A two-stage damage location recognition method based on guided-wave signal sparse decomposition and damage localization relates to the field of ultrasonic non-destructive testing. The present invention aims to resolve a problem that an ultrasonic guided-wave signal analysis result is not accurate enough because a dictionary design method and a signal sparse decomposition algorithm of sparse representation in ultrasonic guided-wave signal overlapping wave packet recognition are not perfect enough. According to the method, a penalty term is used to make coefficient vectors as sparse as possible,a possibility that noise matches dictionary atoms is significantly reduced; a dictionary matrix is designed by using a guided-wave propagation model, wherein problems of guided-wave frequency dispersion, multiplex modes and modal transformation are considered, and an overlapping waveform is recognized in a linear decomposition manner. The method has advantages over a conventional signal processing method. A sparse optimization solution algorithm based on sparse Bayesian learning is used, which has unique advantages in the aspect of handling with under-determined linear problems such as sparserepresentation, and has better robustness for noise.