Two novel technologies are presented. The first one detects malicious packets by simply examining a single packet. This technique is based on word embedding and Long Short-Term Memory. The experimental results show that the detection accuracy is higher than 99.4%. The second technique is a flow-based mechanism but requires to examine only two packets in a flow. It uses a convolutional neural network and an automatic encoder. It can achieve accuracy higher than 99.8%.

1. This is the first work in the literature to perform packet-based malicious packet detection.
2. For flow-based, we show that it only needs to examine two packets from a flow, and 80 bytes of each packet, the accuracy rate is already higher than 99.8%.
3. The techniques have been tested on IoT attack traffic, our experimental results show the techniques are robust to IoT attacks.

Facing the DDoS attack caused by a large number of IoT devices, the industry urgently needs to deploy technologies to quickly detect malicious traffic on edge computing servers or network devices. The two fast malicious traffic detection technologies developed by the project can effectively block a large number of DDoS attacks of the Internet of Things, which will help the development of the Internet of Things industry.