Abstract: Soil total nitrogen content is one of the core indexes of soil fertility. It is important for agricultural production to rapidly and accurately determine the soil total nitrogen content. A typical rice farmland in Jiangning District, Nanjing was taken as the research object. The checkerboard method was used to select 60 points, and each point was sampled in the 0-30 cm topsoil layer. The spectral reflectance of soil samples in the bands of 450, 560, 650, 730 and 840 nm was obtained by using DJI Phantom 4 Multispectral at the same time. Through the multivariate linear analysis of soil total nitrogen content and spectral reflectance, the unique multicollinearity problem characteristic of the spectral reflectance data was revealed. A prediction model for soil total nitrogen content retrieved from UAV remote sensing images based on ridge regression was constructed. The calculation results show that when the ridge regression coefficient reaches 0.12, the linear regression determination coefficient（R~2） is 0.408, and the variance inflation factors are under 10, and the regression coefficients are significantly different. The inversion model based on ridge regression can give good consideration to both inversion accuracy and spectral data multicollinearity of spectral data. The research results can provide a theore-tical basis for UAV remote sensing diagnosis of soil nitrogen nutrition.