https://hdl.handle.net/20.500.12412/2545 2026-04-30T11:08:56Z https://hdl.handle.net/20.500.12412/7150 Self Calibration of Soil Moisture Sensors for Automatic Irrigation Systems Aranda Brítez, Diego Soil moisture monitoring is essential for efficient irrigation management in precision agriculture, yet sensor calibration remains a major bottleneck, often requiring labor-intensive manual adjustments. Inaccurate soil moisture readings due to sensor drift or manufacturing variability compromise irrigation efficiency and crop health. To address this, this thesis explores automated calibration strategies that combine machine learning and data assimilation techniques to enhance the reliability of soil moisture measurements. A first contribution is the development of a deep learning-based self-calibration algorithm that predicts the soil moisture sensor reading at field capacity using short sequences of uncalibrated sensor outputs. This model, built upon a hybrid CNN-DNN architecture, learns from synthetic data generated through HYDRUS-1D simulations across diverse soil textures and realistic hydrological conditions. By embedding soil-specific thresholds such as saturation and field capacity into the learning process, the approach eliminates the need for manual tuning and mitigates sensor drift, offering scalable calibration in real-world scenarios. In parallel, a data assimilation-based framework using the Iterative Ensemble Smoother and Particle Filter has been proposed. This method dynamically updates sensor parameters by integrating real-time field data with physically-based simulations. It enforces physical constraints on key variables, ensuring that updated parameters remain within soil-specific bounds. The approach demonstrates high adaptability to different environmental conditions and sensor error patterns, significantly reducing bias and improving irrigation performance. Both methods were experimentally validated under controlled and field conditions using an IoT-enabled infrastructure and high-precision reference instruments. The results confirmed the practical applicability, accuracy, and complementary strengths of the proposed approaches, establishing a solid foundation for scalable, autonomous soil moisture sensor calibration in smart irrigation systems. 2026-03-01T00:00:00Z https://hdl.handle.net/20.500.12412/7148 Computer vision and neural networks applied to the growth of crops Martínez, Fátima Belén This thesis presents an intelligent crop monitoring alternative aimed at automating visual analysis. It comprises a low-cost, low-power hardware system that operates algorithms capable of determining specific crop parameters, such as crop height, wheat spiking and stubble detection, as well as identifying the contours of leafy vegetables, in crops such as lettuce and chard. The hardware integrates a Raspberry Pi for algorithmic processing and a LoPy for power management and data transmission via LoRaWAN. The device schedules image capture based on seasonal light variations to compensate for its lack of automatic light adaptation. For height estimation, a kernel-based algorithm was developed, combining stereoscopic vision with reference-based techniques. The Hough Transform enables the identification of crop boundaries in dense plantations, followed by 3D reconstruction using stereoscopic vision to measure height. Alternatively, a reference-based approach circumvents the calibration challenges associated with stereoscopic vision in field conditions. For the detection of spiked and stubble areas, methods from the YOLOv8 and YOLOv10 family were employed, trained using proprietary images captured in low resolution with the hardware proposed in this study and previously annotated using Roboflow. For leafy vegetable contours for lettuce, a custom metric was developed to balance computational cost and accuracy, and was used to evaluate the performance of different detection methods. The YKMS method (YOLOv10 + K-means + Superpixel) enhances precision by utilising bounding box labels, while YOLOv8 and Detectron2 were also evaluated. Additionally, a combined approach named YKMSY8 (YOLOv10 + K-means + Superpixel + YOLOv8) was applied to lettuce. In the case of chard, the YKMS method was used; however, YOLOv10 was replaced by YOLOv11, demonstrating its adaptability to irregular crop shapes. The results confirm the feasibility of these techniques, which provide a cost-effective and real-time crop monitoring system to optimise decision-making and sustainability. The following outcomes were obtained: • The device was designed with a low-cost approach, prioritising minimal energy consumption and ensuring continuous operation in the field for three months with two captures per day. • Regarding height estimation, using the kernel-based algorithm combined with stereo vision, tests conducted on banded planting crops yielded a 3% error in height calculation. Similarly, an alternative method that combines the kernel-based algorithm with a reference-based approach, applied to wheat crops, also demonstrated a 3% height estimation error. • For the detection of the spiked and stubble area, YOLOv8 and YOLOv10 were employed. YOLOv8 achieved a recall of 71.1% and a precision