Uncertainty-Aware Wildfire Smoke Density Classification from Satellite Imagery via CBAM-Augmented EfficientNet with Evidential Deep Learning
A new study presents a probabilistic framework for assessing wildfire smoke severity from satellite imagery. The model categorizes smoke into Light, Moderate, and Heavy classes while providing uncertainty estimates. It achieves high accuracy in classification and demonstrates effective localization of smoke regions.
- ▪The proposed model uses a CBAM-augmented EfficientNet architecture to classify smoke density.
- ▪It achieves a weighted test accuracy of 93.8% on a dataset of 16,298 satellite patches.
- ▪The model provides both epistemic and aleatoric uncertainty estimates in a single forward pass.
Opening excerpt (first ~120 words) tap to expand
Computer Science > Computer Vision and Pattern Recognition arXiv:2605.15894 (cs) [Submitted on 15 May 2026] Title:Uncertainty-Aware Wildfire Smoke Density Classification from Satellite Imagery via CBAM-Augmented EfficientNet with Evidential Deep Learning Authors:Ranjith Chodavarapu View a PDF of the paper titled Uncertainty-Aware Wildfire Smoke Density Classification from Satellite Imagery via CBAM-Augmented EfficientNet with Evidential Deep Learning, by Ranjith Chodavarapu View PDF HTML (experimental) Abstract:Rapid and accurate wildfire smoke severity assessment from satellite images is essential for emergency response, air quality modeling, and human health risk management.
…
Excerpt limited to ~120 words for fair-use compliance. The full article is at arXiv cs.AI.
Discussion
0 commentsMore from arXiv cs.AI
