Search Results (63)

Sugarcane yield prediction plays a pivotal role in enabling farmers to monitor crop development and optimize cultivation practices, guiding harvesting operations for sugar mills. In this study, we established three experimental fields, which were planted with three main sugarcane cultivars in Guangxi, China, respectively, implementing a multi-gradient fertilization design with 39 plots and 810 sampling grids. Multispectral imagery was acquired by unmanned aerial vehicles (UAVs) during five critical growth stages: mid-tillering (T1), late-tillering (T2), mid-elongation (T3), late-elongation (T4), and maturation (T5). Following rigorous image preprocessing (including stitching, geometric correction, and radiometric correction), 16 VIs were extracted. To identify yield-sensitive vegetation indices (VIs), a spectral feature selection criterion combining gray relational analysis and correlation analysis (GRD-r) was proposed. Subsequently, three supervised learning algorithms—Gradient Boosting Decision Tree (GBDT), Random Forest (RF), and Support Vector Machine (SVM)—were employed to develop both single-stage and multi-stage yield prediction models. Results demonstrated that multi-stage models consistently outperformed their single-stage counterparts. Among the single-stage models, the RF model using T3-stage features achieved the highest accuracy (R² = 0.78, RMSE_V = 7.47 t/hm²). The best performance among multi-stage models was obtained using a GBDT model constructed from a combination of DVI (T1), NDVI (T2), TDVI (T3), NDVI (T4), and SRPI (T5), yielding R² = 0.83 and RMSE_V = 6.63 t/hm². This study highlights the advantages of integrating multi-temporal spectral features and advanced machine learning techniques for improving sugarcane yield prediction, providing a theoretical foundation and practical guidance for precision agriculture and harvest logistics. Full article

Accurate estimation of rapeseed yield is crucial for harvesting decisions and improving efficiency and output. Machine learning (ML) models driven by remote sensing data are widely used for yield prediction. This study explores the generality of feature-based rapeseed yield prediction models across different varieties and years. Seven vegetation indices (VIs) and twenty-four texture features (TFs) were calculated from UAV-based imagery. Pearson’s correlation coefficient was used to assess variable sensitivity at different growth stages, and the variable importance score (VIP) from the random forest (RF) model was used for feature selection. Three ML regression methods—RF, support vector regression (SVR), and partial least squares regression (PLSR)—were applied using the single-stage VI, selected multi-stage VI, and multivariate VI-TFs for yield prediction. The best yield model was selected through cross-validation and tested for temporal fit using cross-year data. Results showed that the multi-stage VI and RF model achieved the highest accuracy in the training dataset (R² = 0.93, rRMSE = 7.36%), while the multi-stage VI and PLSR performed best in the test dataset (R² = 0.62, rRMSE = 15.20%). However, this study demonstrated that the addition of TFs could not enhance the robustness of rapeseed yield estimation. Additionally, the model updating strategy improved the RF model’s temporal fit, increasing R² by 25% and reducing the rRMSE to below 10%. This study highlights the potential of the multi-stage VI for rapeseed yield prediction and offers a method to improve the generality of yield prediction models over multiple years, providing a practical approach for meter-scale yield mapping and multi-year prediction. Full article

Cotton (Gossypium hirsutum L.), as a significant economic crop, has undergone significant modernization in planting methods, and its smart irrigation management relies heavily on accurate cotton water content (CWC) estimation. Existing ground-based methods for measuring CWC are constrained by their limited scope and high monitoring costs. Although the development of unmanned aerial vehicle (UAV) technology has provided a new opportunity for large-scale CWC measurements, there remains a gap in the study of CWC estimation in cotton using multi-source and multi-stage data. In this study, we used UAV-based data, including texture features, vegetation indices, and a heat index, and applied four machine learning algorithms, i.e., partial least-squares regression (PLSR), support vector regression (SVR), random forest regression (RFR), and extreme gradient boosting (XGB), to estimate CWC. The findings demonstrate that in a single growth stage, the boll setting stage performs the best, and multi-source and multi-stage inputs can improve the accuracy of CWC estimation, with the best performance of XGB (R² = 0.860). Overall, this study highlights that the synergistic use of multi-source and multi-stage data can effectively improve CWC estimation in cotton, suggesting UAV-based data will lead to a brighter future for precision agriculture. Full article

