Abstract: Abstract: During the storage process, honeysuckle easily undergoes discoloration and mildew under the influence of temperature, humidity and microorganisms, which leads to a significant decrease of its medicinal efficacy and economic value, and even harms the health of consumers. Hence, it is necessary to monitor the quality of honeysuckle during storage. Chlorogenic acid (CGA), as the main active ingredient, is an important indicator to evaluate the quality of honeysuckle. In order to realize rapid and effective detection of CGA content in honeysuckle, 500 hyperspectral images of honeysuckle during different storage periods were collected by hyperspectral imaging (HSI) system, and then CGA content values were measured by high performance liquid chromatography (HPLC) method. Average spectral information extracted from the hyperspectral images and corresponding CGA values were used to build HSI detection models. Because of the non-uniformity of sample surface, baseline drift of instrument, random noise and light scattering, the collected hyperspectral images contained some redundant information, which could reduce the accuracy of modeling. In order to improve the prediction accuracy and efficiency of the model, six spectral preprocessing methods were used to improve the signal-to-noise ratio of the original spectrum, including Savizky-Golay filter (SG), moving average, standard normal variable (SNV), baseline correction (BC), multiplicative scatter correction (MSC), orthogonal signal correction (OSC). Comparing the effects of different pretreatment methods by establishing partial least squares regression (PLSR) models, the SNV-PLSR model obtained the best prediction accuracy with determination coefficient (R2) of 0.976 6 and root mean square error (RMSE) of 0.271 1% in prediction set, and SNV was identified as the best pretreatment method for further analysis. In order to simplify the calibration model, the uninformative variable elimination (UVE), successive projections algorithm (SPA), competitive adaptive reweighted sampling (CARS), the combination of UVE and CARS (UVE-CARS), and the combination of UVE and SPA (UVE-SPA) were used to extract characteristic wavelengths from the pre-processed spectrum by SNV method. And UVE, CARS, SPA, UVE-CARS and UVE-SPA selected 192, 51, 17, 26, 9 characteristic wavelengths from the full spectrum. Then, based on the full spectrum data and the selected characteristic variables by five variable screening methods, the linear PLSR and the non-linear BP neural network model were established. The performance of all the models were evaluated by the index of R2 for calibration set and prediction set, (RMSE) for calibration set and prediction set. The results showed that UVE-CARS algorithm could effectively eliminate useless information variables from full spectrum, and 26 characteristic wavelengths were selected from full spectrum by UVE-CARS algorithm, and the established model based on UVE-CARS algorithm had high accuracy, which was considered as the best feature wavelength screening method. The prediction results of the non-linear BP model were better than that of the linear PLSR model. In all BP model, the prediction accuracy of UVE-CARS-BP was the highest with R2 of 0.978 4 and RMSE of 0.250 3% in prediction set, respectively, and it was proved that the non-linear model was more suitable for the prediction of CGA content in honeysuckle. In conclusion, HSI technology combined with SNV-UVE-CARS-BP model can realize the rapid and non-destructive prediction of CGA content in honeysuckle during storage.