Method, system, terminal and medium for three-dimensional morphology recognition and reconstruction of lunar soil particles

Abstract

The present invention discloses a method, system, terminal and medium for three-dimensional morphology recognition and reconstruction of lunar soil particles, and relates to the field of three-dimensional reconstruction technology. The method includes: obtaining a set of lunar soil particles; photographing the set of lunar soil particles to obtain a two-dimensional picture of the set of lunar soil particles; segmenting the two-dimensional picture to obtain a particle picture of each lunar soil particle; inputting the particle picture of each lunar soil particle into a trained dimensional reconstruction model to obtain reconstructed three-dimensional morphology data; the dimensional reconstruction model includes an encoder and a decoder; the encoder is used to extract several feature maps of different dimensions according to the input particle picture; the decoder is used to perform three-dimensional deconvolution on each feature map to obtain reconstructed three-dimensional morphology data. The present invention only needs one camera to obtain a two-dimensional picture of the lunar soil particle set, and combined with the trained dimensional reconstruction model, it can realize end-to-end dimensional reconstruction, obtain three-dimensional morphology data of lunar soil particles, and reduce the time and cost of obtaining three-dimensional morphology information.

Description

Technical Field

[0001] The present invention relates to the field of three-dimensional reconstruction technology, and in particular to a method, system, terminal and medium for three-dimensional morphology recognition and reconstruction of lunar soil particles. Background Art

[0002] The lunar soil covering the lunar surface is the main carrier of resources, covering almost all resources such as solid minerals and water ice. The morphology of lunar soil particles is complex and different from that of earth particles. The morphology of lunar soil particles contains a lot of information. However, due to the small size of lunar soil particles, the three-dimensional morphology information of lunar soil particles is currently mainly obtained through computed tomography (CT), white light interferometry, three-dimensional laser scanning and other technologies, which are costly and not suitable for batch acquisition.

[0003] Therefore, the existing technology still needs to be improved and developed. Summary of the invention

[0004] The technical problem to be solved by the present invention is to provide a three-dimensional morphology recognition and reconstruction method, system, terminal and medium for lunar soil particles in response to the above-mentioned defects of the prior art, aiming to solve the problem in the prior art that it is difficult to obtain three-dimensional morphology information of lunar soil particles conveniently and quickly.

[0005] The technical solution adopted by the present invention to solve the problem is as follows:

[0006] In a first aspect, an embodiment of the present invention provides a method for identifying and reconstructing the three-dimensional morphology of lunar soil particles, the method comprising:

[0007] Acquire a lunar soil particle set, wherein the lunar soil particle set includes a plurality of lunar soil particles;

[0008] photographing the lunar soil particle set to obtain a two-dimensional image corresponding to the lunar soil particle set;

[0009] Segmenting the two-dimensional image to obtain particle images corresponding to the lunar soil particles;

[0010] For each of the lunar soil particles, the particle image of the lunar soil particle is input into the trained dimensionality-raising reconstruction model to obtain the reconstructed three-dimensional morphology data corresponding to the lunar soil particle; wherein the dimensionality-raising reconstruction model includes an encoder and a decoder; the encoder is used to extract a number of feature maps of different dimensions according to the input particle image; the decoder is used to perform three-dimensional deconvolution on each of the feature maps to obtain the reconstructed three-dimensional morphology data.

[0011] In one embodiment, segmenting the two-dimensional image to obtain particle images corresponding to the lunar soil particles includes:

[0012] Converting the two-dimensional image into a grayscale image, and converting the grayscale image into a binary image by using a preset threshold;

[0013] The binary image is segmented to obtain particle images corresponding to each of the lunar soil particles.

[0014] In one implementation, the encoder is specifically configured to:

[0015] The input particle image is subjected to feature extraction through a two-dimensional convolutional layer and a maximum pooling layer, so that the size of the particle image gradually decreases, while the feature dimension gradually increases, and several feature maps of different dimensions are obtained.

[0016] In one implementation, the decoder is specifically configured to:

[0017] Generate a number of input voxels according to each of the feature maps, perform three-dimensional deconvolution on each of the input voxels through a three-dimensional deconvolution layer so that the voxel size gradually increases, and obtain a number of output voxels;

[0018] Reconstructed three-dimensional shape data is generated according to each of the output voxels.

[0019] In one embodiment, the method for generating the trained dimensionality-raising reconstruction model is:

[0020] Acquire a training data set, wherein the training data set includes a plurality of training samples and real labels corresponding to the training samples; the training samples are particle images, and the real labels are real three-dimensional morphology data corresponding to the particle images;

[0021] Inputting one of the training samples into the dimension-raising reconstruction model to obtain reconstructed three-dimensional shape data corresponding to the training sample;

[0022] Calculate a binary cross entropy loss value and a mean square error loss value according to the true label corresponding to the training sample and the reconstructed three-dimensional shape data;

[0023] Determining a loss function value of the dimensionality-raising reconstruction model according to the binary cross entropy loss value and the mean square error loss value;

[0024] Determine whether the loss function value meets the preset training requirements, and if not, update the dimension-raising reconstruction model according to the loss function value;

[0025] Continue to execute the step of inputting one of the training samples into the dimension-raising reconstruction model until the loss function value meets the training requirement, thereby obtaining the trained dimension-raising reconstruction model.

