The identification of rice seeds is crucial for agriculture production. An inverse Fourier transform (IFT) method based on laser-induced breakdown spectroscopy (LIBS) is proposed to identify five kinds of rice seeds. The LIBS data of the samples were preprocessed by inverse fast Fourier transform (IFFT), and the time-domain signals of rice seeds were obtained. The back propagation (BP) neural network was used to establish full spectrum, segmented spectrum, time-domain full spectrum and time-domain segmented spectrum discrimination models. Compared with the original spectrum, the time-domain spectrum can significantly improve the identification accuracy. The time-domain full-spectrum identification accuracy reached 95.28%, and the time-domain segmented spectrum identification accuracy reached 94.36%, whose identification time was only a few seconds. The results demonstrate that LIBS detection technology combined IFFT and BP neural network is fast and accurate, which provides a new idea for batch detection of rice seeds.