Gustavo Adolfo Mesias

Mr. Gustavo Adolfo Mesias

orcid

Research Focus & Metrics

Academic Career Length

5Years (2022-Present)

(Calculated once Scilit publication data is synced: current year − year of first publication + 1)

Representative Publications

View All Publications
18 August 2024 in Remote Sensing
Verified
Journal Article
Open access
PDF file
GMGustavo Adolfo Mesias
Gustavo Adolfo Mesias
José M. PeñaJosé M. Peña
José M. Peña
Ana I. De CastroAna I. De Castro
Ana I. De Castro
IBIrene Borra-Serrano
Irene Borra-Serrano
JDJosé Dorado
José Dorado
Early detection of weeds is crucial to manage weeds effectively, support decision-making and prevent potential crop losses. This research presents an innovative approach to develop a specialized cognitive system for classifying and detecting early-stage weeds at the species level. The primary objective was to create an automated multiclass discrimination system using cognitive computing, regardless of the weed growth stage. Initially, the model was trained and tested on a dataset of 31,002 UAV images, including ten weed species manually identified by experts at the early phenological stages of maize (BBCH14) and tomato (BBCH501). The images were captured at 11 m above ground level. This resulted in a classification accuracy exceeding 99.1% using the vision transformer Swin-T model. Subsequently, generative modeling was employed for data augmentation, resulting in new classification models based on the Swin-T architecture. These models were evaluated on an unbalanced dataset of 36,556 UAV images captured at later phenological stages (maize BBCH17 and tomato BBCH509), achieving a weighted average F1-score ranging from 94.8% to 95.3%. This performance highlights the system’s adaptability to morphological variations and its robustness in diverse crop scenarios, suggesting that the system can be effectively implemented in real agricultural scenarios, significantly reducing the time and resources required for weed identification. The proposed data augmentation technique also proved to be effective in implementing the detection transformer architecture, significantly improving the generalization capability and enabling accurate detection of weeds at different growth stages. The research represents a significant advancement in weed monitoring across phenological stages, with potential applications in precision agriculture and sustainable crop management. Furthermore, the methodology showcases the versatility of the latest generation models for application in other knowledge domains, facilitating time-efficient model development. Future research could investigate the applicability of the model in different geographical regions and with different types of crops, as well as real-time implementation for continuous field monitoring.

6 December 2024 in Data in Brief
Verified
Journal Article
Open access
GMGustavo Adolfo Mesias
Gustavo Adolfo Mesias
José M. PeñaJosé M. Peña
José M. Peña
Ana I. De CastroAna I. De Castro
Ana I. De Castro
JDJosé Dorado
José Dorado
Identifying weed species at early-growth stages is critical for precision agriculture. Accurate classification at the species-level enables targeted control measures, significantly reducing pesticide use. This paper presents a dataset of RGB images captured with a Sony ILCE-6300L camera mounted on an unmanned aerial vehicle (UAV) flying at an altitude of 11 m above ground level. The dataset covers various agricultural fields in Spain, focusing on two summer crops: maize and tomato. It is designed to enhance early-season weed classification accuracy by including images from two phenological stages. Specifically, the dataset contains 31,002 labeled images from the early-growth stage—maize with four unfolded leaves (BBCH14) and tomato with the first flower bud visible (BBCH501)—as well as 36,556 images from a more advanced-growth stage—maize with seven unfolded leaves (BBCH17) and tomato with the ninth flower bud visible (BBCH509). In maize, the weed species include Atriplex patula, Chenopodium album, Convolvulus arvensis, Datura ferox, Lolium rigidum, Salsola kali and Sorghum halepense. In tomato, the weed species include Cyperus rotundus, Portulaca oleracea and Solanum nigrum. The images, stored in JPG format, were labeled in orthomosaic partitions, with each image corresponding to a specific plant species. This dataset is ideally suited for developing advanced deep learning models, such as CNNs and ViTs, for early classification of weed species in maize and tomato crops using UAV imagery. By providing this dataset, we aim to advance UAV-based weed detection and mapping technologies, contributing to precision agriculture with more efficient, accurate tools that promote sustainable and profitable farming practices.


