The invention belongs to the technical field of radar, and discloses a multiple-input multiple-output array radar staring imaging method based on tensor compression perception. The method includes the steps of configuring a transmit-receive array, setting each emission array element in the emission array to emit random frequency hopping signals, setting the referential array element position coordinate of the array, the transmit-receive array and the coordinates of a target scene, establishing a signal reception model of a radar reception array in a same range gate, matching and filtering the reception data, converting the problem of restoring the scattering coefficient of a single-range gate into solving sparse optimization, reconstructing the target scene in the single-range gate for the sparse optimization of signal reception model by using an improved 2D-SL0 high-efficient rapid imaging method, reconstructing the scattering coefficient matrix in each range gate by using the rapid image reconstruction algorithm, and finally obtaining the multi-range gate target scene image. The algorithm complexity is reduced, the image reconstruction time is shortened, and rapid imaging of non-sparse large scene object is realized.