Abstract: When detecting in high resolution and largegrazingangle situation, radar has a larger equivalent backscattering area. Thus, most of the sea clutter energy is projected to a few of distance units, which leads to inhomogeneity of the energy distribution. Therefore, the “abnormal unit” with a sudden increase in power appears, which makes the background environment of the reference window of a detector complex and variable. Consequently, the detection probability is reduced and the false alarm rate is increased. In order to solve this problem, positive definite matrices are constructed by referring to the covariance matrix of the sliding window and its matrix norm is solved to estimate the clutter power intensity. The support vector machine is used to improve the traditional constant falsealarm rate (CFAR) detector. An improved CFAR detector based on support vector machine which is trained by the reference widow power estimated by positive definite matrices is obtained. The experimental results show that the new detector has stable detection performance in the uniform clutter and multitarget environment, and the performance of false alarm control at the clutter edge is good.

Key words: largegrazingangle, constant falsealarm rate (CFAR) detector, covariance matrix, support vector machine （SVM）