Alfalfa is a perennial legume forage crop harvested several times a year, and identifying the primary factors that influence stem cutting effectiveness could substantially improve harvesting efficiency. While stem cutting directly impacts mechanical energy consumption and operational efficiency, existing research has struggled to clarify the complex interactions between simultaneously acting multiple blade variables. To address these limitations, we developed a dataset focusing on maximum cutting force for alfalfa stems by evaluating the blade’s approach angle, shear angle, sliding angle, cutting speed, and sharpness index through one-factor-at-a-time (OFAT) experiments and fractional factorial and Box–Behnken designs. Random forest, extra trees, and CatBoost regression models were then trained on this dataset and compared with a conventional second-order response-surface model. CatBoost outperformed the other models, achieving a coefficient of determination of 0.7013 and a root-mean-square error of 0.5723 on the Yeo–Johnson transformed scale, corresponding to 2.885 N on the original force scale. Variable importance and SHAP analyses identified stem cross-sectional area as the most important predictor of maximum cutting force overall and, among the controllable blade design variables, the sharpness index and approach angle as the dominant predictors. Among the geometric blade angle pairs, the shear angle × approach angle interaction yielded the widest predicted force range. While the individual importance of the shear angle was low, it showed significant interaction effects when combined with other blade variables that substantially altered predictions. The CatBoost model was employed as the fitness function of a genetic algorithm to minimize cutting force, yielding optimal design ranges of 40.2–42.3° for shear angle, 33.9–40° for approach angle, 24.1–29.54° for sliding angle, 400–404.41 mm/min for cutting speed, and 186.19–233.04 g for the sharpness index. These ranges are preliminary, model-predicted optima obtained under quasi-static conditions of 100–500 mm/min and require experimental validation before application to field harvesters.
To address the issue that mismatches between operating conditions and control parameters during combine harvester operations lead to increased grain breakage rate, loss rate, and impurity rate, this study proposes a reinforcement learning–based optimization control method for operational parameters, supported by self-developed grain quality and loss rate sensors. A reinforcement learning simulation environment for the threshing and cleaning system of the combine harvester is established. On this basis, the effects of four types of reward function components—optimal value, constant-offset optimal value, reference-offset optimal value, and minimum power consumption—as well as their combinations, on the control effectiveness of the reinforcement learning algorithm are systematically analyzed. Simulation results demonstrate that the combination of multiple reward functions improves the performance of the algorithm within the range of operating conditions evaluated in this study. The four-combined reward reduces impurity rate by 6.6
The objective of this study was to address the challenge of reliable backfat thickness estimation in pig carcasses, often hindered by muscle coverage, fat deposits, and bloodstains, by proposing a robust and efficient dual-stage deep learning network. An enhanced U-Net architecture was proposed, integrated with chest cavity–guided localization to form a dual-stage network. A custom image acquisition system was developed to collect carcass images under real slaughterhouse conditions for dataset construction. The baseline U-Net was enhanced with a dual-attention mechanism (DANet) and residual connections to improve structural edge perception of key anatomical regions. For precise localization, multiple edge detection methods (Roberts, Sobel, Laplacian, and Canny) were compared to extract the chest cavity contour and identify the terminal point of the sternum, defining the optimal measurement site. A secondary segmentation stage, guided by the localized chest cavity, was then employed to focus on critical regions and enhance detection accuracy. The proposed ARU-PCSN network achieved an average IoU of 97.12
This research aimed to examine the effects of slow freezing pretreatment (SFT) on the drying process of sliced ginger, focusing on the drying kinetics and color change, energy consumption, microstructure, and the characteristics of brewed ginger. The ginger slices underwent pretreatment by slow freezing for various durations (0, 24, 48, and 72 h) before hot air drying at 60 °C. The performance of six drying kinetic models and four color change models was evaluated, and the most representative model was chosen. The brewed ginger was characterized by color and phytochemical properties. The results exhibited that the SFT produced cavities among the cell walls in the ginger slices microstructure. These cavities increased the drying rate, thereby lowering the total energy consumption (from 3.44 to 3.18 kWh/kg of fresh ginger). Moreover, they enhance the browning reaction during freezing, triggering the ginger’s color change. The 72 h of freezing produced the highest color differences (ΔE). Therefore, the SFT for 24 h was found to be the best treatment for accelerating the drying rate while minimizing color change in ginger. The two-term exponential and fractional conversion models were the most suitable to represent the drying kinetics and color change of ginger slices, respectively. Potassium was the most abundant mineral found in the ginger slices. In addition, higher drying rates and larger cavities induced by SFT enhanced the total phenol content (TPC) and antioxidant activity in brewed ginger, but decreased the pH and extractable matter. The SFT for 24 h has been shown to increase the drying rate, reduce energy consumption, minimize the alteration of ginger color, and preserve the phytochemical properties of brewed gingers.