of 79.5%, while YOLOv10 attained a recall of 70% and a precision of 77%. • To determine the contour of lettuce, neural networks such as Detectron2, YOLOv8, and a proprietary method based on YOLOv10, termed YKMS, were applied. Additionally, a hybrid approach combining YKMS with YOLOv8, referred to as YKMSY8, was introduced. A performance metric was presented to evaluate the efficiency of each neural network, incorporating both contour identification accuracy and computational cost. The following results were obtained: an accuracy of 81.9% was achieved by YOLOv8, 84% was achieved by Detectron2, 70.26% was achieved by YKMS, and 87.3% was achieved by YKMSY8. Mean error rates of 3.3% were recorded for YOLOv8, 3.9% for Detectron2, and 5.2% for both YKMS and YKMSY8. Mean inference times were recorded as 1.01s for YOLOv8, 4.12s for Detectron2, 4.07s for YKMS, and 0.46s for YKMSY8. • Preliminary results on chard were obtained with the YKMS methodology, modified to enable the detection of multiple objects within the same image, with YOLOv11 employed instead of YOLOv10. 2026-03-01T00:00:00Z https://hdl.handle.net/20.500.12412/7062 Analysis and implementation of advanced control and estimation algorithms applied to a fleet of autonomous marine surface vehicles Morel Otazu, Thalia Alicia This doctoral thesis presents a comprehensive analysis, development, and experimental validation of advanced control and state estimation algorithms specifically tailored for coordinated operation of autonomous marine surface vehicles (ASVs). Motivated by the practical constraints of implementing robust navigation and control strategies on low-cost, underactuated catamaran-type vessels, this research addresses critical challenges associated with limited sensor instrumentation, environmental disturbances, and complex marine vehicle dynamics. Initially, a robust system identification methodology was developed and experimentally validated to accurately characterise the dynamic response of ASVs, capturing essential vessel dynamics from basic position and orientation sensors without requiring direct velocity and acceleration measurements. Both static and dynamic propulsion models were identified, yielding highly reliable and precise representations of the vessel’s motion across a broad operational range, thereby providing a solid foundation for subsequent estimation and control frameworks. Subsequently, the thesis explored advanced state estimation techniques, specifically extended state observers (ESOs), capable of accurately reconstructing unmeasured states and disturbances from inherently noisy measurements. Three distinct ESOs were rigorously evaluated experimentally, including a novel zonotopic observer integrated with agrey-boxinputgainidentificationmethod, demonstratingsuperior robustness and practical effectiveness in dealing with system uncertainties and varying operational conditions. Building upon these estimation methodologies, an enhanced zonotopic observer framework was further proposed, featuring real-time online input gain identification and direct integration of yaw rate sensor measurements. This approach delivered improved resilience against parameter variations and actuator degradation, markedly enhancing predictive performance and reliability under dynamic marine environments. The culmination of this research is the development of a hierarchical control architecture that integrates high-level flocking algorithms, mid-level path-following guidance, and low-level dynamic control into a unified operational framework. This innovative approach effectively manages fleet coordination, collision avoidance, and environmental disturbances, while explicitly handling actuator limitations. Extensive simulations and quasi-experimental validations using the CyberShip II and Yellowfish ASV platforms within ROS-based realistic marine environments have demonstrated cohesive fleet behaviour, significant reductions in navigation errors, and reliable performance under diverse conditions. Through the integration of robust modelling, adaptive estimation, and hierarchical control strategies, this thesis validates the central hypothesis that advanced f leet-level algorithms can be successfully implemented in affordable ASVs equipped with minimal sensor configurations. The methodologies and insights provided herein contribute substantially to the field, bridging critical gaps between theoretical developments and practical applications in autonomous maritime operations, thus laying the groundwork for future research and broader real-world deployment of coordinated autonomous marine vehicle fleets. 