Security Performance Prediction Method of Artificial Intelligence of Things Based on Lightweight MS-Net Network

Emerging technologies such as artificial intelligence and big data have made numerous Internet of things (IoT) applications possible. In particular, the Artificial Intelligence of Things (AIoT) has the potential to promote the digitization and intelligent connection of all things. However, the openness and diversity of AIoT makes data information vulnerable to security attacks which can lead to a disruption of mobile communication networks. The complexity of real-time data security events requires accurate prediction of AIoT security performance. In this paper, a secure communication system model based on decode-and-forward (DF) relaying is proposed and its security performance is analyzed. Expressions for the secrecy outage probability (SOP) are derived, and these are used to evaluate the security performance. For this purpose, an intelligent SOP prediction algorithm based on MS-Net is proposed. MobileNet and SqueezeNet networks are used to design an improved lightweight MS-Net model, which is composed of a depth separable convolution block and a fire module in parallel. The fire module is used to reduce the number of parameters in the first branch, and the depth-separable convolution block is employed in the second branch instead of the standard convolution. This can adapt to nonlinear characteristic in the AIoT safety data and reduce energy consumption. Afterwards, the convolutional block attention module(CBAM) attention mechanism is used to improve the model's ability to capture features. The proposed algorithm provides better AIoT security performance than other algorithms. In particular, the mean squared error (MSE) is 68.1% better than that of RegNet.