Towards Fire Identification Model in Satellite Images Using HPC Embedded Systems and AI.

Forest fires and environmental disasters that are rarely avoided due to Forest fires are environmental disasters is a crucial problem to resolve with High Performance Computing (HPC) due to the real-time need to avoid the reaction of the control agencies and the community. One of the strategies to support early warnings related to forest fires is using space technology and realtime image treatment. However, the large amount of data given by the satellite images, the cost of the satellite technology, and the difficulty of accessing remote places information make it challenging to deal with the problem. This project presents the development of a solution that fights fires through identification supported using artificial intelligence (AI), mainly Convolutional Neural Networks (CNN) and Computer Vision (CV). Space technology captures images in various spectral frequency ranges by optical instruments onboard artificial satellites. In addition, the solution deploys on a low-cost and easily accessible open-source embedded system, which allows its scope to be extended for use on mobile device applications such as robots, and uncrewed aerial vehicles, among others. This paper reflects the progress achieved within the project, mainly the creation of an open-source dataset of satellite images for fire classification, the election, conditioning, and training of the CNN.