Background/Objectives: This study presents a comparative analysis of the multistage diagnosis of Alzheimer’s disease (AD), including mild cognitive impairment (MCI), utilizing two distinct types of biomarkers: blood gene expression and clinical biomarker samples. Both of these samples, obtained from participants in the Alzheimer’s Disease Neuroimaging Initiative (ADNI), were independently analyzed utilizing machine learning (ML)-based multiclassifiers. This study applied novel machine learning-based data augmentation techniques to gene expression profile data that are high-dimensional, low-sample-size (HDLSS) and inherently highly imbalanced. The investigation obtained the highest multiclassification performance to date in the multistage diagnosis of Alzheimer’s disease utilizing the blood gene expression profiles of Alzheimer’s Disease Neuroimaging Initiative (ADNI) participants. Based on the performance results obtained, and other factors such as early prediction capabilities, this study compares the efficacies of the two types of biomarkers for multistage diagnosis. This study presents the sole investigation in which multiclassification-based AD stage diagnosis was conducted utilizing blood gene expression data. We obtained the best multiclassification result in both modalities of the ADNI data in terms of F1-score and were able to identify new genetic biomarkers. Methods: The combination of the XGBoost and SFBS (Sequential Floating Backward Selection) methods was used to select the features. We were able to select the 95 most effective gene probe sets out of 49,386. For the clinical study data, eight of the most effective biomarkers were selected using SFBS. A deep learning (DL) classifier was used to identify the stages—cognitive normal (CN), mild cognitive impairment (MCI), and Alzheimer’s disease (AD)/dementia. DL, support vector machine (SVM), gradient boosting (GB), and random forest (RF) classifiers were used for the AD stage detection from gene expression profile data. Because of the high data imbalance in genomic data, borderline oversampling/data augmentation was applied in the model training and original samples for validation. Results: Utilizing clinical data, the highest ROC AUC scores attained were 0.989, 0.927, and 0.907 for the identification of the CN, MCI, and dementia stages, respectively. The highest F1 scores achieved were 0.971, 0.939, and 0.886. Employing gene expression data, we obtained ROC AUC scores of 0.763, 0.761, and 0.706 for the CN, MCI, and dementia stages, respectively, and F1 scores of 0.71, 0.77, and 0.53 for CN, MCI, and dementia, respectively. Conclusions: This represents the best outcome to date for AD stage diagnosis from ADNI blood gene expression profile data utilizing multiclassification techniques. The results indicated that our multiclassification model effectively manages the imbalanced data of a high-dimension, low-sample-size (HDLSS) nature to identify samples of the minority class. MAPK14, PLG, FZD2, FXYD6, and TEP1 are among the novel genes identified as being associated with AD risk. Full article

The Transductive Support Vector Machine (TSVM) is an effective semi-supervised learning algorithm vulnerable to adversarial sample attacks. This paper proposes a new adversarial attack method called the Multi-Stage Dual-Perturbation Attack (MSDPA), specifically targeted at TSVMs. The MSDPA has two phases: initial adversarial samples are generated by an arbitrary range attack, and finer attacks are performed on critical features to induce the TSVM to generate false predictions. To improve the TSVM’s defense against MSDPAs, we incorporate adversarial training into the TSVM’s loss function to minimize the loss of both standard and adversarial samples during the training process. The improved TSVM loss function considers the adversarial samples’ effect and enhances the model’s adversarial robustness. Experimental results on several standard datasets show that our proposed adversarial defense-enhanced TSVM (adv-TSVM) performs better in classification accuracy and adversarial robustness than the native TSVM and other semi-supervised baseline algorithms, such as S³VM. This study provides a new solution to improve the defense capability of kernel methods in an adversarial setting. Full article

