Research On Nuclear Radiation Contaminated Image Enhancement Based On Total Variation And Sparsity Representation

Nuclear energy has been extensively used because of its low environmental pollution,high economic efficiency,and safe reliability.The machine vision is an effective method to monitor the scene information of nuclear energy operation and to assist control and diagnosis.In the nuclear-radiation environment,the high-energy particles hit the arrays of CCD and CMOS video sensor composed of semiconductor materials.It results in the random pulse plaques in the obtained image,which makes radiation scene images blurred and degraded.Total variation and sparse representation can effectively utilize the prior knowledge of radiation interference image,eliminate the plaques and recover the real information of the nuclear-radiation scene image.In the paper,characteristics of radiation contaminated image were analyzed,corresponding total variational and sparse representation models were utilized to enhance the radiation interference scene image,and effective optimization solutions were deduced.The research content includes the following aspects.1.The paper studied the mechanism of generating radiation plaque noises.The statistical models of the number of nuclear-radiation plaques and the interval time of adjacent patches were established by statistical analysis of the series system of nuclear radiation generator,transmitter,and receiver.The intensity and the size of plaque noise were analyzed based on the electric effect of particles of the sensor.In order to ensure the quantitative evaluation of image enhancement algorithms,a method of generating pseudo radiation plaque noise was proposed based on a pseudo-random matrix and random walk according to its statistical property.2.Utilizing the nonlinear characteristic of the total variation that was suitable for detecting radiation plaques,a new method of radiation-plaque detection method based on variational frequency spectrum was proposed.Correlations between the variational spectrum and the variation of image plaque shape and gray were analyzed.The total variational spectrum was sensitive to the variation of image size and gray and coincidentally changed with the rotation and shaping of image.Using adaptive threshold shrinkage of medium variational frequency,the radiation plaques can be detected.3.Considering the pulse characteristic of radiation-plaque noise,the maximum posterior probability was used and constructed a variational model composed of fidelity term and regularization term for removing radiation-plaque noises.In order to better retain the details of the image,the l₁-l₂ norm was used to restrict the fidelity term,and the value of the weight parameter was adjusted by the intensity of noise.Hence the constructed TV-l₁-l₂ model was appropriate for removing radiation-plaque noise.The ADMM method was used to optimize this model,and its performance at different noise levels was analyzed.The experimental rsults prove that the TV-l₁-l₂ model had better performance than other similar variational models in subjective and objective evaluations.4.To overcoming the defect that the nonlocal variational model did not consider the boundary region of plaque noises and caused a discontinuity in the boundary region of the radiation-noise patch.The robin boundary was used to describe the boundary region of radiation-plaque noises,then an adaptive fidelity nonlocal total variation model for repairing radiation plaques which keeps continuity in the boundary area was established,and the ADMM iteration for optimizing the model was given.The results showed that the adaptive fidelity nonlocal total variation model had better performance than that of the local-based variational model,and improved the performance of NLTV.5.In order to further improve the visual quality of the radiation scene image,the group sparse was used to represent the orthogonal transformation coefficient of similar patches in adjacent frames,and a new nonlocal group sparse optimal filter was proposed.This method used a more accurate similar-block search algorithm for researching similar patched,and then the Bayesian framework used to estimate the optimal filter parameters.The results showed that the nonlocal group sparse optimal filter improved the visual image quality of the radiation scene image.Moreover,the image-enhancement performance was better than that of some related nonlocal-based enhancement algorithms.6.Considering the components of radiation-interfered video have different functional features in the time domain and the space domain,a video enhancement model based on variational and group sparse technology was proposed.The smoothness,sparsity,and continuity of the two components of radiation-interfered video in the space domain and the time domain were analyzed,and different functional space was used to describe the plaque-noise image and the original image.The ADMM process for optimizing the model was deduced.In the compared experiment,the variational and group sparse representation model could eliminate the noise and separate the real scene image from radiation contaminated video,and outperforms VRF3D and FDR models both in terms of evaluation indices and image visual quality.