In order to solve the influence of mixed Poisson- Gaussian noise (MPGN) on power grid inspection image, an image denoising algorithm based on generalized Anscombe trans-form (GAT) is proposed to improve the details of the image. Since the actual image noise distribution is unknown, the Gaussian noise variance is a necessary parameter of the GAT. Accurate estimation of Gaussian noise variance becomes one of the keys to improve denoising performance. Therefore, a fast Gaussian noise variance estimation method is adopted. After performing the GAT, we adopt the Stein's unbiased risk estimator-linear expansion of threshold (SURE-LET) method for one-pass image denoising considering the computational cost and time cost of practical applications. Finally, inverse GAT is used to obtain the denoised image. Experimental results show that the proposed method achieves higher PSNRs and SSIMS, and can significantly improve the visual effect of power grid patrol image.