Privacy protected internet of unmanned aerial vehicles for disastrous site identification

Abstract

AbstractDisastrous site identification through the internet of UAVs is a current research area that aims to improve data sharing by connecting servers to the internet. Internet‐connected unmanned aerial vehicles (UAVs) for aerial image classification necessitated the sharing of large datasets between the number of connected flying machines. The artificial intelligence model training to predict images need to compromise data privacy, consume a lot of energy, and high requirement of data communication. This article aims to propose and implement federated deep learning trained aerial image classification for disastrous sites through the internet of UAVs. The proposed model provides privacy‐preserving and resource efficient deep learning by sharing only models rather than large datasets of images. Multisystem client–server federated learning UAV architecture was implemented and comparatively analyzed on basis of parameters namely, training–testing accuracy, train‐testing loss, RAM‐CPU utilization.