Forest Fire Image Classification Through Decentralized DNN Inference

In the realm of Natural Disaster Management (NDM), timely communication with local authorities is paramount for an effective response. To achieve this, multi-agent systems play a pivotal role by proficiently identifying and categorizing various disasters. In the field of Distributed Deep Neural Network (D-DNN) inference, such approaches often require DNN nodes to transmit their results to the cloud for inference, or they necessitate the establishment of a fixed topology network to enable inference directly on the edge, a practice prone to security risks. In this work, we propose a decentralized inference strategy tailored for fire classification tasks. In this approach, individual DNN nodes communicate within a network and enhance their predictions by considering other DNN node inference outputs that contribute to improving their individual performance. The overall coordination of the system on a specific decision is achieved through a consensus protocol, which acts as a universally accepted inference rule adopted by all DNN nodes operating within the system. We present a comprehensive experimental analysis, of the forest-fire classification task, focusing on enhancing both individual DNN node performance and the stability of the consensus protocol.