Prompt detection of leaf wetness with high accuracy is essential for effective disease management in crops, especially strawberries, where diseases like Botrytis fruit rot are major threats to yield. Traditional flat-plate sensors have limitations that reduce the reliability of advisory systems like the Strawberry Advisory System (SAS). This study aimed to overcome these challenges by developing an innovative and reliable vision-based detection system. This study proposed a hybrid deep learning model combining a Convolutional Neural Network (CNN) with a Long Short-Term Memory (LSTM) network. A ConvLSTM model was designed to classify surface wetness as either dry or wet by analyzing sequences of images from a high-resolution camera. The CNN effectively extracted spatial features to detect minute droplets, while the LSTM captured temporal dependencies to understand patterns in droplet evolution. The model also employed Convolutional Block Attention Modules (CBAM) to enhance feature relevance and used Focal Loss to address the class imbalance problem. The model achieved 97
Controlled atmosphere (CA) storage technologies extend the shelf life of agricultural products by precisely regulating oxygen (O2) and carbon dioxide (CO2) concentrations, utilizing devices such as nitrogen (N2) generators, CO2 scrubbers, and ethylene (C2H4) scrubbers. This study aims to analyze the gas regulation characteristics of a CA storage room, focusing on the efficiencies of gas control equipment under various operational conditions. Specifically, the efficiencies of O2, CO2, and C2H4 removal mechanisms in a CA storage system were investigated, with emphasis on the impact on gas exchange processes and storage stability. Experiments were conducted in an airtight CA storage room (12.4 m3) under controlled conditions, with manipulation of O2, CO2, and C₂H₄ concentrations. The pull-down time for O2 reduction was measured to evaluate gas control efficiency, and the effects of CO2 and C2H4 removal processes were assessed under different operational modes. The N2 supply flow rate significantly influenced the completion time for gas regulation; higher flow rates (3 Nm3/h versus 1 Nm3/h) reduced regulation time by 49.8
This study analyzes frost-damage risks to horticultural crops under changing climatic conditions and reviews practical protection strategies. This review synthesizes climatological and observational data, physiological research on cold hardiness and crop-specific critical temperatures, meteorological studies of frost formation, and recent advances in frost-protection technologies. It also incorporates studies on frost- and damage-probability estimation and short-term frost prediction. Frost damage varies widely with crop species, developmental stage, and local microclimate, with buds and flowers being the most vulnerable organs. Passive strategies, including site selection, soil and canopy management, and cultivar choice, can reduce exposure risk. Active methods—such as heaters, wind machines, and over-canopy sprinkling—can effectively mitigate frost injury when operated under appropriate frost conditions. Recent empirical cooling-rate models, dew-point-based approaches, and energy-balance formulations provide practical tools for forecasting minimum temperatures and estimating freeze-injury likelihood based on crop-specific critical thresholds. However, their effectiveness depends strongly on frost type and site-specific conditions. No single frost-protection strategy ensures adequate prevention. Integrated approaches combining microclimate monitoring, site-specific data, probabilistic risk assessment, predictive tools, and adaptive management are required to stabilize yields under climate variability. Effective frost protection therefore requires engineering solutions tailored to local conditions, supporting growers and engineers in sustaining agricultural productivity.
