Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

Next day fire prediction via semantic segmentation

Konstantinos Alexis, Stella Girtsou, Alexis Apostolakis, Giorgos Giannopoulos, Charalampos Kontoes

TL;DR

This paper reformulates the next day fire prediction task as a semantic segmentation task on images, starting from the previous problem formulation as a binary classification task on instances and reformulating the problem as a semantic segmentation task on images.

Abstract

In this paper we present a deep learning pipeline for next day fire prediction. The next day fire prediction task consists in learning models that receive as input the available information for an area up until a certain day, in order to predict the occurrence of fire for the next day. Starting from our previous problem formulation as a binary classification task on instances (daily snapshots of each area) represented by tabular feature vectors, we reformulate the problem as a semantic segmentation task on images; there, each pixel corresponds to a daily snapshot of an area, while its channels represent the formerly tabular training features. We demonstrate that this problem formulation, built within a thorough pipeline achieves state of the art results.

Next day fire prediction via semantic segmentation

TL;DR

This paper reformulates the next day fire prediction task as a semantic segmentation task on images, starting from the previous problem formulation as a binary classification task on instances and reformulating the problem as a semantic segmentation task on images.

Abstract

In this paper we present a deep learning pipeline for next day fire prediction. The next day fire prediction task consists in learning models that receive as input the available information for an area up until a certain day, in order to predict the occurrence of fire for the next day. Starting from our previous problem formulation as a binary classification task on instances (daily snapshots of each area) represented by tabular feature vectors, we reformulate the problem as a semantic segmentation task on images; there, each pixel corresponds to a daily snapshot of an area, while its channels represent the formerly tabular training features. We demonstrate that this problem formulation, built within a thorough pipeline achieves state of the art results.
Paper Structure (15 sections, 1 equation, 2 figures, 3 tables)

This paper contains 15 sections, 1 equation, 2 figures, 3 tables.

Table of Contents

  1. Introduction
  2. Background and related work
  3. Problem definition
  4. Related work
  5. Methodology
  6. Feature extraction
  7. Dataset selection and augmentation
  8. Semantic segmentation network training and validation
  9. Semantic segmentation architecture: U-Net
  10. Cross-validation and model selection
  11. Experimental evaluation
  12. Evaluation setting
  13. Results
  14. Conclusion
  15. Acknowledgements

Figures (2)

  • Figure 1: The implemented semantic segmentation pipeline.
  • Figure 2: Predicted fire masks on test tiles. Images on the first and third columns correspond to the actual fire annotations, while on the second and fourth columns the corresponding predicted masks are shown. Images on the first row showcase predictions made by the U-Net-sh2 variation proposed in this work on random 2019 test tiles. In the second row, predictions made by the U-Net-sh1 on random 2020 test tiles are shown.