In the field of soil digital mapping, precision agriculture and soil resource investigation, the study of aerial hyperspectral data to provide scientific prediction results by aerial hyperspectral have become the focus of research, especially in the case of black soil rich in nutrients in Northeast China. The data source is CASI-1500 aerial hyperspectral imaging system with a spectral range of 380~1 050 nm, and spatial resolution of 1.5 m. 59 soil samples were collected from the Jiansanjiang area in Heilongjiang, and the contents of organic matter, total nitrogen, total phosphorus and total potassium were obtained. In addition, the eps-regression support vector machine model, BP neural network and PLS1 least square regression model are selected to establish the machine learning model of spectrum and content. A support vector machine (SVM) method is used to extract the total nitrogen, total phosphorus and total potassium in aerial hyperspectral data by evaluating the prediction accuracy of the 3 models. The information of organic matter is retrieved by neural network. The results revealed that the date computed by the spectral statistic, spectral characteristics and spectral values is a kind of effective spectrum of training data, which can reflect the soil comprehensive reflectance situation. The neural network method is the most accurate method for the extraction of organic matter and total potassium. The errors are 1.21% and 0.81% respectively. The accuracy is the highest in the extraction of total nitrogen and total phosphorus information by support vector machines (SVM). The comprehensive accuracy of aerial hyperspectral extraction of soil nutrients was evaluated. The extraction errors of organic matter, total nitrogen, total phosphorus and total potassium were 5.25%, 6.05%, 2.74% and 8.90%, respectively, and the total phosphorus retrieval accuracy was the highest.