Conference Agenda

Maintaining uniform crop growth requires precise monitoring of crop-proximity microclimate, including temperature, relative humidity, and CO₂. However, fixed sensors are limited to installation points, making it difficult to capture vertical canopy distributions and localized variations caused by ventilation, air circulation, and shading. To address this, this study uses a quadruped robot to measure within-canopy microclimate conditions across a greenhouse at high spatial resolution. The data are visualized as height-resolved three-dimensional (3D) maps to identify spatiotemporal patterns in the crop-proximity zone.

The platform includes a vertical lifting mechanism to collect measurements at the crop base, growing point, and upper canopy. As the robot follows a predefined path, it samples both sides of the aisle. Greenhouse geometry is mapped using LiDAR-based SLAM, and robot pose is estimated from joint states and foot contact data. These are synchronized with sensor measurements to generate crop-specific 3D maps. Performance is evaluated against fixed sensors, while 2D heatmaps and height-dependent gradients (ΔT, ΔRH, ΔCO₂) analyze environmental non-uniformity and daily microclimate changes.

Acknowledgement: This work was carried out with the support of the “New Agricultural Climate Change Response System Construction Project (Project No. RS-2025-02283290)” of the Rural Development Administration of the Republic of Korea.

The development of accurate digital twins of fruit trees offers new opportunities for training, decision-making, different labors simulation. This study aims to generate a full three‑dimensional reconstruction of an olive tree using a commercial depth camera as a low‑cost and field‑deployable sensing solution. A total of 24-point cloud captures were acquired around the tree at 15° intervals and a fixed distance of 4 m. The point cloud data obtained in .PLY format from each viewpoint were subjected to a processing workflow that included noise reduction, geometric filtering, and the estimation of local surface properties. A multi‑view alignment procedure was then applied to transform all captures into a common coordinate system and progressively merge them into a unified 3D representation of the tree. The resulting reconstruction produced a coherent 360° model of tree trunk and scaffolds, preserving geometric detail suitable for downstream analysis. The fused point cloud demonstrates sufficient accuracy and completeness to support future applications such as virtual training for pruning operations, simulation of phytosanitary treatments, and the development of interactive digital twins for precision agriculture. These results confirm the feasibility of using portable depth cameras for detailed 3D reconstruction of individual trees in real orchard conditions.

Canopy structure in the fruiting zone plays a key role in grapevine performance, as it affects fruit exposure, bunch microclimate, and grape composition. Its characterization is therefore of major interest for vineyard monitoring and precision viticulture. This study evaluates the use of a field robotic platform equipped with a LiDAR sensor for the non-destructive assessment of vine canopy architecture under field conditions.

The objective was to determine whether variables derived from three-dimensional point clouds acquired by the robot can be related to canopy traits and agronomic variables measured in the field. LiDAR data were collected along vine rows. The resulting point clouds were used to characterize vine architecture. Special attention was given to the relationship between robot-derived measurements and field observations in the fruiting zone.

Preliminary results indicate that the study approach can detect relevant differences in canopy structure and provide variables associated with grape quality-related parameters. In particular, volume of the canopy showed a significant relationship with soluble solids content (°Brix), suggesting that LiDAR-based canopy characterization may be useful for non-destructive grape quality assessment

In conservation and regenerative agriculture, subsoiling is used to relieve soil compaction without inverting soil layers, preserving soil structure and minimizing disturbance. However, subsoiler tips are exposed to severe abrasive wear due to soil–tool interaction, making durability a key factor in operational efficiency.

This study evaluated the effect of different tungsten carbide reinforcement strategies on the wear resistance and structural robustness of subsoiler tips under real field conditions. A comparative field experiment was conducted on several configurations: an unreinforced reference tip and reinforced variants produced using brazed tungsten carbide inserts, conventional hardfacing and laser cladding. Wear progression was monitored through periodic measurements of tip length and thickness, while soil properties and operating parameters were recorded to support data interpretation.

Results showed clear differences among reinforcement methods. Tips with brazed tungsten carbide inserts achieved the lowest wear rates and best preserved their functional geometry over time. Hardfaced tips showed intermediate performance, whereas laser-cladded tips offered no significant improvement over the unreinforced solution.

