Conference Agenda

Modern agriculture emphasizes the development of efficient methodologies to maximize crop yields while minimizing resource input. To ensure optimal productivity, sensor-based approaches—such as real-time growth monitoring and automated pest and diseas detection—are increasingly being implemented. Among these, the morphological analysis of crops serves as a critical indicator for assessing overall health and physiological status.

This study proposes a methodology for estimating the morphological traits of butterhead lettuce—specifically height, width, volume, and fresh weight—using 3D point clouds acquired from LiDAR sensors. The primary objective is to compare these morphological characteristics between a control group and a group grown under drought-stress conditions to evaluate the impact of environmental stressors on plant development.

Significant differences observed between the two groups suggest that drought stress deeply affects the structural development of butterhead lettuce. Furthermore, this study demonstrates that LiDAR-based laser scanning offers substantial potential for automated crop monitoring. By enabling precise and non-contact measurements, this approach minimizes the risk of cross-contamination and physical damage associated with manual work, providing a robust solution for efficient crop management in limited growing environments.

LiDAR-based methods are widely used to estimate crop variables such as canopy height, leaf area index, and aboveground biomass. These methods require predefined parameter values, such as voxel size, extinction coefficient, minimum point density, and the number of vertical strata. In most studies, these parameter values are calibrated using linear regressions against ground-truth measurements. While effective, this strategy relies on measured data, potentially limiting the broader applicability of LiDAR metrics. To avoid this calibration step, LiDAR metrics with and without field calibration were tested in a wheat field experiment. This study quantifies (i) the sensitivity of LiDAR-derived metrics to variations in parameter values, (ii) the change in prediction performance associated with parameter variations, and (iii) the ability of more complex models to integrate multiple parameterized versions of the same metric while maintaining predictive performance under parameter uncertainty. Preliminary analyses showed that Partial Least Squares Regression models using the full set of LiDAR-derived metrics outperform models based on individually evaluated metrics. This approach may avoid the need to preselect individual metrics based on their correlation with field measurements.

Reliable phenotypic trait estimation from 3D point clouds is fundamental to digital phenotyping. However, incomplete or sparse point clouds often compromise measurement accuracy, particularly for crops with complex morphological structures. To address this challenge, a point cloud completion model for shiitake mushroom fruiting bodies with distinct cap–stem architectures was developed based on the AdaPoinTr framework. The model integrates a data augmentation strategy to simulate realistic missing regions, an Attention-Enhanced EdgeConv module to improve local geometric feature learning, and a Multi-Scale Fusion Geometry-Aware Attention module to capture structural features at different scales. These designs improve the reconstruction of incomplete point clouds while preserving fine local geometry. Experimental results showed superior reconstruction performance over baseline models, including a 7% improvement in F-Score@1%. Phenotypic trait analysis based on the reconstructed point clouds also achieved high accuracy, with pileus diameter prediction reaching an R² of 0.96 and an RMSE of 3.22 mm, while the other key traits produced R² values above 0.70.

Monitoring apple tree phenology is essential for understanding plant development and optimizing crop management. Although remote sensing offers an alternative to time-consuming field observations, its application to apple orchards remains limited. The objective of this study was to assess the potential of remote sensing data in estimating phenological phases of apple orchards in southern Brazil. The study was conducted in the municipality of Vacaria, Rio Grande do Sul, which is part of the Science Center for Development of Digital Agriculture (CCD-AD/Semear Digital). Field phenological observations were collected during the 2024/2025 growing season. Time series of the Normalized Difference Vegetation Index (NDVI) derived from Sentinel-2 images were obtained and phenological metrics were extracted using the TIMESAT software. Overall, NDVI captured the development of apple trees well, with a slightly earlier harvest date than observed in the field. The metrics estimated by TIMESAT were associated with key phenological phases, with RMSE of 10, 16, and 11 days for bloom, harvesting, and leaf fall, respectively. Despite the differences, the estimated length of the season showed good agreement with field observations (RMSE = 10 days). These findings highlight the potential of remote sensing to reliably characterize apple phenology and support crop monitoring and decision-making.

Unmanned Aerial Vehicle (UAV) multispectral imaging is widely used in precision agriculture, where accurate retrieval of crop reflectance is critical for quantitative field monitoring. However, multispectral cameras typically operate in automatic exposure (AE) mode, in which exposure parameters automatically vary with scene brightness during flight. This leads to inconsistent exposure settings among images and reduces radiometric calibration stability. Additionally, AE offers limited controllability and cannot be adjusted to target brightness. This study examines the limitations of AE and demonstrates the need for fixed exposure (FE) in agricultural remote sensing. A band-wise exposure optimization strategy based on target digital number (DN) values was developed to determine optimal FE settings for each band during flight. This approach ensures that crop canopies and calibration panels remain within valid dynamic ranges, minimizing saturation and radiometric noise. Radiometric calibration accuracy was evaluated by comparing the FE-based object-based empirical line method (O-ELM) with the conventional AE-based calibration method. Results indicate that FE-based O-ELM reduces radiometric error by approximately 50% compared with the conventional AE-based method, significantly improving radiometric precision and inter-band consistency. The proposed strategy provides practical guidance for UAV multispectral data acquisition and supports applications such as crop phenotyping, stress detection, and precision field management.

The regulatory requirements with slurry spreading in Spain make it necessary to know its nutrient content to record the quantities applied. The Spanish market for slurry tanks is led by national companies. These domestically manufactured tanks use electrical conductivity (EC) as an affordable way to measure the nutrient content of slurry. Near infrared spectroscopy (NIRS) technology has also demonstrated the ability to deliver accurate results to measure nutrient ingredients in slurry for application equipment. This work compares EC and NIRS sensors mounted in the slurry tanker for the determination of nutrient content during application under real farm conditions. An experimental slurry tanker equipped with the two sensors was developed and tested in different dairy farms in Galicia (northwest Spain). Chemical analyses of cattle slurry samples for dry matter (DM), total nitrogen (TN), ammonium nitrogen (NH₄-N), total phosphorus (TP) and total potassium (TK), were performed at Mabegondo Agricultural Research Centre (CIAM). The results of laboratory analysis were compared with the measurements taken by the EC and NIRS sensors during the slurry application. The tests performed showed that calibrations need to be improved so that the sensors can deliver accurate results for the cattle slurry types used in the study.