[0026] In one implementation, obtaining a training data set includes:

[0027] Acquire a preprocessed training lunar soil particle set; wherein the preprocessing includes impurity removal, screening, and drying; the training lunar soil particle set includes a plurality of training lunar soil particles;

[0028] The training lunar soil particle set is photographed to obtain a two-dimensional image data set; wherein the two-dimensional image data set includes particle images corresponding to each of the training lunar soil particles;

[0029] The training lunar soil particle set is placed in a fixed container for three-dimensional scanning to obtain a three-dimensional morphology aggregate;

[0030] Performing particle segmentation on the three-dimensional morphology aggregate to obtain a three-dimensional morphology data set; wherein the three-dimensional morphology data set includes real three-dimensional morphology data corresponding to each of the training lunar soil particles;

[0031] The training data set is obtained according to the two-dimensional image data set and the three-dimensional shape data set.

[0032] In one embodiment, the method further comprises:

[0033] The reconstructed three-dimensional morphological data of the lunar soil particles are visualized, wherein the reconstructed three-dimensional morphological data can be rotated and the background can be rendered.

[0034] In a second aspect, an embodiment of the present invention further provides a system for recognizing and reconstructing the three-dimensional morphology of lunar soil particles, the system comprising:

[0035] A particle acquisition module, used to acquire a lunar soil particle set, wherein the lunar soil particle set includes a plurality of lunar soil particles;

[0036] An image acquisition module, used for photographing the lunar soil particle set to obtain a two-dimensional image corresponding to the lunar soil particle set;

[0037] An image segmentation module is used to segment the two-dimensional image to obtain particle images corresponding to each of the lunar soil particles;

[0038] A three-dimensional reconstruction module is used to input the particle image of each lunar soil particle into a trained dimensionality-raising reconstruction model to obtain reconstructed three-dimensional morphology data corresponding to the lunar soil particle; wherein the dimensionality-raising reconstruction model includes an encoder and a decoder; the encoder is used to extract a number of feature maps of different dimensions based on the input particle image; the decoder is used to perform three-dimensional deconvolution on each of the feature maps to obtain reconstructed three-dimensional morphology data.

[0039] In the third aspect, an embodiment of the present invention also provides a terminal, which includes a memory and one or more processors; the memory stores one or more programs; the program includes instructions for executing any of the three-dimensional morphology recognition and reconstruction methods of lunar soil particles as described above; and the processor is used to execute the program.

[0040] In a fourth aspect, an embodiment of the present invention further provides a computer-readable storage medium on which a plurality of instructions are stored, wherein the instructions are suitable for being loaded and executed by a processor to implement the steps of any of the above-described methods for three-dimensional morphology recognition and reconstruction of lunar soil particles.

[0041] Beneficial effects of the invention: The embodiment of the invention only needs one camera to obtain a two-dimensional picture of a set of lunar soil particles. Combined with the trained dimensionality increase reconstruction model, it can quickly realize end-to-end dimensionality increase reconstruction and obtain the reconstructed three-dimensional morphology data of the lunar soil particles, which greatly reduces the time and cost of obtaining three-dimensional morphology information of lunar soil particles and is suitable for batch acquisition application scenarios. BRIEF DESCRIPTION OF THE DRAWINGS

[0042] In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the drawings required for use in the embodiments or the description of the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments recorded in the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying creative work.

[0043] Figure 1 It is a flow chart of the three-dimensional morphology recognition and reconstruction method of lunar soil particles provided in an embodiment of the present invention.

[0044] Figure 2 It is a schematic diagram of the effect of binarization processing provided by an embodiment of the present invention.

[0045] Figure 3 It is a data processing schematic diagram of an encoder and a decoder provided in an embodiment of the present invention.

[0046] Figure 4 It is a schematic diagram of the functional modules used in the dimensionality-increasing reconstruction model training provided in an embodiment of the present invention.

[0047] Figure 5 It is a module schematic diagram of the three-dimensional morphology recognition and reconstruction system of lunar soil particles provided in an embodiment of the present invention.

[0048] Figure 6 It is a principle block diagram of a terminal provided by an embodiment of the present invention. DETAILED DESCRIPTION

[0049] The present invention discloses a method, system, terminal and medium for three-dimensional morphology recognition and reconstruction of lunar soil particles. In order to make the purpose, technical solution and effect of the present invention clearer and more specific, the present invention is further described in detail with reference to the accompanying drawings and examples. It should be understood that the specific embodiments described here are only used to explain the present invention and are not used to limit the present invention.

