Automatic IoT device identification: a deep learning based approach using graphic traffic characteristics

IoT device identification is an effective security measure to track different devices, helping analyze and defend against potential vulnerabilities of various IoT devices. However, existing IoT device identification works mainly use hand-designed features generated from relevant prior knowledge in the field, resulting in additional labor costs, low efficiency, and loss of some potential features. In addition, most of these works only identify known devices in the training set, without considering unknown devices. In this paper, we propose a quick and efficient IoT device identification method. Our method employs the convolutional neural network and converts raw network traffic into images as the model input, automatically extracting features from images instead of manually extracting features. Our method can identifies device types including unknown device types, and detects abnormal traffic of devices. We achieve over 98% accuracy on public datasets with few time consume, demonstrating the accuracy and practicality of our method.

Automated summary

IoT device identification is an effective security measure that helps analyze and defend against potential vulnerabilities. This paper proposes a quick and efficient method that uses a convolutional neural network to convert raw network traffic into images, automatically extracting features from them.