Aristotle University Of Thessaloniki (AUTH)
Μοντελοποίηση ιδιοτήτων εδαφών από υπερφασματικές υπογραφές με χρήση τεχνικών υπολογιστικής νοημοσύνης
Abstract
dc:descriptionThis doctoral dissertation focuses on the estimation of the physicochemical properties of soil samples through their diffuse reflectance spectrum in the visible--infrared electromagnetic spectrum captured in the laboratory. The overarching objective entails the development of novel methodologies which fall under the general framework of explainable artificial intelligence, in order to involve an interpretability aspect. These interpretation capabilities that the models possess enable researchers and end-users to understand, appropriately trust, and manage them, because it allows the models to explain their inference process, to identify their merits and weaknesses, and to demonstrate how they would respond in the future or in new input data, thus safeguarding their transferability. To this end, the dissertation focuses on the development of methodologies which belong to the computational intelligence domain. More concretely, initially a family of evolutionary fuzzy systems is proposed. One of their unique characteristic is the high interpretability degree, given that they employ simple yet understandable models, which are made up from rules that resemble the human deductive reasoning process. Every member of this family of systems showcases concrete advantages over the other evolutionary fuzzy systems that have been proposed in the literature, and it is capable of tackling problems of large dimensionality in the input space as well achieving the necessary balance between the accuracy of prediction and the structural complexity of the developed rule bases. Following that, several machine learning techniques and a deep learning model are proposed which are designed having as a goal the achievement of enhanced accuracy of prediction, but also at the same time integrating interpretability capacities which assist in providing understandable results and explaining the models' reasoning process. Within the frame of the above, the combined use of different spectral pre-processing techniques is proposed, which aims to make use of the complementary information contained within the different spectral sources that originate from them. The proposed techniques are applied in large soil spectral libraries that have been developed in the past years.
Degree
thesis:*- Grantor dc:publisher
- Aristotle University Of Thessaloniki (AUTH)
- Year dc:date
- 2020
Author and committee
dc:creator, dc:contributor.*- Authors dc:creator
-
- Tsakiridis, Nikolaos
- Τσακιρίδης, Νικόλαος
Subjects
dc:subject × 34- Ερμηνεύσιμη τεχνητή νοημοσύνη
- Ασαφής λογική
- Εξελικτικοί αλγόριθμοι
- Εξελικτικά ασαφή συστήματα βασισμένα σε κανόνες
- Μηχανική μάθηση
- Βαθιά μάθηση
- Φασματοσκοπία σε ορατό και υπέρυθρο
- Υπερφασματικά δεδομένα
- Εδαφική φασματοσκοπία
- Φασματοσκοπικές βιβλιοθήκες εδαφών
- Παρατήρηση γης
- Explainable artificial intelligence
- Fuzzy logic
- Evolutionary algorithms
- Evolutionary fuzzy rule-based systems
- Machine learning
- Deep learning
- Visible and infrared spectroscopy
- Hyperspectral data
- Soil spectroscopy
- Soil spectral libraries
- Earth observation
- Γεωπονία, Δασολογία και Αλιεία
- Γεωπονικές Επιστήμες και Κτηνιατρική
- Εδαφολογία
- Επιστήμη Ηλεκτρολόγου Μηχανικού, Ηλεκτρονικού Μηχανικού, Μηχανικού Η/Υ
- Επιστήμες Μηχανικού και Τεχνολογία
- Αυτοματισμοί και Συστήματα ελέγχου
- Agriculture, Forestry and Fisheries
- Agricultural and Veterinary Sciences
- Soil Science
- Electrical Engineering, Electronic Engineering, Information Engineering
- Engineering and Technology
- Automation and control systems
Rights
- Language dc:language
- gre
Identifiers
dc:identifier.*- Identifier
- 10.12681/eadd/49756