A Digital Materialization Method and System for Molded Fossils

Abstract

The present invention proposes a digital materialization method and system for cast fossils, belonging to the technical field of CT imaging. The method includes: obtaining a CT image of the cast fossil; generating a three-dimensional model of the cast fossil based on the CT image of the cast fossil; extracting local features of the three-dimensional model of the cast fossil, and performing feature combination on the local features, and classifying the cast fossil according to the overall features after combination; wherein, the local features of the three-dimensional model of the cast fossil include: nodule structure features, radiating ridge structure features, and medullary cavity pore structure features; generating a digitally materialized cast fossil based on the three-dimensional model of the cast fossil and the classification result of the cast fossil. The present invention can efficiently and objectively perform CT scanning, image denoising, semantic segmentation, three-dimensional modeling, fossil classification, and 3D printing materialization on cast fossils.

Description

Technical Field

[0001] The present invention relates to the technical field of CT imaging, and particularly to a digital materialization method and system for cast fossils. Background Art

[0002] Paleontology is a discipline that studies the history of life on Earth based on fossils. Fossils refer to the remains, relics, or traces of organisms in geological history that are preserved in rocks. Fossils can be classified into four major categories according to their preservation characteristics: body fossils, cast fossils, trace fossils, and chemical fossils. Body fossils are formed by the preservation of all or part (especially the hard parts) of the ancient organism's body itself. Cast fossils are the imprints and recasts left by ancient organism remains in rock layers or surrounding rocks. Trace fossils refer to the activity traces and relics left by ancient organisms during their lifetime that are preserved in rock layers. Chemical fossils refer to the amino acids, fatty acids, and other organic substances that are still preserved in rock layers under certain specific conditions after the organic components of ancient organisms have decomposed without the preservation of the ancient organism remains.

[0003] For one type of cast fossil, since its original bones have been partially or completely decomposed during burial, only the cavities preserved in the surrounding rock remain. These cavities preserve the original distribution of the cast fossils in the surrounding rock, which is important information in paleontological research. In addition, statistical analysis of the original distribution of cast fossils can uncover a large amount of rich geological information. Especially in oil and gas exploration, geological workers can provide important basis for finding and determining the potential direction of source rock layers by statistically analyzing the distribution of cast fossils in strata.

[0004] However, usually this type of cast fossil is a cavity structure preserved in the surrounding rock and cannot be repaired by traditional mechanical methods, so it is difficult to obtain the original distribution of the cast fossil in the surrounding rock. In recent years, thanks to the widespread use of micro-CT technology in non-destructive testing of fossils, paleontologists can obtain more comprehensive three-dimensional information of fossils non-destructively through micro-CT scanning, including the three-dimensional microscopic structure from the inside to the surface of the fossils. At the same time, 3D printing technology is entering all walks of life and playing an increasingly important role, making it possible for paleontologists to combine CT scanning and 3D printing technology to replicate the original distribution of cast fossils, thereby providing a reference basis for geological workers in oil and gas exploration. Summary of the Invention

[0005] In order to realize the digital materialization of this type of cast fossil with partially or completely decomposed original bones, the present invention provides a digital materialization method and system for cast fossils, which can efficiently and objectively perform CT scanning, image denoising, semantic segmentation, three-dimensional modeling, fossil classification, and 3D printing materialization on cast fossils.

[0006] The technical content of the present invention includes:

[0007] A digital materialization method for mold fossils, the method comprising:

[0008] Obtaining a CT image of the mold fossil;

[0009] Generating a three-dimensional model of the mold fossil based on the CT image of the mold fossil;

[0010] Extracting local features of the three-dimensional model of the mold fossil, combining the local features, and classifying the mold fossil according to the combined overall features; wherein, the local features of the three-dimensional model of the mold fossil include: nodule structure features, radiating ridge structure features, and medullary cavity pore structure features;

[0011] Generating a digitally materialized mold fossil based on the three-dimensional model of the mold fossil and the classification result of the mold fossil.

[0012] Further, generating the CT image of the mold fossil includes:

[0013] Cutting a rock sample according to a set size to obtain a mold fossil;

[0014] Using an X-ray detector to perform 360° step-by-step imaging of the mold fossil at different angles;

[0015] Reconstructing the images at each angle and using a filtered back-projection algorithm to obtain the CT image of the mold fossil.

[0016] Further, generating a three-dimensional model of the mold fossil based on the CT image of the mold fossil includes:

[0017] Denosing the CT image of the mold fossil to obtain a CT image with a high signal-to-noise ratio;

[0018] Performing semantic segmentation on the CT image with a high signal-to-noise ratio to obtain a semantic segmentation result of the mold fossil;

[0019] Performing isosurface extraction and surface volume rendering on the semantic segmentation result to generate a three-dimensional model of the mold fossil.

