Conference Agenda

Precision livestock farming has increasingly incorporated three-dimensional sensing technologies to obtain non-invasive morphometric measurements in cattle. Among these technologies, Light Detection and Ranging (LiDAR) sensors and RGB-D cameras can generate three-dimensional information about animal body shape. This study aimed to evaluate the potential of different data acquisition methods—manual field measurements, a LiDAR sensor, and an Intel RealSense D435i depth camera—for predicting dairy cow body mass using dorsal morphometric variables. The experiment was conducted in Ijaci, Minas Gerais, Brazil, with 15 cows evaluated after the first milking of the day (05:00 a.m.). Four morphometric variables were considered: thoracic width (TW), abdominal width (AW), rump width (RW), and dorsal length (DL). The highest correlation with body mass was observed for RW_field (r = 0.80). Multiple regression models were fitted using the two most correlated variables for each method: RW_field + AW_field (R² = 0.64; RMSE = 40.6 kg), TW_LiDAR + RW_LiDAR (R² = 0.51; RMSE = 47.3 kg), DI_D435i_2D + AW_D435i_2D (R² = 0.59; RMSE = 43.3 kg), and RW_D435i_3D + DI_D435i_3D (R² = 0.50; RMSE = 47.8 kg). The results indicate that different sensing technologies capture distinct morphometric aspects related to body mass, highlighting their potential for automated monitoring systems in digital and precision livestock farming.

With the expansion of large-scale broiler production, indoor microclimate control has become essential for maintaining productivity, especially in regions with strong seasonal temperature variation such as South Korea. Inadequate ventilation can cause heat stress, gas accumulation, and production losses. However, most commercial broiler houses still rely on operator-dependent or simplified rule-based control strategies, and only a few scientifically validated algorithms are available. This study developed an intelligent ventilation control algorithm based on heat–energy balance analysis to estimate ventilation demand more accurately. One year of operational data from an experimental farm showed that the conventional strategy maintained optimal temperature conditions for only 74% of the production period, whereas the proposed algorithm increased this to 92%. The algorithm was further tested in two broiler houses during bird ages of 20-34 days. The proposed control reduced daytime indoor temperature by 1.5-2.0 ºC, alleviating heat stress. Cooling pad operation increased eightfold, while tunnel fan operation decreased by 52%, resulting in net energy savings. Mortality was reduced by 16.5%, indicating improved production performance. These results demonstrate that the proposed algorithm can enhance thermal conditions and energy efficiency in broiler houses.

Precision-agriculture monitoring using UAV imagery has advanced rapidly, yet many studies still deliver single-field or single-task outputs, limiting operational decision making across adjacent fields managed as one unit. This study proposes an auto-mated, GIS-ready monitoring pipeline for a multi-field block that integrates (i) field-boundary extraction at the parcel level, (ii) within-parcel detection and geolocation of crops and weeds, and (iii) multi-layer agronomic information delivery at the region-of-interest (ROI) level. UAV data are processed into georeferenced orthomosaics, from which individual field parcels are delineated to generate parcel-wise management zones. Within each parcel, crop instances of Chinese cabbage and white radish, together with major weed regions, are detected and mapped to spatial coordinates. For each detected ROI, the framework is designed to attach expandable attribute layers—such as growth stage, pest/disease status, vegetation indices (e.g., NDVI-based health indicators), and phenotypic/biophysical traits (area, height, and fresh weights)—enabling users to query multiple information layers by selecting ROIs on an interactive map interface. Overall, this work presents a standardized, modular UAV–GIS automation architecture that links parcel segmentation, crop/weed localization, and ROI-level layered information delivery, enabling scalable site-specific decision support.