As expected, wear was more severe in sandy, highly abrasive soils. Overall, the study demonstrates that the tungsten carbide application method is crucial for enhancing subsoiler tip durability, extending service life, reducing maintenance and supporting sustainable subsoiling practices.

The agricultural machinery market offers various fertilizer spreaders, including those designed for organic fertilizers. Many machines operate using pre-prepared application maps based on vegetation indices, enabling variable-rate fertilization according to spatial crop requirements. While synthetic fertilizers have a fixed and known chemical composition, eliminating the need for real-time verification, organic fertilizers—particularly solid ones—exhibit significant variability in nutrient content.

In liquid manure application, systems for real-time composition analysis and dose adjustment are already available. However, precise application of solid organic fertilizers remains challenging due to their heterogeneous structure and variable density. Large void spaces between fertilizer particles hinder accurate measurements, especially when using optical probes that require direct and stable contact with the material.

This research focused on developing a device and method for accurate real-time application of solid organic fertilizers in manure spreaders. Nutrient content is assessed using a contact VIS–NIR spectroscopic probe supported by dedicated calibration models developed under simulated operating conditions. A key innovation is an electro-hydraulically controlled compaction mechanism that forms and stabilizes a several-centimeter-thick fertilizer layer pressed against the probe window during measurement. Prototype testing under near-operational conditions demonstrated significant reductions in prediction errors for NPK estimation.

Recent advances in optical sensing technologies are fostering the development of innovative, affordable tools for precision agriculture, enhancing the monitoring of crop physiological status across spatial and temporal scales. Within the framework of the PRIN2022 O4A project (Optics for Agrifood), an ongoing research initiative, a compact and cost-effective hyperspectral imaging system is being developed for in-field assessment of grapevine water status and ripening dynamics.

O4A aims to design and validate optical sensing solutions tailored to agrifood applications, with particular attention to robustness, portability, and economic sustainability. The proposed hyperspectral system operates in the visible–near infrared range and enables the extraction of spectral indices correlated with key indicators of water stress and berry composition. Preliminary tests have shown promising potential in discriminating different levels of vine water status and grape maturity under controlled conditions.

The project is currently in progress, and its main experimental campaign is scheduled for summer 2026, when field trials will be conducted in vineyards to consolidate calibration models and validate system performance across different environmental conditions. By integrating optical design, sensor optimization, and advanced data analysis, O4A seeks to bridge the gap between laboratory-grade spectroscopy and practical agricultural applications, supporting more informed and sustainable decision-making in viticulture.

In mechanized harvesting, grain losses limit operational efficiency, being influenced by regulation failures and crop variability. Technologies embedded in modern harvesters automatically adjust operational parameters in real time to minimize losses. The objective was to evaluate the effects of an automatic harvester regulation system during soybean harvesting, comparing conditions with the system activated and deactivated. Losses were quantified using coated rings positioned on the ground after the harvester passed, totaling 72 samples. The mass of each sample was subsequently weighed and determined. Data were submitted to analysis of variance using the F test. Average total machine losses were 1.10±0.11% with the system activated and 1.15±0.14% when the system was deactivated, representing a relative reduction of 5.20%. Losses in the internal mechanisms averaged 0.26% and 0.32% for the activated and deactivated systems, respectively, with a relative reduction of 19%. Although no statistically significant differences were detected between treatments, operation with the system activated resulted in lower coefficients of variation (10.40%) compared to the deactivated (12.10%). Under the conditions evaluated, the automatic harvester regulation system tended to mitigate internal losses and reduce operational variability, suggesting greater adaptation of the harvester to crop conditions.

