Deep Learning Method for forest fire detection and simulation of wildfire expansion using sentinel-2 images

Abstract

One of the main threats that the world faces right now is natural disasters, which every year cause irreparable damage to both the human and natural environments. Forest fires have emerged as one of the most important threats that humanity faces in the 21st century. On the other hand, Convolutional neural networks as a part of machine learning have gained a lot of attention over the last years due to the rise of artificial neural networks. Convolu-tional neural networks are distinguished by their high accuracy and ability to recognize diffi-cult image patterns. In addition to that, cellular automata have been widely used for forest fire spread simulation as it can easily adapted to dynamic environments with spatiotemporal development. This thesis aims to integrate the emerging evolution of artificial intelligence with cellular automata in order to detect forest fire outbreaks and simulate forest fire spreads by using sentinel-2 images. The results of this study show that the inception V3 model, as part of convolutional neural networks, can detect very well fire outbreaks with low data availability and that cellular automata can simulate the spread of the forest fire very accurately.