High-speed maize sowing often reduces the filling stability of pneumatic seed-metering devices, thereby limiting seeding precision. To address this problem, this study systematically investigates the influence of seed posture on filling performance and proposes an optimized stepped seed-hole structure. The mechanical behavior of maize seeds during the filling process was analyzed, and a flow-confinement strategy was introduced to improve the attachment stability of postures prone to detachment. A computational fluid dynamics (CFD) model of a positive-pressure seed-metering device was developed to examine the coupled effects of seed-hole geometry, seed posture, air pressure, and forward speed. High-speed imaging experiments were conducted to quantify seed posture distributions under different structural configurations, and optimal operating parameters were identified. The results show that optimizing the stepped seed-hole parameters significantly increases the aerodynamic drag acting on unstable seed postures, thereby reducing both missed and multiple seeding. After optimization, the qualified index exceeded 99.3% at 8–12 km/h and remained above 97% at 12–16 km/h. This study provides a new structural optimization approach for improving the stability of pneumatic seed-metering devices under high-speed conditions.

The domestic laying hen industry in Korea is undergoing structural changes and faces increasing vulnerability to heat stress due to climate change, highlighting the need for data-driven environmental management. This study evaluated the rearing environment in laying hen houses by measuring temperature and humidity in aisles and cage interiors and analyzing their spatial distributions. Field measurements showed higher humidity on the second floor than on the first floor, indicating moisture accumulation in the upper layer. On the first floor, average temperatures were 27.8 °C in central rows and 28.3 °C in sidewall rows, while on the second floor they were 27.6 °C and 26.7 °C, respectively. The highest relative humidity occurred in second-floor sidewall rows (76.5%), with outdoor humidity exceeding 83% in the same area, revealing ventilation imbalance and localized stagnation zones. A three-dimensional CFD model using half of the domain was developed and validated against measured temperature and humidity inside and outside cages. An air-deflector-assisted ventilation strategy utilizing the existing manure duct was proposed to promote air diffusion, induce upward airflow, and improve the temperature and humidity distribution inside cages. CFD results indicate that the proposed system can effectively mitigate stagnant zones and reduce microclimatic heterogeneity.

The aim of this study was to evaluate the effect of greenhouse-integrated semi-transparent photovoltaic (STPV) modules on physiological parameters of cherry tomato, namely plant height, number of flowers, number of fruits, and total yield. A comparative experiment was carried out in a greenhouse divided into two cultivation zones, where 21 cherry tomato plants were cultivated in each treatment (42 plants in total): (i) an STPV-covered area and (ii) a conventional transparent-cover control area. The same cherry tomato cultivar, planting density, irrigation/fertigation schedule, and crop management practices were applied in both zones. Plant height was measured at regular intervals, while flowering and fruit set were recorded by counting flowers and fruits per plant during the reproductive period. Total yield was determined by harvesting and weighing marketable fruits across the production cycle.

The results of this study showed that plants grown under STPV exhibited differences in vegetative development, as reflected in plant height, and a measurable change in reproductive performance, with variation in flower production and fruit number compared with the control. Consequently, total yield also differed between treatments, indicating that shading intensity and light quality associated with STPV integration can affect yield-forming components in greenhouse tomato cultivation.

As space exploration continues to expand, space agriculture has emerged as a key technology for next-generation missions, and plant growth chambers serve as essential platforms. Previous studies have cultivated crops under various spaceflight conditions. However, in microgravity environments such as those inside spacecraft, buoyancy-driven natural convection does not occur as it does on Earth. As a result, the leaf boundary layer thickens, inhibiting gas exchange and potentially causing physiological disorders. Therefore, forced ventilation is crucial to mitigate the leaf boundary layer in a microgravity environment. In addition, maintaining temperature and air velocity within adequate and uniform ranges is necessary for stable plant growth. In this study, an HVAC system for a plant growth chamber in microgravity was designed to provide a uniform and suitable environment around lettuce while alleviating the leaf boundary layer. To simulate the internal environment, a 630 × 630 × 630 mm³ chamber, a cultivation station, and a simplified lettuce model were constructed. Computational Fluid Dynamics (CFD) simulations were conducted to evaluate environmental uniformity under various ventilation rates and inlet–outlet configurations in a microgravity environment, leading to the identification of an optimal design. These findings contribute to establishing fundamental technologies for stable food production in space.

