Double-Step deep learning framework to improve wildfire severity classification

Wildfires are dangerous events which cause huge losses under natural, humanitarian and economical perspectives. To contrast their impact, a fast and accurate restoration can be improved through the automatic census of the event in terms of (i) delin- eation of the affected areas and (ii) estimation of damage severity, using satellite images. This work proposes to extend the state- of-the-art approach, named Double-Step U-Net (DS-UNet), able to automatically detect wildfires in satellite acquisitions and to associate a damage index from a defined scale. As a deep learning network, the DS-UNet model performance is strongly dependent on many factors. We propose to focus on alternatives in its main architecture by designing a configurable Double-Step Framework, which allows inspecting the prediction quality with different loss-functions and convolutional neural networks used as backbones. Experimental results show that the proposed framework yields better performance with up to 6.1% lower RMSE than current state of the art.