Near--Real-Time Conflict-Related Fire Detection in Sudan Using Unsupervised Deep Learning
Researchers have developed a new unsupervised deep learning method for near-real-time detection of conflict-related fires in Sudan. The approach utilizes a lightweight Variational Auto-Encoder (VAE) model with 4-band Planet Labs satellite imagery, achieving detection within 24 to 30 hours. This method outperforms existing techniques in precision, recall, and F1-score, particularly in imbalanced fire-detection scenarios, and offers a scalable solution for monitoring war-affected regions. AI
IMPACT Provides a novel, efficient method for monitoring conflict zones using satellite imagery and AI.