Drones or unmanned aerial vehicles (UAVs) have become an effective tool for applying phytosanitary treatments, especially in precision agriculture and crops that are difficult to access. UAVs allow for localized and selective applications, adapting the dose and spray pattern to the actual conditions of the crop and the environment. Their effectiveness depends on several factors, among which flight height is particularly important. This study analyzes how spray quality varies as a function of three flight heights (2, 2.5, and 3 meters). To this end, trials were conducted in rice fields, with three replicates, using a DJI Agras T30 drone and water-sensitive paper. The water-sensitive paper was analyzed using ImageJ 1.54g software, which provided data on the coverage, density, and size of the droplets in each treatment. The results were organized in Microsoft Excel, and statistical analysis was performed in RStudio v.4.4.0. The results showed that flying at 2 m offered the best combination of coverage uniformity and treatment efficacy, while at higher altitudes, drift increased and effectiveness decreased. These conclusions highlight the importance of adjusting operating parameters in drone applications, not only to improve the efficiency of pesticide use, but also to move towards a more sustainable model of agriculture.

In silvopastoral agroforestry systems (AFS), trees and forage plants are integrated on the same land area. In temperate climates, trees of the genera Populus, Juglans, or Salix are often cultivated. The presence of trees in a treeless agriculture creates heterogeneity and alters the microclimate, which leads to changes in the productivity of accompanying crops, especially in the vicinity of trees. International studies have determined forage yields in AFS with results ranging from 21% lower to 44% higher values compared to areas without tree influence. In this study, both the microclimatic effects on forage yield and annual growth of trees depending on the design of the overall system have been investigated at practice scale in Germany. Using this data, the land equivalent ratio was determined as a measure of the productivity of the AFS. Trees in AFS are typically used as woodchips for bioenergy, while timber production for material application, in which trees reach a height of 10-25 m, are neglected. However, due to the excellent supply of light, water and nutrients for trees in AFS, these systems are also particularly suitable for agricultural timber production, as approx. 30% higher yields compared to plantation timber production can be achieved.

Early and reliable detection of Mal Secco disease, caused by Plenodomus tracheiphilus, remains a major challenge for citrus production across the Mediterranean basin. Conventional diagnostics based on culturing or molecular assays are accurate but time-consuming, destructive, and poorly suited for large-scale or real-time monitoring. Volatile organic compound (VOC) profiling offers a promising non-invasive alternative for early detection and precision agriculture. This study analyzed VOC emissions from symptomatic and asymptomatic citrus twig fragments previously validated by culturomics and molecular methods. Principal Component Analysis (PCA) revealed clear separation between healthy and infected samples, highlighting the diagnostic potential of VOC signatures. Supervised modeling with Partial Least Squares Discriminant Analysis (PLS-DA) achieved 91% correct classification on external test samples (RMSEC 0.38; RMSECV 0.51; mean sensitivity and specificity 0.8). A Random Forest model yielded a mean AUC of 0.985 in training (sensitivity 0.962; specificity 0.968) and 0.781 in testing. These results demonstrate that VOC profiling combined with machine learning provides a robust framework for rapid, non-destructive Mal Secco detection. Ongoing in vivo analyses aim to validate and optimize the system under field conditions and support future integration into a real-time detection platform.

This study presents a UAV-RGB framework for individual-plant-level (IPL) growth mapping of Chinese cabbage and white radish in a commercial field to support crop monitoring and precision-agriculture decision-making. A field experiment was conducted in Muan, Jeollanam-do, during the 2017 growing season. UAV images were acquired (20 m AGL) and processed to generate RGB orthomosaics and DSM/DTM products with GCP/RTK-based calibration and radiometric correction. Fresh weight was estimated at the IPL using a linear regression model with two UAV-derived variables: vegetation fraction (VF; ExG + Otsu) and plant height (PH; DSM–DTM). The IPL models achieved R² = 0.94 for Chinese cabbage and R² = 0.86 for white radish, and robustness for deployment was improved via two-point normalization. To avoid manual labeling, we developed a zero-shot IPL crop identification pipeline combining ExG–Otsu and segment anything integrating segmentation, grid-based overlap-aware splitting, and contrastive language–image pre-training-based removal of non-crop, achieving 98.4% accuracy for cabbage and improving radish identification from 65.1% to 95.6%. Finally, georeferenced IPL estimates were converted into yield/growth maps and interpolated via kriging, enabling integration with terrain attributes and crop layers for management-zone delineation, anomaly detection, and site-specific prescriptions.

