In recent years,grassland survey and monitoring is mainly based on sa tellite remote sensing spectral images,but the low resolution and high cost lim it its applications.The high-resolution hyperspectral images obtained at a clos e range can make up for the lower resolution of remote sensing,which is current ly less studied.Therefore,in this study,we propose an efficient and accurate identification method for grassland forage hyperspectral images based on varianc e selection and Gaussian Naive Bayes,integrating hyperspectral imaging technolo gy with machine learning.First,hyperspectral images of grassland and visible-n ear infrared spectrum (400~1000nm) were collected using a hy perspectral imagin g system,the effective information in the features was optimized by the dimensi onality reduction method based on variance selection.Then,Gaussian naive bayes (Gaussian NB) and support vector machine (SVM) were used to establish the recogn it ion models with the K-fold cross-validation method.Finally,the model was evalu ated by Kappa coefficient,OA,test time and other indicators.In the preprocess ing phase,five methods were compared:multiplicative scatter correction (MSC), standard normal variable transformation (SNV),normalization,Savitzky-Golay smo othing filter (SG),and moving window smoothing spectral matrix (Nirmaf).Among them,MSC preprocessing improves the signal-to-noise ratio and guarantees the a ccuracy and stability of the prediction model.In feature selection and extractio n,the principal component analysis whitening based on variance selection (V-pca w) method was adopted,and number of optimal feature variables was set 2based o n the threshold and principal component.Compared with the principal component a nalysis (PCA),V-pcaw′s average of the overall classification accuracy and Kapp a coefficient increased by 2.995% and 0.05025,respectively.In the sa me case,the GaussianNB model and the SVM model are compared.In the GaussianNB mo del,the grass spectrum processed by MSC has the top recognition performance afte r V-pcaw feature extraction with the least running time. The values of OA,Kappa coefficient and test time are 99.33%,0.99and 0.002022s,resp ectively.The results showed that the algorithm based on variance selection and G aussian Naive Bayes enhanced the feature expression ability of hyperspectral ima ges of grassland forage effectively.and thus high efficient indentification of f orage species could be realized.