To better preserve the detailed information of nuclear environment image after noise reduction, we propose a noise reduction method for anti-nuclear radiation image based on hybrid second-order total variation. The method combines non-convex second-order total variation with overlapping group sparse regularization, where non-convex second-order total variation denoises the image, and overlapping group sparse regularization is used to remove artifacts caused by total variation. alternating direction method of multiplier (ADMM) and the augmented Lagrange multiplier method are used to optimize and solve the global problem, and the basic denoised image is obtained after several iterations. Finally, the basic denoised image after multiple denoising is subjected to difference iteration, so that the gray value of the large-scale jump in the nuclear radiation image is closer to the gray value of the original image. Before the experimental verification, according to the characteristics of nuclear noise, an algorithm is designed to simulate the nuclear noise patches. Through experiments on datasets collected in a real nuclear environment and simulated nuclear noise datasets, indicators such as peak signal-to-noise ratio (PSNR) and structural similarity (SSIM) and the visual effect after processing shows that the algorithm is better than the contrast algorithm in preserving the details of the image.