This study investigates the effects of soil moisture content and mechanical impact on soil detachment efficiency from garlic roots to optimize post-harvest processing systems and provide quantitative design criteria for mechanical soil removal equipment. Garlic samples (n = 180) with two moisture levels representing practical drying conditions (12
Exceeding permissible gas concentrations in poultry houses causes stress in birds, affects their respiration, reduces egg production, and increases the risk of disease. Therefore, monitoring gas concentrations in poultry houses is critical for hen health and productivity. This study presents a low-cost Internet of things (IoT) and machine learning (ML) system for monitoring and predicting critical air quality parameters in a laying hen house, to enable proactive environmental management in agricultural settings. This study also compared the environmental differences between closed (mechanically controlled) and open (naturally ventilated) systems and determined the feasibility of predicting gas concentrations (CO2, MCG, H2S) using measurable temperature (Temp.) and humidity (RH) data. The methodology involved deploying a multi-sensor system (DHT22, MQ-2, MQ-136, MG-811) in a laying hen house, Ministry of Agriculture, Poultry Research Directorate in Iraq, over 25 days. The collected dataset of 1200 samples was divided into 960 samples (80
Intercropping improves agricultural sustainability by enhancing biodiversity and optimizing resource use, but its adoption is limited by labor-intensive and crop-specific seeding methods. The primary aim of this study was to develop and evaluate a mechatronic-controlled seed metering mechanism that enables precise, real-time, and crop-specific seed placement for intercropping, addressing the challenges of variable seed spacing and seed rate. A novel seed metering system was designed, comprising a compartmentalized seed box, metering rollers for large, medium, and small seeds, and an electronic control system. The control unit included a rotary encoder, stepper motors with drivers, a microcontroller, and a Bluetooth module. A custom Android application enabled real-time input of seeding parameters. The system was programmed using the Arduino IDE and evaluated in the laboratory using a sticky belt setup. Tests were conducted across varying belt speeds (2–5 km/h), cell shapes (circular, elliptical, triangular), and cell counts. Key performance metrics— variation in seed spacing (VSS), missing index (MI), multiple index (MU), quality of feed index (QFI), and precision index (PI)— were analyzed. Optimization was performed using response surface methodology. Elliptical cells showed superior performance with the lowest VSS and MU and the highest QFI. Circular cells had the lowest MI but the highest MU, while triangular cells exhibited intermediate results. Optimized configurations achieved up to 99.5
This study compared the agricultural tractor emission inventory systems in Korea, the United States, and Europe to identify discrepancies between inventory reference values (CAPSS, EPA, EEA) and field measurements obtained using a Portable Emission Measurement System (PEMS) under real-world operating conditions. Emission data derived from PEMS-based studies were compiled to summarize emission factors (EFs) and load factors (LFs) across various agricultural tasks. These values were systematically compared with the default parameters used in each national inventory. Additionally, four estimation scenarios were developed—inventory defaults, measured LF, measured EF, and combined measured EF and LF—to calculate and contrast emissions during idling, driving, plowing, and rotary tillage. Measured LFs ranged from 0.2 to 0.9, varying significantly by task, even among tractors with similar rated power. Relative to inventory defaults, Korea (CAPSS = 0.48) underestimated LF by approximately 25
Field experiments for optimising tractor–implement systems are often constrained by high costs, long durations, and variability in weather and soil conditions. To address these limitations, this study developed a draft force prediction model using the discrete element method (DEM) combined with field-measured soil properties. Soil samples collected from cultivated and non-cultivated fields were tested to determine shear strength, water content, and bulk density. The measured properties were incorporated into a calibrated Edinburgh elasto-plastic adhesion (EEPA) contact model to reproduce the reaction forces generated during soil–tool interactions in sandy loam. Soil characteristics are measured in two distinct fields: Field A (non-cultivated soil) and Field B (cultivated soil), with the Edinburgh elasto-plastic adhesion (EEPA) contact model applied to replicate the reaction forces observed during soil–tool interactions in sandy loam. The measured bulk density, water content, and shear strength were used to calibrate