Abstract: Ground based synthetic aperture radar (GBSAR) has the advantages of high deformation measurement accuracy, large area non-contact monitoring and all-weather monitoring. It is one of the main technical means of submillimeter deformation monitoring for working slope and dump of open pit mine. The differential phase change caused by multipath effect in the process of slope deformation monitoring will be wrongly identified as deformation, and the accuracy of deformation results is an important basis for triggering the early warning process. Aiming at the problem of low accuracy of deformation recognition, this paper studies the expression method of differential interferometric phase sequence features, and proposes a time-series differential phase classification method for ground-based SAR Based on attention network model. Based on the phase change trend and regional range, the abrupt change region and the gradual change region are distinguished, and the real deformation distribution is predicted by the model. The experimental results show that the attention network model can accurately extract the deformation distribution, and effectively reduce the error interference caused by multipath effect.