Operational adjustment of furrow opener mechanisms in direct seeding is a critical factor. The study quantified the draft efficiency of double disc and shank furrow openers as affected by the distance to the cutting disc and forward speed. A 2×3×4 factorial design was adopted, considering furrow opener type, distance from this to the cutting disc and speed. Draft force was measured using a load cell and soil disturbance was characterized with a microprofilometer, enabling determination of specific operational resistance (kN m⁻²) and specific draft force (kN m⁻¹). The shank required 40% greater draft per unit of soil disturbance than the disc, with averages of 231,07 kN m⁻² and 166,92 kN m⁻², respectively. Per unit of working depth, the disc required 15% more, averaging 26,06 kN m⁻¹ compared to 22,69 kN m⁻¹. The combination of maximum speed and maximum distance increased specific resistance to 310,54 kN m⁻², approximately 60% above other conditions. The double disc performed better per disturbed area, while the shank was more efficient per unit of working depth. The interaction between distance and speed exerted stronger influence on draft demand than opener type, underscoring operational adjustment as a primary determinant of energetic performance.

We analyze 65 years of certified combine harvester homologation data in Italy, drawing on a registry of 1,913 models from 19 manufacturers between 1960 and 2025. The dataset, maintained by Federunacoma, was reconstructed from family-specific code patterns covering eight distinct homologation series. After removing seven anomalous power values and harmonizing 158 BRAND–MODEL pairs with conflicting specifications via median imputation, the cleaned sample spans a continuous power range from 16 to 515 kW. Ten of the eleven brands with sufficient longitudinal coverage show positive power trends. Massey Ferguson recorded the steepest increase (+2.40 kW per year, p = 0.001), followed by Case-IH and Claas (both exceeding +2.1 kW/yr, p < 0.001). New Holland stands apart as the only brand with a statistically significant negative coefficient (−0.46 kW/yr, p = 0.036), a finding that runs counter to the sector-wide trend and that we discuss in terms of portfolio strategy. A two-way ANOVA on brand and historical period confirms a highly significant interaction effect (F = 6.94, p < 0.001), indicating that the trajectory of power growth has not been uniform across manufacturers. A linear mixed model places the intra-class correlation at 0.35, meaning roughly one third of total power variance is attributable to brand identity rather than to year of homologation.

The study examines 45,233 tractor homologation records registered in Italy between 1960 and 2025, drawing on the Federunacoma dataset to trace how the national market has shifted across six decades. After a standardized data-cleaning procedure, models were divided into five power classes and analyzed through OLS regression, ANOVA, k-means clustering, and a generalized additive model to capture non-linear trends over time. Three findings stand out. The market has polarized: median registered power rose from roughly 40 kW in 1960 to 80 kW in 2025, but the distribution widened rather than simply shifted upward, with the Compact and High-Power segments gaining ground at the expense of intermediate classes. This trajectory varies considerably across brands — OLS slopes range from +0.92 kW per year for John Deere to −0.13 kW per year for McCormick, while CASE-IH, Deutz-Fahr and Fiat show statistically anomalous patterns explained by portfolio composition rather than genuine technical stagnation. Finally, cluster analysis identifies four stable market archetypes — Small-Specialist, Mid-Range, Premium-General, and High-Power Innovator — whose membership has remained largely unchanged across successive emission-standard cycles, suggesting that competitive positioning in this industry owes more to brand heritage than to regulatory pressure.

Soil loosening requires high energy demand, making technologies that increase operational efficiency essential. A recent development is the automatic shank spacing adjustment system for chisel plow, enabling variable rate operation. However, validating such equipment is necessary given the complexity of variable rate operations. This study investigated the technical feasibility of variable rate chisel plow operation using field tests and fuel consumption simulations. The experiment was conducted in a commercial area using an 11-shank chisel plow at 5 km h⁻¹. Shank response time was determined by the distance between hydraulic actuator engagement at shank stabilization at 40 cm. Fuel consumption was estimated by mathematical modeling based on ASABE standards, considering total operational resistance and required drawbar power. The total transition zone totaled 9.3 m until steady-state operation. Operational simulation in a 142 ha area demonstrated that three variable rate management zones, defined by target depths of 30, 40, and 50 cm, resulted in an estimated savings of 920L of diesel, representing 15% reduction compared to fixed rate operation. Excessive management zones compromised operational efficiency due to accumulated areas. The implement demonstrated technical feasibility for variable rate application, providing fuel consumption reduction and greater operational efficiency when associated with adequate logistical planning.