the DEM soil model by adjusting particle mass and surface energy until the simulated forces between soil particles aligned with the experimentally observed values. Using the calibrated soil model, draft-force simulations were performed at tillage depths of 8, 12, and 16 cm under the same operating conditions as the reference field. The predicted draft forces were then integrated with field-measured workload data—including drawbar power, slip ratio, and total power requirement—to calculate tractive efficiency for each depth. Finally, the optimal tillage depth for each soil type was determined as the depth that maximised tractive efficiency while maintaining draft forces within the operable range of field conditions. The results indicate that Field A exhibited a higher bulk density and shear strength owing to long-term natural compaction compared to Field B. Tractive efficiency calculations for varying tillage depths (8, 12, and 16 cm) revealed that Field A achieved 60.98
Vacuum seed-metering devices (VSMDs) are essential components of precision seeders, responsible for distributing seeds uniformly to maximize crop yield. Although different VSMD models share similar key components, differences in the design of these components can significantly influence performance and, consequently, seeding uniformity. Therefore, it is crucial to understand how the design characteristics of these key components affect their operation. This study evaluated the impact of the key component design differences of four VSMD models on their performance and seeding uniformity. For each VSMDs, the following procedures were conducted: (1) detailed characterization of the key components, including the vacuum chamber, seed plate, seed-cleaning device, and seed-deflector; (2) measurement of vacuum pressure distribution within each device; and (3) development of a machine vision algorithm to determine and analyze seed trajectories, as well as to evaluate seeding uniformity. All results were then compared with seeding uniformity outcomes to identify the impact of each key component on the performance of each VSMD. The evaluation revealed that: (1) the VSMDs showed clear differences in their key components; (2) analysis of pressure distribution demonstrated that chambers with rounded walls allow for higher and more stable vacuum pressure levels; (3) seed fall trajectories with launch angles close to 0° showed less variation regardless of operating conditions; (4) secondary seed-cleaning devices effectively remove excess seeds; (5) seed-deflectors reduce trajectory dispersion under appropriate operating conditions; and (6) the developed machine vision algorithm proved to be reliable, facilitating the evaluation of seeding uniformity. Design differences in key components significantly affected VSMD performance. Although all devices reached qualified index values close to 98
Grain quality is typically assessed during the reception and shipment stages at storage and processing units through physical classification. However, traditional methods based on visual inspection are often subjective, imprecise, and time-consuming. This study aimed to evaluate the use of near-infrared spectroscopy (NIRS) and hyperspectral sensors combined with machine learning algorithms to determine the physicochemical properties and classify the quality of white, parboiled, black, and red rice. Spectral data were acquired using a FieldSpec 4 Jr spectroradiometer covering the 350–2500-nm range, with analysis focused on the 350–750-nm and 1400–2100-nm intervals. These spectral windows correspond to the most informative absorption features associated with O–H, C–H, and N–H bonds, directly related to the moisture, protein, starch, and lipid contents of rice grains. Samples were classified according to regulatory standards, and representative samples were subdivided into 100 sub-samples of 20 g each. Distinct nutritional and physicochemical profiles were observed among rice types. Principal component analysis (PCA) effectively discriminated compositional characteristics, emphasizing the higher nutritional value of pigmented rice. Hyperspectral signatures revealed distinct spectral differences among rice types based on their physicochemical composition. The combination of NIRS, hyperspectral sensors, and machine learning algorithms achieved high accuracy across all evaluation metrics, with the J48, SL, RF, and SVM models delivering the best performance in rice quality classification. The integration of NIR spectroscopy, hyperspectral sensing, and machine learning models provides a rapid, non-destructive, and highly accurate approach for assessing the physicochemical quality of rice in storage and processing facilities. This methodology demonstrates strong potential for enhancing efficiency, reproducibility, and objectivity in grain quality monitoring, offering a data-driven alternative to traditional visual inspection methods.
