The heterogeneous or complex clutter in real environment is usually encountered by detection of weak target. It may be difficult for track-before-detect (TBD) based on modeling of clutter to detect weak target, when the signal-to-clutter ratio (SCR) is low, or prior information and clutter distribution is difficult to obtain. A new detection algorithm is proposed for real data of unknown strong clutter. The improved greatestof constant false alarm rate (CFAR) is used to preprocess the clutter according to the features of unknown distribution, clutter edge, etc. Then twodimensional data of Doppler-distance is compressed to avoid accumulation of clutter in the same unit after association of multiframe. Finally, the dynamic programmingbased TBD algorithm is improved in the time-distance space to avoid expansion of clutter in different range units and to eliminate false alarms after jointed multiframe. Experimental results show that the proposed algorithm has higher detection probability and lower estimation error than conventional TBD methods.