For infrared detection target signal uncertainty problem, a robust matched filter algorithm was presented. Based on Parceval theorem,the uncertain spectral model of infrared target signal was established by qualitative analyzing the target signal spectrum distortion source, and signal spectrum distortion associated factors were analyzed. The saddle-point solution for Signal-to-Noise of robust matched filtering was constructed using the minimax method, and the most negative signal spectrum was obtained with Genetic Algorithm of minimax bilevel optimization. Then transfer function of the robust matched filter was designed based on the classical matched filter. The simulation results show that Signal-to-Noise of robust matched filtering under low background noise can significantly eliminate the Signal-to-Noise over-sensitivity to the abnormal change of infrared input signal spectrum ,and spectral Signal-to-Noise compensatio in signal spectral is the highest to nenergy distortion, followed by correlation peak distortion, spectral related distortion to at least.