[0020] Further, the denosing the CT image of the mold fossil to obtain a CT image with a high signal-to-noise ratio includes:

[0021] Denosing the CT image of the mold fossil using a filter to obtain a CT image with a high signal-to-noise ratio; wherein, the filter includes: a median filter or a mean filter;

[0022] Or,

[0023] A CT image denoising network based on deep learning is used to denoise the CT image of the cast fossil to obtain a high signal-to-noise ratio CT image; wherein the CT image denoising network based on deep learning includes: a Noise2Noise network model or a Noise2Void network model.

[0024] Furthermore, the semantic segmentation of the high signal-to-noise ratio CT image to obtain the semantic segmentation result of the molded fossil includes:

[0025] Get the grayscale mean of the molded fossil , grayscale mean of surrounding rock and the overall grayscale value of high signal-to-noise ratio CT images ;

[0026] By calculating the optimal threshold The between-class variance , get the optimal threshold ;in, is the probability of a molded fossil, is the probability of surrounding rock;

[0027] Applying the best threshold Performing binary image segmentation on the high signal-to-noise ratio CT image to obtain a binary mold-cast fossil CT image;

[0028] After removing the external small particles and internal small gaps in the binarized cast fossil CT image, the seed points of the cast fossil are selected to merge adjacent pixels with similar attributes and combined with manual circle selection to group pixels with similar properties into semantic regions, thus obtaining the semantic segmentation result of the cast fossil.

[0029] Furthermore, the semantic segmentation results are subjected to isosurface extraction and surface volume rendering to generate a three-dimensional model of the molded fossil, including:

[0030] Obtain the three-dimensional volume data corresponding to each semantic segmentation result;

[0031] Find the voxels that intersect the isosurface in the three-dimensional volume data and obtain the voxel edges;

[0032] Applying linear interpolation to solve the intersection of the isosurface and the voxel edge;

[0033] All intersection points are connected to extract isosurfaces to generate a three-dimensional model of the molded fossil.

[0034] Furthermore, the extracting of features of the three-dimensional model of the cast fossil and classifying the cast fossil based on the features includes:

[0035] Preprocessing the three-dimensional model of the molded fossil;

[0036] The overall features of the preprocessed three-dimensional model of the mold fossil are extracted by using a VGG16 network model with transfer learning; wherein, the extraction of the features of the preprocessed three-dimensional model of the mold fossil by using the VGG16 network model with transfer learning includes:

[0037] Using the front part of the VGG16 network model, a low-level feature map of the three-dimensional model of the mold fossil is obtained, and the low-level feature map contains the edge structure information of the three-dimensional model of the mold fossil;

[0038] Using the middle part of the VGG16 network model to combine the low-level feature maps to obtain the local features of the three-dimensional model of the mold fossil;

[0039] Using the back part of the VGG16 network model to combine the local features to obtain the overall features of the three-dimensional model of the mold fossil;

[0040] The overall features from multiple perspectives are fused by using the long short-term memory architecture in the recurrent neural network model;

[0041] An SVM multi-classifier is selected to classify the fused features to obtain the classification result of the mold fossil.

[0042] Further, the preprocessing of the three-dimensional model of the mold fossil includes:

[0043] Performing pose normalization on the three-dimensional model of the mold fossil by using the PCA method;

[0044] And,

[0045] Applying an average curvature map to enhance the local geometric shape features of the three-dimensional model of the mold fossil;

[0046] And,

[0047] Using a multi-view rendering tool to collect multi-view views of the three-dimensional model of the mold fossil.

[0048] Further, based on the three-dimensional model of the mold fossil and the classification result of the mold fossil, generating a digital entity of the mold fossil further includes:

[0049] Repairing the three-dimensional model of the mold fossil, and the repair includes: reversing reverse triangular patches, stitching gaps, filling holes, deleting bad edges, crossing and overlapping triangular patches, and interfering shells;

[0050] Importing the repaired three-dimensional model of the mold fossil into 3D printing slicing software, assigning a transparent material to the surrounding rock, and assigning a color material to the mold fossil based on the classification result of the mold fossil to obtain a physical three-dimensional model of the mold fossil.

[0051] A digital materialization system for cast fossils, the system comprising:

[0052] An image acquisition module for acquiring CT images of cast fossils;

[0053] A model generation module for generating a three-dimensional model of a cast fossil based on the CT image of the cast fossil;

[0054] A fossil classification module for extracting local features of the three-dimensional model of the cast fossil, performing feature combination on the local features, and classifying the cast fossil according to the overall features after combination; wherein, the local features of the three-dimensional model of the cast fossil include: nodule structure features, radiating ridge structure features, and medullary cavity pore structure features;

[0055] A digital materialization module for generating a digitally materialized cast fossil based on the three-dimensional model of the cast fossil and the classification result of the cast fossil.

