial lung diseases, though larger validation studies integrating clinical and biological data are warranted. Full article

Femtosecond laser processing has become a powerful approach for high-precision micro- and nanofabrication in transparent materials, owing to its ultrashort pulse duration and minimized thermal effects. However, the limited predictability of processing depth remains a major obstacle to practical applications. Here, we present a morphology prediction framework for femtosecond Bessel laser processing of ZnS that integrates plasma physics modeling with deep learning. Through combined experimental measurements and plasma physics simulations, the influence of laser pulse energy on electron density evolution and material removal depth is systematically investigated. The results reveal the dominant roles of multiphoton ionization, avalanche ionization, and free-electron dynamics in deep-volume processing, and demonstrate the strong sensitivity of the processing morphology to the plasma distribution. Conventional plasma models can accurately reproduce the ablation diameter, yet exhibit significant limitations in predicting the processing depth. We propose a physics data-based framework for femtosecond Bessel beam processing, which integrates a depth-residual regression network conditioned on the peak electron density distribution to effectively learn and compensate for systematic modeling errors in plasma-based simulations. This strategy leads to excellent agreement between predicted and experimental processing depths and three-dimensional morphologies under various energy conditions. The model achieves a mean absolute error (MAE) of 4.9 nm at the pixel level for 3D crater reconstruction. Under rigorous crater-grouped cross-validation with Leave-One-Group-Out evaluation, the model achieves a mean R² of 0.74 across 8 independent craters, demonstrating reliable generalization to unseen energy conditions. These results demonstrate that incorporating physical priors into data-driven learning provides an effective pathway to overcoming accuracy limitations in modeling complex laser–matter interactions. This approach offers a reliable tool for quantitative prediction and parameter optimization in deep femtosecond laser processing of transparent materials and enabling highly controllable and reproducible micro- and nanofabrication for advanced photonic and three-dimensional optical applications. Full article

Accurate estimation of battery state of health (SoH) is an important aspect for improving the reliability, safety, and operating efficiency of an energy storage system. This study presents a unified deep learning pipeline for prediction, latent feature extraction, and anomaly detection. A convolution neutral network autoencoder is used to learn compact latent features from a dataset (NASA battery datasets, i.e., B0005, B0006, B0007, and B0018). These features serve as inputs to random forest and linear regression models, which are further compared with the CNN and GRU. The system is evaluated using leave-one-group-out cross-validation to ensure robustness across different batteries. Latent space quality is studied using PSA, t-SNE, and UMAP analyses. Furthermore, clustering performance is measured using the Silhouette Score, and anomalies are detected using reconstruction error and the Isolation Forest technique. The obtained results show that the AE+RF model achieves the best performance, with a 0.0285 root mean square value (RMSE) and a 0.0109 mean absolute error (MAE), with a high 0.96 coefficient of determination (R²). It is evident that AE+RF shows high prediction accuracy and model reliability. The results show that latent features improve prediction accuracy, helping to clearly separate normal and abnormal patterns, providing a robust and accurate approach to battery SoH estimation that is suitable for battery management system applications. Full article

