Tropical dry forest land use/land cover change detection using semi-supervised deep learning algorithms and remote...

Area of Science:

• Environmental Science
• Remote Sensing
• Computer Science

Background:

• Tropical dry forests (TDFs) are vital but challenging to map with remote sensing due to spectral similarities and landscape complexities.
• Data-scarce regions face limitations with traditional land use and land cover (LULC) classification methods.

Purpose of the Study:

• To develop a novel semi-supervised deep learning (DL) framework for accurate TDF change detection using fused SAR and optical satellite imagery.
• To address the challenges of mapping TDFs in data-scarce environments with limited labeled data.

Main Methods:

• A semi-supervised DL framework combining unsupervised pseudo-labeling and a custom Y-Net architecture was developed.
• The framework fuses synthetic aperture radar (SAR) and optical satellite imagery for enhanced change detection.
• The model was evaluated for its performance in detecting TDF changes with varying amounts of labeled data.

Main Results:

• The proposed framework achieved state-of-the-art results with a mean overall accuracy of 95.3% and a mean Intersection over Union (mIoU) of 88.1%.
• The semi-supervised approach maintained over 90% accuracy even with only 60% labeled data, demonstrating robustness.
• The framework successfully mapped TDF changes in Colombia's Cauca River Valley between 2017 and 2021.

Conclusions:

• The novel semi-supervised DL framework offers a robust solution for TDF change detection, especially in data-limited regions.
• This approach advances remote sensing applications for environmental monitoring, conservation, and sustainable resource management.