[0050] It will be understood by those skilled in the art that, unless expressly stated, the singular forms "one", "said", and "the" used herein may also include plural forms. It should be further understood that the term "comprising" used in the specification of the present invention refers to the presence of the features, integers, steps, operations, elements and / or components, but does not exclude the presence or addition of one or more other features, integers, steps, operations, elements, components and / or groups thereof. It should be understood that when we refer to an element as being "connected" or "coupled" to another element, it may be directly connected or coupled to the other element, or there may be intermediate elements. In addition, the "connection" or "coupling" used herein may include wireless connection or wireless coupling. The term "and / or" used herein includes all or any unit and all combinations of one or more associated listed items.

[0051] It will be understood by those skilled in the art that, unless otherwise defined, all terms (including technical and scientific terms) used herein have the same meaning as generally understood by those skilled in the art in the art to which the present invention belongs. It should also be understood that terms such as those defined in general dictionaries should be understood to have meanings consistent with the meanings in the context of the prior art, and will not be interpreted with idealized or overly formal meanings unless specifically defined as herein.

[0052] In view of the above-mentioned defects of the prior art, the present invention provides a method for recognizing and reconstructing the three-dimensional morphology of lunar soil particles, the method comprising: obtaining a set of lunar soil particles, wherein the set of lunar soil particles comprises a number of lunar soil particles; photographing the set of lunar soil particles to obtain a two-dimensional image corresponding to the set of lunar soil particles; segmenting the two-dimensional image to obtain a particle image corresponding to each of the lunar soil particles; for each of the lunar soil particles, inputting the particle image of the lunar soil particle into a trained dimensional reconstruction model to obtain the reconstructed three-dimensional morphological data corresponding to the lunar soil particle; wherein the dimensional reconstruction model comprises an encoder and a decoder; the encoder is used to extract a number of feature maps of different dimensions according to the input particle image; the decoder is used to perform three-dimensional deconvolution on each of the feature maps to obtain the reconstructed three-dimensional morphological data. The present invention only requires one photographing device to obtain a two-dimensional image of the lunar soil particle set, and combined with the trained dimensional reconstruction model, it can quickly realize end-to-end dimensional reconstruction and obtain the reconstructed three-dimensional morphological data of the lunar soil particles, which greatly reduces the time and cost of obtaining three-dimensional morphological information of lunar soil particles, and is suitable for batch acquisition application scenarios.

[0053] like Figure 1 As shown, the method specifically comprises the following steps:

[0054] Step S100: Obtain a lunar soil particle set, wherein the lunar soil particle set includes a plurality of lunar soil particles.

[0055] Specifically, the lunar soil particle set in this embodiment is composed of multiple lunar soil particles. In actual application scenarios, after obtaining the original lunar soil, it is necessary to perform pretreatment such as impurity removal, screening, and drying on the lunar soil to remove impurities in the lunar soil and reduce the viscosity of the lunar soil for subsequent analysis and processing. The lunar soil particle set is established by using the lunar soil particles that are finally retained after pretreatment for subsequent three-dimensional morphology reconstruction.

[0056] Step S200: photograph the lunar soil particle set to obtain a two-dimensional image corresponding to the lunar soil particle set.

[0057] Specifically, in this embodiment, it is necessary to reconstruct the three-dimensional morphology through the two-dimensional image. Therefore, after the lunar soil particle set is established, it is necessary to shoot the lunar soil particle set to obtain a two-dimensional picture of the lunar soil particle set. In actual application, the lunar soil particle set is placed on a preset background plate, and a camera or other imaging device is used to aim at the lunar soil particle set. By adjusting the device parameters of the camera, it is ensured that the outline and shape of each lunar soil particle can be clearly captured to obtain a two-dimensional picture of the lunar soil particle set.

[0058] Step S300: segment the two-dimensional image to obtain particle images corresponding to each of the lunar soil particles.

[0059] Specifically, in order to save time costs, this embodiment does not take a separate photo of each lunar soil particle, but takes a unified photo of all lunar soil particles, and then uses image segmentation technology to segment the local image corresponding to each lunar soil particle separately, and generates particle images corresponding to each lunar soil particle. In actual application, the two-dimensional image obtained based on the lunar soil particle set shows the distribution of multiple lunar soil particles on a plane. Each lunar soil particle in the two-dimensional image is individually identified and separated through a preset image recognition and segmentation algorithm to obtain a particle image of each lunar soil particle. The particle image shows the morphology of the lunar soil particles, which is convenient for the subsequent reconstruction of the three-dimensional morphology of the lunar soil particles.

[0060] In one implementation, segmenting the two-dimensional image to obtain particle images corresponding to the lunar soil particles includes:

[0061] Converting the two-dimensional image into a grayscale image, and converting the grayscale image into a binary image by using a preset threshold;

[0062] The binary image is segmented to obtain particle images corresponding to each of the lunar soil particles.

