Study on nonlinear filters in detection

摘要

There are a kind of non-linear fillers in the active detection problem solution which can weaken the bigger and strengthen the smaller samples, that can be called as Gaussianization, and to improve performance of subsequent correlation test. An explicit definition about this kind of nonlinear Gaussianization filter is given at first. In succession, the two typical nonlinear filters are proposed and studied. One is U-filter, based on the probability density function and its derivate. The other is G-filter, based on the cumulative distribution function and its inverse. Instances with lake trial data are illustrated to test these two methods' performance.