Aristotle University Of Thessaloniki (AUTH)
Κατανεμημένα συστήματα και αλγόριθμοι συλλογής μετρήσεων, λήψης αποφάσεων και απεικόνισης γεωαναφερόμενης πληροφορίας με εφαρμογές στην αμπελοκαλλιέργεια
Abstract
dc:descriptionThis doctoral thesis focuses on the development of a distributed system, called thingsAI, that can collect sensor measurements at the vine-branch level. Additionally, within the framework of this thesis, services and algorithms were developed to support precision cultivation proposals for viticulture. Within the framework of this thesis and to apply precision agriculture in the viticulture sector, particularly for monitoring biotic and abiotic stresses caused by climate change and diseases, a distributed precision system was designed, tested, and piloted. Additionally, new distributed networks, such as middleware data-collection protocols, are using low-energy wireless technologies, including Wi-Fi, BLE, and LoRaWAN. The new intelligent algorithms were based on the classification of problem occurrences. Through the proposed thingsAI real-time surveillance system at the vine level, it can accurately determine the locations of downy mildew incidents to target interventions that aim to narrow down downy mildew expansion. In this way, the resilience of the viticultural variety is ensured while protecting the environment from unnecessary, excessive universal interventions. The system's interaction with other periodic surveillance systems, such as cameras and drones, was also examined. Corresponding algorithms for spatial and temporal localization of downy mildew were studied within the framework of this thesis, with the ultimate goal of early classification of the phenomenon and its application to other diseases such as botrytis and vine water stress. In the context of adding intelligence to the thingsAI system, three different deep learning algorithms were implemented: 1) Stranded-NN, 2) Fuzzy stranded-NN, and 3) Stranded-LSTM. These artificial intelligence (AI) algorithms are deep learning algorithms. They can be used for geodetically choreographed classification via GPS and high-frequency sensor measurements, while stranded-LSTM can also be used as a prediction algorithm. For the detection of downy mildew disease from camera and drone visual material, relevant location recording and real-time detection services were designed and implemented using existing convolutional deep learning methods, such as ResNet models, and fast detection methods supported by YOLOv5 and YOLOv8. Furthermore, a model for detecting water stress from images was designed and implemented using specialized concentrators and autonomous vineyard towers. These towers record water stress levels using soil electrical conductivity sensors from beacon devices that transmit the measurements via BLE (Bluetooth Low Energy). Subsequently, these sensors were modified to operate either over Wi-Fi or LoRaWAN. The dataset was recorded in the wine-growing area of Zitsa, Epirus, Ioannina, during a downy mildew pandemic to train these models. On it, pixel-level identification of downy mildew vine disease occurrence at the image level (annotation) was performed. This dataset was made available as open data within the framework of this thesis and is available to the community for observation and the development of additional downy mildew detection methods. Additionally, within the framework of this thesis, the methodology for recording and locating the position of optical material from drone cameras, storing them in the cloud via a data streaming service, indexing the positions using GPS information from the drone images, and locating the disease using deep learning models was implemented.
Degree
thesis:*- Grantor dc:publisher
- Aristotle University Of Thessaloniki (AUTH)
- Year dc:date
- 2026
Author and committee
dc:creator, dc:contributor.*- Authors dc:creator
-
- Kontogiannis, Sotirios
- Κοντογιάννης, Σωτήριος
Subjects
dc:subject × 19- Κατανεμημένα συστήματα
- Κατανεμημένα Δίκτυα Ευφυών Πρακτόρων
- IoT
- Αλγόριθμοι βαθείας μάθησης
- Γεωαναφορά
- Κατανεμημένοι αλγόριθμοι
- Distributed systems
- Distributed networks
- Deep learning algorithms
- Georeferencing
- Distributed algorithms
- Επιστήμη Ηλεκτρονικών Υπολογιστών και Πληροφορική
- Φυσικές Επιστήμες
- Πληροφοριακά συστήματα
- Τεχνητή νοημοσύνη
- Computer and Information Sciences
- Natural Sciences
- Information Systems
- Artificial Intelligence
Rights
- Language dc:language
- gre
Identifiers
dc:identifier.*- Identifier
- 10.12681/eadd/61192