With the explosive growth of blockchain platforms and applications, security threats of blockchain also occur frequently. As a decentralized application deployed on the blockchain, smart contracts help the blockchain realize safe and efficient information storage, asset management, and value transfer. Therefore, smart contracts play a vital role in the security of the blockchain. In recent years, security threats against smart contracts have increased, not only causing huge economic losses but also impacting the credit system of the blockchain. Therefore, many researchers have carried out corresponding research on the security threats of smart contracts. Common threat detection methods include formal verification, symbolic execution, fuzzing, etc. Most of these methods are only for known threats, while there is not much work on detecting unknown threats. In order to better deal with unknown threats, we present a review of the typical smart contract security events in recent years, analyze the security threats from contract coding, Ethereum virtual machine, and blockchain characteristics. Further, we compare and summarize the latest unknown threat detection methods. Then, to address the problem that very few unknown threat samples are available, a detection method based on a few-shot learning is proposed.