Detection of Node Capture under Asynchronous Sleep Mode Based on Recurrent Neural Network

In recent years, wireless sensor networks (WSNs) have found numerousapplications in industrial manufacturing and people’s daily lives. However, security risks associated with the use of WSNs have also become increasingly pronounced. An attacker launching an internal attack on a WSNmust first physically capture several nodes, i.e., take control of the target nodes by acquiring, cracking, and analyzing important information carried by the target nodes, thus laying the groundwork for subsequent attack steps. Therefore, physical node capture is a critical step in an internal attack on aWSN. Detecting behaviorsthat indicatephysical capture of nodes provides an early warning of anetwork attack, allowing steps to be taken to prevent further attacks from being launched from the captured nodes. This paper proposes an RNN (recurrent neural network)-based detection method that can be used to detect node capture in WSNs with asynchronous sleep mode at an early stage (i.e., before captured nodes rejoin the network).Thus, the methodenables early detection of network attacks. During the decision-making process, a common monitoring mechanism that relies on cooperation between neighboring nodes is employed to improve detection accuracy. The proposed method obtains the sensor nodes’ states and makes a judgment with the help of RNN, achieving accurate detection of node capture under the condition of unsynchronized clocks. Simulation results demonstrate the proposed method’s capability to achieve high detection accuracy.