Abstract: To address the problems of target masking, threshold elevation, and uncontrolled edge false alarms encountered by conventional constant false alarm rate (CFAR) detectors in ground surveillance radar under intertwined environments of non-Gaussian heavy-tailed clutter and dense multi-target interference. This paper proposes a robust CFAR detection technology using MAD-based asymmetric trimming. The proposed method employs the median and MAD as robust statistical references for local background，introducing an upper trimming factor α and a lower trimming factor β to construct a dynamic trimming mechanism with independently adjustable upper and lower bounds. Specifically, upper-bound trimming is used to suppress high-intensity outliers while preserving the natural long-tail samples of heavy-tailed clutter as much as possible, whereas lower-bound trimming is introduced to mitigate background underestimation caused by low-power trailing samples when the sliding window crosses clutter edges. To address the difficulty of analytically deriving the detection threshold after nonlinear trimming, a Monte-Carlo statistical method is adopted to obtain the offline threshold factor, which is then combined with a lookup table for fast online invocation. Simulation results show that the proposed algorithm introduces only a small CFAR loss in homogeneous heavy-tailed backgrounds, significantly suppresses target masking under multi-target interference, and exhibits superior false-alarm recovery capability and statistical stability in clutter-edge scenarios, and indicates its feasibility and potential application value for engineering implementation.