In order to detect small star point targets in star images, an adaptive clutter suppression algorithm based on kernel Rayleigh quotient quadratic correlation filter was proposed. The star image simulation method was adopted to generate optical axis point randomly, produce training samples according to the twodimensional Gaussian model,  extract improved speed-up robust features(SURF), and learn to build kernel Rayleigh quotient quadratic correlation filter(KRQQCF). In order to detect the target quickly, for the image to be detected, the spectral residual method was used to detect salient regions probably containing targets. Then the improved 5-dimension SURF feature of the salient regions was extracted. Finally, the target was recognized using KRQQCF, and  the clutter and noise were suppressed effectively which improved the SNR. Experimental results indicate that the proposed algorithm is fast, effective and robust.