The invention discloses a single image super-resolution reconstruction method based on multi-directional feature prediction a priori. The invention mainly comprises the following steps: a feature prediction depth convolution neural network is constructed; according to the four directional features, the constructed network is trained separately. The trained network model is used to predict the multi-directional features of the input low-resolution image. Using the four kinds of directional features predicted in the previous step, a multi-directional feature prediction a priori is constructed. Construction of Super-resolution Reconstruction Cost Function Based on Depth Convolution Neural Network Feature Prediction and Degradation Constraint; the TFOCS technique is used to optimize the reconstruction cost function, and the high resolution image estimation is obtained. The single image super-resolution reconstruction method of the invention can obtain good subjective and objective effects,and has good anti-noise performance. Therefore, the invention is a high-performance single image super-resolution reconstruction method, and has extremely high application potential in the fields ofaerospace, traffic monitoring, medical imaging, film and television entertainment and the like.