18 March 2026 in Congresos UEX, actas de congresos
Verified
Journal Article
Open access
JMJuan Diego Mena
Juan Diego Mena
GMGustavo Adolfo Mesias
Gustavo Adolfo Mesias
Jose G. GuerraJose G. Guerra
Jose G. Guerra
Darwin Castillo MallaDarwin Castillo Malla
Darwin Castillo Malla
José M. PeñaJosé M. Peña
José M. Peña
IBIrene Borra-Serrano
Irene Borra-Serrano
La pérdida de biodiversidad en los agrosistemas asociada a la intensificación agrícola, al uso de agroquímicos y a la fragmentación del hábitat, compromete los servicios ecosistémicos. En este contexto, la gestión y monitorización de la flora arvense en márgenes florales surge como una estrategia eficaz para evaluar la diversidad y riqueza de especies atractivas para diferentes grupos de polinizadores y fauna auxiliar. En la finca experimental “La Poveda” (ICA‑CSIC, Madrid) se desarrolló un experimento multipropósito durante los años 2023 y 2024 para monitorizar márgenes florales a 15 m de altura, empleando vehículos aéreos no tripulados (UAV). Se realizaron seis vuelos por campaña entre abril y julio, coincidiendo con el pico de floración de 11 especies objetivo pertenecientes a las familias Brassicaceae, Asteraceae y Fabaceae. Las particiones de las ortofotos (teselas de 640×640 px) se etiquetaron siguiendo dos esquemas: (a) unidad floral y (b) conjunto floral. El conjunto de datos resultantes se utilizó para entrenar al modelo Ultralytics YOLOv26 bajo una configuración experimental homogénea y técnicas de aumento de datos. Los mejores resultados se obtuvieron con la variante del modelo y aumento de datos bajo el esquema (b), alcanzando métricas de validación robustas en la precisión promedio media de objetos ( = 0.534), en precisión (P= 0.528) y en exhaustividad ( = 0.558). Estos resultados destacan la relevancia de definir adecuadamente la unidad de etiquetado en conjuntos de datos florísticos de muy alta resolución y ofrece directrices metodológicas para la monitorización de la biodiversidad en agricultura.

27 July 2023
Verified
Book Chapter
Open access
VBVicente Burchard-levine
Vicente Burchard-levine
HNHéctor Nieto
Héctor Nieto
GMGustavo Adolfo Mesias
Gustavo Adolfo Mesias
JDJosé Dorado
José Dorado
Ana I. De CastroAna I. De Castro
Ana I. De Castro
José M. PeñaJosé M. Peña
José M. Peña
Precision agriculture is a reality in agriculture and is playing a key role as the industry comes to terms with the environment, market forces, quality requirements, traceability, vehicle guidance ...

