Doctoral Thesis
Machine learning-based frost detection in plants from infrared thermography
Doctor of Philosophy (PhD), Murdoch University
2023
Abstract
Frost damage significantly reduces crop production. Current frost damage assessment practices are heavily dependent on traditional temperature logger data and manual inspection of the crops after a suspected frost event. Such an approach can be labour-intensive and lead to delayed management decisions. Although there are a few studies on image-based frost damage assessment using computational methods, these are limited to post-frost analysis.
On the other hand, freezing in plants is a complex and dynamic process. Environmental variables, such as temperature fluctuations, plant’s physiology, and the presence of ice nucleating agents, are important factors that influence freezing in plants. This thesis has investigated the use of machine learning (ML) models to capture the thermodynamic patterns of freezing based on infrared thermography (IRT) imagery. This thesis has made several major contributions: First, we present a comprehensive literature review of computational learning-based frost detection approaches in plants.
Second, sequential IRT images are used to visualise the latent heat release during the freezing inside the plants. Therefore, we hypothesise that with the sequential IRT images, the thermodynamic patterns of freezing events can be detected from the thermal properties in plants during phase transitions. We formulate the methodology as an adjacent temperature-based image classification problem. To prove the concept, we conducted a pilot study on a dataset of IRT images of ice-nucleating bacterium droplets collected in an ice-bath in a laboratory setting. Inspired by the successful results on detecting ice nucleation and freezing points in this study, we extended this study to detect freezing patterns in crop plants from field-collected data and demonstrated that crystallisation and supercooling events can be detected despite the adversaries in natural conditions.
Third, a vertical temperature gradient develops from soil to canopy due to the heat loss from the soil and canopy boundary during a frost event. We hypothesise that the relationship between the canopy, plant and ground temperatures can be an early indicator of frost. We demonstrate that ML models can learn the distinct patterns on frost and non-frost nights and detect the comprehensive coldness scale based on the temperatures conducive to frost formation of a certain severity degree.
Our study provides important insights into establishing a solid foundation for ML-based frost detection for developing automatic and real time on-field frost monitoring systems in agricultural industry applications.
Details
- Title
- Machine learning-based frost detection in plants from infrared thermography
- Authors/Creators
- Sayma Shammi
- Contributors
- Ferdous Sohel (Supervisor) - Murdoch University, Centre for Crop and Food InnovationDean Diepeveen (Supervisor)Sebastian Zander (Supervisor) - Murdoch University, School of Information TechnologyMichael G. K. Jones (Supervisor) - Murdoch University, Centre for Crop and Food Innovation
- Awarding Institution
- Murdoch University; Doctor of Philosophy (PhD)
- Identifiers
- 991005645270207891
- Murdoch Affiliation
- School of Information Technology; Centre for Crop and Food Innovation
- Resource Type
- Doctoral Thesis
UN Sustainable Development Goals (SDGs)
This output has contributed to the advancement of the following goals:
Metrics
92 File views/ downloads
122 Record Views