To overcome the shortcoming of bright speck phenomenon in four order totally variational model and the non-unique optimal solution in sparse regularization model of regularization image super-resolution reconstruction, an infrared remote sensing image super-resolution reconstruction model based on total generalized variation regularization is proposed in combination with the actual demand of infrared remote sensing image super-resolution reconstruction. The advantages and feasibility are analyzed with the concept of zero-order tensor space and the relaxation solution. Combined with the self-fissility of this model, the reconstruction model is split into two sub-problems by alternating direction multiplier method. The conjugate gradient method and the fast Fourier transform method are used to solve the sub-problem in numerical solution process, respectively. From the analysis of the testing results, the proposed model has a significant improvement in the resolution of the reconstructed image for both the simulated image and the real image. The objective evaluation is better than the method used in the literature, in which the peak signal to noise ratio can are increased by 1, 0.02 and 0.1 unit, respectively.