Abstract
Many strategies have been developed to monitor the volume of volume of Above Ground Biomass (AGB) in forest areas as a fundamental step for managing carbon concentration. This study explores the use of use of Light Detection and Ranging (LiDAR) data obtained through Unmanned Aerial Vehicles (UAVs) to estimate height values in a vegetation colony composed of oaks (Quercus pyrenaica Willd.) in northern Portugal. The extraction of pertinent information from LiDAR data was facilitated by using the LAStools extension within the Quantum Geographic Information System (QGIS) software framework. The generated raster and image information were used to calculate the height values of the vegetation. Following this extraction, the information was meticulously organized into datasets, which were then employed in Deep Learning (DL) algorithms. The VGG16 model was selected as the underlying framework for the present study. Height predictions were made using dimensions of 16 x 16, 32 x 32, and 64 x 64 pixels for the Red, Green and Blue (RGB) images. The data was estimated and compared using both the standard format of the VGG16 model and a superficially adapted version of its convolution layers. The algorithm's efficacy was validated by comparing the forecast results with the data obtained from QGIS, which revealed minimal discrepancies. It was observed that using 64 x 64 pixel scale images yielded enhanced accuracy, resulting in reduced values for the Mean Absolute Error (MAE). The study demonstrates the viability of applying DL techniques to accurately capture information about a forest area using RGB images.

Abstract
Olive cultivation is a pillar of Mediterranean agriculture, deeply rooted in both tradition and economic importance. This paper presents a novel two-phase methodology for the automated preprocessing of olive leaf images to facilitate accurate cultivar classification. Leveraging the state-of-the-art YOLO11 framework, two models (YOLO11n and YOLO11s) were employed for detection and segmentation tasks. A comprehensive dataset, combining in-situ captured images with publicly available data, was meticulously annotated using both manual and semi-automatic processes. The detection model identifies individual olive leaves, while the segmentation model isolates the leaves by replacing the background with a uniform white, thereby simulating laboratory conditions. Experimental results demonstrate that YOLO11n outperforms YOLO11s in terms of mean Average Precision and F1-score, confirming the feasibility of deploying the system on mobile devices for real-time, in-field classification.

Abstract
This study explores the impact of hyper-parameter optimization on the performance of convolutional neural networks (CNNs) for olive cultivar classification using transfer learning. Pre-trained ImageNet models such as VGG16, InceptionV3, and ResNet50 were adapted to a proprietary dataset, with VGG16 selected for detailed evaluation. Key hyper-parameters, including layer count, neurons per layer, dropout rate, learning rate, and batch size, were tuned using random search. The best configuration achieved a validation accuracy of 87.5%, significantly outperforming the control model. Sensitivity analyses with Morris and Sobol methods identified the number of layers as the most influential factor, followed by dropout and learning rates through interaction effects. These findings demonstrate the importance of tailoring CNN architecture and regularization settings to the problem domain. These results underscore the importance of tuning architectural depth and regularization mechanisms for performance optimization. As a practical guideline, models with fewer layers and intermediate dropout levels demonstrated higher robustness and generalization, offering an effective strategy for adapting CNNs to agricultural classification tasks.

Abstract

Abstract
This article presents the development of a tethered fixed-wing tail-sitter VTOL (Vertical Take-Off and Landing) Unmanned Aerial Vehicle system. The design focuses on improving energy efficiency by utilizing the wings to harness wind power, similar to a kite, while maintaining VTOL functionality. A distinguishing feature is the purpose-built autopilot system, with custom hardware and software components specifically engineered for this application. The study presents the system identification process for obtaining five MIMO (Multiple-Input Multiple-Output) transfer functions that characterize the dynamics between roll-yaw commands and responses, including the tether angle feedback. To address the inherent coupling effects and uncertainties in the system, robust mixed sensitivity (H-infinity) MIMO controllers are developed. The controllers were validated through both simulations and experimental flights, demonstrating effective performance in handling cross-coupling effects and maintaining stability under various operating conditions. According to flight test findings, the system can precisely manage the tether angle while adjusting for ground effect disturbances. This allows for accurate tethered navigation, a stable attitude, and the maintenance of an adequate yaw heading.

Abstract
This paper presents a novel Nonlinear Model Predictive Controller (NMPC) architecture for trajectory tracking of omnidirectional robots. The key innovation lies in the method of handling constraints on maximum velocity and acceleration outside of the optimization process, significantly reducing computation time. The controller uses a simplified process model to predict the robot’s state evolution, enabling real-time cost function minimization through gradient descent methods. The cost function penalizes position and orientation errors as well as control effort variation. Experimental results compare the performance of the proposed controller with a generic Proportional-Derivative (PD) controller and a NMPC with integrated optimization constraints. The findings reveal that the proposed controller achieves higher precision than the PD controller and similar precision to the NMPC with integrated constraints, but with substantially lower computational effort. © The Author(s), under exclusive license to Springer Nature Switzerland AG 2025.