12 May 2025
Verified
Thesis
Open access
GMGustavo Adolfo Mesias
Gustavo Adolfo Mesias
Crop protection is essential to ensure agricultural production against threats such as weeds, diseases and pests. The evolution towards Agriculture 5.0 has incorporated advanced technologies such as artificial intelligence (AI), robotics and deep learning (DL), which enable smart and sustainable crop management. One of the most critical challenges in crop protection is the early and accurate identification of weed species, which allows for effective control during the most sensitive stage of these weeds. In addition, site-specific weed management in precision agriculture requires detection of these species for targeted mapping. However, this process faces difficulties due to the morphological similarities between crops and weeds in early growth stages. Technologies such as unmanned aerial vehicles (UAVs), advanced sensors and DL are essential in precision agriculture as they improve both, the efficiency and sustainability of farming practices. The main objective of this research was to develop an automatic system for early identification of weed species using UAV-captured imagery and cutting-edge AI algorithms. To achieve this, advances in weed identification using AI were first reviewed, creating a taxonomy that includes both traditional and advanced neural network-based approaches. In addition, representative datasets of high quality UAV imagery were developed, in which weed species were labeled in their early growth stages and under various conditions. Using these databases, the study focused on the multiclass classification of weeds through convolutional neural networks (CNNs). To this end, frameworks such as TensorFlow and PyTorch were used, optimizing the handling of large volumes of data and improving computational efficiency. Convolutional neural networks, architectures such as VGG16, ResNet152, Inception-ResNet-v2, EfficientNet-B0 and YOLOv8, as well as vision transformers (ViT) such as ViT-Base and Swin-T were analyzed in the classification task. These models demonstrated their effectiveness in identifying weeds under various environmental conditions, achieving accuracies above 90%. However, the variability in performance with unbalanced datasets complicated the identification of less represented species. Comparison between CNN-based models and ViT revealed that architectures such as YOLOv8 and Swin-T offer balanced performance in the accurate classification of weeds at the species level in early stages. Additionally, the integration of DL models for object detection, such as Faster R-CNN, YOLOv8m, DETA and DETR with geographic information systems enabled the automated generation of geo-referenced weed distribution maps. The use of techniques such as generative adversarial networks (GANs) to generate synthetic images has shown improvements in model generalization across different phenological states of weeds, significantly increasing accuracy while also raising computational costs. The integration of UAV imagery and cutting-edge DL algorithms enables early and accurate identification of weed species, as well as the generation of georeferenced maps essential for informed decision-making. These maps facilitate the visualization of weed distribution, contributing to sustainable management strategies and reducing herbicide use. This research, pioneering the detection, classification and mapping of weeds using low-altitude UAV imagery, highlights the need to develop efficient classifiers and detectors for targeted control using advanced technologies. RESUMEN La protección de cultivos es esencial para garantizar la producción agrícola frente a amenazas como malas hierbas, enfermedades y plagas. La evolución hacia la Agricultura 5.0 ha incorporado tecnologías avanzadas como la inteligencia artificial (IA), la robótica y aprendizaje profundo (DL), permitiendo una gestión inteligente y sostenible de los cultivos. Uno de los desafíos más críticos en la protección de cultivos es la identificación temprana y precisa de especies de malas hierbas, lo cual permite realizar un control efectivo en la etapa más sensible de las especies arvenses. Además, el manejo localizado de malas hierbas en el marco de la agricultura de precisión requiere la detección de estas especies para la elaboración de mapas específicos. Sin embargo, este proceso enfrenta dificultades debido a las similitudes morfológicas entre cultivos y malas hierbas en etapas tempranas de crecimiento. Tecnologías como los vehículos aéreos no tripulados (UAV), sensores avanzados y DL son fundamentales en la agricultura de precisión, ya que mejoran tanto la eficiencia como la sostenibilidad de las prácticas agrícolas. El objetivo principal de esta investigación fue desarrollar un sistema automático para la identificación temprana de especies de malas hierbas utilizando imágenes capturadas por UAV y algoritmos de IA de última generación. Para ello, en primer lugar, se revisaron los avances en la identificación de malas hierbas mediante IA, creando una taxonomía que abarca tanto métodos tradicionales como avanzados basados en redes neuronales. Además, se desarrollaron bases de datos representativas con imágenes de alta calidad obtenidas por UAV, en las cuales se etiquetaron especies de malas hierbas en las primeras etapas de crecimiento y en diversas condiciones. Utilizando estas bases de datos, el estudio se centró en la clasificación multiclase de malas hierbas mediante redes neuronales convolucionales (CNN). Para ello, se utilizaron frameworks como TensorFlow y PyTorch, lo que permitió optimizar el manejo de grandes volúmenes de datos y mejorar la eficiencia computacional. Se evaluaron modelos de CNN como VGG16, ResNet152, Inception-ResNet-v2, EfficientNet-B0 y YOLOv8, así como transformadores de visión (ViT) como ViT-Base...