Abstract: It is necessary to distinguish between the live and the dead insects effectively for counting the density of the storage insects accurately. A 900-1700nm near-infrared hyperspectral imaging system was developed to detect live insects in stored wheat. The Sitophilus oryzae was killed by using low-temperature sudden death method with liquid nitrogen, and then the hyperspectral images were acquired over the period of time 0-7 day after the death of insects. The relative spectral reflectance of the insects increased gradually with the duration of the death time. Then the spectral curve of the insects became stable on the fifth day after the death. 110 hyperspectral images whose wavelength was from 1320 to 1680 nm were analyzed by the neighbor wavelength index, and the optimal characteristic wavelength to distinguish the live and the dead was 1417.2 nm. The region-growing method for identifying the live insects was proposed based on the area ratio of the two thresholds for connecting regions. And the insect should be judged to be alive if the area ratio was higher than 0.5. The results showed that the training samples and the testing samples of the live and the dead insects were all correctly identified since the second day after the death. This research provides a basis for the real-time detection and classification of stored-gain live insects based on computer vision technology.