Due to the complicated structure of microwave radiometric imaging system and the massive amount of data collection in one snapshot, it is difficult to achieve the high spatial resolu- tion image by conventional microwave radiation imaging method based on the Nyquist sampling. In this paper, according to the priori information of the compressible multi-structural informa- tion of microwave radiation image, we adopt the Fourier random observation matrix to sparsely project the microwave radiation image, reducing the amount of data collection, and lowering the complexity of the system. Considering the multi-structural information of microwave radiation image, such as multi-sparsity in different domains, piecewise smoothness, etc, it is difficult to sparsely represent microwave radiation image in complex scene by the traditional orthogonal ba- sis, but the piecewise smoothness ingredients of microwave radiation image meet the constraint condition of total variation, and the mixed orthogonal basis can sparsely represent the sparsity information of microwave radiation image. We make use of the sparse representation of the mixed orthogonal basis and the constraint condition of total variation regularization to construct the reconstruction model of microwave radiation image based on adaptive multi-structural spar- sifying dictionary learning. We integrate dictionary learning technique to adaptively learn the multi-structural sparsifying dictionary, and the sparsifying dictionary is adapted to sparsely rep- resent the microwave radiation image, and solve the convex programming problem of microwave radiation image reconstruction, and design the reconstruction method. The simulation results show that reconstructing microwave radiation image by the proposed algorithm achieves better reconstruction performance than that by the DLMRI algorithm.