Abstract: In order to develop a method to identify the field maturity of tobacco leaves based on hyperspectral technology, the hyperspectral information of tobacco leaves with different field maturities were collected with a hyperspectral imager. The applicabilities of five data preparation methods, First derivative (1^stD), multiplicative scatter correction (MSC), standard normal variable transformation (SNV), Savitzky-Golay (SG) smoothing, first derivative + SG smoothing, and two modeling algorithms, support vector machine (SVM) and BP neural network (BPNN), in the development of discriminant model for field maturity of tobacco leaves, and the best characteristic variables for the development of discriminant model were selected using genetic algorithm (GA). The results showed that: 1) Hyperspectral reflectance of tobacco leaves with different field maturities were significantly different, especially within the wavelength range of 550-675 nm, and the reflectance increased with the increase of leaf maturity. 2) Among ten combinations between five spectral data preparation methods and two modeling algorithms, the combination of SNV and SVM was the optimal one. 3) Genetic algorithm was used to select 19 characteristic spectral bands reflecting the differences of field maturities of tobacco leaves between 400 and 1 000 nm, most of which correlated to the characteristic spectra of plastid pigments of tobacco leaves. 4) An SVM discriminant model for the field maturity of tobacco leaves was established by taking characteristic bands as the input variables. The prediction accuracy rate reached 95% with the F1 score of 0.95, and the average precision rate and recall rate were both higher than 95%. Hyperspectral information can accurately reflect the differences of field maturity of tobacco leaves. The combination of SNV data preparation method and SVM algorithm can be used to establish discriminant model with excellent performance for field maturity of tobacco leaves.