Abstract: For fast, non-destructively and real-time monitoring of nicotine content in upper leaves of flue-cured tobacco in the field, experiments were conducted with two tobacco cultivars (Cuibi No.1 and Yunyan 87) at five fertilizer levels across different tobacco growing regions. A UAV remote sensing platform equipped with a hyperspectral camera was used to acquire canopy hyperspectral data at three post topping growth stages (topping stage and mature stages) on lower and middle leaves. The original canopy spectra were preprocessed by single or combined preprocessing methods, and the characteristic bands for nicotine in upper leaves across the three growth stages were screened by successive projection algorithm (SPA) and random frog algorithm (RF). Partial least squares regression (PLSR), support vector machine (SVM), ridge regression (RR), kernel ridge regression (KRR) and gradient boosting regression tree (GBR) models were established using both all spectra bands and characteristic bands to predict nicotine content in upper leaves across the three growth stages. The results showed that: 1) Appropriate preprocessing methods improved model accuracy for nicotine content in upper leaves across the growth stages, and the first derivative combined with the standard normal variate (D1-SNV) showed optimal prediction performance for nicotine content in upper leaves across the three stages. 2) RF and SPA were able to screen out 19 and 13 characteristic bands respectively. Of these two techniques, the characteristic bands screened by RF were relatively more evenly distributed and included the visible and near infrared bands, while the characteristic bands screened by SPA were less evenly distributed and the number of characteristic bands in the 400-700 nm range was lower. 3) In general, the D1-SNV-RF-KRR model achieved optimal prediction performance with R², RMSE and RPD of the test set of 0.872, 0.228 and 2.796 respectively, indicating higher accuracy and stability. This study provides a methodological foundation for large-scale real-time monitoring of nicotine content in upper leaves of flue-cured tobacco using UAV-based hyperspectral remote sensing technology after topping.