DL Based System for On-Board Image Classification in Real Time, Applied to Disaster Mitigation

Natural phenomena such as earthquakes that cause the destruction of bridges and can cause tsunamis, among other phenomena and disasters, are on the rise as a result of climate change, which is affecting our planet. This is because of the rise in average global temperature and the development of extreme rains, which causes rivers to overflow. The use of satellites for terrestrial observation has made it possible to keep tabs on events throughout the world in recent years. However, these satellite systems aren't without their drawbacks; one is that cloud cover may make it difficult to accurately record photographs. It is done based on the clouds present, and another drawback is the visiting time that the satellites have, which means that the area that you want to record the image is not always available, so you have to wait for the day and the time in which the satellite orbit can pass through the area of interest that the image is to be recorded. Drones, which fly at low altitudes and do not have cloud problems, are one such alternative, solving a problem of satellites, on cloud cover. In light of this, in the present work, a methodology based on low-cost systems is proposed for on-board processing in drones of such data in the event of an emergency, such as after the effects of rain or an earthquake. Methodology involves processing the image in an embedded device, which is coupled to the drone and connected to the camera, in such a way that the information that is sent to the ground station, concentrates on the affected area; this allows for faster decision-making and lessens the impact of the natural disaster. Therefore, it is shown by testing of the embedded computer based on a Raspberry pi 3 for detecting changes in land cover and transmitting this information to an earth station.