Chisel pow with automatic shank spacing adjustment systems represent a technological advancement for mechanized soil loosening, enabling precise real-time operational adjustments. However, the influence of the shank spacing-depth ratio on the soil disturbance profile is not yet fully understood. The objective of the study was to evaluate the soil disturbance profile of a chisel plow with an automatic adjustment system as a function of spacing-depth ratio and tractor forward speed. The experiment was conducted in a commercial area using a 3x3 factorial design, with three spacing-depth ratios (1.00×, 1.25× e 1.50×) and three forward speeds (1.11, 1.38 e 1.66 m s-1), evaluating the soil disturbance and elevated areas, soil swelling, roughness modification, and residue flow. The 1.25× ratio at 1.66 m s-1 provided the largest disturbance area (843,50 cm²) with the lowest soil swelling (6,34 cm²), representing the most technically efficient combination. The 1.0× ratio resulted in the greatest roughness modification (344% at 1.11 m s-1) and critically compromised residue flow, with frequent blockages. The automatic shank spacing adjustment capability proved decisive for the simultaneous optimization of the soil disturbance variables and residue flow.

Lodging is a major challenge in cereal crop production, reducing yield, grain quality, and harvest efficiency. Methods for detecting and quantifying lodging using LiDAR sensors require validation data based on known canopy geometry and lodging conditions. This study developed a controllable artificial wheat canopy designed to support the development and validation of LiDAR-based lodging analysis methods. The artificial canopy was constructed from paper, tubing, and wire mounted to wooden boards. The modular design allowed systematic manipulation of canopy orientation and lodging angles under controlled conditions. LiDAR data were collected using a Livox Mid-70 sensor positioned approximately 2.5 m above ground in a near-nadir configuration. Multiple lodging scenarios were created by adjusting the canopy rows to predefined angles. Preprocessing included noise filtering based on Livox tags, ground normalization, and plot-level cropping. Canopy vertical profiles were derived from height distributions to analyze vertical canopy structure. Surface-based lodging characterization was explored using gridfit interpolation and alpha-shape surface reconstruction. The approach successfully distinguished between different lodging configurations and enabled estimation of lodging angle distributions from reconstructed canopy surfaces. The controlled experimental dataset and analysis framework provide a foundation for improving LiDAR-based lodging detection methods in crop phenotyping and agricultural research.

Chisel plows are essential implements for soil tillage and crop establishment. Each field condition requires specific shank adjustment for adequate operation. However, manual adjustments are time-consuming and physically demanding. New systems automate this task through electro-hydraulic actuators with reported efficiency gains. This study quantified the operational gain of an automatic adjustment system compared to traditional manual adjustment in chisel plows. An automatic and a manually adjusted chisel plow, both with 11 shanks, were evaluated across spacing intervals of 300-600 mm. Manual adjustment was performed by two operators using appropriate tools. Time was recorded using a stopwatch in both cases. Mean adjustments time for the adjustment system was 12 s per 100 mm, with time not proportional to spacing due to the hydraulic actuator engagement delay. A 400-600 mm adjustment required 16 s. Manual adjustment for 11 shanks required approximately 173 min by two operators, totaling 346 man-minutes compared to 0.39 man-minutes for the automatic system. Non-anotomical chassis positioning and bolt-tightening imposed misaligned body movements, compromising operator health and safety. The automatic system was 887 times faster than manual regulation, eliminating auxiliary labor and ergonomic risks, demonstrating the transformative potential of automatic electro-hydraulic systems for agricultural mechanization.

As the importance of simulation in engineering and science continues to grow, its application has expanded across a wide range of fields. In particular, increasing attention has been paid to the accurate modeling of granular materials such as soil, which exhibit complex mechanical behavior. Because the performance of agricultural machinery is directly affected by soil–machine interactions during operations such as tillage and seeding, it is essential to understand their mechanical relationship. The Discrete Element Method (DEM) is commonly used to simulate soil behavior.

