Dendrobium officinale has high commercial value and nutritional value. In this study,130 samples were taken as research samples from Wenshan in Yunnan,Jinxiu in Guangxi,Huoshan in Anhui and Taizhou in Zhejiang. The Raman spectrawere obtained by a portable Raman spectrometer under a 785 nm laser. Then, the total flavonoid content of Dendrobium officinale was determined by NaNO2-Al (NO3)3-NaOH colorimetry. With each normalized Raman spectral data as input,different preprocessing methods included SG,SNV and MSC are used to preprocess the spectral data. Partial least squares (PLS),support vector machine (SVM) and convolution neural network short and long-term memory neural network (CNN-LSTM) models are used as a comparison,and competitive adaptive reweighting sampling (CARS) is used as wavelength selection method to compare different machine learning models. In addition,the following prediction quality indicators were used: correction set and correlation coefficient of the test set (Rc，Rp),root mean square error of correction set and root mean square error of prediction set (RMSEC,RMSEP) to evaluate the performance of the prediction model of total flavone content in Dendrobium officinale. The results showed that the prediction accuracy of the CNN-LSTM method was the highest,with Rc and Rp of 0.983 and 0.964,RMSEC and RMSEP of 0.032 and 0.047 mg·g-1,respectively. The SNV-CNN-LSTM deep learning model based on Raman spectroscopy is accurate,reliable,and robust,superior to traditional machine learning models (PLS,SVM). In this study,Raman spectroscopy combined with the CNN-LSTM model was used to predict the content of total flavonoids in Dendrobium officinale with the characteristics of fast and non-destructive,which overcame the shortcomings of traditional physical and chemical identification methods. This method can distinguish the quality of Dendrobium officinale,accelerate the industrialization of Dendrobium officinale in the market of medicinal and edible homologous plants,build its brand and increase its influence. At the same time,this technology can also be applied to consumers and market supervision departments,with broad prospects.