Back to search

University of Thessaly (UTH)

Investigation and development of machine learning algorithms for analysis of large volumes of data

Abstract

dc:description

The current era is defined by the abundancy of data. Globally, the accumulation of data is exponentially increasing each year, leading to huge amounts of data, containing useful information. Machine learning, a set of data-driven approaches for developing mathematical models, based on self-learning algorithms, has found fertile ground to grow. The adaptation of machine learning has been also vast not only due to the recent data availability, but also because of the easy access to high performance hardware. In most domains, the implementation of machine learning approaches brought previously unattainable solutions to complex problems and overall high performances to their operations, leading to a significant paradigm shift. However, a gap has been identified in the domain of Agriculture, the only domain that is steadily behind in technological advancements, even though it is highly significant and belongs in the primary production. The need for fast technological adaptation to agriculture is grave, because of the rapid increase in global population, leading to overpopulation and the adverse effect of climate change.The problem at hand is the development of a methodology for proper identification of diseases on trees, that are located within orchards of high valued crops. Agricultural operational environments are highly complex, and the disease detection is problem based on vision, therefore large amounts of imaging data and sophisticated machine learning algorithms are employed to tackle all specific issues. Specifically, a significant number of data has been collected from various walnut orchards and has been manually labelled by expert agronomists. Additionally, multiple convolutional neural network architectures are utilized, since they are highly capable of extracting features and useful information from complex images. The proposed methodology comprises of three consecutive tasks, aiming to offer a holistic solution towards the identification of infected trees, in order to mitigate the disease’s spread and preserve the crop production and yield. The first task tackles the issue of tree localization within orchards with the use of semantic segmentation, from images taken by unmanned aerial vehicles, under a large variety of conditions and throughout all seasons. Once the proper localization is complete, unmanned ground vehicles receive the information to autonomously navigate through the orchard and inspect each tree individually, leading to the next phase. The second step concerns the detection of disease infected leaves, specifically anthracnose, in images of walnut trees, a high value crop. Final step is to properly classify the images of disease-infected leaves and approximate the level of infection in each tree. Outcome of this methodology is a variability map of the orchard and the extent of the disease spread within it, which can then be used by automatic precision spraying systems or manual operations to treat the trees with the minimum possible resources.Each task of the proposed methodology has been thoroughly investigated, tested and evaluated for application in real-life agricultural environments. Aim of the present thesis is to develop a methodology that is not just applicable in real-life operational environments, but also achieves high performance, robustness, and generalization, a demanding task due to the complex and variable nature of the agricultural environments. All targets have been met for each one of the tasks, leading to the publication of three scientific papers in esteemed peer-reviewed journals, proving the scientific soundness of all approaches. Future goals of the proposed methodology include, but are not limited to, further development based on the potential of the methodology, and direct implementation in real-life farming systems.

Degree

thesis:*
Grantor dc:publisher
University of Thessaly (UTH)
Year dc:date
2022

Author and committee

dc:creator, dc:contributor.*
Authors dc:creator
  • Anagnostis, Athanasios
  • Αναγνώστης, Αθανάσιος

Subjects

dc:subject × 13

Rights

Language dc:language
eng

Identifiers

dc:identifier.*
Identifier
10.12681/eadd/52685

Chain of custody

source
Harvested from
Greek National Archive of PhD Theses
Base URL
phdtheses.ekt.gr/eadd_oai/request
Last updated
2026-11-18
Source record
OAI-PMH GetRecord
citation

Anagnostis, Athanasios; Αναγνώστης, Αθανάσιος. Investigation and development of machine learning algorithms for analysis of large volumes of data. University of Thessaly (UTH), 2022. http://hdl.handle.net/10442/hedi/52685