To achieve rapid，non-destructive and large-scale field measurement of nitrogen content in flue-cured tobacco leaves，monitor the nitrogen nutritional status of tobacco in a timely manner and develop precise fertilization strategies，utilized a DJI drone equipped with a hyperspectral camera to capture canopy images of flue-cured tobacco and extract spectral reflectance data. Three algorithms，namely Successive Projections Algorithm，Least Absolute Shrinkage and Selection Operator and Elastic Net，were employed to reduce the dimensionality of the original spectral data，extract nitrogen-sensitive bands，and calculate vegetation indices. Subsequently，three machine learning algorithms were applied to construct inversion models for leaf nitrogen content. The results showed that the combination of Successive Projection Alogrithm for dimensionality reduction and Random Forest for modeling achieved the best performance，with a coefficient of determination of 0.711 and a root mean square error of 5.138 mg/g. The second-best performance was achieved by Successive Projection Alogrithm combined with Multiple Linear Regression，with a coefficient of determination of 0.70 and an root mean sqvare error of 5.234 mg/g. This study provided effective technical support for smart nutrient management in flue-cured tobacco fields.