University of Twente Student Theses


Deep learning-based polarimetric SAR and optical image fusion to map oil palm along rivers

Fauzan, Muhammad Afif (2023) Deep learning-based polarimetric SAR and optical image fusion to map oil palm along rivers.

[img] PDF
Abstract:The cultivation of vegetable oil crops such as oil palm often occurs illegally in ecologically significant areas such as river banks despite being protected by law. More lands including river banks are expected to be converted into oil palm plantations to meet the growing demand for palm oil, but the magnitude of this problem is currently undocumented. In this study, a complex-valued deep learning classification model to combine complex-valued Synthetic Aperture Radar (SAR) and optical images was developed for mapping oil palm plantations. The proposed model was compared with several baseline classification models. The result suggested that the complex-valued neural network trained with complex-valued Sentinel-1 SAR alone achieved the highest accuracy with an F1-score of 0.972. This model was applied to classify multi-temporal SAR images from 2017 to 2021 to produce multi-temporal oil palm plantation distribution maps on the riparian zones. It was found that oil palm cultivation on river banks increased with an average rate of nearly 4% per year. By 2021, the total oil palm plantation areas on river banks reached 8500 Ha which accounted for 29% of river banks in the study area. However, as the model failed to detect sparse and open-canopy oil palm plantations, the result should be interpreted carefully as solely productive oil palm plantation areas indicated by their closed-canopy characteristic once the trees reach a mature age and start producing palm fruit bunch.
Item Type:Essay (Master)
Faculty:ITC: Faculty of Geo-information Science and Earth Observation
Programme:Geoinformation Science and Earth Observation MSc (75014)
Link to this item:
Export this item as:BibTeX
HTML Citation
Reference Manager


Repository Staff Only: item control page