Researched On γ Radiation Image Denoising Method Based On Spatiotemporal Correlation Features

Tasks such as status monitoring and maintenance inspection of irradiation facilities in nuclear energy development and nuclear technology applications have put forward urgent needs for clear visualization of γ radiation scenes.However,in the γ radiation environment,the CMOS image sensor(CIS)is directly or indirectly affected by high-energy γ photons,and there are additive speckle noises such as Compton noise,airglow noise and induced noise caused by space charge change,as well as multiplicative noise such as darkness,insufficient contrast,and color cast caused by response degradation of CIS.Aiming at the multiple interference of addtive noise and multiplicative noise in γ radiation scene images,under the guiduance of noise formation mechanism,a γ radiation image denoising method based on spatiotemporal correlation features is proposed,which extracts relible information from temporal adjacent frames and within image achieve clear visualization of γ radiation scene images.The main research contents of the project are shown as follow.(1)Research on the construction of composite noise model of radiation images.In this paper,the relationship between the light and electromagnetic waves,the photoelectric effects and the mathematical models of photoelectric imaging are firstly analyzed.Then,the composite noise model of γ radiation image is constructed according to the mode of action of high-energy γ photons and matter,the interference process and effects of γ photons act on CIS,and the formation mechanism of γ radiation additive noise and multiplicative noise,and the spatiotemporal characteristics of γ radiation noise are also analyzed.Finally,the datasets of this paper are analyzed,including γ radiation noises simulation methods,the simulated γradiation dataset and the captured real γ radiation dataset.(2)Temporal correlation method of γ radiation images in dynamic scenes.In this paper,the temporal association accuracy of γ radiation images is improved from suppressing noise feature expression and enhancing the saliency of scene information,respectively.Firstly,spcekle splitting and pre-denoising methods are used to suppress the expression of noise features.Secondly,the ability of features expression of scene information is improved by extracting and enhancing the structural features.Then,the Fourier-Merlin method is used to obtain the translation,rotation and scale transformation parameters between images.Finally,the transformation parameters between images are optimized by using the strategies of random sampling,group association and consistency constraint.With the help of the proposed correlation method,this paper achieved the alighment accuracy that the scale error is less than0.01,the rotation error is less than 0.4。,and the translation error is less than 0.6 pixels,resulting in subpixel-level image alignment on γ radiation images.(3)Additive noise removal method based on correlated shallow temporal features and spatial features.Based on the correlated temporal images,this paper proposed an additive noise removal method based on the spatiotemporal correlation characteristics of γ radiation noise.In the non-spatiotemporal correlation pixels,the noisy pixels are repaired by by crossneighborhood filtering and non-local means filtering,while in the spatiotemporal correlation pixels,the spatiotemporal outlier method further combined to achieve the clear visualization of the γ radiation image.With the help of proposed additive noise removal method,the quality ofγradiation images had been obviously improved.Concretely,the peak signal noise ratio are boosted by 6.2d B-8.17 d B and the structure similarity are increased by 0.32-0.47.(4)Multiplicative noise removal method based on multivariate nonlinear mapping.In view of the multiplicative noise removal requirements,this paper firstly improves the overall brightness of scene images through logarithmic mapping.Secondly,a two-dimensional Gamma function is proposed to adaptively stretch image contrast.Finally,the color distortion of γradiation scene image is corrected by color level remapping and the proposed BayerGuided AWB method.Experimental results show that the proposed method can increase the brightness information entropy by 0.45-0.9,the contrast by 0.012-0.027,and the color cast by0.35-1.86,which further enhances the clarity effects of γ radiometric images.Based on the effective analysis of research contents,this subject has conducted comparison experiments with a variety of typical γ-radiation image denoising methods in recent years on real γ radiation images and simulated noise images of different scenes and different radiation areas,respectively.The effects of speckle noise removal are quantitatively evaluated by peak signal-to-noise ratio and structural similarity,and the global distortion elimination results are quantitatively evaluated by quantitative metrics of brightness,contrast and color cast.The experimental results show that the proposed method achieves the excellent results in both visual characterization and metrics quantification,which proves the effectiveness of the γ radiation image denoising method based on spatiotemporal correlation features.