Wheeled multifunctional motorized crossing frames represent a new type of crossing equipment for high-voltage transmission line construction. The initial design is too conservative, having a large safety margin and high material redundancy. Therefore, it is necessary to study a lightweight design version. However, as the structure constitutes an assembly consisting of multiple components, it also exhibits relatively high complexity. In a lightweight design, optimizing multi-component and multi-size parameters can lead to structural interference and separation, seriously affecting the smooth progress of design optimization. Therefore, an optimization design method of a multi-parameter complex assembly structure is proposed to solve this problem. Firstly, the typical stress conditions of the wheeled multifunctional motorized crossing frame were analyzed using its structural model. Then, a finite element model of the beam was established in ANSYS 2021 R1 Workbench, and the mechanical characteristics were analyzed. The results show that the arm support is the key load-bearing component and has significant optimization potential. Subsequently, functional mapping relationships were established among the 14 dimension parameters of the arm support, reducing the number of design variables to six and successfully avoiding component separation or interference during optimization. Through global sensitivity analysis, the height, thickness, and length of the arm body were screened out as the core optimization parameters from six initial design variables. Then, 29 groups of sample points were generated via central composite design (CCD), and a response surface model reflecting the relationships among the arm body’s dimensional parameters, total mass, maximum stress, and maximum deformation was established using the Kriging method. Leave-one-out cross-validation (LOOCV) was performed, and the coefficients of determination (R²) for model fitting were all higher than 0.995, indicating extremely high prediction accuracy. Taking mass and deformation minimization as the optimization objectives, the MOGA algorithm was adopted to perform multi-objective optimization and determine the optimal engineering parameters. Simulation verification was conducted on the optimized arm support, and an eigenvalue buckling analysis was performed simultaneously to verify structural stability. Finally, the proposed optimization method was experimentally verified through mechanical performance tests of the full-scale prototype under symmetric and eccentric loads. The results show that the mass of the optimized arm support is reduced from 217.73 kg to 189.8 kg, with a weight reduction rate of 12.8%. Under an eccentric load of 70,000 N, the maximum deformation of the arm support is 8.9763 mm, the maximum equivalent stress is 314.86 MPa, and the buckling load factor is 6.08, all of which meet the requirements for structural stiffness, strength, and buckling stability. The maximum error between the experimental and finite element results is only 4.64%, verifying the accuracy and reliability of the proposed method. The proposed optimization methodology, validated on a wheeled multifunctional motorized crossing frame, serves as a transferable paradigm for the lightweight design of complex assemblies with coupled dimensional constraints, thereby offering a general reference for the structural optimization of multi-component transmission line equipment, construction machinery, and other multi-component engineering systems. Full article

Background: Osteotomy around the knee (OAK) is a joint-preserving surgery for knee osteoarthritis, yet some patients experience suboptimal outcomes. Preoperative identification of high-risk patients remains challenging. This study aimed to develop a machine learning model to predict clinical outcomes after OAK using preoperative gait acceleration data from inertial measurement units (IMUs). Methods: This multicenter prospective study enrolled patients undergoing OAK. Preoperative gait was recorded using synchronized IMUs placed on the lumbar spine and tibia. Lumbar and tibial signals were used for gait-cycle segmentation, while wavelet-based time–frequency features were extracted from tibial acceleration only. Outcomes were defined by achievement of the minimal clinically important difference in ≥3 KOOS subscales at 2-year follow-up (Good vs. Poor). Continuous wavelet transform features (5–20 Hz) were summarized as mean and standard deviation across six stance subphases. A Random Undersampling Boost classifier was trained and evaluated using nested leave-one-subject-out cross-validation. A sensitivity analysis using logistic regression confirmed that the IMU-based prediction score was independently associated with outcome after adjustment for baseline KOOS (p = 0.047). Results: Of 67 enrolled patients, 37 were classified as Good and 30 as Poor outcome. For machine learning analysis, 1173 tibial acceleration gait-cycle waveforms were usable. The model achieved an AUC of 0.744 (95% CI, 0.610–0.860) using a median of 15 features (range, 5–25) with sensitivity of 0.69 and specificity of 0.72. The most informative predictors were the mean magnitude in the 5–8 Hz band during loading response (0–17%) and variability in the 5–8 Hz band during late stance (67–83%). No significant differences in baseline demographics or radiographic parameters were found between outcome groups. Conclusions: Preoperative IMU-derived gait acceleration features showed moderate-to-good discrimination between outcome groups and may support preoperative risk stratification and individualized perioperative management. Full article

