| Autor: |
Meseci, Elif, Ozcan, Caner, Ozdemir, Dilara, Dilmac, Muhammet |
| Předmět: |
|
| Zdroj: |
Arabian Journal of Geosciences; Nov2023, Vol. 16 Issue 11, p1-11, 11p |
| Abstrakt: |
Classification of synthetic aperture radar (SAR) images is very important for analyzing these images. The developing remote sensing allows many high-dimensional SAR images to be recorded and interpreted. However, due to the growth in data sizes, the features increase, and analyzing becomes difficult. Therefore, deep learning algorithms capable of automatic feature extraction are needed. This study proposes SAR DenseNet-based Ensemble Network (SARDE-Net), an ensemble deep learning network based on DenseNet architectures, for the classification task. The high-dimensional real-world SAR image was taken from the TerraSAR-X image archive. Before the SAR image is transferred to our model, it is split into 100 x 100 patches and categorized into five classes. In addition, Sparsity-Driven Despeckling (SDD) filter is applied for denoising to increase the capability of proposed method on patch classification. Our method classifies denoised-SAR images obtained with a patch-based approach with its state-of-the-art components. SARDE-Net was compared to other deep learning classifiers in recent literature and achieved the highest results with 98.77% accuracy, 98.81% precision, 98.64% recall, and 98.72% f1-score metrics. It has been confirmed that our model can also be applied to large datasets containing complex images. [ABSTRACT FROM AUTHOR] |
| Databáze: |
Complementary Index |
| Externí odkaz: |
|
Full text from SpringerLink