IoT device recognition based on variational autoencoder and multiscale convolutional network

With the development of IoT technology, IoT devices are widely used in various fields of production and life, but they also bring severe challenges to equipment asset management and security management. It is particularly important to understand IoT devices in the network through device identification to ensure their normal operation. At present, most of the identification through network data does not consider the imbalance of samples between device traffic classes. Aiming at the problem of imbalanced data in most real world, an IoT device identification method based on variational autoencoder (VAE) and multi-scale convolutional network is proposed. First, the traffic is segmented into sessions, cleaned, and fixed-length intercepted; secondly, VAE is used to expand the IoT device data with less traffic to partially enhance the unbalanced data set; finally, a multi-scale convolutional neural network (MSCNN) model is constructed, and convolution kernels of different sizes in MSCNN are used to extract multi-scale features to improve the learning ability and training accuracy of the model. Experimental results show that the accuracy, recall, precision and F1-score indicators of this method on two data sets are all above 98%, which verifies the effectiveness of this method and can accurately identify IoT devices.