Phenology can confound rice nitrogen diagnosis based on SPAD readings because leaf greenness and nitrogen concentration change nonlinearly with development. We tested whether physiological time, expressed as growing degree days (GDD), can reduce this developmental bias and improve the portability of SPAD-based diagnosis. We analyzed 1141 observations from 20 independent field experiments across five sites, spanning japonica, indica, and hybrid cultivars and nitrogen fertilizer treatments (0–300 kg N ha⁻¹). SPAD was measured on up to five leaf-from-top positions (LFT1–LFT5) and used to predict leaf nitrogen concentration (LNC), plant nitrogen concentration (PNC), and nitrogen nutrition index (NNI). Across group-wise cross-validation by experiment, adding GDD to SPAD consistently improved cross-environment accuracy (mean R² up to 0.75 for LNC and 0.79 for PNC) and markedly weakened residual trends along GDD. Multiplicative SPAD×GDD degraded performance, while explicit interaction terms provided little gain over a simple additive SPAD + GDD form. Interpretable analyses further showed that diagnostic information is concentrated in mid-canopy leaves and shifts with physiological time. Combining GDD with a two-leaf SPAD protocol retained most accuracy for concentration targets, supporting a time-aligned and field-practical approach for robust nitrogen diagnosis. Full article

The strong heterogeneity of clastic reservoirs and the phenomenon of similar log responses for different lithologies (i.e., “same spectrum, different rocks”) significantly weaken feature separability. Furthermore, distribution shifts between different wells cause traditional models to suffer from severe generalization bottlenecks in cross-well applications. To address this critical challenge, this paper proposes a dual-driven framework comprising “Multivariate Feature Enhancement + Dynamic Ensemble”. At the feature level, physics-informed enhancement and multi-scale statistics are introduced to construct a Multivariate high-dimensional feature system, thereby strengthening the representation of geological patterns. At the model level, a sample-aware Dynamic Confidence-Weighted Ensemble (DCWE) strategy is designed to achieve sample-wise adaptive decision-making based on prediction uncertainty, fundamentally breaking through the limitations of fixed weights in static ensembles. This method combines the complementary advantages of Gradient Boosting Decision Trees (GBDT) and deep sequence networks, enabling the simultaneous capture of local textural variations and continuous trends across depths. Based on rigorous Leave-One-Group-Out (LOGO) cross-validation, the proposed framework achieves a maximum accuracy of 84.58%. It significantly reduces the misclassification rate in lithology transition zones and for minority class samples, while maintaining the geological continuity of prediction results. These results verify the significant advantages of the proposed method in cross-well generalization scenarios. Full article

The coconut water market continues to expand, but industrial supply is constrained by the high perishability of fresh coconut water and the need for stabilization routes that preserve quality. This study optimized crossflow microfiltration of coconut water using a silicon carbide (SiC) ceramic membrane, high permeability, chemical/thermal robustness, and cleanability, and assessed the techno-economic feasibility of a green coconut biorefinery producing microfiltered coconut water and coconut pulp. Pressure and temperature were modeled and optimized using a face-centered design (FCD) and artificial neural networks coupled with a genetic algorithm (ANN–GA), considering permeate flux and fouling index (p < 0.05). Both approaches converged to the same operating point, and experimental validation at 75 kPa and 30 °C achieved 605.32 ± 15.34 L h⁻¹ m⁻² and 82.79 ± 1.35% at VRR = 1. Sample-level fit statistics favored ANN (higher R² and lower sample-level errors), whereas condition-wise grouped cross-validation (leave-one-condition-out) indicated higher predictivity and lower RMSE_CV for the quadratic FCD/RSM models across experimental conditions, highlighting response-dependent generalization within the investigated domain. Fouling analysis indicated concentration polarization as the main resistance contribution and a flux-decline behavior best described by the intermediate blocking mechanism. A SuperPro Designer^® simulation over a 20-year project life indicated economic feasibility under baseline assumptions (Internal rate of return—IRR = 23.80%, Net present value—NPV = US$733,761, payback = 2.96 years), with profitability remaining attractive under ±10% selling-price variation. Overall, the process optimization and modeling outcomes align with the economic case, reinforcing the potential of this biorefinery concept for industrial deployment. Full article

