Citri grandis fructus immaturus is a local Chinese medicinal material with a long history of medicinal use in Guangdong Province, because the higher the price of the product with the older the production year, the phenomenon of shoddy charging is often in the market. The study used hyperspectral imaging technology combined with the Harris Eagle optimized(HHO) kernel extreme learning machine(KELM) to identify four sets of different years of citri grandis fructus immaturus. In this study, 193 orange-red tire section samples were collected in four years, and hyperspectral images of 400~1 000 nm were collected. Firstly, the original reflection spectra of orange-red tire sections were analyzed by principal component analysis (PCA), and then Savitzky-Golay smoothing (S-G), multiple scattering correction (MSC),and standard normal variable exchange (SNV) were used to pretreat the sample spectra and establish KELM model, and found that the discrimination accuracy of the sample spectra treated by SNV was the highest, reaching 99. 24% of the training set and 95. 56% of the test set. Further, use of competitive adaptive weighting algorithm (CARS) and Monte Carlo Information-Free Variable Elimination (MCUVE) to select the characteristic wavelength of the sample spectrum; Finally, the discriminant model is established by KELM, and the HHO is used to optimize the KELM parameter selection and compare the modeling effect. The results show that the discrimination effect based on HHO-KELM is 0. 76%~4. 44% higher than that of KELM. The redundancy of feature band information obtained by MCUVE screening is significantly reduced, The accuracy is improved, and the optimal accuracy can reach 100% of the training set and 100% of the test set, so the use of hyperspectral imaging technology can realize the non-destructive identification of citri grandis fructus immaturus in different years.