[0063] Specifically, in order to facilitate image processing, this embodiment will first convert the two-dimensional image into a grayscale image, and then obtain the corresponding binary image (such as Figure 2 Finally, the binary image is segmented to obtain a binary particle image of each lunar soil particle.

[0064] In one implementation, segmenting the binary image to obtain particle images corresponding to the lunar soil particles includes:

[0065] Using a preset threshold segmentation algorithm, or an edge detection algorithm, or a region growing algorithm to identify boundary data corresponding to each of the lunar soil particles in the binary image;

[0066] The binary image is segmented according to the boundary data of each lunar soil particle to obtain particle images corresponding to each lunar soil particle.

[0067] Step S400: for each of the lunar soil particles, the particle image of the lunar soil particle is input into the trained dimensionality-raising reconstruction model to obtain the reconstructed three-dimensional morphology data corresponding to the lunar soil particle; wherein the dimensionality-raising reconstruction model includes an encoder and a decoder; the encoder is used to extract a number of feature maps of different dimensions according to the input particle image; the decoder is used to perform three-dimensional deconvolution on each of the feature maps to obtain the reconstructed three-dimensional morphology data.

[0068] Specifically, in order to facilitate the reconstruction of the three-dimensional morphology of lunar soil particles, this embodiment pre-constructs and trains a dimensional reconstruction model. The input data of the dimensional reconstruction model is the particle image of the lunar soil particles, and the output data is the reconstructed three-dimensional morphology data of the lunar soil particles. The dimensional reconstruction model can learn the complex mapping relationship from input to output and the dimensional reconstruction method from two-dimensional to three-dimensional through a large amount of training data. Therefore, the trained dimensional reconstruction model can generate relatively accurate reconstructed three-dimensional morphology data based on the particle image of the lunar soil particles. The functional implementation of the dimensional reconstruction model is closely related to its model structure. The model structure of the dimensional reconstruction model is mainly composed of an encoder and a decoder. Among them, the encoder is responsible for receiving the input particle image, and performing feature recognition and extraction on the particle image, and inputting the extracted feature map into the decoder. For example, the feature map output by the encoder is n*m*x, where n and m are the size of the feature map (generally n and m are the same value, that is, m=n, and the feature map is a square), and x is the number of feature maps or the number of dimensions. The decoder is responsible for reconstructing the extracted feature map using three-dimensional deconvolution to obtain the corresponding reconstructed three-dimensional morphology data.

[0069] In one implementation, the encoder is specifically used for:

[0070] The input particle image is subjected to feature extraction through a two-dimensional convolutional layer and a maximum pooling layer, so that the size of the particle image gradually decreases, while the feature dimension gradually increases, and several feature maps of different dimensions are obtained.

[0071] Specifically, the encoder in this embodiment is responsible for receiving the input two-dimensional particle image, and extracting image features of different dimensions in the particle image through a series of two-dimensional convolutional layers and maximum pooling layers, such as different dimensional information such as depth, texture, and shape, and finally representing the extracted image features in the form of multiple feature maps. In practical applications, such as Figure 3 As shown in the figure, after the particle image is input into the encoder, the encoder will gradually reduce the image size and gradually increase the feature dimension, and finally obtain the feature map of the particle image.

[0072] In one implementation, in addition to the two-dimensional convolution layer and the maximum pooling layer, the encoder also includes a residual connection layer between two adjacent layers.

[0073] Specifically, after the particle image is input into the encoder, it will be processed by the data of the neural network layers such as the two-dimensional convolution layer, the maximum pooling layer and the residual connection layer. Among them, the two-dimensional convolution layer slides on the particle image through the convolution operation, performs the dot product operation element by element, extracts the local features of the particle image, and generates a feature map. The maximum pooling layer selects the maximum value in the pooling window as the output, retains the most significant features, and reduces the size of the feature map and the computational complexity. The maximum pooling layer not only reduces the data dimension, but also enhances the invariance of the model to input changes, which can effectively improve the robustness of the model. The residual connection layer is set between two adjacent neural network layers. The input data of the residual connection is the output data of the previous layer, and the output data of the residual connection is the data processed by the current layer (such as convolution, activation and other operations) plus the input data itself. The residual connection layer introduces direct connections across layers, so that the gradient can be back-propagated more easily, solving the problem of gradient disappearance.

[0074] For example, the calculation formula and feature map changes of the two-dimensional convolution layer, maximum pooling layer, and residual connection layer included in the encoder are as follows:

[0075] The calculation formula of the two-dimensional convolutional layer is:

[0076] ;

[0077] The feature map changes to:

[0078] ;

[0079] ;

[0080] in, is the pixel value of the input feature map; is the convolution kernel weight; is bias; is the convolution kernel size; is the output feature image pixel value; the input image size is ; is the padding size; is the stride; the output image size is .