[0056] A storage medium having a computer program stored thereon, the computer program, when executed by a processor, implementing any of the above methods.

[0057] An electronic device comprising a memory and a processor, wherein the memory stores a computer program, and the processor is configured to run the computer program to execute any of the above methods.

[0058] Compared with the prior art, the present invention has the following advantages:

[0059] The present invention uses an unsupervised learning method to denoise and improve the signal-to-noise ratio of the CT images of cast fossils and their surrounding rocks even when noiseless CT images of cast fossils cannot be obtained. Without damaging the cast fossils, the anatomical structure and distribution of the cast fossils inside the surrounding rocks can be obtained, and the types and quantities of the cast fossils can be counted. By combining CT scanning and 3D printing technologies, the surrounding rocks can be made transparent to clearly display the actual preservation scenario of the cast fossils in the surrounding rocks, reproduce the burial positions of different types of cast fossils and different parts of the same type, and thus provide a reference basis for oil and gas exploration by geological workers. BRIEF DESCRIPTION OF THE DRAWINGS

[0060] Figure 1 is a flowchart showing the method for digital materialization of cast fossils provided by an embodiment of the present invention.

[0061] Figure 2 is a flowchart showing the process of obtaining the semantic region of each cast fossil provided by an embodiment of the present invention.

[0062] Figure 3 is a flowchart showing the process of obtaining the statistics of the types and quantities of cast fossils provided by an embodiment of the present invention.

[0063] Figure 4 It is a pie chart showing the types and quantities of mold fossils counted in the embodiments of the present invention.

[0064] Figure 5 It is a visualization display diagram of the in-situ preservation of mold fossils before and after transparent treatment in the embodiments of the present invention.

[0065] Figure 6 It is a block diagram of the method and device for digital materialization of mold fossils provided by the embodiments of the present invention. Detailed implementation manners

[0066] In order to make the purpose, solutions and advantages of the present invention clearer, taking the experiments conducted on early vertebrate mold fossils as an example, the present invention will be further described in detail. It should be understood that the specific embodiments described herein are only used to explain the present invention and are not used to limit the present invention.

[0067] The method for digital materialization of mold fossils of the present invention includes steps such as obtaining CT images of mold fossils, denoising CT images of mold fossils based on deep learning, surface volume rendering of CT images of mold fossils based on marching cubes, counting the types and quantities of three-dimensional models of mold fossils based on deep learning, and materializing mold fossils.

[0068] Specifically, as Figure 1 shown, the present invention has the following steps:

[0069] Step 1: Obtain CT images of mold fossils.

[0070] According to the CT image acquisition standard of mold fossils, the present invention uses micro-CT to obtain CT images of mold fossils, thereby reducing the influence of noise and artifacts in the CT images of mold fossils. The format of the CT images of mold fossils can be DICOM files or ordinary image formats BMP / TIF files that are not DICOM.

[0071] In one example, the mold fossil is collected from the vertebrate microfossils in the Early Devonian of Yunnan, China. By selecting the rock samples containing the fossil-rich layer, the rock samples are cut into cuboids with a length of 1 cm, a width of 1 cm, and a height of approximately 5 cm using a small cutting machine. The cuboid size is obtained through multiple repeated experiments and summarized according to the actual size of the mold fossil. This size can not only ensure the high efficiency of CT scanning of the fossil but also ensure that the CT data of the fossil has sufficient spatial resolution for creating a three-dimensional model of the mold fossil to meet the requirements of identifying the microscopic structure of the mold fossil. After mechanical processing of the mold fossil, the mold fossil can enter the micro-CT scanning process, and CT images of the mold fossil are obtained using a 225 kV micro-industrial CT. Among them, the parameter settings of the 225 kV micro-industrial CT are: voltage: 100 kV; current: 100 μA; exposure time: 1000 ms; number of superimposed frames: 2 frames; number of projection images: 1440; voxel size: 5.96 μm; filter: 1 mm thick aluminum.

[0072] Step 2: Generate a three-dimensional model of the mold fossil based on the CT image of the mold fossil.

[0073] Step 2.1: Denoise the CT image to obtain a CT image with a high signal-to-noise ratio.

