Abstract: State estimation and sensor attack detection methods based on unknown input observers are proposed to address cyber-physical systems' vulnerabilities to attack. Considering that the system network is composed of several subsystems, an overall system model is introduced. The system with unknown input and senor attack is transformed into a singular system without attack signal by considering the senor attack as an augmented system state. A new input observer is designed to estimate the system state when the unknown input and unknown noise signals are in the system. Under the premise that the attack signal can be detected, the observer gain matrix is determined by the minimum covariance matrix of residual to identify the attack signal when at least one subsystem is attacked. The detection accuracy is improved by setting an alarm threshold. Finally, simulation examples show that the proposed method is feasible by selecting a variety of parameters.