An optical fiber technique of impact load identification method of composite structure based on non-negative Bayesian regularization is proposed in terms of the problems existing in impact load identification. For example， the quality of sensors cannot be ignored and the identification result has negative component. Firstly， the fiber Bragg grating （FBG） sensor is integrated into the composite tube structure to obtain its dynamic response signal. Secondly， the impact load identification model is established to represent the impact load identification problem as deconvolution problem of discrete-time. Taking into account the non-negative characteristics of the impact load， its posterior probability density function is obtained through Bayesian hierarchical model. then the maximum posterior probability solution for the impact load is obtained by maximizing probability density function. The results show that the simultaneous measurement of impact response at multiple sensing points can be achieved by FBG sensors， and the non-negative Bayesian regularization method can effectively overcome the shortcomings of traditional Tikhonov regularization， adaptively determine the algorithm parameters， eliminate the negative components which have no physical meaning. Therefore， the reconstructed time history of impact load can coincide with the actual time history.