[0074] In step 1, the GPU-accelerated FDK algorithm can convert the 1,440 collected projection images into 1,536 tomograms with 2048×2048 pixels. The CT image of the mold fossil is directly input into the trained deep learning-based CT image denoising network, that is, the Noise2Noise network model with the optimal weight coefficients saved, to obtain a CT image of the mold fossil with a high signal-to-noise ratio, realizing the denoising of the CT image of the mold fossil. The CT image denoising network uses an unsupervised Noise2Noise network model based on the UNet architecture. The Noise2Noise network model trains the network model using CT image pairs with different levels of noise and the L2 loss function, and finally applies the Noise2Noise network model with the optimal weight coefficients saved to generate a CT image with a high signal-to-noise ratio.

[0075] In one example, when training the denoising Noise2Noise network model, by inputting CT image pairs of mold fossils with different levels of noise as the experimental dataset, the deep learning-based CT image denoising network of the mold fossil is trained, verified, and tested. Load the original weight file of the pre-trained UNet architecture, use the L2 loss function to discriminate CT images of mold fossils with different levels of noise and perform backpropagation. Train the deep learning-based CT image denoising network of the mold fossil, adjust the denoising accuracy of the Noise2Noise network model, and save the parameter weights of the optimized Noise2Noise network model.

[0076] The Noise2Noise network model is established based on the UNet architecture. This network model consists of three parts: an encoder, a decoder, and skip connections. The network model contains 18 convolutional layers with a 3×3 convolutional kernel and 1 convolutional layer with a 1×1 convolutional kernel, and the stride is set to 1. Except for the last convolutional layer, the remaining convolutional layers are all equipped with the Relu activation function. The encoder part can convert the input cast fossil CT image into a multi-channel feature map. Its redundant information can be downsampled through a pooling layer with a stride of 2. After passing through 4 pooling layers, the feature map is input into the decoder part. The skip connections can connect the feature maps at the same level in the encoder and decoder, thereby improving the signal-to-noise ratio of the cast fossil CT image. The decoder part upsamples the feature map 4 times, thereby restoring the feature map. Finally, the decoder outputs the denoised cast fossil CT image through a convolutional layer with a 1×1 convolutional kernel.

[0077] Step 2.2: Perform semantic segmentation on the denoised CT image to obtain the semantic regions of each cast fossil.

[0078] The present invention mainly uses the cone-beam FDK algorithm of the filtered back-projection framework to implement the pre-reconstruction work of the cast fossil CT image, and applies the Noise2Noise network model based on the UNet architecture to reduce the noise of the cast fossil CT image.

[0079] In this embodiment, the Otsu method is used to confirm the within-class variance of the cast fossils and the surrounding rocks in the overall denoised CT image, and the optimal threshold is determined according to the maximum value of the traversed within-class variance for threshold segmentation of the binary image of the overall CT image. The optimal threshold can divide the gray histogram of the overall CT image into two categories (surrounding rock and cast fossil), making the variance between the two categories the largest. The gray value of the overall CT image in this embodiment is in the range of 0 - 65535 levels, a 16-bit gray image. Generally, when the gray distribution histogram of the overall CT image is bimodal, the optimal threshold should fall at the trough position between the two peaks. The calculation formula for the optimal threshold of the overall CT image is as follows:

[0080]

[0081] where, is the threshold; is the between-class variance when the threshold is ; , are the probabilities of the cast fossil and the surrounding rock respectively; and are the average gray values of the cast fossil and the surrounding rock respectively; is the gray value of the overall CT image.

[0082] Such as Figure 2As shown, the optimal threshold is applied The overall CT image is subjected to binary image segmentation to select the cast fossil as the region of interest. Using morphological operation methods, external small particles and internal small gaps in the binary cast fossil CT image are removed through operations such as erosion and dilation, and the main structures in the binary cast fossil CT image are retained, that is, the early vertebrate microfossils that are partially or fully decomposed. The region growing algorithm is used to combine adjacent pixels with similar attributes by selecting seed points of the cast fossil and, in combination with manual selection, to gather pixels with similar properties to form a semantic region, realizing the semi-automatic semantic segmentation of the cast fossil CT image.

[0083] Step 2.3: Perform isosurface extraction and surface volume rendering on the semantic segmentation result to generate a three-dimensional model of the cast fossil.

[0084] The present invention applies a surface volume rendering method for cast fossil CT images based on marching cubes. The isosurface extraction and surface volume rendering are performed on the semantic segmentation result of the cast fossil CT image to generate an isothreshold triangular patch model of the cast fossil CT image. The marching cubes algorithm is one of the most commonly used surface volume rendering methods. By finding the voxels that intersect the isosurface in the three-dimensional volume data, the intersection points of the isosurface and the edges of the intersecting voxels are solved using linear interpolation, and finally, all the intersection points are connected to extract the isosurface, generating an isothreshold triangular patch model of the CT image. The marching cubes algorithm can extract the voxels with the same gray threshold in the cast fossil CT image stack and connect them in a topological form to form the cast fossil triangular patches.

