Aiming at that it is difficult to solve the 1/f noise re moving and the loss of signal for fiber Bragg gratting (FBG) sensing signal,the compressed sensing (CS) is used to process the sensing signal in this paper.But CS takes the sparsity as prio ri condition.The sparsity is determined with innovation method proposed in this paper,a new algorithm which is called group twice orthogonal matching pursuit (GtOMP) based on twice orthogo nal matching pursuit (TOMP) and group orthogonal matching pursuit (GOMP) is proposed,and the number of atoms in each i teration as well as the number of iteration is obtained.First of all,the calculated and normalized correlation coefficients in descending order used with the method of saturation value to determine the sparsity.And the number of atoms each iteration is confi rmed by using smoothness.In the end,the proposed method is applied to reconstru ct the FBG signal.Simulatin results show that,compared with the similar TOMP algorithm,the algorithm in this paper not only reduces the running time,but also decreases the reconstruction error of 6~20%,and compared with other algorit hms,signal noise ratio (SNR) of the signal after processing by the proposed method can be improved by more than 27dB.