Abstract: CPSs incorporate computation units into the original industrial control systems, which also enable attackers to launch attacks through the communication channels. This paper considers the problem that attackers maliciously tamper sensor measurements in the channels such that the computation unit cannot make right decisions based on these measurements. In order to detect such deception attacks on CPS control layer, an improved Kalman filter algorithm is proposed, which has a relatively better performance on detecting various deception attacks, and can recover the tampered data to some extent. The improved algorithm is utilized to detect the measurements of the reactor pressure in the Tennessee Eastman process. Besides, the results are also compared with those of the original algorithm. It shows through the experiments that the improved algorithm is superior to the original one in the detection and recover on bias and geometric attacks.