University of Twente Student Theses
Deep-learning based analysis of satelites image time series for mapping forest regeneration in amazon rainforest
Hatangimana, F. (2024) Deep-learning based analysis of satelites image time series for mapping forest regeneration in amazon rainforest.
|
PDF
3MB |
| Abstract: | The Amazon Rainforest, crucial for global climate regulation and biodiversity, faces significant threats from deforestation and degradation. Traditional monitoring methods often lack precision and scalability, failing to capture the complex temporal dynamics of forest ecosystems. This study addresses these gaps by developing and implementing a deep-learning-based model to map forest regeneration areas in the Amazon Rainforest using satellite image time series. The methodology involved data pre-processing, model development, and performance evaluation using metrics such as accuracy, precision, recall, and F1-score. Precision-recall and ROC curves (AUC-ROC) were also employed to assess model performance. Time-series analysis is essential for understanding the temporal patterns of forest regeneration. We found that the hybrid transformer architecture outperformed the standard transformer model in distinguishing regenerated areas from other classes. The hybrid transformer model demonstrated superior performance, achieving an overall accuracy of 86.36% compared to the traditional transformer's 85.48%. The model achieved an F1-score of 0.863. When analysing longer periods, the hybrid transformer achieved an overall accuracy of 86.88%, a recall of 0.86, and a precision of 0.86. According to the most accurate model, secondary forest occupies 6.4% of the research area and has a mapping accuracy of 86.38%, which aligns with previous studies. In conclusion, the hybrid transformer model is a valuable tool for conservation and management, providing precise and reliable maps of forest regeneration. Future research should continue investigating hybrid transformers that utilize both spatial and temporal data and explore more advanced deep-learning architectures, such as Long Short-Term Memory (LSTM) networks. additional features such as the Normalized Difference Vegetation Index (NDVI), which is crucial for differentiating vegetation from other classes Additionally, integrating other additional features such as the Normalized Difference Vegetation Index (NDVI), which is crucial for differentiating vegetation from other classes. Keywords: Deep-learning models, Forest regeneration, Satellite image time series, Performance metrics. |
| Item Type: | Essay (Master) |
| Faculty: | ITC: Faculty of Geo-information Science and Earth Observation |
| Subject: | 38 earth sciences |
| Programme: | Geo-informatics MSc (60031) |
| Link to this item: | https://purl.utwente.nl/essays/103435 |
| 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