[0081] The calculation formula of the maximum pooling layer is:

[0082] ;

[0083] The feature map of the maximum pooling layer changes to:

[0084] ;

[0085] ;

[0086] in, is the pixel value of the input feature map; is the output feature image pixel value; is the pooling window size; the input image size is ; is the padding size; is the stride; the output image size is .

[0087] The calculation formula of the residual connection layer is:

[0088] ;

[0089] in, is the weight; It is the output after convolution and other processing on the input x; the residual link does not involve the size change of the feature map.

[0090] In one implementation, the decoder is specifically configured to:

[0091] A number of input voxels are generated according to each of the feature maps, and three-dimensional deconvolution is performed on each of the input voxels through a three-dimensional deconvolution layer so that the voxel size gradually increases to obtain a number of output voxels; and reconstructed three-dimensional morphological data is generated according to each of the output voxels.

[0092] Specifically, the decoder reconstructs the feature map learned by the encoder. The function of the decoder is closely related to the 3D deconvolution layer it contains. The 3D deconvolution layer can restore the low-resolution feature map to a higher resolution. The size of the feature map output by the deconvolution operation will be larger than the size of the input feature map (such as Figure 3 As shown in the figure). In practical applications, the 3D deconvolution layer will operate in three dimensions (length, width, and height). The voxel size gradually increases with the transmission of the 3D deconvolution layer, and the low-dimensional feature map is reconstructed into high-dimensional data. Finally, the required size of the particles is reconstructed to complete the morphology reconstruction.

[0093] Furthermore, the decoder is specifically used for:

[0094] Converting the feature map extracted by the encoder into voxel form, wherein each voxel is a cube in three-dimensional space and contains feature information;

[0095] Input each voxel into the three-dimensional deconvolution layer, perform a series of deconvolution operations on the input voxel through the three-dimensional deconvolution layer, gradually increase the size of the voxel, and obtain a new set of output voxels;

[0096] The output voxels are integrated into a complete three-dimensional shape to obtain the reconstructed three-dimensional shape data.

[0097] First, the dimensions of the feature map extracted by the encoder are rearranged and reshaped to convert them into voxel form. These voxels can be regarded as small cubes in three-dimensional space. Each voxel contains certain feature information to meet the input requirements of the three-dimensional deconvolution layer. Each voxel is input into the three-dimensional deconvolution layer, which inserts gaps between the input data and then applies the convolution kernel to achieve three-dimensional reconstruction. Specifically, the three-dimensional deconvolution layer performs a series of deconvolution operations on the input voxels. The resolution of the voxels will be improved after each deconvolution operation, thereby gradually increasing the size of the voxels. Among them, the deconvolution operation slides on the voxels through a learnable filter (kernel) and generates new voxels through element-by-element product and sum operations. After each deconvolution operation, new output voxels are generated. These voxels contain rich spatial information and help to reconstruct the three-dimensional shape more accurately. The output voxels generated by the last deconvolution operation are integrated together to form a complete three-dimensional shape, that is, the final reconstructed three-dimensional shape data is obtained. The reconstructed three-dimensional shape data can be represented in the form of a grid or a voxel grid.

[0098] For example, the calculation formula of the three-dimensional deconvolution layer contained in the decoder and the change of feature map size are as follows:

[0099] The calculation formula of the 3D deconvolution layer is:

[0100] ;

[0101] Feature map changes:

[0102] ;

[0103] ;

[0104] ;

[0105] in, is the input feature voxel value; is the convolution kernel weight; is bias; is the convolution kernel size; is the output feature image pixel value; the input voxel size is ; is the padding size; is the stride; the output voxel size is .

[0106] In one implementation, the method for generating the trained dimensionality-raising reconstruction model is:

[0107] Acquire a training data set, wherein the training data set includes a plurality of training samples and real labels corresponding to the training samples; the training samples are particle images, and the real labels are real three-dimensional morphology data corresponding to the particle images;

[0108] Inputting one of the training samples into the dimension-raising reconstruction model to obtain reconstructed three-dimensional shape data corresponding to the training sample;

[0109] Calculate a binary cross entropy loss value and a mean square error loss value according to the true label corresponding to the training sample and the reconstructed three-dimensional shape data;

[0110] Determining a loss function value of the dimensionality-raising reconstruction model according to the binary cross entropy loss value and the mean square error loss value;

[0111] Determine whether the loss function value meets the preset training requirements, and if not, update the dimension-raising reconstruction model according to the loss function value;

[0112] Continue to execute the step of inputting one of the training samples into the dimension-raising reconstruction model until the loss function value meets the training requirement, thereby obtaining the trained dimension-raising reconstruction model.