[0085] In one example, the visualization toolkit VTK is used to provide a support environment for generating the three-dimensional model of the cast fossil. The present invention uses the vtkMarchingCubes class in VTK to implement the surface volume rendering of the cast fossil CT image. The output three-dimensional model file of the cast fossil is usually in formats such as STL, OBJ, PLY, etc. The cross-platform application tool Blender is used to optimize the three-dimensional model of the cast fossil, improve the rendering quality and visualization effect. The three-dimensional model of the cast fossil is composed of voxels containing color and measurement value information. The appearance of the three-dimensional model can be changed arbitrarily (such as color, transparency, etc.) and can be used for research work on three-dimensional data visualization. Currently, the STL file, as one of the most popular three-dimensional model format files for sharing, can be read by three-dimensional data visualization software such as Meshlab or ImageJ and can also be applied to digital-physical tasks such as 3D printing.

[0086] Step 3: Extract the local features of the three-dimensional model of the cast fossil, perform feature combination on the local features, and classify the cast fossil according to the combined overall features.

[0087] Input the three-dimensional model of the cast fossil to be classified into the trained classification network of the three-dimensional model of the cast fossil based on deep learning to obtain the classification result of the three-dimensional model of the cast fossil, realize the identification of the three-dimensional model of the cast fossil, output the identification result of the three-dimensional model of the cast fossil, and count the types and quantities of the cast fossils. This step mainly includes: three-dimensional model data preprocessing, feature extraction, feature fusion, modeling classification, and type and quantity statistics, etc.

[0088] The three-dimensional model data preprocessing includes pose normalization, average curvature map calculation, and multi-view rendering function. The present invention uses the PCA method to perform pose normalization on the three-dimensional model of the cast fossil, applies the average curvature map to enhance the local geometric shape features of the three-dimensional model of the cast fossil, and uses the multi-view rendering tool to collect multi-perspective views of the three-dimensional model of the cast fossil.

[0089] The three-dimensional model feature extraction is mainly realized by using transfer learning technology, and the VGG16 network model trained in the ImageNet dataset is used as the transfer network. As Figure 3 shown, the low-level feature maps output by the front part of the transfer network, such as the edge structure information of the cast fossil. The intermediate-level feature maps in the middle part of the transfer network are the recombined of the low-level features, and the intermediate features are close to the local attributes of the cast fossil, such as the nodule points, radiating ridges, and medullary cavity pores of the cast fossil and other structural features. The high-level feature maps output by the rear part of the transfer network are the recombined of the intermediate features to obtain the overall features, and the overall features can highly summarize the overall attributes of the cast fossil. Finally, the long short-term memory architecture in the recurrent neural network model is used to realize the multi-view overall feature fusion of the three-dimensional model.

[0090] The three-dimensional model modeling classification selects a support vector machine multi-classifier to perform modeling classification on the three-dimensional model of the cast fossil. A series of hyperplanes are calculated by the support vector machine multi-classifier to distinguish the predicted class of the three-dimensional model of the cast fossil from other classes. The hyperparameter of each hyperplane is represented as w j , and the set of hyperparameter vectors is represented as W = {w 1 , w 2 , …, w k}, where k represents the types of cast fossils in the experimental dataset. The SVM classifier needs to calculate k hyperparameters to complete the classification task of the three-dimensional model of the cast fossil. The quadratic optimization method is applied to solve the unknown hyperparameter vector set W of the SVM multi-class classifier.

[0091] When training the classification network model, combined with the main shape features of the mold fossils, the three-dimensional models of the mold fossils are labeled, summarized, and classified using the corresponding genus and species names to obtain genus and species name labels. The three-dimensional models of the mold fossils and the corresponding genus and species name labels are input as the experimental data set to train, validate, and test the network model. Load the original weight file of the pre-trained improved MVCNN network model. By comparing the classification results of the three-dimensional models of the mold fossils with the true labels and continuously regressing, train the three-dimensional model classification network of the mold fossils based on deep learning, adjust the accuracy of the three-dimensional model classification of the mold fossils, and save the parameter weights of the optimized improved MVCNN network model.