2026-01-01T00:00:00Z https://hdl.handle.net/20.500.12412/7059 Estudio e intensificación de la hidrólisis enzimática de celulosa derivada de residuos agroalimentarios bajo consideraciones de transporte Álvarez González, Celia El desarrollo de biorrefinerías de segunda generación ha recibido especial atención en los últimos años debido a su potencial para reducir la dependencia de recursos fósiles y disminuir el impacto ambiental asociado a las materias primas no renovables, evitando el uso de alimentos o sus ingredientes. En el marco de estas biorrefinerías, la biomasa lignocelulósica constituye una de las materias primas más abundantes para la obtención de azúcares mediante la hidrólisis enzimática de la celulosa. Los azúcares obtenidos actúan como plataformas intermedias para la producción de gran variedad de productos de interés industrial, tales como biocombustibles, biopolímeros y otros compuestos de alto valor añadido mediante procesos fermentativos, enzimáticos, catalíticos o térmicos. Entre los residuos lignocelulósicos existentes, destaca la paja de arroz por su abundancia, su alto contenido de celulosa y su potencial de valorización. La hidrólisis enzimática de la celulosa se lleva a cabo mediante celulasas, en las que las endoglucanasas y exoglucanasas actúan liberando celobiosa, mientras que la β-glucosidasa convierte la celobiosa en glucosa. Sin embargo, la eficiencia de la hidrólisis enzimática se ve limitada debido a la estructura compleja de la paja de arroz, que confiere recalcitrancia y dificulta el acceso a las enzimas. Además, para que estos procesos sean viables a escala industrial, es imprescindible trabajar con altas cargas de sólido durante la hidrólisis enzimática, de manera que se incremente la concentración final y la productividad de glucosa y se reduzcan los volúmenes de procesamiento y de vertido. Este enfoque genera importantes desafíos relacionados con la mezcla, difusión y eficiencia catalítica. Para superar las limitaciones asociadas a la recalcitrancia de la biomasa, es necesario aplicar un pretratamiento eficaz que modifique su estructura y aumente la accesibilidad de las enzimas durante la hidrólisis enzimática. De igual manera, es necesario contar con información precisa sobre la actividad catalítica de los cócteles enzimáticos, así como conocimiento sobre la cinética del proceso. Finalmente, la elección del reactor resulta crucial para superar las limitaciones de mezcla sólido-líquido que se generan al emplear altas cargas de sólidos. La presente tesis, titulada “Estudio e intensificación de la hidrólisis enzimática de celulosa derivada de residuos agroalimentarios bajo consideraciones de transporte”, tiene como objetivo principal estudiar los procesos que afectan al proceso de hidrólisis enzimática de la paja de arroz. Para ello, se emplea un enfoque integrado que combina el pretratamiento alcalino de la paja de arroz, la evaluación de la actividad catalítica de cócteles enzimáticos comerciales, el estudio de la cinética del proceso de sacarificación y el estudio del efecto de la carga de sólido utilizando diferentes sistemas de reacción. La finalidad de este enfoque es optimizar la conversión de celulosa a glucosa empleando bajas concentraciones de enzima, bajo condiciones viables a escala industrial. Primero se llevó a cabo la caracterización de los cócteles enzimáticos comerciales Biogazyme 2X (B2X) y β-glucosidase-1000 (BG) de ASA Spezialenyzme GmbH. Se observó que el cóctel B2X presentaba una elevada actividad endoglucanasa y exoglucanasa, mientras que el cóctel BG presentaba una alta actividad β-glucosidasa. Posteriormente, se llevó a cabo la caracterización de la paja de arroz, y se determinó que su contenido de celulosa era del 28,6 %. Este dato sirvió como base de cálculo para evaluar el pretratamiento alcalino y la posterior hidrólisis enzimática. Una vez caracterizados tanto los cócteles enzimáticos como la paja de arroz, se llevó a cabo el estudio del pretratamiento de la paja de arroz, evaluando diferentes variables de proceso, considerando tanto la influencia de las condiciones de pretratamiento como las concentraciones de enzima en los tiempos de hidrólisis enzimática. El objetivo del estudio consistió en identificar una ventana de operación que permitiera aplicar un pretratamiento suave, utilizar bajas concentraciones de enzima y alcanzar rendimientos de glucosa cercanos al 90 % en 8 horas para cargas bajas de sólido en la sacarificación. Para complementar este estudio, se llevaron a cabo análisis cinéticos, composicionales y estructurales, incluyendo XRD, SEM, microscopía confocal de fluorescencia y ensayos de adsorción enzimática. Estos análisis se hicieron con el fin de relacionar las modificaciones de la estructura del sustrato con la accesibilidad de las enzimas y la eficiencia del proceso de hidrólisis enzimática. Asimismo, se evaluó la posibilidad de reutilizar el licor negro generado durante el pretratamiento y de llevar a cabo el escalado del proceso, con el fin de explorar la viabilidad y eficiencia del pretratamiento a mayor escala. Con unas condiciones de pretratamiento de 1 % de NaOH (p/v), 73 °C y 4,5 h, se obtuvo un sólido (S8) con aproximadamente un 60 % de celulosa y una estructura completamente desfibrilada. Respecto a la hidrólisis enzimática, se alcanzó una conversión del 93 % en 8 horas, empleando una dosis de enzima de 6,5 mgproteína/gSS y una carga de sólido del 2,4 %. Estos resultados demostraron que existían unas condiciones mínimas necesarias para que el sólido comenzara a ser reactivo (0,5 % NaOH, 50 °C y 1 h), y que el tiempo de pretratamiento fue un parámetro clave, ya que influyó de manera significativa tanto en la velocidad inicial como en la conversión final del proceso.; En abierto se puede consultar la parte no embargada de la Tesis Doctoral 2026-01-01T00:00:00Z