[0113] Specifically, in the preparation stage of model training, it is necessary to collect particle images with real labels to establish a training data set. The training samples in the training data set are two-dimensional particle images, and the real labels are the real shape / morphology data of the particles in three-dimensional space. The training data set is used to train the dimensional reconstruction model to learn the intrinsic mapping law of two-dimensional to three-dimensional. In the training stage of the model, a training sample is input into the dimensional reconstruction model, and the dimensional reconstruction model outputs a reconstructed three-dimensional morphological data. In order to evaluate the performance of the model, it is necessary to compare the reconstructed three-dimensional morphological data output by the model with the real label to calculate the model loss. In this embodiment, two loss functions, namely, binary cross entropy loss value and mean square error loss value, are used to jointly calculate the loss function value of the model. For example, the weighted sum of the binary cross entropy loss value and the mean square error loss value can be used as the loss function value of the dimensional reconstruction model. During the training process, the loss function value of the model is iteratively calculated, and it is determined whether the value meets the preset training requirements. For example, the training requirements can be that the loss function value converges to a preset threshold or the number of iterations reaches a preset number. If the training requirements are not met, the model parameters need to be adjusted based on the loss function value to continue converging the model loss until the training requirements are met. This embodiment can minimize the difference between the three-dimensional morphology of the particles reconstructed by the dimensional reconstruction model and the real data set through model training, so that the model can learn the mapping law of dimensionality increase.

[0114] For example, the loss function of the dimensional reconstruction model is used to measure the difference between the real 3D shape data and the reconstructed 3D shape data. The calculation method of the reconstruction loss of the traditional algorithm is as follows:

[0115] ;

[0116] in, is the binary cross entropy loss.

[0117] Different from the reconstruction loss of the traditional algorithm, the reconstruction loss of the dimensional reconstruction model in this embodiment also adds , so as to more effectively learn the dimension-upgraded features of irregular particles. The calculation method of the loss function of the dimension-upgraded reconstruction model in this embodiment is as follows:

[0118] ;

[0119] Among them, the loss function of the dimensional reconstruction model is It consists of two parts. The first part is , is the binary cross entropy loss, which is an effective binary classification loss function for classifying predicted voxel values ​​0 and 1. is the weight parameter used to change The weight ratio of . The calculation method is as follows:

[0120] ;

[0121] The second part is , is the mean squared error loss, which is used to evaluate the difference between the predicted value and the true value. The calculation method is as follows:

[0122] ;

[0123] in, The number of voxels representing real particles, Each voxel value representing a real particle, Represents the voxel value corresponding to the predicted particle position. and The smaller it is, the closer the predicted particles are to the real particles. control, The smaller the , the better the reconstruction result.

[0124] In one implementation, obtaining a training data set includes:

[0125] Acquire a preprocessed training lunar soil particle set; wherein the preprocessing includes impurity removal, screening, and drying; the training lunar soil particle set includes a plurality of training lunar soil particles;

[0126] The training lunar soil particle set is photographed to obtain a two-dimensional image data set; wherein the two-dimensional image data set includes particle images corresponding to each of the training lunar soil particles;

[0127] The training lunar soil particle set is placed in a fixed container for three-dimensional scanning to obtain a three-dimensional morphology aggregate;

[0128] Performing particle segmentation on the three-dimensional morphology aggregate to obtain a three-dimensional morphology data set; wherein the three-dimensional morphology data set includes real three-dimensional morphology data corresponding to each of the training lunar soil particles;

[0129] The training data set is obtained according to the two-dimensional image data set and the three-dimensional shape data set.

[0130] Specifically, Figure 4 As shown, in order to realize model training, this embodiment can pre-establish three different functional modules, namely, a lunar soil data set acquisition module, a two-dimensional image acquisition module, and a three-dimensional morphology acquisition module. The lunar soil data set acquisition module is used to obtain particles with a particle size range that meets the three-dimensional scanning by screening, drying, and other treatments on the lunar soil particles, and remove impurities therein to obtain a lunar soil particle set. The two-dimensional image acquisition module is used to photograph the lunar soil particle set to obtain a two-dimensional image. For example, the lunar soil particles can be photographed using a two-dimensional photographing device or a mobile phone. In order to ensure the comprehensiveness of the two-dimensional image data set, the photographed particles are randomly placed and then photographed. Each time the image is photographed, only one shooting angle is used to obtain a single-view image. The two-dimensional image data set obtained by the shooting is then pre-processed into a binary image to obtain a pre-processed two-dimensional image data set. The 3D shape acquisition module is used to: place the lunar soil particles in a fixed container, perform a 3D scan on the lunar soil particles in the fixed container through CT scanning, obtain a 3D shape aggregate of the lunar soil particles, and then use the shape intelligent recognition and segmentation algorithm to intelligently segment the 3D shape aggregate based on the threshold to obtain the 3D shape of a single particle, and summarize to obtain a 3D shape dataset. A training dataset is established through a 2D image dataset and a 3D shape dataset.

[0131] In one implementation, the method further includes:

[0132] The reconstructed three-dimensional morphological data of the lunar soil particles are visualized, wherein the reconstructed three-dimensional morphological data can be rotated and the background can be rendered.

