A CNN-based method for forest classification using compact PolSAR images

The primary intention of this study is to explore the ability of convolutional neural networks (CNNs) for forest classification using Compact Polarimetric (CP) data. Due to the phenomenal performance of the CNNs, more and more studies have tended to apply CNN-based methods to classify polarimetric s...

Full description

Saved in:
Bibliographic Details
Published in:Arabian journal of geosciences 2025, Vol.18 (1), Article 21
Main Authors: Ebrahimi, Sahar, Ebadi, Hamid, Aghabalaei, Amir
Format: Article
Language:English
Subjects:
Artificial neural networks
C band
Classification
Earth and Environmental Science
Earth Sciences
Modes
Neural networks
Original Paper
Polarimetry
Radar imaging
Radarsat
SAR (radar)
Support vector machines
Synthetic aperture radar
Citations: Items that this one cites
Online Access:Get full text
Tags: Add Tag
No Tags, Be the first to tag this record!
Description
Summary:The primary intention of this study is to explore the ability of convolutional neural networks (CNNs) for forest classification using Compact Polarimetric (CP) data. Due to the phenomenal performance of the CNNs, more and more studies have tended to apply CNN-based methods to classify polarimetric synthetic aperture radar (PolSAR) images. In this study, three strategies were applied for this purpose. The first strategy involved designing and applying a CNN-based network to the Full Polarimetry (FP) mode of RADARSAT-2 C band, the simulated CP modes, and the reconstructed Pseudo Quad (PQ) modes. The results of these different modes were then compared with each other. In the second strategy, we compared the outcomes obtained from the first strategy with those from the Wishart classifier and the support vector machine (SVM) used in previous studies. Finally, the last strategy combined the CP modes to improve the classification outcomes further. Results showed that the CNN network outperformed other methods by using the CP modes for forest classification, and combining π/4 and DCP_L modes provided higher overall accuracy.
ISSN:1866-7511
1866-7538
DOI:10.1007/s12517-024-12163-4