Accurate crop yield prediction is crucial for optimizing cultivation practices and informing breeding decisions. Integrating UAV-acquired multispectral datasets with advanced machine learning methodologies has markedly refined the accuracy of crop yield forecasting. This study aimed to construct a robust and versatile yield prediction model for multi-genotyped oat varieties by investigating 14 modeling scenarios that combine multispectral data from four key growth stages. An ensemble learning framework, StackReg, was constructed by stacking four base algorithms—ridge regression (RR), support vector machines (SVM), Cubist, and extreme gradient boosting (XGBoost)—to predict oat yield. The results show that, for single growth stages, base models achieved R² values within the interval of 0.02 to 0.60 and RMSEs ranging from 391.50 to 620.49 kg/ha. By comparison, the StackReg improved performance, with R² values extending from 0.25 to 0.61 and RMSEs narrowing to 385.33 and 542.02 kg/ha. In dual-stage and multi-stage settings, the StackReg consistently surpassed the base models, reaching R² values of up to 0.65 and RMSE values as low as 371.77 kg/ha. These findings underscored the potential of combining UAV-derived multispectral imagery with ensemble learning for high-throughput phenotyping and yield forecasting, advancing precision agriculture in oat cultivation. Full article

Breeding high-photosynthetic-efficiency wheat varieties is a crucial link in safeguarding national food security. Traditional identification methods necessitate laborious on-site observation and measurement, consuming time and effort. Leveraging unmanned aerial vehicle (UAV) remote sensing technology to forecast photosynthetic indices opens up the potential for swiftly discerning high-photosynthetic-efficiency wheat varieties. The objective of this research is to develop a multi-stage predictive model encompassing nine photosynthetic indicators at the field scale for wheat breeding. These indices include soil and plant analyzer development (SPAD), leaf area index (LAI), net photosynthetic rate (Pn), transpiration rate (Tr), intercellular CO₂ concentration (Ci), stomatal conductance (Gsw), photochemical quantum efficiency (PhiPS2), PSII reaction center excitation energy capture efficiency (Fv’/Fm’), and photochemical quenching coefficient (qP). The ultimate goal is to differentiate high-photosynthetic-efficiency wheat varieties through model-based predictions. This research gathered red, green, and blue spectrum (RGB) and multispectral (MS) images of eleven wheat varieties at the stages of jointing, heading, flowering, and filling. Vegetation indices (VIs) and texture features (TFs) were extracted as input variables. Three machine learning regression models (Support Vector Machine Regression (SVR), Random Forest (RF), and BP Neural Network (BPNN)) were employed to construct predictive models for nine photosynthetic indices across multiple growth stages. Furthermore, the research conducted principal component analysis (PCA) and membership function analysis on the predicted values of the optimal models for each indicator, established a comprehensive evaluation index for high photosynthetic efficiency, and employed cluster analysis to screen the test materials. The cluster analysis categorized the eleven varieties into three groups, with SH06144 and Yannong 188 demonstrating higher photosynthetic efficiency. The moderately efficient group comprises Liangxing 19, SH05604, SH06085, Chaomai 777, SH05292, Jimai 22, and Guigu 820, totaling seven varieties. Xinmai 916 and Jinong 114 fall into the category of lower photosynthetic efficiency, aligning closely with the results of the clustering analysis based on actual measurements. The findings suggest that employing UAV-based multi-source remote sensing technology to identify wheat varieties with high photosynthetic efficiency is feasible. The study results provide a theoretical basis for winter wheat phenotypic monitoring at the breeding field scale using UAV-based multi-source remote sensing, offering valuable insights for the advancement of smart breeding practices for high-photosynthetic-efficiency wheat varieties. Full article