[0133] Specifically, the reconstructed three-dimensional morphological data of lunar soil particles can reflect the three-dimensional structure and surface details of lunar soil particles. The reconstructed three-dimensional morphological data is imported into the preset visualization software to generate a visualized three-dimensional model of lunar soil particles for users to view. Secondly, the visualization software can also provide interactive functions. Users can freely rotate the three-dimensional model through input devices such as a mouse or a touch screen, so that users can view the morphological characteristics of lunar soil particles from different angles. In addition, in order to increase the visualization effect and immersion, a background can be added to the visualized three-dimensional scene. The background can be a solid color, gradient, starry sky, moon or other environmental images related to lunar soil particles. In addition, rendering technologies such as ambient occlusion and global illumination can also be supported to simulate the lighting conditions of the real world. Finally, according to user needs, quantitative information such as annotations and dimension lines can be added to the three-dimensional model, so that users can more intuitively know the size, shape and other geometric characteristics of lunar soil particles.

[0134] For example, the present embodiment can also pre-establish a fourth functional module, namely a three-dimensional morphology display module, for visualizing the reconstructed three-dimensional morphology data set, displaying the two-dimensional images of the lunar soil particles and the reconstructed three-dimensional morphology, the reconstructed three-dimensional morphology can be rotated, and the background can be rendered. The four functional modules are used to achieve two-dimensional to three-dimensional, end-to-end output.

[0135] The advantages of the present invention are:

[0136] 1. Through artificial intelligence algorithms, the morphology of three-dimensional particles can be obtained using only two-dimensional images, realizing the dimensional reconstruction of a single view. Not only can the reconstructed three-dimensional morphology of high-quality lunar soil particles be obtained conveniently and efficiently, but the acquisition cost is also greatly reduced. This solves the current problem of difficulty in obtaining the three-dimensional morphology of lunar soil particles.

[0137] 2. Studying the scanned data can fully utilize the value of the data and avoid the need to obtain the three-dimensional morphology of lunar soil particles through tedious and costly laboratory operations every time.

[0138] 3. In the face of the ever-increasing types of lunar soil particles in the learning process, a more comprehensive and richer lunar soil particle database can be established, and the model's reconstruction ability is also stronger, realizing a positive cycle. The amount of data in the two-dimensional photo data set of lunar soil particles is increasing day by day, which can effectively support the learning and development of the model.

[0139] Based on the above embodiments, the present invention also provides a three-dimensional morphology recognition and reconstruction system for lunar soil particles, such as Figure 5 As shown, the system comprises:

[0140] A particle acquisition module 01 is used to acquire a lunar soil particle set, wherein the lunar soil particle set includes a plurality of lunar soil particles;

[0141] An image acquisition module 02 is used to photograph the lunar soil particle set to obtain a two-dimensional image corresponding to the lunar soil particle set;

[0142] An image segmentation module 03 is used to segment the two-dimensional image to obtain particle images corresponding to each of the lunar soil particles;

[0143] The three-dimensional reconstruction module 04 is used to input the particle image of each lunar soil particle into the trained dimensionality-raising reconstruction model to obtain the reconstructed three-dimensional morphology data corresponding to the lunar soil particle; wherein the dimensionality-raising reconstruction model includes an encoder and a decoder; the encoder is used to extract a number of feature maps of different dimensions according to the input particle image; the decoder is used to perform three-dimensional deconvolution on each of the feature maps to obtain the reconstructed three-dimensional morphology data.

[0144] Based on the above embodiment, the present invention further provides a terminal, whose principle block diagram can be shown as follows: Figure 6 As shown. The terminal includes a processor, a memory, a network interface, and a display screen connected through a system bus. Among them, the processor of the terminal is used to provide computing and control capabilities. The memory of the terminal includes a non-volatile storage medium and an internal memory. The non-volatile storage medium stores an operating system and a computer program. The internal memory provides an environment for the operation of the operating system and the computer program in the non-volatile storage medium. The network interface of the terminal is used to communicate with an external terminal through a network connection. When the computer program is executed by the processor, a three-dimensional morphology recognition and reconstruction method of lunar soil particles is implemented. The display screen of the terminal can be a liquid crystal display screen or an electronic ink display screen.

[0145] Those skilled in the art will understand that Figure 6 The principle block diagram shown in the figure is only a block diagram of a partial structure related to the scheme of the present invention, and does not constitute a limitation on the terminal to which the scheme of the present invention is applied. The specific terminal may include more or fewer components than those shown in the figure, or combine certain components, or have a different arrangement of components.

[0146] In one implementation, the terminal stores one or more programs in its memory and is configured to be executed by one or more processors. The one or more programs include instructions for performing a three-dimensional morphology recognition and reconstruction method for lunar soil particles.

