Research On Image Denoising In Nuclear Radiation Scenes

When a nuclear accident occurs,due to the excessive radiation dose in the nuclear environment,humans cannot enter the site for follow-up operations.As one of the few machines that can enter the nuclear accident site,nuclear emergency operation robots play an important role in solving the accident.Some nuclear emergency operation robots are equipped with an image acquisition module,and the collected image data is an important basis for the robot to perceive the internal environment.However,the high-energy particles in the nuclear environment will have radiation effects with the semiconductor materials in the image acquisition module,resulting in a large number of nuclear noise bright spots in the acquired images.In this paper,corresponding noise reduction algorithms are proposed for grayscale nuclear noise and color nuclear noise,respectively.The specific research contents are as follows:1.From the perspective of the formation of nuclear noise,the shape,area and other characteristics of nuclear noise are summarized.A random walk algorithm is designed to simulate the grayscale kernel noise,and the simulated grayscale kernel noise dataset is obtained.Then,the relationship between color nuclear noise and additive noise is verified through experiments,and the color nuclear noise image pair is obtained by using multi-frame stacking solution.2.For the grayscale images disturbed by nuclear radiation,a second-order hybrid total variational noise reduction algorithm is proposed.The algorithm greatly suppresses the complex noise in the nuclear environment image while preserving the edge texture information of the image.Experiments are carried out on the real grayscale nuclear noise data set and the simulated grayscale nuclear noise data set.Compared with the contrastive variational algorithm,the noise reduction algorithm achieves the highest peak signal-to-noise ratio(PSNR)and structural similarity(SSIM).,the maximum value can reach 31.61 d B and0.889,which are 2.62 d B and 0.084 higher than the comparison variational algorithm.At the same time,the algorithm can better preserve the details of the image in the subjective visual effect.3.For color images disturbed by nuclear radiation,a denoising network based on deep learning is proposed.This network adopts parallel design,and the main modules are up and down sampling module,multi-scale feature extraction module,residual feature extraction module,and spatial attention module.The network performance is adjusted and improved by using the Mish activation function and the mixed loss function.The experimental results show that the noise reduction network is better than the contrast noise reduction network in processing time,and the peak signal-to-noise ratio(PSNR)and structural similarity(SSIM)can reach 34.59 d B and 0.941 at most,which are better than the comparison algorithm by2.65 d B and 0.023..In addition,excellent results are also achieved in the experiment of removing synthetic mixed noise.The peak signal-to-noise ratio(PSNR)and structural similarity(SSIM)can reach up to 32.97 d B and 0.943,which are 1.42 d B and 0.021 higher than the comparison algorithm.It is proved that the noise reduction network also has strong adaptability.