Research On Artifact Reduction And Porosity Calculation Of Rock And Mine CT Image Based On Deep Learning

Computed Tomography(CT)is a non-destructive testing technology that can present the internal fault structure information of objects in two-dimensional and three-dimensional ways.It has been applied in the geological field in recent years.Rock and mineral geological samples are the carrier of mineral resources.Based on CT technology,the pore structure of rock and mineral geological samples can be accurately described,which has important reference value for studying the system structure of metal logenic belts.However,due to the presence of high-density metallic substances in rock and mineral samples,large dark areas or black and white radial streaks(metal artifacts)appear in reconstructed CT images.The existence of metal artifacts will seriously reduce the quality of CT images and affect the calculation of relevant parameters of rock and mineral samples(such as porosity).Therefore,it is undoubtedly of great theoretical significance and practical value to study how to suppress metal artifacts in CT images.The research work of this thesis is based on the industrial CT detection project of geological samples provided by the Chinese Academy of Geological Sciences(subject number: 2019YFC0605203),starting from rock and mineral samples,using industrial CT technology,mainly focusing on rock and mineral samples metal artifact reduction and There are two aspects of porosity calculation of rock and mineral samples.Based on the deep learning method,a metal artifact reduction network that combines projection domain enhancement and image domain enhancement is studied,and the metal artifact reduction processing and analyzed,and its porosity was calculated.The main research contents of this thesis include:(1)Aiming at the key technical problems such as the introduction of new artifacts when the traditional Metal Artifact Reduction(MAR)algorithm performs artifact suppression,this thesis proposes a metal artifact reduction algorithm for CT images based on a dual-domain adaptive network.(DDA-CNN-MAR).The network is based on the Residual Encoder-Decoder Network(RED-CNN)network and consists of four parts:the projection domain enhanced image,the projection layer,the image domain enhancement network and the back projection layer.The "projection layer" represents the projection operator,which maps the input image to the projection domain by means of projection,and obtains the sinogram of the image;the "back projection layer" represents the back projection operator,which maps the sinogram of the projection domain through the inverse The projection method is mapped to the image domain to obtain the reconstructed image;the projection domain enhancement network and the image domain enhancement network are both RED-CNN network structures,the projection domain enhancement network input is the sinusoidal domain projection;the image domain enhancement network input is the image domain the image to be enhanced.LI-NMAR is an interpolation reduction module,which can perform interpolation reduction on the input image in the sine domain.The input of the model includes an image with metal artifacts,which becomes a sinogram after passing through the projection layer.After passing through LI-NMAR,the projection domain enhancement network and the image domain enhancement network are first and then input,and finally the corrected CT image is obtained.The generated images were quantitatively analyzed with the help of three evaluation indicators: SSIM,PSNR,and MSE.The experimental results show that compared with RED-CNN-MAR,the SSIM value and PSNR value of the algorithm proposed in this thesis are increased by 1.97%and 3.61% respectively,and the MSE value is decreased by 43.16%,and the visual effect is also the best.(2)Based on the corrected CT images,the research on porosity calculation of rock and mineral samples was carried out.Firstly,ROI edge detection was performed on CT images of rock and mineral samples,and the edge detection of CT images was quantitatively evaluated by introducing parameters such as peak signal-to-noise ratio,number of main edge chains,ROI entropy,and better edge detection results were obtained.Furthermore,according to the characteristics of the large difference between solid matter and pore composition in CT images of rock and mineral samples,an improved maximum inter-class variance threshold(OTSU)segmentation method is proposed in this thesis.Compared with the traditional OTSU threshold segmentation method,the algorithm proposed in this thesis can prevent the threshold distortion problem in the traditional OTSU algorithm and realize the effective segmentation of rock and mineral samples.(3)A set of rock and mineral sample porosity calculation and analysis software based on micro-focus CT images was designed,which realized the calculation of rock and mineral sample porosity and other parameters.