Corn syrup is a cost-effective sweetener ingredient for the food industry. In producing syrup from corn, process control to enhance and/or maintain a constant dextrose equivalent value (DE) is a constant challenge, especially in semi-automated/batch production settings, which are common in small to medium-size factories. Existing work has focused on continuous process control to keep parameter values within a setpoint. The machine learning method applied is for time series data. This study focuses on building process control models to enable semi-automation in small to medium-size factories in which the data are not as time dependent. Correlation coefficients were used to identify key process parameters that contribute to feed pH value and DE. Artificial neural network (ANN), support vector machine (SVM), and linear regression (LR) models were built to predict feed pH and DE. The results suggest (1) model accuracy ranges from 91% to 96%; (2) the ANN models yielded about 1% to 3% higher accuracy than the SVM and LR models and the prediction accuracy is robust even with as few as six data sets; (3) both the SVM and ANN models have noise tolerant properties, but ANN has a higher noise tolerance than SVM; (4) SVM performance can be hindered when using high-dimensional data sets; (5) the LR model yields higher variation in accuracy prediction than ANN and SVM; (6) distribution fitting is a good approach for generating data; however, fidelity of fitting can greatly impact accuracy; and (7) multi-stage models yield higher accuracy than single-stage models, but there are pros and cons to each approach. Full article

The long-distance multi-stage pressurized pump station water delivery system involves numerous valve closure parameters, complicating the rapid identification of an optimal valve closure scheme that satisfies multiple transient flow oscillation protection requirements. A hydraulic transient model was established based on transient flow calculation theory to address this challenge. Decision biases were identified using the Analytic Hierarchy Process and the Entropy Weight Method. A multi-objective optimization model, incorporating Support Vector Regression (SVR) and the Beluga Whale Optimization (BWO) algorithm, iteratively searches for optimal schemes under different biases. The results indicate that Support Vector Regression exhibits optimal performance, while Beluga Whale Optimization demonstrates excellent performance. The optimal schemes obtained from the multi-objective optimization model meet the transient flow protection requirements of the water delivery system. The study demonstrates that this model effectively solves the multi-objective optimization problem for water hammer protection in multi-stage pressurized pump station water delivery systems. Full article

Alzheimer’s Disease (AD) poses a significant health risk globally, particularly among the elderly population. Recent studies underscore its prevalence, with over 50% of elderly Japanese facing a lifetime risk of dementia, primarily attributed to AD. As the most prevalent form of dementia, AD gradually erodes brain cells, leading to severe neurological decline. In this scenario, it is important to develop an automatic AD-detection system, and many researchers have been working to develop an AD-detection system by taking advantage of the advancement of deep learning (DL) techniques, which have shown promising results in various domains, including medical image analysis. However, existing approaches for AD detection often suffer from limited performance due to the complexities associated with training hierarchical convolutional neural networks (CNNs). In this paper, we introduce a novel multi-stage deep neural network architecture based on residual functions to address the limitations of existing AD-detection approaches. Inspired by the success of residual networks (ResNets) in image-classification tasks, our proposed system comprises five stages, each explicitly formulated to enhance feature effectiveness while maintaining model depth. Following feature extraction, a deep learning-based feature-selection module is applied to mitigate overfitting, incorporating batch normalization, dropout and fully connected layers. Subsequently, machine learning (ML)-based classification algorithms, including Support Vector Machines (SVM), Random Forest (RF) and SoftMax, are employed for classification tasks. Comprehensive evaluations conducted on three benchmark datasets, namely ADNI1: Complete 1Yr 1.5T, MIRAID and OASIS Kaggle, demonstrate the efficacy of our proposed model. Impressively, our model achieves accuracy rates of 99.47%, 99.10% and 99.70% for ADNI1: Complete 1Yr 1.5T, MIRAID and OASIS datasets, respectively, outperforming existing systems in binary class problems. Our proposed model represents a significant advancement in the AD-analysis domain. Full article

