Abstract: To accurately estimate the leaf area index (LAI) of tobacco, a field experiment was carried out in a tobacco experimental plot in Xuanzhou District, Xuancheng City, Anhui Province, China. Three key sampling time points, namely the booming stage (T1), budding stage (T2), and early maturity stage (T3), were selected for synchronous data collection, including multispectral imagery, Light Detection and Ranging (LiDAR) data, and field-measured LAI data. Tobacco Vegetation Indices (VIs) and Texture Features (TIs) were extracted from the multispectral imagery, and the Canopy Height Model (CHM) was derived from the LiDAR data. Feature selection was performed via feature importance analysis. LAI inversion models for tobacco at different sampling time points were established using five algorithms: Back Propagation Neural Network (BPNN), Random Forest Regression (RFR), Support Vector Regression (SVR), Convolutional Neural Network (CNN), and Long Short-Term Memory (LSTM) network. The accuracy of all models was evaluated to identify the optimal LAI inversion model. The results showed that: ① Compared with TIs, VIs had a stronger correlation with tobacco LAI. Among models built with single-type features, VIs-based models outperformed TIs-based models overall, and the combination of VIs and TIs further improved the inversion accuracy. ② CHM data effectively enhanced the accuracy of LAI inversion, and the magnitude of improvement was highly coupled with the correlation between LAI and CHM. Specifically, significant improvement was observed at T1 and T2, while only negligible improvement was found at T3. ③ The CNN model achieved the best LAI inversion performance with the combined input of CHM, VIs and TIs, yielding coefficients of determination (R²) of 0.822, 0.847 and 0.820 at the three sampling time points, respectively. This study provides a methodological reference for tobacco growth monitoring and yield estimation based on multi-source remote sensing data.