This study evaluated UVC-induced fluorescence imaging as an alternative to ATP bioluminescence testing for detecting meat residues on food-contact surfaces. Objectives included assessing classification accuracy, correlation with ATP RLUs, key predictors, and model robustness. Fluorescence imaging at 275 nm and ATP assays produced 957 images, from which 23 intensity, texture, and edge features were extracted. A TabNetClassifier trained with cross-validation and SMOTETomek resampling achieved a Matthews correlation coefficient and Cohen’s Kappa of 0.77. GEE analysis showed no significant effect of resolution, gain, or exposure on sensitivity or specificity, with borderline impact for wood surfaces (p = 0.08). Texture features, especially Local Binary Patterns, were most predictive. Fluorescence imaging with machine learning is a reliable alternative to ATP testing, offering consistent performance and minimal parameter sensitivity. Future work should refine calibration for specific surfaces and address practical deployment issues, including environmental variability, UVC safety, and integration with automated systems.
This study investigated the impact of sampling frequency on the spatial interpolation accuracy of tractor performance data and identified an optimal frequency range that balances data processing efficiency with spatial reliability. A 78 kW tractor equipped with sensors and a data acquisition system was tested in two experimental fields. Eight performance variables—engine torque (ET), engine speed (ES), engine power (EP), engine fuel rate (EFR), travel speed (TS), draft force (DF), traction power (TP), and axle power (AP)—were recorded at 100 Hz and subsequently downsampled to 50, 20, 10, 5, and 1 Hz. Spatial interpolation was performed using the Kriging method in ArcGIS Pro, and prediction accuracy was evaluated with the coefficient of determination, normalized root mean square error, and a composite performance index (CPI). Most variables maintained stable spatial interpretation performance, recording R2 values above 0.830 at sampling frequencies of 10 Hz or higher. ET gradually declined as the sampling frequency decreased, with a pronounced drop in performance observed below 10 Hz. Traction-related variables (DF, TP) were sensitive to downsampling, exhibiting substantial decreases in R2 and notable increases in NRMSE at 5 Hz and lower. Frequencies of 10 Hz or higher ensured stable performance while balancing data efficiency, with 10–20 Hz identified as the optimal interval for efficient mapping. These findings highlight the critical role of sampling frequency in spatial analysis and provide practical guidelines for precision agriculture.
Measuring strategies subjected to three-point hitch (TPH) forces and moments of supervised and robotized tractor–implement (manned and unmanned systems) are systematically reviewed in this paper. The review paper not only covers historical perspective research, but also sheds light into prospective research directions. In case of the perspective researches, 644 documents officially published over the past seventy years (1955–2025) have been completely analyzed. Obtained results demonstrate that four types of the TPH dynamometer (mountable electrical, portable electrical, mountable load cell, and portable load cell dynamometer) have been frequently developed by researchers. The dynamometers comprised different transducers (load cell and strain gauge) and fixtures (three-point linkage, supporting frame, extended octagonal ring, and connecting element). Meanwhile, it has been inferred that the portable load cell dynamometer was highly developed and employed by the researchers than other types of the dynamometer. In case of the prospective researches, the development of novel smart dynamometers must be considered. Microwave interferometric radar, three-dimensional laser scanning, radiated acoustic, self-inductance eddy current, vision system, fiber Bragg grating, and vibration method should be prescribed for the development of contact or noncontact transducers of the smart dynamometers. In this regard, the employment of Internet of Things technology through fifth or sixth generation of cellular communication is an acceptable candidate for the enhancement of data transferring between the transducer and data acquisition unit of the dynamometer. Overall, the measuring strategy of the dynamometer should be adapted to IEEE 1451 and ISO 11783 standards.