In Nuclear Power Plants (NPPs), online condition monitoring and the fault detection of structures, systems and components (SSCs) can aid in guaranteeing safe operation. The use of data-driven methods for these tasks is limited by the requirement of physically consistent outcomes, particularly in safety-critical systems. Considering the importance of regulating valves (e.g., safety relief valves and main steam isolation valves), this work proposes a multistage Physics-Informed Neural Network (PINN) for fault detection in such components. Two stages of the PINN are built by developing the process model of the regulating valve, which integrates the basic valve sizing equation into the loss function to jointly train the two stages of the PINN. In the 1st stage, a shallow Neural Network (NN) with only one hidden layer is developed to estimate the equivalent flow coefficient (a key performance indicator of regulating valves) using the displacement of the valve as input. In the 2nd stage, a Deep Neural Network (DNN) is developed to estimate the flow rate expected in normal conditions using inputs such as the estimated flow coefficient from the 1st stage, the differential pressure, and the fluid temperature. Then, the residual, i.e., the difference between the estimated and measured flow rates, is fed into a Deep Support Vector Data Description (DeepSVDD) to detect the occurrence of faults. Moreover, the deviation between the estimated flow coefficients of normal and faulty conditions is used to interpret the consistency of the detection result with physics. The proposed method is, first, applied to a simulation case implemented to emulate the operating characteristics of regulating the valves of NPPs and then validated on a real-world case study based on the DAMADICS benchmark. Compared to state-of-the-art fault detection methods, the obtained results from the proposed method show effective fault detection performance and reasonable flow coefficient estimation, thus guaranteeing the physical consistency of the detection results. Full article

The reachable set of controlled dynamical systems is the set of all reachable states from an initial condition over a certain time horizon, subject to operational constraints and exogenous disturbances. In astrodynamics, rapid approximation of reachable sets is invaluable for trajectory planning, collision avoidance, and ensuring safe and optimal performance in complex dynamics. Leveraging the connection between minimum-time trajectories and the boundary of reachable sets, we propose a sampling-based method for rapid and efficient approximation of reachable sets for finite- and low-thrust spacecraft. The proposed method combines a minimum-time multi-stage indirect formulation with the celebrated primer vector theory. Reachable sets are generated under two-body and circular restricted three-body (CR3B) dynamics. For the two-body dynamics, reachable sets are generated for (1) the heliocentric phase of a benchmark Earth-to-Mars problem, (2) two scenarios with uncertainties in the initial position and velocity of the spacecraft at the time of departure from Earth, and (3) a scenario with a bounded single impulse at the time of departure from Earth. For the CR3B dynamics, several cislunar applications are considered, including L1 Halo orbit, L2 Halo orbit, and Lunar Gateway 9:2 NRHO. The results indicate that low-thrust spacecraft reachable sets coincide with invariant manifolds existing in multi-body dynamical environments. The proposed method serves as a valuable tool for qualitatively analyzing the evolution of reachable sets under complex dynamics, which would otherwise be either incoherent with existing grid-based reachability approaches or computationally intractable with a complete Hamilton–Jacobi–Bellman method. Full article

