Abstract: In order to observe the status of tobacco leaves exactly in real time and set appropriate curing parameters accordingly in the process of flue-curing, a CNN-LSTM model for identifying tobacco flue-curing stages based on Convolutional Neural Network and Long Short-Term Memory Neural Network was developed. An image acquisition device composed of industrial cameras and light sources was used to collect tobacco image data during curing, and the acquired image data were categorized into 10 curing stages according to the color and texture changes of tobacco leaves. The CNN-LSTM model was verified and compared with three conventional recognition models: CNN, ResNet50 and Inception-ResNet v2. The results showed that: 1) For the 10 flue-curing stages, the recognition accuracies of the training and verification sets of the CNN-LSTM model reached 99.55% and 98.70%, respectively. The recognition accuracy and average F1 score of the test set reached 99.58% and 99.36%, respectively, which were 1.17 and 1.72 percentage points higher than those of the Inception-ResNet v2 model. 2) The number of samples which were incorrectly judged by the CNN-LSTM model was far less than that by the other three models, and most of them fell in adjacent curing stages. This method supports the rapid and accurate identification of tobacco flue-curing stages.