[0092] The creation process of the three-dimensional model classification network of mold fossils based on deep learning provided by the embodiments of the present invention. Input the three-dimensional models of the mold fossils in the training set, validation set, and test set of the experimental data set into the data preprocessing module to ensure the pose normalization of the three-dimensional models of the mold fossils through translation, scaling, and rotation operations. Enhance the local microscopic structure of the three-dimensional models of the mold fossils through the mean curvature, and use the application program developed by combining OpenGL and OpenCV to collect multi-view images of the three-dimensional models of the mold fossils. Input the multi-view images of the three-dimensional models of the mold fossils into the improved MVCNN network model training network. First, train the VGG16FT network model for feature extraction, with the number of iterations set to 100, the optimizer as the Adam function, the batch size set to 4, the initial learning rate set to 1e-4, and the loss function of the VGG16FT network model for feature extraction as the cross-entropy function. Then, train the LSTM network model for the overall feature fusion of the multi-view of the three-dimensional model of the fossil, with the number of iterations set to 100, the optimizer as the Adam function, the batch size set to 36, the initial learning rate set to 1e-3, and the loss function of the LSTM network model for feature fusion as the cross-entropy function. Finally, train the SVM multi-classifier, with the number of iterations set to 3500 and the regularization parameter set to 0.1, to obtain the optimal three-dimensional model classification network of the mold fossils based on deep learning. Use the three-dimensional model classification network of the mold fossils based on deep learning to count the types and quantities of the mold fossils in the surrounding rock. The surrounding rock mainly includes the three-dimensional models of the bone pieces, teeth, and scales of five early ancient fish species (agnathans, placoderms, chondrichthyans, acanthodians, and osteichthyans). Among them, agnathans (Galeaspida, 105; Parasemionotus asiaticus, 57; Parasemionotus liaokuoensis, 100; Parasemionotus wangi, 100), placoderms (Placoderms, 106), osteichthyans (Guiyu, 145; Youngolepis, 163; Psarolepis, 170), chondrichthyans (Guo's fish, 888), acanthodians (Acanthodians, 34; Dorsalspina, 142). The detailed statistical situation is as Figure 4 shown.

[0093] Step 4: Generate a digitally materialized cast fossil based on the 3D model of the cast fossil and the classification results of the cast fossil.

[0094] Import the whole surrounding rock and cast fossil into 3D model optimization software Materialies Magics or RapidForm XOV. In this embodiment, Materialies Magics is taken as an example. Successively perform repair operations on the 3D model, such as reversing reverse triangular patches, stitching gaps, filling holes, deleting bad edges, crossing and overlapping triangular patches, and interfering shells. Import the repaired 3D model of the surrounding rock and cast fossil into the slicing software GrabCAD print. Assign a transparent color material to the surrounding rock, such as Fullcure720 photosensitive resin. Different types of cast fossils can be assigned different colors, such as VeroBlue, VeroYellow, VeroCyan, VeroMagenta, and VeroGray photosensitive resins. Optimize the direction and position of the 3D model on the tray to shorten the printing time, and adjust the scaling ratio of the 3D model to achieve the best effect of observing the distribution of the cast fossil in the surrounding rock.

[0095] In summary, aiming at the problem that geologists cannot restore the true distribution of cast fossils buried in the strata without damage, the present invention collects CT images of cast fossils through micro-CT, uses an unsupervised Noise2Noise network model based on the UNet architecture to remove the noise of the CT images of cast fossils, completes the semantic segmentation of the CT images of cast fossils in a semi-automated manner, classifies the cast fossils using a deep learning-based 3D model classification network for cast fossils and conducts statistics on the types and quantities, and realizes the materialization of cast fossils by applying 3D printing technology. The present invention can intuitively display the in-situ preservation of cast fossils to geologists from all aspects and angles such as digital visualization and materialization. As Figure 5 shown in the visualization display diagrams of the in-situ preservation of cast fossils before and after the surrounding rock is made transparent in the embodiment. In the left figure, the transparency threshold of the surrounding rock is set to 0, and the internal distribution of the cast fossil cannot be displayed. In the right figure, the transparency threshold of the surrounding rock is set to 50%, showing the distribution of different types of cast fossils inside the surrounding rock. The present invention is expected to point out potential directions for the determination of geological ages, the division and correlation of strata, and the search for mineral resources.

[0096] Please refer to Figure 6 , which shows a block diagram of a cast fossil digital materialization system provided by an embodiment of the present invention. The device includes the following several modules: an image acquisition module 610, a model generation module 620, a fossil classification module 630, and a digital materialization module 640.

[0097] The image acquisition module 610 is used to acquire CT images of cast fossils;

[0098] A model generation module 620, configured to generate a three-dimensional model of a cast fossil based on the CT image of the cast fossil;

[0099] A fossil classification module 630, configured to extract local features of the three-dimensional model of the cast fossil, perform feature combination on the local features, and classify the cast fossil according to the overall features after combination; wherein, the local features of the three-dimensional model of the cast fossil include: nodule structure features, radiating ridge structure features, and medullary cavity pore structure features;

[0100] A digital entity generation module 640, configured to generate a digitally entityified cast fossil based on the three-dimensional model of the cast fossil and the classification result of the cast fossil.

