In many practical applications, signals to be detected are unknown nonlinear frequency modulated (FM) and are corrupted by strong noise. The phase histories of the nonlinear FM signals are assumed to be unknown smooth functions of time, which are usually poorly modeled or cannot be modeled at all by a small number of parameters. Because of the lack of phase model, a nonparametric detection method is proposed based on successive fractional Fourier transform and double-characters detection. The detection process goes in three steps. First, an image is constructed by the fractional Fourier transforms of successive angles in one period. Then, the threshold procedure is utilized to transform the image into a binary image. After the multiple median filtering, the binary image is refined where the isolated noise pixels are removed. Finally, two complementary features are extracted from the refined image, and a double-characters detector is proposed to decide whether the target is present or not. The simulation experiments to three polynomial phase signals with different orders and a sinusoidal phase signal show that the proposed detection method is effective and robust.