A4 Refereed article in a conference publication
Detection of Deforestation Using Prisma Hyperspectral and Deep Learning (1DCNN) in the Amazon Forest
Authors: Gupta, Rajit; Ruokolainen, Kalle; Tuomisto, Hanna
Editors: N/A
Conference name: IEEE International Geoscience and Remote Sensing Symposium
Publication year: 2024
Journal: IEEE International Geoscience and Remote Sensing Symposium proceedings (IGARSS)
Book title : IGARSS 2024: 2024 IEEE International Geoscience and Remote Sensing Symposium
First page : 3704
Last page: 3707
ISBN: 979-8-3503-6033-2
eISBN: 979-8-3503-6032-5
ISSN: 2153-6996
eISSN: 2153-7003
DOI: https://doi.org/10.1109/IGARSS53475.2024.10642196
Web address : https://ieeexplore.ieee.org/document/10642196
Accurate detection of deforestation and logging activities is useful to monitor large scale damages in the Amazon forests. In this study, we focused on the use of deep learning based one dimensional convolutional neural network (1D-CNN) with Hyperspectral Precursor of the Application Mission (PRISMA) hyperspectral data for the detection of deforestation in the Amazon Forest. The PRISMA data was pre-processed to remove noisy bands, water absorption and some blue spectrum bands. Three main classes were identified and sampled as ground truth for classification: forest, deforestation and waterbodies. 1D-CNN were parameterised to obtain a classified map and then accuracy assessment was performed. Model achieved a very high overall accuracy of 98.92%, confirming that the method can be used for accurate mapping of deforestation.
Funding information in the publication:
Funding was provided by Research Council of Finland (grant 351460 to Hanna Tuomisto). This study was carried out using PRISMA products, of the Italian Space Agency (ASI), delivered under an ASI License to use.
