Spectral images contain abundant space information and spectral information, which can provide important information support for space-based early warning detection. However, the huge amounts of data also brings great challenge for hardware. The traditional treatment of first sampling and then compressing based on Nyquist sampling not only can’t solve the problem of mass-data fundamentally, but also causes wasting of sources. To solve this problem, we propose a spectral imaging and reconstruction method based on spatial compressive sampling and spectral Karhunen-Loève (KL) transform by using the sparsity of single-band images and the spectral redundant of spatial encoded data. A two-dimensional composite regular reconstruction model based on 1 -norm and total variation is constructed for single band images, and an inference algorithm named two-dimensional compound regularized projection gradient (2D-CRPG) is then proposed for the model by combining the projection gradient method with the soft-threshold operator. The results show that the spectral imaging and reconstruction method based on spatial compressive sampling and KL transform can effectively reduce the cost of data sampling, and thus can benefit the spectral imaging of space-based early warning detection. The 2D-CRPG reconstruction algorithm can effectively preserve structural information of spectral images, thus the original spectral image can be reconstructed at a limited sampling rate.