It will make damage detection based on lamb wave more effectively to obtain accurate damage-scattered signal. Meanwhile, one of the key obstacles in the process of extracting damage-scattered signal is the fact that damage-scattered signal is often overlapped with many other reflected signals from structural features such as the boundary. To solve overlap, the baseline subtraction approach which is sensitive to external factors such as structure and environment is adopted in many existing methods to obtain damage-scattered signal. Due to the signal after incident wave in the damage signal captured by sensor is not an arbitrary signal, but rather a superposition of several reflected waves, so as long as the superposition of reflected waves which is most similar to the target signal is got, the target signal is considered to be interpreted successfully, and consequently, damage-reflected signal is got. Therefore, an approach is proposed based on the maximum likelihood for damage signal decomposition. Damage-reflected signals and boundary-reflected signals are simulated on the basis of analysis of lamb wave propagation characteristics, and then the various parameters of the synthesis of the above signals are optimized through genetic algorithm based on similarity index in order to make the similarity between the synthetic signal and the target signal reach the maximum. Finally, damage is localized through Time of Flight (ToF). The experimental results on an aluminum plate indicate that damage- reflected signal can be accurately extracted from the aliasing signal by means of the proposed method.