The application of remote sensing, which is non-destructive and cost-efficient, has been widely used in crop monitoring and management. This study used a built-in multispectral imager on a small unmanned aerial vehicle (UAV) to capture multispectral images in five different spectral bands (blue, green, red, red edge, and near-infrared), instead of satellite-captured data, to monitor soybean growth in a field. The field experiment was conducted in a soybean field at the Mississippi State University Experiment Station near Pontotoc, MS, USA. The experiment consisted of five cover crops (Cereal Rye, Vetch, Wheat, Mustard plus Cereal Rye, and native vegetation) planted in the winter and three fertilizer treatments (Fertilizer, Poultry Liter, and None) applied before planting the soybean. During the soybean growing season in 2022, eight UAV imaging flyovers were conducted, spread across the growth season. UAV image-derived vegetation indices (VIs) coupled with machine learning (ML) models were computed for characterizing soybean growth at different stages across the season. The aim of this study focuses on monitoring soybean growth to predict yield, using 14 VIs including CC (Canopy Cover), NDVI (Normalized Difference Vegetation Index), GNDVI (Green Normalized Difference Vegetation Index), EVI2 (Enhanced Vegetation Index 2), and others. Different machine learning algorithms including Linear Regression (LR), Support Vector Machine (SVM), and Random Forest (RF) are used for this purpose. The stage of the initial pod development was shown as having the best predictability for earliest soybean yield prediction. CC, NDVI, and NAVI (Normalized area vegetation index) were shown as the best VIs for yield prediction. The RMSE was found to be about 134.5 to 511.11 kg ha⁻¹ in the different yield models, whereas it was 605.26 to 685.96 kg ha⁻¹ in the cross-validated models. Due to the limited number of training and testing samples in the K-fold cross-validation, the models’ results changed to some extent. Nevertheless, the results of this study will be useful for the application of UAV remote sensing to provide information for soybean production and management. This study demonstrates that VIs coupled with ML models can be used in multistage soybean yield prediction at a farm scale, even with a limited number of training samples. Full article

Accurately estimating the breast muscle weight of broilers is important for poultry production. However, existing related methods are plagued by cumbersome processes and limited automation. To address these issues, this study proposed an efficient method for predicting the breast muscle weight of broilers. First, because existing deep learning models struggle to strike a balance between accuracy and memory consumption, this study designed a multistage attention enhancement fusion segmentation network (MAEFNet) to automatically acquire pectoral muscle mask images from X-ray images. MAEFNet employs the pruned MobileNetV3 as the encoder to efficiently capture features and adopts a novel decoder to enhance and fuse the effective features at various stages. Next, the selected shape features were automatically extracted from the mask images. Finally, these features, including live weight, were input to the SVR (Support Vector Regression) model to predict breast muscle weight. MAEFNet achieved the highest intersection over union (96.35%) with the lowest parameter count (1.51 M) compared to the other segmentation models. The SVR model performed best (R² = 0.8810) compared to the other prediction models in the five-fold cross-validation. The research findings can be applied to broiler production and breeding, reducing measurement costs, and enhancing breeding efficiency. Full article

As threat vectors and adversarial capabilities evolve, Cloud-Assisted Connected and Autonomous Vehicles (CCAVs) are becoming more vulnerable to cyberattacks. Several established threat analysis and risk assessment (TARA) methodologies are publicly available to address the evolving threat landscape. However, these methodologies inadequately capture the threat data of CCAVs, resulting in poorly defined threat boundaries or the reduced efficacy of the TARA. This is due to multiple factors, including complex hardware–software interactions, rapid technological advancements, outdated security frameworks, heterogeneous standards and protocols, and human errors in CCAV systems. To address these factors, this study begins by systematically evaluating TARA methods and applying the Spoofing, Tampering, Repudiation, Information disclosure, Denial of service, and Elevation of privileges (STRIDE) threat model and Damage, Reproducibility, Exploitability, Affected Users, and Discoverability (DREAD) risk assessment to target system architectures. This study identifies vulnerabilities, quantifies risks, and methodically examines defined data processing components. In addition, this study offers an attack tree to delineate attack vectors and provides a novel defense taxonomy against identified risks. This article demonstrates the efficacy of the TARA in systematically capturing compromised security requirements, threats, limits, and associated risks with greater precision. By doing so, we further discuss the challenges in protecting hardware–software assets against multi-staged attacks due to emerging vulnerabilities. As a result, this research informs advanced threat analyses and risk management strategies for enhanced security engineering of cyberphysical CCAV systems. Full article