In this study, direct shear tests and angle of repose tests were performed on dry soil to obtain calibration targets for DEM. Influential DEM parameters were first screened using a Plackett–Burman design, and then optimized using a Box–Behnken design within the Response Surface Methodology (RSM) framework. The optimized parameters reproduced the experimental results within acceptable error bounds, supporting the validity of the calibrated DEM model. In future work, the derived parameters will be applied to simulate interactions between agricultural implements and soil, and the model accuracy will be assessed through comparison with experimental data.

Accurate identification and counting of macaúba palms (Acrocomia aculeata) are important for agricultural planning and biomass production systems. This study evaluates the influence of different optimization algorithms on the training performance of a deep learning model for automated detection of macaúba palms using imagery acquired by a Remotely Piloted Aircraft (RPA). High-resolution aerial images were processed to generate an orthomosaic, which was subdivided into 3,074 tiles of 640 × 640 pixels to create the training dataset. The YOLOv10n object detection architecture was trained for 100 epochs using six optimization algorithms: SGD, Adam, AdamW, NAdam, RAdam, and RMSProp. Model performance was evaluated using mean Average Precision at 50% IoU (mAP50), mean Average Precision across IoU thresholds (mAP50–95), precision, recall, and training time. The best predictive performance was obtained with SGD, reaching mAP50 of 0.767 and mAP50–95 of 0.267, while RAdam showed the best overall efficiency considering predictive metrics and computational cost. The selected model was applied to an orthomosaic subdivided into 3,465 tiles, detecting 11,342 macaúba individuals compared with 12,827 plants identified through reference counting, resulting in an overall accuracy of 88.42%. These results demonstrate the potential of deep learning and RPA imagery for automated crop monitoring and inventory.

Implements with automatic adjustment of tine spacing and working depth represent a significant technological advance for mechanized soil decompaction, enabling real-time operational adjustments and variable rate application. However, the stability and uniformity of the process under actual commercial operating conditions still need to be investigated. The objective was to evaluate the stability and uniformity of the decompaction process as a function of the spacing between tines and the operating speed, using statistical process control charts as a monitoring tool. The experiment was conducted with an 11-tine chisel plow with an automatic adjustment system, in a factorial scheme with three speeds and three spacing ratios between rods and a target depth of 40 cm. The depth was measured at 15 equidistant points for central and lateral rods. The central rods operated on average 35% above the target depth (54 cm), while the lateral rods recorded an average depth 20% below the target (32 cm). The control charts identified special causes of variation in the central stems, traced to modifications in the implement made without technical recommendation, demonstrating the sensitivity of the method to detect external interference in the process and validate the operational quality of innovative implements under commercial conditions.

Agricultural mechanization in different production systems has become increasingly efficient with the adoption of modern machinery, enhancing field capacity and improving operational quality. This study evaluated energy efficiency and operational parameters to reduce costs and minimize soil impacts. Tests were conducted in two areas with flat and undulating terrain using five Class II tractors (A, B, C, D, and E) from different manufacturers, each with an engine power of 58.88 kW. In each area, performance was assessed at a travel speed of 5 km h⁻¹ under two PTO configurations: standard and economy. A 4,000 L sprayer was coupled to each tractor to represent a typical field operation. The variables analyzed were (i) fuel consumption and (ii) operational performance. Individual control charts were used to quantify process variability and stability over time. The monitoring approach indicated that Tractor A had the lowest engine-speed variability. Tractor D exhibited the most stable engine speed and the lowest fuel-consumption variability, whereas Tractor E showed the highest fuel consumption. Overall, this evaluation procedure enabled the identification and selection of tractors with superior energy and operational performance under the tested conditions.

