Detection of Dead Victims at Volcanic Disaster Location based on Drone and LoRa

Abstract

When a natural disaster occurs, the process of evacuating victims after a disaster must be carried out immediately to reduce the risk due to the late evacuation process. In this evacuation process, the Search and Rescue (SAR) team played a big role in addition to focusing on the safety of victims and also paying attention to their own safety. Meanwhile, due to the large area of the disaster the search took quite a long time. In this paper, research was carried out to facilitate the evacuation process at the location of volcanic disasters by using drones carrying electronic devices to find the whereabouts of victims on the surface. This process of detecting and classifying is based on the Convolutional Neural Network (CNN) method with the MobileNetV2 model. We train the data set of disaster victims using batch sizes 16 and 32 with epochs of 40, 80 and 100. The resulting model has the greatest accuracy of 0.81 and f1-score of 0.86.