Aristotle University Of Thessaloniki (AUTH)
Παρακολούθηση της κατάστασης των καλλιεργειών με τη χρήση Διαδικτύου των Πραγμάτων και μεθόδων Τεχνητής Νοημοσύνης
Abstract
dc:descriptionIn modern agriculture, monitoring crops using conventional methods presents significant challenges, as these methods are often limited by a lack of accuracy and difficulty in collecting and analyzing data in real-time. Dependence on traditional techniques, such as on-site observation and sampling, can lead to delays in decision-making, inadequate resource utilization, and lower crop efficiency. Furthermore, these methods are subjective and rely heavily on the knowledge and experience of the farmworker. This makes timely and objective detection of issues, such as plant diseases, difficult, which can negatively impact productivity and the sustainability of agricultural activities. These challenges can be addressed using modern sensors and precision agriculture systems. Specifically, Internet of Things (IoT) sensors enable online crop monitoring, providing continuous and reliable real-time data streams that can be transmitted instantly through data transmission technologies and protocols. This connectivity between sensors and IoT plat-forms allows farmers to make timely decisions, improving crop performance and reducing costs. Additionally, when these sensors are combined with artificial intelligence (AI) methods, they offer even greater advantages, as algorithms can analyze data, detect patterns, and predict problems before they occur. This combination increases the accuracy and effective-ness of crop monitoring, enabling more sustainable and efficient management of agricultural resources. This dissertation presents five applications where IoT technologies are combined with AI methods to address the challenges of modern agriculture. Each application serves as an ex-ample of how the integration of these technologies can provide innovative solutions for crop monitoring and management. Initially, a thorough literature review is conducted, covering the existing literature and the methods used in this work (Chapters 1 and 2). The first application (Chapter 3) involves the use of a spectrophotometer to detect viral load in young tomato plants, using machine learning methods for early detection of plant infection by the ToCV virus at a presymptomatic stage, considering two scenarios regarding the use of outliers in the training of algorithms. This is followed by three applications for monitoring the health of a vineyard. In the first ap-plication (Chapter 3A), two transfer learning scenarios are created, with different philosophies of training and model validation, using three different deep learning algorithms. The training and validation of the models are performed on images from online databases, and testing is carried out on RGB images from the experimental vineyard. Then (Chapter 3B), the most successful transfer learning technique from Chapter 3A is tested for training lightweight models that can be easily applied to microcontrollers, such as the Raspberry Pi, for real-time disease detection, also supported by a lightweight semantic segmentation algorithm to achieve greater accuracy in disease detection. This microcontroller can function autonomously or as an IoT device when mounted on an unmanned ground vehicle, with the ability to process data through ROS software and send the model's response to a decision support system. Finally (Chapter 3C), data obtained from disease detection algorithms through a multispectral camera mounted on a UAV, ground hyperspectral camera, and images collected in the vineyard are combined into a data fusion algorithm to enhance detection capabilities. The final application focuses on predicting the infection of a winter wheat field by the Fusarium Head Blight fungus (Chapter 5), using data fusion from satellite imaging, IoT sensors measuring temperature and humidity, and soil data. Features from the different data sources are applied as input to machine learning algorithms. From the above, it is evident that in Chapters 3 and 4 (including all subchapters), the problem being addressed is a classification problem (the first with machine learning, the second with deep learning), while in Chapter 5, the problem is quantitative, applying regression methods. In both cases, highly satisfactory results have been achieved, exceeding 90% accuracy or 0.9 in the coefficient of determination, depending on the problem. This research highlighted the potential of IoT and artificial intelligence technologies in monitoring crop conditions, offering new opportunities for precise and timely management. The results demonstrate that these technologies can be effectively integrated into Decision Sup-port Systems, enhancing decision-making and efficiency in agriculture. Furthermore, it is suggested that future research should explore the use of digital representations through digital twins, which could offer even more comprehensive and tailored solutions for managing agricultural systems in the future.
Degree
thesis:*- Grantor dc:publisher
- Aristotle University Of Thessaloniki (AUTH)
- Year dc:date
- 2024
Author and committee
dc:creator, dc:contributor.*- Authors dc:creator
-
- Morellos, Antonios
- Μορέλλος, Αντώνιος
Subjects
dc:subject × 28- Γεωργία ακριβείας (ΓΑ)
- Τηλεπισκόπηση και ΓΣΠ
- Τεχνητή νοημοσύνη (ΤΝ)
- Διαδίκτυο των πραγμάτων
- Ανίχνευση ανωμαλιών
- Ανίχνευση αντικειμένων
- Πρόβλεψη με νευρωνικά δίκτυα
- Ασύρματο Δίκτυο Αισθητηρίων
- Βαθειά μάθηση
- Μηχανική μάθηση
- Ασθένειες φυτών
- Precision agriculture (PA)
- Remote sensing - G.I.S.
- Artificial intelligence
- Internet of things (IoT)
- Anomaly detection
- Object detection
- Prediction with Neural Networks
- Wireless sensor networks
- Deep learning
- Machine learning
- Plant diseases
- Γεωπονία, Δασολογία και Αλιεία
- Γεωπονικές Επιστήμες και Κτηνιατρική
- Γεωπονία, άλλοι τομείς
- Agriculture, Forestry and Fisheries
- Agricultural and Veterinary Sciences
- Agriculture, miscellaneous
Rights
- Language dc:language
- gre
Identifiers
dc:identifier.*- Identifier
- 10.12681/eadd/58748