Satellite Data for Forest Fire Detection using Deep learning

Wildfires are devastating natural disasters that cause damage in the earth's ecosystem. Many detection and mapping systems which are created use a lot of tools, including artificial intelligence methods and good human observations. One of the most used systems is satellite. Remote sensing imagery is widely used for forest fire detection (FFD) due to their large zone coverage, which uses traditional and deep learning methods. In the last decade deep learning techniques have given promising results in remote sensing problems. This study uses Landsat-8 images dataset and deep convolutional network method for FFD. The network used in this paper has a special characteristic which is using simultaneously multiple kernels with different sizes. In this work, to improve the performance of forest fire detection, we have used several data in the deep learning input layer: bands 2, 6 and 7 of Landsat-8, and Forest Fire Index value, which is powerful index for FFD. Three different scenarios were used in this study with 3 network configurations for each one, resulting in 9 total distinct models, using multiple kernels of 3x3, 5x5 and 7x7. Landsat-8 images dataset and deep neural network model used in this paper have given good results in detecting forest fires of distinct shapes and different sizes in multiple difficult tests.