[0101] For the specific execution process, beneficial effects, etc. of the system module, please refer to the introduction in the above method embodiments, and details are not described here.

[0102] In an exemplary embodiment, a computer device is further provided. The computer device includes a memory and a processor. A computer program is stored in the memory and is loaded and executed by the processor to implement the above digital entityification method for a cast fossil.

[0103] In an exemplary embodiment, a computer-readable storage medium is further provided, on which a computer program is stored. When the computer program is executed by a processor, the above digital entityification method for a cast fossil is implemented.

[0104] In an exemplary embodiment, a computer program product is further provided. When the computer program product runs on a computer device, the computer device is enabled to execute the above digital entityification method for a cast fossil.

[0105] The above is only one embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent replacements, improvements, etc. made within the spirit and principle of the present invention shall be included within the protection scope of the present invention.

Claims

1. A method for digital materialization of a molded fossil, characterized in that: The method comprises: Obtaining a CT image of a mold-cast fossil; wherein the mold-cast fossil is a mold-cast fossil with a cavity structure, and obtaining a CT image of the mold-cast fossil includes: Cut the rock sample according to the set size to obtain a rectangular molded fossil; The X-ray detector is used to image the molded fossil at different angles in a 360° step-by-step manner; The images at each angle are reconstructed, and the CT images of the molded fossil are obtained using a filtered back-projection algorithm; Based on the CT image of the die-cast fossil, a three-dimensional model of the die-cast fossil is generated; wherein, based on the CT image of the die-cast fossil, the three-dimensional model of the die-cast fossil is generated, including: Denoising the CT image of the die-cast fossil to obtain a high signal-to-noise ratio CT image; wherein the denoising the CT image of the die-cast fossil to obtain a high signal-to-noise ratio CT image includes: By inputting pairs of cast fossil CT images with different levels of noise as experimental datasets; The Noise2Noise network model based on the UNet architecture is unsupervisedly trained on the experimental data set; wherein the Noise2Noise network model is established based on the UNet architecture, and the Noise2Noise network model consists of three parts: an encoder, a decoder, and a skip connection, including 18 convolutional layers with 3×3 convolution kernels and 1 convolutional layer with 1×1 convolution kernel, the step size is set to 1, and except for the last convolutional layer, the remaining convolutional layers all have Relu activation functions; Based on the trained Noise2Noise network model, the CT images of the molded fossils were used to obtain high signal-to-noise ratio CT images; Perform semantic segmentation on high signal-to-noise ratio CT images to obtain semantic segmentation results of molded fossils; The semantic segmentation results are subjected to isosurface extraction and surface volume rendering to generate a three-dimensional model of the molded fossil; Extracting local features of the three-dimensional model of the mold-cast fossil, combining the local features, and classifying the mold-cast fossil according to the combined overall features; wherein the local features of the three-dimensional model of the mold-cast fossil include: tumor point structure features, radial ridge structure features, and medullary cavity foramen structure features; Based on the three-dimensional model of the cast fossil and the classification results of the cast fossil, a digitally materialized cast fossil is generated; wherein, based on the three-dimensional model of the cast fossil and the classification results of the cast fossil, a digitally materialized cast fossil is generated, including: Repairing the three-dimensional model of the molded fossil, the repairing comprising: inverting reverse triangular facets, stitching gaps, filling holes, deleting bad edges, crossing and overlapping triangular facets, and interfering with shells; The repaired three-dimensional model of the cast fossil is imported into the 3D printing slicing software, and the surrounding rock is given a transparent material, and the cast fossil is given a color material based on the classification result of the cast fossil, so as to obtain a solid three-dimensional model of the cast fossil.

2. The method according to claim 1, characterized in that The semantic segmentation of the high signal-to-noise ratio CT image to obtain the semantic segmentation result of the molded fossil includes: Get the grayscale mean of the molded fossil , grayscale mean of surrounding rock and the overall grayscale value of high signal-to-noise ratio CT images ; By calculating the optimal threshold The between-class variance , get the optimal threshold ;in, is the probability of a molded fossil, is the probability of surrounding rock; Applying the best threshold Performing binary image segmentation on the high signal-to-noise ratio CT image to obtain a binary mold-cast fossil CT image; After removing the external small particles and internal small gaps in the binarized cast fossil CT image, the seed points of the cast fossil are selected to merge adjacent pixels with similar attributes and combined with manual circle selection to group pixels with similar properties into semantic regions, thus obtaining the semantic segmentation result of the cast fossil.

