Abstract: To achieve nondestructive and efficient monitoring in changes of physicochemical indexes of tobacco leaves during flue-curing process, a lightweight MobileNetV2_CBAM deep transfer learning regression model was proposed and tested. The model employed the pre-trained MobileNetV2 as its feature extraction backbone network and incorporated a convolutional block attention module (CBAM) and a deep regression head to enhance feature expression capability and nonlinear fitting ability. The proposed model was used to predict the contents of moisture, starch, reducing sugar, and total sugar in tobacco leaves during flue-curing, and the results showed that: 1) Simultaneously incorporating CBAM and a deep regression head into the MobileNetV2 model significantly improved predictive performance. 2) The MobileNetV2_CBAM model achieved 0.561 G floating-point operations (FLOPs) with total parameters of 15.43 M. The coefficients of determination (R²) for 4 predicted physicochemical indexes ranged from 0.865 1 to 0.906 5, the mean absolute error (MAE) ranged from 0.062 2 to 0.983 1, the root mean square error (RMSE) ranged from 0.092 7 to 2.043 5, residual prediction deviation (RPD) ranged from 2.594 3 to 4.646 2, and frames per second (FPS) ranged from 100.578 0 to 121.254 4. This model achieved a good balance between predictive performance, processing speed and deployment cost. 3) Across different flue-curing stages, the MobileNetV2_CBAM model achieved an explanatory variance score ranging from 0.73 to 0.91 for the predicted results of the 4 physicochemical indexes, demonstrating strong predictive capability. 4) In complex backgrounds, the MobileNetV2_CBAM model still had good predictive ability for the four indexes with an R² range of 0.788 5-0.870 1. This study provides a theoretical reference for online monitoring of physicochemical indexes during tobacco flue-curing.