Gaming Disorder (GD) is becoming more widely acknowledged as a behavioral addiction characterized by impaired control and functional impairment. While resting-state impairments are well understood, the neurophysiological dynamics during active gameplay remain underexplored. This study identified task-based occipital EEG biomarkers of GD and assessed their diagnostic utility. Occipital EEG (O1/O2) data from 30 participants (15 with GD, 15 controls) collected during active mobile gaming were used in this study. Spectral, temporal, and nonlinear complexity features were extracted. Feature relevance was ranked using Random Forest, and classification performance was evaluated using Leave-One-Subject-Out (LOSO) cross-validation to ensure subject-independent generalization across five models (Random Forest, KNN, SVM, Decision Tree, ANN). The GD group exhibited paradoxical “spectral slowing” during gameplay, characterized by increased Delta/Theta power and decreased Beta activity relative to controls. Beta variability was identified as a key biomarker, reflecting altered attentional stability, while elevated Alpha power suggested potential neural habituation or sensory gating. The Decision Tree classifier emerged as the most robust model, achieving a classification accuracy of 80.0%. Results suggest distinct neurophysiological patterns in GD, where increased low-frequency power may reflect automatized processing or “Neural Efficiency” despite active task engagement. These findings highlight the potential of occipital biomarkers as accessible and objective screening metrics for Gaming Disorder. Full article

Background and Objectives: Accurate breast tumor segmentation in dynamic contrast-enhanced MRI (DCE-MRI) is crucial for treatment planning, therapy monitoring, and quantitative studies of breast cancer response. However, deep learning models often have worse performance when applied to new hospitals because scanner hardware, acquisition protocols, and patient populations differ from those in the training data. This study investigates how such center-related domain shift affects automated breast DCE-MRI tumor segmentation on the multi-center MAMA-MIA dataset. Methods: We trained a standard 3D U-Net for primary tumor segmentation under two evaluation settings. First, we constructed a random patient-wise split that mixes cases from the three main MAMA-MIA center groups (ISPY2, DUKE, NACT) and used this as an in-distribution reference. Second, we designed a balanced leave-one-center-out cross-validation (LoCoCV) protocol in which each center is held out in turn, while training, validation, and test sets are matched in size across folds. Performance was assessed using the Dice similarity coefficient, 95th percentile Hausdorff distance (HD95), sensitivity, specificity, and related overlap measures. Results: On the mixed-center random split, the best three-channel model achieved a mean Dice of about 0.68 and a mean HD95 of about 19.7 mm on the held-out test set, indicating good volumetric overlap and boundary accuracy when training and test distributions match. Under balanced LoCoCV, the one-channel model reached a mean Dice of about 0.45 and a mean HD95 of about 41 mm on unseen centers, with similar averages for the three-channel variant. Compared with the random split baseline, Dice and sensitivity decreased, while HD95 nearly doubled, showing that boundary errors become larger and segmentations less reliable when the model is applied to new centers. Conclusions: A model that performs well on mixed-center random splits can still suffer a substantial loss of accuracy on completely unseen institutions. The balanced LoCoCV design makes this out-of-distribution penalty visible by separating center-related effects from sample size effects. These findings highlight the need for robust multi-center training strategies and explicit cross-center validation before deploying breast DCE-MRI segmentation models in clinical practice. Full article

Automatic modulation classification (AMC) under low signal-to-noise ratio (SNR) and complex channel conditions remains a significant challenge due to the trade-off between robustness and efficiency. This study proposes a lightweight temporal convolutional network (TCN) and Mamba fusion architecture designed to enhance modulation recognition performance. In the modulation signal denoising stage, a non-local adaptive thresholding denoising module (NATM) is introduced to explicitly improve the effective signal-to-noise ratio. In the parallel feature extraction stage, TCN captures local symbol-level dependencies, while Mamba models long-range temporal relationships. In the output stage, their outputs are integrated through additive layer-wise fusion, which prevents parameter explosion. Experiments were conducted on the RadioML 2016.10A, 2016.10B, and 2018.01A datasets with leakage-controlled partitioning strategies including GroupKFold and Leave-One-SNR-Out cross-validation. The proposed method achieves up to a 3.8 dB gain in the required signal-to-noise ratio at 90 percent accuracy compared with state-of-the-art baselines, while maintaining a substantially lower parameter count and reduced inference latency. The denoising module provides clear robustness improvements under low signal-to-noise ratio conditions, particularly below −8 dB. The results show that the proposed network strikes a balance between accuracy and efficiency, highlighting its application potential in real-time wireless receivers under resource constraints. Full article