Precision seeding put forward higher requirements for the seed quality. However, existing grading methods of seeds are destructive and unsuitable for sorting seeds based on internal characteristics, and a classification method based on the seed vitality is required. An algorithm of image acquisition and characteristics identification and classification of tomato seeds was proposed based on hyperspectral technique and image processing technology. Totally 170 grains of tomato seeds were randomly selected as the research object, and the ratio of calibration set and validation set was about 3∶1. Images of tomato seeds were collected by hyperspectral acquisition system which was composed of high performance lighting CCD camera, line scanning spectrometer, oriel instruments, image acquisition card and computer. The resolution of the camera and range of spectrometer were respectively 1376 pixels×1040 pixels and 400~1100nm. An average spectrum of interest region of each seed could be obtained. Then the results of seed vigor were obtained from the standard germination test. And the characteristic wavelengths of tomato seed vigor were acquired by successive projections algorithm (SPA), including 535nm, 577nm, 595nm, 654nm, 684nm, 713nm, 744nm, 768nm, 809nm and 840nm. The images under above characteristic wavelengths were preprocessed by bilateral filtering, Ostu and morphological transformation to gain the seed eigenvalues including area, circularity and average gray. The classification thresholds were calculated according to the eigenvalues and vitality results of the calibration set based on statistical regularity, and prediction analysis of validation set was carried out. The results showed that both of average of area and gray had significant difference between viable seeds and non-viable seeds, while the difference of circularity between viable seeds and non-viable seeds was insignificant. Classification accuracy of calibration and validation sets was above 85% in eight characteristics wavelengths. And 713nm gave the best result, the accuracy of the calibration and validation set were 93.75% and 90.48%, respectively. The results provided a new method for rapid nondestructive grading of tomato seeds, and lay the foundation for the development of tomato seed grading equipment based on seed vitality.