To address the challenges of cumbersome operation and spatiotemporal discontinuities in traditional laboratory-based water quality testing methods, this study proposes a total nitrogen inversion method based on hyperspectral imaging. Taking the Shiwuli River in the Chaohu Basin as the research object, near ground hyperspectral data serve as the data source. The hyperspectral data of the water samples are resampled to match the resolution of unmanned aerial vehicle (UAV) hyperspectral data. The random frog (RF) algorithm is used to extract the characteristic bands of the total nitrogen mass concentration in the water samples. The particle swarm optimization (PSO)-backpropagation neural network (BPNN) algorithm is then used to construct a total nitrogen inversion model, enabling the inversion of total nitrogen mass concentration in water bodies using UAV hyperspectral images. These results indicate that the feature bands 465.1, 495.2, 756.2, 830.1 nm, and 847.7 nm, extracted using the RF algorithm, align with the sensitive band range of total nitrogen. The established PSO-BPNN inversion model has a prediction coefficient of determination (R²) of 0.862 and a root mean square error (RMSE) of 0.405 mg/L for the training set, while the test set yields a prediction R² of 0.711 and RMSE of 0.640 mg/L. The RMSE of the test set is significantly reduced compared with those by the BPNN and partial least squares models. Applying this model to UAV hyperspectral imaging enables rapid inversion of the spatial distribution characteristics of total nitrogen, with the relative deviation between the inversion values at verification points and the measured values remaining below 5.50%. These findings demonstrate that the model exhibits a certain degree of generalization and strong practical applicability.