Moringa oleifera Lam. is an emerging multipurpose crop valued for its nutraceutical seeds, oil production, and adaptability to semi-arid and Mediterranean environments. Accurate estimation of fruit and seed yield is essential for optimizing harvest scheduling and improving crop management within precision agriculture systems. This study evaluates the potential of three-dimensional canopy analysis based on Light Detection and Ranging (LiDAR) technology for estimating fruit and seed yield in Moringa oleifera. UAV-mounted LiDAR surveys were conducted during the fruiting stage to acquire high-density point cloud data describing canopy structure. From these datasets, Digital Surface Models (DSM), Digital Terrain Models (DTM), and Canopy Height Models (CHM) were generated to derive structural parameters including plant height, canopy volume, and canopy density. Field measurements of fruit number, pod weight, and seed yield were simultaneously collected for calibration and validation of LiDAR-derived metrics. Statistical analyses revealed strong correlations between canopy structural parameters and production variables (R² > 0.80), with canopy volume emerging as the most reliable predictor of fruit and seed yield. The results demonstrate that LiDAR-based canopy modeling provides a reliable and non-destructive approach for estimating Moringa oleifera productivity and supporting data-driven crop management in Mediterranean precision agriculture systems.

The accurate estimation of coffee plant height from handheld LiDAR point clouds depends on the method used to classify the ground and generate the Digital Terrain Model (DTM). This study compared two ground filtering algorithms, Cloth Simulation Filter (CSF) and Progressive Morphological Filter (PMF), using DTM interpolation based on a Triangulated Irregular Network (TIN), point cloud normalization, and plant height metrics extracted within rotated rectangular windows defined from field measurements. The evaluated metrics were the maximum height of the normalized point cloud, the 99th height percentile, the mean of the upper 5% of heights, and the maximum pixel value of the Canopy Height Model (CHM), calculated as the difference between the Digital Surface Model (DSM) and the DTM. The best results were obtained with PMF using the 99th percentile (RMSE = 0.101 m; MAE = 0.086 m) and the maximum CHM pixel value (RMSE = 0.117 m; MAE = 0.087 m), whereas the best CSF result was obtained with the maximum height of the normalized point cloud (RMSE = 0.139 m; MAE = 0.108 m). PMF was therefore the most suitable method for ground classification, providing the most accurate coffee plant height estimates.

This study aimed to integrate remote sensing data and machine learning algorithms to develop models for estimating the peanut maturity index (PMI). Peanut pods were collected in a commercial field cultivated with the IAC 503 cultivar to determine PMI. The dataset comprised 200 observations, corresponding to 40 sampling points evaluated at five moments throughout the crop cycle. PlanetScope satellite imagery and field assessments were obtained 30 days before harvest. The input variables included spectral bands and vegetation indices (NDVI, GNDVI, MNLI, and SAVI). For modeling, the algorithms K-Nearest Neighbors (KNN), Support Vector Regression (SVR), Ridge Regression (Ridge), Random Forest (RF), XGBoost (XGB), Gradient Boosting (GB), and Decision Tree (DT) were used. All evaluations were integrated into a single dataset, which was used to train and test each algorithm, with 80% of the data allocated for training and 20% for testing. The KNN, SVR, and RF algorithms showed similar performance, with KNN standing out by achieving an R² of 0.89 and an MAE of 0.07. The results demonstrate the potential of integrating remote sensing and machine learning for peanut maturity estimation, supporting decision-making regarding the optimal harvest time.

Reliable density measurements in non-helical X-ray computed tomography (CT) require the mitigation of geometric artifacts. This study developed an automated spatial alignment and masking procedure to correlate reconstructed gray values from an axial cone beam CT (CBCT) with gravimetric density across diverse wood species.

A set of 298 wood cubes from 12 different species was analyzed. Samples were oriented within the Field of View (FOV) to minimize reconstruction artifacts. A custom algorithm was developed to automate the extraction of average density values:
1. Alignment: Singular Value Decomposition (SVD) of the voxel cloud determined the rotation matrix for alignment.
2. Masking: Euclidean Distance Transform (EDT) generated centered cube masks to sample interior volumes and avoid edge-artifacts.
3. Correlation: Mean gray values were compared against gravimetric density using linear regression.

The EDT-SVD-based approach enabled consistent data extraction across all 298 samples. Linear regression between the masked gray values and gravimetric density showed a high coefficient of determination (R² = 0.9933) with a relative standard error of 0.5% for the slope.

This study demonstrates that an axial CBCT setup, combined with automated spatial processing, is a viable tool for non-destructive wood density assessment. The SVD-EDT framework is suitable for high-throughput research and industrial quality control.