Due to the influence of the sky background, the background light of the wavelet spot image of the Shake-Hartmann wavefront sensor is enhanced, resulting in the location of the centroid of the wavelet spot cannot be accurately extracted, and the detection accuracy of the wavefront sensor is reduced. Due to the influence of system assembly error and lens diffraction limitation, the system will produce a large vignetting, resulting in an uneven distribution of skylight background noise in the focal plane of the sub-aperture. Once the background light distribution is not uniform, the conventional denoising method can't accurately remove the noise.For the adaptive optical system under non-uniform background, a noise removal method which can adapt to various background light distribution characteristics is needed. By using this method, the background noise can be accurately removed, and the accuracy of spot centroid extraction and wave front recovery can be improved.The paper suggests utilizing Bezier surfaces to fit the background noise, and the fitting result is subtracted as the true noise value to achieve the purpose of separating noise. When working during the day or in an environment with strong background light interference, affected by background noise, the wavefront detection error of the adaptive optical system increases and the correction ability decreases, limiting its work efficiency. Under a strong skylight background, the interference to the detector is mainly additive interference caused by strong background light. According to the law of radiation transfer, the background light intensity of the skylight is related to the off-axis angle of the solar beam relative to the optical axis and the scattering angle after passing through the optical element. After sunlight enters the optical system, it passes through multi-layer optical path turns and is affected by system assembly errors and lens diffraction effects, resulting in uneven distribution of background light detected by the CCD detector. The Bezier surface can fit the unevenly distributed background noise well. This method avoids the selection of fitting basis functions, and selecting control points at certain intervals can effectively reduce the weight of the influence of target point information on surface fitting. Taking the diffraction limit of the light spot as the interval, select control points in two mutually perpendicular directions in the image to construct a Bezier surface. The obtained Bezier surface is used as the true value of the skylight background light intensity distribution, and this surface function is subtracted from the spot array image detected by the CCD to achieve the purpose of separating the skylight background. Then centroid extraction and wavefront recovery are carried out.This method can effectively remove strong background light interference in the spot array image and adapt to unevenly distributed background light intensity changes. It has the advantages of simple implementation, strong adaptability and good effect. By comparing the centroid positioning accuracy and wavefront recovery residual RMS of multiple sub-spot images under different background light environments, it is confirmed that the method proposed in this article can effectively eliminate background interference in the spot array image, and improved accuracy of wavefront restoration. Experiments further confirmed that the centroid calculation accuracy of this method is greatly improved compared with other methods. The wavefront restoration residual RMS obtained by the method proposed in this article is improved by 33% compared with the traditional method.Through simulation and experiments, it is proved that the method proposed in this article can effectively separate the background light intensity from the target signal. This method avoids the selection of fitting basis functions, is more sensitive to changes in the background light intensity distribution that needs to be separated, has stronger adaptability to background light intensity non-uniformity, and is more conducive to the integrated application of the algorithm in actual systems. Experiments have shown that this method can effectively restore the wavefront in a strong background light interference environment with a back signal ratio ranging from 30 to 120, and the recovery effect is further improved compared with traditional methods.