On-farm meteorological sensors are essential for capturing site-specific microclimatic conditions that often differ from regional observations because of topography and local crop management. These differences can strongly affect the timing and usefulness of agrometeorological risk models, making locally measured data crucial for operational warning systems. We present a modular framework that converts field measurements into decision-support alerts. The system integrates data acquisition from SOFIS® sensor stations, automated quality control, daily aggregation, and the calculation of derived indicators such as degree-day accumulation and event-based indices based on temperature, relative humidity, rainfall, and leaf-wetness duration. Outputs are delivered through simple dashboards and exportable tables to support routine technical reporting.

The framework is being applied in apple and pear orchards, vineyards, olive orchards, and tomato crops. Warning modules target major pests and diseases, including Plasmopara viticola, Uncinula necator, Cydia pomonella, Empoasca vitis, Jacobiasca lybica, Bactrocera oleae, Alternaria spp., Guignardia bidwellii, and Phytophthora infestans. A key feature is the validation of warnings against local ground-truth data, including trap captures and structured field observations provided by growers. This feedback is used to refine biofix dates, thresholds, accumulation periods, and risk classes each season, improving accuracy, reducing false alarms, and strengthening practical decision support.

Accurate monitoring of dairy cow behavior is essential for improving animal welfare, health management, and productivity in precision livestock farming (PLF). Although recent deep learning approaches have shown strong performance in classifying individual cow behaviors, most methods treat behaviors as isolated events and overlook temporal transitions between them. However, patterns such as frequent standing–lying transitions or prolonged restlessness can indicate physiological conditions including estrus, lameness, or discomfort. This study proposes a deep learning framework to model temporal behavior transitions in dairy cows using video data. First, spatiotemporal features are extracted from video sequences using advanced video recognition models. These features are used to predict behavior sequences for each cow over time. Transition probabilities between consecutive behaviors are then calculated to construct a behavior transition matrix, capturing temporal dependencies and revealing biologically meaningful activity patterns. To further enhance performance, sequence modeling techniques such as Long Short-Term Memory (LSTM) networks and Transformer architectures are explored to improve prediction accuracy. By integrating temporal transition dynamics into vision-based monitoring systems, the proposed approach aims to improve both predictive capability and interpretability. The findings could demonstrate that temporal behavior modeling can support early health assessment and informed decision-making in modern dairy farming systems.

Agricultural research plays a crucial role in addressing the growing challenges faced by farmers, as accurate crop growth prediction is essential for sustainability, stability, and efficient resource management. The complexity of predicting plant growth arises from highly variable environmental and cultivation conditions. This study overcomes these challenges by integrating ensemble learning, deep neural networks, and optimization techniques to create a robust predictive framework capable of handling multiple growth‑influencing parameters. The proposed system is evaluated using standard performance metrics and comparative analyses, offering insights into the strengths and limitations of each modeling approach.

Focusing on chili-cultivation, the research compares hybrid machine learning models for predicting plant growth in both soil and soilless environments. By employing advanced algorithms that capture intricate interactions among environmental, nutrient, and temporal variables, the study analyzes 30 days of experimental data to assess prediction accuracy, computational efficiency, and scalability. The results demonstrate the superior potential of hybrid models in precision agriculture applications. Experimental findings show accuracies of 98.85% and 99.14% for leaf growth, 99.14% and 96.80% for flower growth, and 57.20% and 96.00% for fruit growth in soilless and soil‑based systems, respectively. The study concludes that soilless cultivation supports faster crop yield compared to traditional soil-based methods.

