Abstract: Compressive Sensing (CS) imaging requires signal to satisfy the sparsity condition, multiple underground targets will reduce the sparsity of signal in spatial domain and eventually caused the imaging defocus and less effective. Expanding the detect area may ensure the sparsity but makes the calculation amount increase, and leads to real-time disadvantage. A CS imaging method is proposed by previous extracting potential target position according to GPR detect wave feature. By filtering the noise data, sliding matrix to determine the level of the target location, and then to extreme value search for the several A-Scan data of the assured location, which can extract the target location information of imaging area, so that the imaging redundant dictionary can be created by only considering the elements where the potential targets locate, and the rest of elements for no target position can be eliminated directly. It reduces the elements which the redundant matrix building required, and consequently lower the calculated amount of CS solving. The algorithm focuses on the potential target area instead of imaging for the whole area, therefore the precision and real-time performance of imaging is both guaranteed. The experimental results show that the algorithm is feasible and effective.