3. The method according to claim 1, characterized in that The semantic segmentation results are extracted by isosurface extraction and surface volume rendering to generate a three-dimensional model of the molded fossil, including: Obtain the three-dimensional volume data corresponding to each semantic segmentation result; Find the voxels that intersect the isosurface in the three-dimensional volume data and obtain the voxel edges; Applying linear interpolation to solve the intersection of the isosurface and the voxel edge; All intersection points are connected to extract isosurfaces to generate a three-dimensional model of the molded fossil.

4. The method according to claim 1, characterized in that: The step of extracting the features of the three-dimensional model of the molded fossil and classifying the molded fossil based on the features comprises: Preprocessing the three-dimensional model of the molded fossil; The VGG16 network model of transfer learning is used to extract the overall features of the pre-processed three-dimensional model of the molded fossil; wherein the features of the pre-processed three-dimensional model of the molded fossil extracted by the VGG16 network model of transfer learning include: Using the front part of the network in the VGG16 network model, a low-level feature map of the three-dimensional model of the molded fossil is obtained, wherein the low-level feature map includes edge structure information of the three-dimensional model of the molded fossil; The low-level feature maps are combined using the middle part network in the VGG16 network model to obtain local features of the three-dimensional model of the molded fossil; The local features are combined by using the latter part of the network in the VGG16 network model to obtain the overall features of the three-dimensional model of the molded fossil; The long short-term memory architecture in the recurrent neural network model is used to fuse the overall features of multiple perspectives; The SVM multi-classifier is selected to classify the fused features to obtain the classification result of the mold cast fossil.

5. The method according to claim 4, characterized in that The three-dimensional model of the molded fossil is preprocessed, including: The PCA method is used to normalize the posture of the three-dimensional model of the molded fossil; and, Using the mean curvature map to enhance the local geometric features of the three-dimensional model of the molded fossil; and, A multi-view rendering tool is used to collect multi-view views of the cast fossil three-dimensional model.

6. A digital materialization system for molded fossils, characterized in that: The system comprises: An image acquisition module is used to acquire a CT image of the molded fossil; wherein the acquisition of the CT image of the molded fossil includes: Cut the rock sample according to the set size to obtain a rectangular molded fossil; The X-ray detector is used to image the molded fossil at different angles in a 360° step-by-step manner; The images at each angle are reconstructed, and the CT images of the molded fossil are obtained using the filtered back-projection algorithm. A model generation module is used to generate a three-dimensional model of the molded fossil based on the CT image of the molded fossil; wherein the generation of the three-dimensional model of the molded fossil based on the CT image of the molded fossil includes: Denoising the CT image of the die-cast fossil to obtain a high signal-to-noise ratio CT image; wherein the denoising the CT image of the die-cast fossil to obtain a high signal-to-noise ratio CT image includes: By inputting pairs of cast fossil CT images with different levels of noise as experimental datasets; The Noise2Noise network model based on the UNet architecture is unsupervisedly trained on the experimental data set; wherein the Noise2Noise network model is established based on the UNet architecture, and the Noise2Noise network model consists of three parts: an encoder, a decoder, and a skip connection, including 18 convolutional layers with 3×3 convolution kernels and 1 convolutional layer with 1×1 convolution kernel, the step size is set to 1, and except for the last convolutional layer, the remaining convolutional layers all have Relu activation functions; Perform semantic segmentation on high signal-to-noise ratio CT images to obtain semantic segmentation results of molded fossils; The semantic segmentation results are extracted by isosurface extraction and surface volume rendering to generate a 3D model of the molded fossil. A fossil classification module is used to extract local features of the three-dimensional model of the die-cast fossil, combine the local features, and classify the die-cast fossil according to the combined overall features; wherein the local features of the three-dimensional model of the die-cast fossil include: tumor point structure features, radial ridge structure features and medullary cavity foramen structure features; A digital materialization module is used to generate a digital materialized die-cast fossil based on the three-dimensional model of the die-cast fossil and the classification result of the die-cast fossil; wherein the digital materialization die-cast fossil based on the three-dimensional model of the die-cast fossil and the classification result of the die-cast fossil is generated, including: Repairing the three-dimensional model of the molded fossil, the repairing comprising: inverting reverse triangular facets, stitching gaps, filling holes, deleting bad edges, crossing and overlapping triangular facets, and interfering with shells; The repaired three-dimensional model of the cast fossil is imported into the 3D printing slicing software, and the surrounding rock is given a transparent material, and the cast fossil is given a color material based on the classification result of the cast fossil, so as to obtain a solid three-dimensional model of the cast fossil.

Images