Abstract:In order to study the difference of protein content in rice from different varieties and places of origin，the collected rice was classified according to varieties and places of origin.Combining the extracted spectral information of the region of interest with the protein content determined by the chemical method，a fullwavelength prediction model was established，and the optimal model was determined as partial least-squares regression （PLSR）by comparison.Successive projection algorithm （SPA）was used to select characteristic bands and build the characteristic wavelength model of PLSR，which has similar performance to the full wavelength model.The hyper-spectral image at characteristic wavelength was extracted，and the spectral data of all the pixels on the extracted feature image was imported into the established SPA-PLSR model.Then the protein content of each pixel was predicted，and the hyper-spectral gray-scale image was pseudo-color processed to obtain the protein content distribution maps of rice from different varieties.The results showed that it was feasible to visualize the protein content distribution in rice by hyper-spectral imaging，which provides a basis for screening varieties and origins of rice in the later stage.