Abstract:

The number of heterogeneous terminals in the Internet of Things is large, the structure is simple, the security protection ability is weak, and it is easy to become the target of attack. Aiming at the difficulties in establishing the evaluation mechanism and low evaluation efficiency when traditional risk assessment methods deal with a large number of changing risk factors, a risk assessment model of IoT terminal based on improved convolutional autoencoder was proposed(Lightweight Convolutional Autoencoder combined with Fully Connected Layers and Classifier Model,LCAE-FC). A lightweight convolutional encoder was combined with a classifier to build a model, which integrated high-dimensional feature learning with the output evaluation probability of order dimensional reduction. The encoder introduced deep separable convolution, and each channel learned the internal structure of generalized behavioral risk. Each output feature was averaged and pooled to retain risk information to the maximum extent. The risk probability value was output by step-dimensionality reduction after the high-dimensional features were abstracted by the fully connected layer and classifier. The experimental results on the N-BaIoT dataset show that the accuracy and F1 value of the proposed model are higher than 99.3%, which has better performance than the traditional CAE, Bi-LSTM and SAE-SBR models.

Key words: internet of things terminal, risk assessment, convolutional automatic encoder, broad behavioral risk factors, depth-separable convolution