The performance of STAP degrades severely when the training samples are not independently and identically distributed(iid).The iid assumption is violated not only by the nonhomogeneity of the ground and the inner moving of the scatterer but also by the targets and other interferences.In some cases

the sample contaminated by a target will affect the estimation of the weight vector effectively and the output SCNR of STAP degrades evidently.In this paper

the target effect on the generalized sidelobe cancellor(GSC) is analysed.It seems that the blocking matrix and the secondary data training can not avoid the effect effectively.And the existing nonhomogeneity detectors can not select the contaminated sample exactly also.So a new NHD criterion and two new NHD methods are proposed.The computer simulation results show good performance can be obtained.Moreover

one of them is computationally easy and suits applications.