Soil Total Nitrogen Spectral Inversion Based on the DANN Model

Soil degradation threatens agricultural sustainability and food security. This study explores the effects of spectral preprocessing methods and machine learning models on predicting soil total nitrogen using the LUCAS2009 dataset. Preprocessing techniques, including first derivative (FD) and detrending (DT), enhanced spectral sensitivity. Machine learning models such as partial least squares regression (PLSR), support vector machines (SVM), random forest (RF), and a novel Density-Aware Neural Network (DANN) were evaluated. DANN achieved the best performance on FD data, with R2 = 0.872, RMSE = 1.362, and RPD = 2.795. These findings provide insights into optimizing soil spectral analysis for precision agriculture.