This study investigates how blade type, forward speed, and tillage depth jointly influence power requirements and soil tillage quality during rotary tillage. While previous studies have mostly assessed these factors independently, limited research has explored their combined effects under field conditions. To address this gap, the performance of four commonly used rotavator blade types is evaluated to identify energy-efficient operating settings that achieve desirable soil tilth quality. This study was conducted on a 1.76 m wide down-cut rotavator with four commercially available blade types (L, C, LJF, and J type) under varying forward speeds and tillage depths in sandy loam soil. Measurements focused on key parameters such as PTO torque, power consumption, soil mean weight diameter (SMWD), and reduction in bulk density. PTO torque and equivalent PTO power increased with higher forward speeds and tillage depths, while soil pulverization (measured by SMWD) and bulk density reduction declined. LJF-type blades showed the highest energy demand (178.02 N·m, 11.65 kW), while C-type blades required the least (143.52 N·m, 9.92 kW). L-type blades offered the best balance between tillage quality and energy efficiency. At a tillage depth of 100 mm, optimal forward speeds were 2.05, 2.28, 1.85, and 1.96 km/h for L, C, LJF, and J-type blades, respectively. L-type blades also recorded the lowest power consumption (9.01 kW), followed by C-type (9.32 kW), indicating better energy efficiency. The study provides a unique multi-factor field-based comparison of commercially available blade types and identifies blade-specific optimized operating conditions for sandy loam soils. The findings support practical decision-making toward improving energy efficiency and soil tilth quality in rotary tillage operations.
The non-destructive quality assessment of fruit and vegetables has become increasingly crucial due to consumer demand for high-quality produce, thus the need to ensure their quality and marketability. Spectral analysis offers a promising approach for assessing internal and external quality attributes. Recent advancements in deep learning (DL) combined with spectral analysis have shown significant potential in providing accurate and non-invasive quality evaluation. This review aims to provide a comprehensive overview of the integration of DL with spectral analysis for the non-destructive quality assessment of fruit and vegetables. It focuses on various DL spectral analysis architectures and models, such as one-dimensional convolutional neural networks (1D-CNN), autoencoders (AE), and long short-term memory (LSTM) networks, used in modeling qualitative and quantitative attributes. The review also explores recent advancements such as model generalization, interpretability, spectral data augmentation, and automated hyperparameter optimization. Furthermore, the applications, future trends, and challenges in DL spectral analysis for evaluating the quality of fruit and vegetables are discussed. The integration of DL with spectral analysis has significantly advanced the non-destructive quality assessment of fruit and vegetables. Key quality attributes such as soluble solids content (SSC), dry matter content (DMC), and moisture content (MC) can be predicted with high accuracy, enhancing the overall quality control processes. DL spectral analysis models have demonstrated better performance in terms of robustness and accuracy compared to conventional methods. Overall, DL spectral analysis has the potential to transform the non-destructive assessment of fruit and vegetable quality, offering substantial benefits for consumers, researchers, and industry practitioners.
Water scarcity and environmental pollution have highlighted the need for efficient water use and drainage in greenhouse hydroponic systems. Evapotranspiration, an indicator of crop water demand, can be measured using the water balance method, based on changes in substrate weight and drainage volume. However, irrigation based on evapotranspiration measurements has not yet been widely adopted in greenhouses. Moreover, single-substrate measurements cannot capture the spatial variations in plant growth and microclimatic conditions across hanging-gutter systems. This study aimed to (1) develop an hourly gutter scale evapotranspiration–based irrigation system for greenhouse tomatoes cultivated in South Korea and (2) evaluate its feasibility and resource-saving effect compared with cumulative solar radiation–based irrigation. An evapotranspiration measurement system comprising three load cells and an ultrasonic level sensor was constructed to measure hourly evapotranspiration using the water balance method. After testing substrate weight and drainage measurements through a preliminary experiment, a cultivation experiment was conducted across two cropping cycles. In the first cycle, the evapotranspiration measurement system was validated under varying environmental and crop growth conditions. In the second cycle, irrigation was performed using 1 an automated system relying on the evapotranspiration measurements from the first cycle. During the first cultivation cycle, the measured daily evapotranspiration at the same growth stage varied from 13.13 to 26.27 kg/day, depending on environmental conditions. Additionally, evapotranspiration increased as crop growth progressed. During the second cycle, the evapotranspiration rate–based irrigation method reduced nutrient solution use by 23.9
