Research On Three-dimensional Reconstruction Of Microstructure Of Rice Seeding Stalks Based On Micro-CT Image

Rice seedling stalks perform a variety of functions such as supporting the grain,transporting water and nutrient salts,storing nutrients and water,adapting to growth,self-healing,and photosynthesis.The multi-scale structural characteristics of rice seedling stalks are an important basis for studying seedling growth characteristics,stressing tolerance,and biomechanical properties.An accurate 3D microphysical model of rice seedling stalk is the basis for in-depth research.As rice seedling stalks are biocomposites with high water content,porous hierarchical structural characteristics,and non-homogeneous anisotropic material properties,previous 3D modeling methods suffer from attrition,non-reproducibility,and low accuracy.In recent years,microscopic modeling based on Micro-Computed Tomography(Micro-CT)has developed rapidly,which can achieve in situ,non-destructive,accurate scanning,and visualization of the internal tissue microstructure of plants and animals without destroying the samples,but the cost of using this technology is large in terms of money and time,and there is an urgent need to find a method to generate 3D microscopic models of the structure of rice seedling stems from Micro-CT images quickly and in large quantities.Therefore,the study of Micro-CT-based microscopic 3D modeling of rice seedlings is of great academic research value and practical significance.This paper addresses the issue of quality and three-dimensional modeling of Micro-CT images of rice seedling stalk cross-sections,using seedling stalks at 38,40,42,and 44 days in the Chuang-Two-You rice as experimental subjects,and has carried out two main studies:(1)When acquiring Micro-CT images of rice seedling stalk cross-sections,to obtain microscopic images of rice seedling stalk cross-sections with high resolution and obvious contrast,an appropriate amount of the contrast agent cesium iodide was added to the rice seedling stalk culture fluid and a low dose of X-rays was selected for Micro-CT scanning,resulting in a large amount of noise and artifacts in the acquired Micro-CT images,which the accuracy of microscopic 3D modeling of rice seedlings by Micro-CT was affected.To address the noise and artifact problems,this paper proposes a Micro-CT image denoising algorithm based on adaptive group sparse residuals for rice seedling stalk cross-sections.First,non-local uniform filtering is used to initially remove the noise from the original Micro-CT images.Then,the original Micro-CT image was divided into multiple image blocks by sliding window,and the target image was determined by combining the structural similarity between the image blocks,and the-Nearest Neighbors(KNN)algorithm was used to adaptively search for similar blocks in the target image to obtain the self-similar block matrix.Finally,the image denoising algorithm based on group sparse residuals and the Lagrange multiplier optimization algorithm were used to find the optimal sparse coefficients.As a result,the accuracy of the similar block group is improved,the detailed features of the image are retained as far as possible,and the noise in the Micro-CT images of rice seedling stalk sections is effectively removed.(2)An improved GAN-based microscopic 3D model of rice seedling stalks is proposed to address the problem of poor robustness of existing Generative Adversarial Networks(GAN)for microscopic 3D modeling of rice seedling stalks in Micro-CT images.Firstly,the Deep Convolutional Generative Adversarial Networks(DCGAN)model of the improved GAN is used to build the GAN models required for the text network model.Secondly,the DCGAN model is improved by removing the last layer of the sigmoid activation function of the original discriminator and replacing the momentum-based optimization algorithm Adam with RMSProp to obtain the network model in this paper.Finally,the loss function of the generative adversarial network model is improved by introducing the Wasserstein distance,which is used to measure the difference between the Micro-CT image of rice seedling stalk structure generated by the generator and the real image,to obtain the loss function of the network model in this paper.This solves the problem that the better the discriminator is,the more serious the generator gradient disappears when GAN models the microscopic 3D of rice seedling stalks in Micro-CT images;the better the discriminator is,the less diverse the generated model is;the training is difficult,and the loss of the generator and discriminator cannot indicate the training process,etc.It improves the accuracy of the 3D estimated model and generates a 3D model of the microscopic structure of rice seedling stalks based on Micro-CT images.