Evaluation of CNN and U-Net Models for Segmenting and Predicting Wildfires from Satellite Images

Abstract

Monitoring the environment and disaster management has become difficult due to increasing wildfires. A system that can predict accurately and effectively. Predicting forest fires from remote sensor data is difficult when there is class imbalance, spatial variability and changing situations. The studies has been done on the different machine learning and deep learning models for predicting wildfire using TS-SatFire dataset. The Random Forest, CNN, U-NET and U-NET++ were evaluated under the same simulated conditions. Deep learning does not have maximum capacity in modelling spatial patterns of wildfire than conventional methods can have. Optimized U-Net gives good results in terms of F1 score 0.7145 and IoU 0.5558. Random Forest also wins the game with low cost The poor spatial localization of the CNN model leads to its ineffective performance. The results imply that real-world wildfire monitoring and management systems must achieve a viable trade-off between prediction efficiency and computational cost.