This study integrated OJIP chlorophyll fluorescence analysis with machine learning classification to evaluate drought stress responses and recovery potential in strawberry. The objectives were to compare photosystem II (PSII) functional impairment across different leaf developmental stages and to identify critical thresholds for drought detection and recovery. PSII responses varied significantly between leaf types. Mature leaves exhibited earlier functional impairment under drought stress, while young leaves showed delayed structural damage but greater sensitivity in early electron transport processes. These differences highlight the importance of leaf position when assessing drought progression and recovery potential. Key OJIP parameters, including O–J phase variables (Vj, Vi, and ΔVOJ) and PSII performance indicators (PIabs, TRo/ABS, ETo/TRo, and REo/RC), effectively characterized the physiological stages of drought response. Machine learning classification accuracy improved when these parameters were combined, indicating stable PSII response patterns as drought progressed. Recovery experiments revealed thresholds distinguishing reversible and irreversible stress. In mature leaves, PIabs recovery potential declined after approximately three days of drought stress, whereas in young leaves, TRo/ABS and REo/RC reliably indicated recovery capacity. These findings demonstrate that integrating fluorescence parameters with machine learning enables quantitative drought detection and provides a useful framework for precision irrigation management in smart agriculture systems.

In ruminant-based extensive livestock systems, pastures are one of the main components and the main feeding source given that they are the most economical resource. Accurate information about pasture dry matter (DM, in kg ha^-1) availability is a key parameter in the manager’s decision, particularly when calculating stocking rates and supplementation needs. This study evaluates an electronic sensor (rising plate meter, RPM) to estimate DM in biodiverse dryland pastures. The study was carried out between December 2023 and December 2025 on three pasture fields, two located in Portuguese Alentejo region (Southern Portugal), “Mitra” and “Tapada dos Números”, and one located in Spanish Extremadura region (Cubillos). The experimental work consisted of sensor measurements, followed by the collection of more than 200 pasture samples, distributed between different dates of the pasture vegetative cycles of 2023/2024 and 2024/2025. The best estimation models for DM were obtained based on measurements carried out in Autumn and Winter (R² > 0.70) and decreased significantly when based on measurements carried out in Spring. These results show the potential for the research and development of proximal and remote sensing tools to support pasture monitoring and animal production management.

Short-cycle hydroponic crops offer a unique opportunity for high-frequency plant monitoring, yet practical frameworks for translating image-based phenotyping into operational decision support remain limited. This study presents an integrated pipeline for real-time crop assessment in controlled environments, combining continuous RGB imaging (5-min intervals) with continuous load-cell measurements (60-s intervals).

Hydroponically grown wheat sprouts (7-day growth cycle) were cultivated under three illumination regimes (60, 350, and 500 µmol m⁻² s⁻¹ PPFD). Across experiments, 3,024 synchronized image–mass observations were collected. Image processing included HSV-based white suppression, automated tray localization, SAM-based instance segmentation, and extraction of 17 interpretable canopy features describing coverage, greenness, structural density, texture, and color balance.

Results demonstrate strong associations between canopy greenness, structural density, and fresh biomass, enabling accurate, non-destructive tracking of growth progression. Tree-based ensemble models achieved the highest predictive performance. LightGBM reached RMSE ≈ 0.015 kg and sMAPE ≈ 0.27%, while Random Forest achieved sMAPE ≈ 0.22%, enabling highly accurate non-destructive biomass estimation.

The proposed approach supports closed-loop hydroponic management by supplying real-time information for adaptive control of lighting, irrigation, and nutrient delivery. This highlights the role of image-based phenotyping as a practical component of intelligent production systems, supporting more resource-efficient controlled-environment agriculture.

Fungal diseases of the grapevine cause significant damages to the quality and productivity of crops. Studying the interaction between grapevine pathologies, and environmental and management factors is essential for developing a sustainable and high-quality wine production. In particular, Esca disease, attributed to a complex of fungi from the genus Fusarium, Phaeoacremonium aleophilum and Phaeomoniella chlamydospora, is gaining growing interest at both national and international levels. It has been observed that some specific characteristics of a vineyard (variety, rootstock, training system, soil, age, etc.) can cause the plant to react differently to the disease. The main objectives of this study are: I) to map the disease distribution on a regional scale; II) to study the agronomic and environmental factors that can influence the spread of the pathogen. Infection outbreaks in the Friuli-Venezia Giulia region (north-eastern Italy) were mapped by considering more than 3,500 surveys taken over 19 years of observations. The relationships between the pathogen spread, some specific vineyard characteristics and the surrounding context were analyzed. Results show a capillary distribution of the pathogen in all viticultural areas of the region, with a particular incidence on white-grape varieties, and a strong correlation with grape variety, training system, soil organic carbon, and landscape structure.