The invention provides a method for optimizing projection modes in image reconstruction based on compression sensing. In the process of reconstructing compressed sensing images, the traditional projection mode usually projects line signals of the image by lines, which resulting in that the row information of the image is wasted, and the sparseness of the image cannot be fully utilized; or, the whole image is pulled into one line signal to be projected, but the projection efficiency is decreased, the complexity in calculation is increased, and the image reconstruction time is prolonged. The method has the advantages that the projection mode is optimized on the basis of traditional single-side line projection, a row and line double-side projection mode with super-low sampling rate is proposed; the sparseness of the global image is fully utilized, the projection efficiency is improved, and the reconstruction accuracy and quality of the image at low sampling rate are improved; especially,under the condition of super-low sampling rate of 1 / 16, compared with the single-side projection mode, the PSNR (peak signal-to-noise ratio) of the reconstructed image of the double-side projection mode is increased by about 4.5dB, and the SSIM (structure similarity) is increased by about 0.1.