Using the spatial-spectral information of the altered license plates collected by hyperspectral lidar system, a twice-threshold method based on spectral feature is designed to recognize the altered license plates according to their retroreflection characteristics. Firstly, the characteristic wavelength of the spectra is selected, then the plate part and the altered material part are separated using the first threshold method according to the reflectance at the wavelength. Secondly, calculate the gradient value of the spectral reflectance curve at each point of the license plate part in the characteristic band, and the second threshold method is used to distinguish the alteration types, such as background to character alteration and character to background alteration. Finally, the correct license plate characters are reconstructed based on the type of the alteration, and the correct license plate image is reconstructed with the fusion of two-dimensional data and the corrected license plate characters, which is sent to the license plate recognition system. Experimental results show that this method can recognize altered license plates correctly regardless of different altered materials or different altered types.