Research On X-ray Absorption Spectrum Correction Based On AAE Algorithm

Compared with the traditional energy integral detector,the photon counting detector can eliminate the dark current noise more effectively,obtain higher quality images,and have stronger imaging ability of material separation.Therefore,it has attracted more and more attention and favor of researchers in the industry.However,photon counting detectors also have some inherent defects,including charge sharing effect,beam hardening effect and pulse stacking effect,which eventually lead to spectral distortion and interference to quantitative detection and analysis.In this paper,we propose a correction method of X-ray absorption spectrum data based on Adversarial Auto-encoders(AAE),which directly maps experimental data to theoretical data in supervised learning mode.Firstly,the experimental spectral data were collected by the X-ray photon counting detector CT system developed by the laboratory,and then the theoretical spectral data were obtained by the special software SPEKTR.The theoretical X-ray absorption spectrum(XAS)data and the laboratory measured spectral data are sent to an autoencoder generation antagonist network for correction of the X-ray absorption spectrum data.Finally,the trained network was used to correct the experimentally measured XAS data.In this paper,the application value of spectral correction is also demonstrated by using the images reconstructed by CT.The filtering back projection algorithm(FBP)is adopted in the CT reconstruction,which is fast in reconstruction and has good robustness.The measured XAS data and the corrected XAS data were used for CT image reconstruction,respectively.The reconstruction quality of the above two CT slice images was compared by five non-reference image evaluation functions and gray variance index of uniform sample reconstruction,so as to verify whether the spectral correction effect can improve the reconstruction image quality.The experimental results show that the distorted spectral data can be corrected to the ideal spectrum by this method,and the mean square error value of the corrected spectral data and the simulated spectral data is 0.000013.The variance of the internal gray scale of the CT reconstructed slice image of the uniform sample before correction is 30.027,and the variance after correction is 16.638.The gray variance of the surrounding air before correction is 134.712,and the variance after correction is 132.507.It shows that spectrum correction can effectively improve the imaging effect of CT.