University of Twente Student Theses
Utilizing Phenocam Imagery and Convolutional Neural Network for Plant Phenological State Prediction
Perez, Maria (2024) Utilizing Phenocam Imagery and Convolutional Neural Network for Plant Phenological State Prediction.
|
PDF
2MB |
| Abstract: | Phenology, the study of biological cycles in plants, is crucial for understanding ecosystem dynamics and the effects of climate change. Climate change has globally altered phenological patterns, thus impacting ecosystem interactions. While phenology research is well-established, its reliance on traditional satellite imagery presents limitations. Satellite data might have low temporal resolution, which restricts the accurate monitoring of phenological events. There is a need for reliable information using high-frequency data to gain deeper insights into these phenological events. We propose to develop a phenological classification model to facilitate the understanding of seasonal dynamics, utilizing PhenoCam imagery known for capturing frequent and detailed observations of vegetation. Specifically, our study utilizes PhenoCam images from two stations in the University of Twente over a four-year period (2020-2023). Convolutional Neural Networks (CNNs) serve as the primary tool for image analysis, due to their ability to automatically learn and extract complex features. This approach aims to achieve accurate classification of phenological states in deciduous vegetation, focusing on spring green-up and summer green peak. This study comprised three main phases. Firstly, we utilized PhenoCam imagery to capture comprehensive representations of phenological stages, complemented by data from other phenological sources. These images were meticulously organized and labelled to facilitate subsequent processing. Various approaches were employed to ensure accurate labelling of the images into their respective phenological stages. Secondly, a Convolutional Neural Network (CNN) model was developed to precisely identify phenological states from the labelled imagery. Lastly, we evaluated the CNN model's performance and compared it with traditional machine learning model. The results demonstrated that the CNN model outperformed a traditional machine learning model (RF) in accurately identifying phenological states. Across all evaluated metrics, the CNN model consistently achieved an accuracy of 0.95. Overall, this study contributes to phenological research by offering a robust framework for predicting vegetation phenological stages using advanced computational techniques. |
| Item Type: | Essay (Master) |
| Faculty: | ITC: Faculty of Geo-information Science and Earth Observation |
| Subject: | 54 computer science |
| Programme: | Geoinformation Science and Earth Observation MSc (75014) |
| Link to this item: | https://purl.utwente.nl/essays/101157 |
| Export this item as: | BibTeX EndNote HTML Citation Reference Manager |
Show download statistics for this publication
Show download statistics for this publicationRepository Staff Only: item control page