Agricultural applications of the Gaussian process (GP) based techniques is considered. A method of classifying crops from multi-temporal Landsat 8 satellite imagery is proposed. The method uses the model of spectral features based on GP regression with constant expectation and square exponential covariance functions. Main steps of the classification procedure and examples of recognition of culture species are represented. The ground based data are used for quantitative validation of the proposed classification method. The highest overall classification accuracy in three classes of crops is 77.78%.