To conduct the discriminant analysis of apple origin, red Fuji apples from Aksu, Jingning, Lingbao and Yantai were used as research objects, and the hyperspectral data of apples were collected in the range of 800-1700 nm. First derivative, second derivative, standard normal transform, multivariate scattering correction (MSC), wavelet transform, smoothing transform, and Fourier transform were used to pretreat the original spectral data. The partial least squares discriminant analysis of the linear model and extreme learning machine (ELM) and the support vector basis (SVM) of the nonlinear model were established. The results showed that the MSC pretreatment method was preferable, and the nonlinear model outperformed the linear model. For the spectral data pretreated by MSC, principal component analysis, continuous projection algorithm, and competitive adaptive weighted sampling (CARS) were used to extract the characteristic bands. Thereafter, the ELM and the SVM discriminant analysis models of the apple-producing area were established based on the characteristic bands. The results show that the CARS method performs the best in the terms of extracting characteristic bands. The accuracy rates of the two machine learning algorithms were 98.75% and 100%, respectively, which can serve as a theoretical reference and experimental basis for apple-producing areas discriminant analysis.