There are still many false alarms in two-value image after CFAR detection in ultra-wideband through-wall radar. These false alarms can be divided into three categories, namely the isolated noise, target side lobe and random strong clutter false alarms. In order to suppress them, a new target recognition algorithm was presented in this paper; it was made up of binary integration, the hierarchical clustering processing and target tracking. The processing results of measured data show that the algorithm can effectively suppress above false alarms, and achieve accurate information of target number and location.