[0147] A person of ordinary skill in the art can understand that all or part of the processes in the above-mentioned embodiment method can be completed by instructing the relevant hardware through a computer program, and the computer program can be stored in a non-volatile computer-readable storage medium. When the computer program is executed, it can include the processes of the embodiments of the above-mentioned methods. Among them, any reference to memory, storage, database or other media used in the embodiments provided by the present invention can include non-volatile and / or volatile memory. Non-volatile memory may include read-only memory (ROM), programmable ROM (PROM), electrically programmable ROM (EPROM), electrically erasable programmable ROM (EEPROM) or flash memory. Volatile memory may include random access memory (RAM) or external cache memory. As an illustration and not limitation, RAM is available in many forms, such as static RAM (SRAM), dynamic RAM (DRAM), synchronous DRAM (SDRAM), double data rate SDRAM (DDRSDRAM), enhanced SDRAM (ESDRAM), synchronous link (Synchlink) DRAM (SLDRAM), memory bus (Rambus) direct RAM (RDRAM), direct memory bus dynamic RAM (DRDRAM), and memory bus dynamic RAM (RDRAM).

[0148] In summary, the present invention discloses a method, system, terminal and medium for three-dimensional morphology recognition and reconstruction of lunar soil particles. The method comprises: obtaining a set of lunar soil particles, wherein the set of lunar soil particles comprises a number of lunar soil particles; photographing the set of lunar soil particles to obtain a two-dimensional image corresponding to the set of lunar soil particles; segmenting the two-dimensional image to obtain particle images corresponding to each of the lunar soil particles; for each of the lunar soil particles, inputting the particle image of the lunar soil particle into a trained dimensional reconstruction model to obtain reconstructed three-dimensional morphological data corresponding to the lunar soil particle; wherein the dimensional reconstruction model comprises an encoder and a decoder; the encoder is used to extract a number of feature maps of different dimensions according to the input particle image; the decoder is used to perform three-dimensional deconvolution on each of the feature maps to obtain reconstructed three-dimensional morphological data. The present invention only requires one camera to obtain a two-dimensional image of the lunar soil particle set, and combined with the trained dimensional reconstruction model, it can quickly realize end-to-end dimensional reconstruction and obtain the reconstructed three-dimensional morphological data of the lunar soil particles, which greatly reduces the time and cost of obtaining three-dimensional morphological information of lunar soil particles, and is suitable for batch acquisition application scenarios.

[0149] It should be understood that the application of the present invention is not limited to the above examples. For ordinary technicians in this field, improvements or changes can be made based on the above description. All these improvements and changes should fall within the scope of protection of the claims attached to the present invention.

Claims

1. A method for recognizing and reconstructing the three-dimensional morphology of lunar soil particles, characterized in that: The method comprises: Acquire a lunar soil particle set, wherein the lunar soil particle set includes a plurality of lunar soil particles; photographing the lunar soil particle set to obtain a two-dimensional image corresponding to the lunar soil particle set; Segmenting the two-dimensional image to obtain particle images corresponding to the lunar soil particles; For each of the lunar soil particles, the particle image of the lunar soil particle is input into the trained up-dimensional reconstruction model to obtain the reconstructed three-dimensional morphological data corresponding to the lunar soil particle; wherein the up-dimensional reconstruction model includes an encoder and a decoder; the encoder is used to extract a plurality of feature maps of different dimensions according to the input particle image; the decoder is used to perform three-dimensional deconvolution on each of the feature maps to obtain the reconstructed three-dimensional morphological data; The encoder is specifically used for: The input particle image is subjected to feature extraction through a two-dimensional convolutional layer and a maximum pooling layer, so that the size of the particle image gradually decreases, while the feature dimension gradually increases, and several feature maps of different dimensions are obtained; The decoder is specifically used for: Generate a number of input voxels according to each of the feature maps, perform three-dimensional deconvolution on each of the input voxels through a three-dimensional deconvolution layer so that the voxel size gradually increases, and obtain a number of output voxels; Generate reconstructed three-dimensional shape data according to each of the output voxels; The method for generating the trained dimensionality-raising reconstruction model is as follows: Acquire a training data set, wherein the training data set includes a plurality of training samples and real labels corresponding to the training samples; the training samples are particle images, and the real labels are real three-dimensional morphology data corresponding to the particle images; Inputting one of the training samples into the dimension-raising reconstruction model to obtain reconstructed three-dimensional shape data corresponding to the training sample; Calculate a binary cross entropy loss value and a mean square error loss value according to the true label corresponding to the training sample and the reconstructed three-dimensional shape data; Determining a loss function value of the dimensionality-raising reconstruction model according to the binary cross entropy loss value and the mean square error loss value; Determine whether the loss function value meets the preset training requirements, and if not, update the dimension-raising reconstruction model according to the loss function value; Continue to execute the step of inputting one of the training samples into the dimension-raising reconstruction model until the loss function value meets the training requirement, thereby obtaining the trained dimension-raising reconstruction model.

2. The method for three-dimensional morphology recognition and reconstruction of lunar soil particles according to claim 1 is characterized in that: The step of segmenting the two-dimensional image to obtain particle images corresponding to the lunar soil particles includes: Converting the two-dimensional image into a grayscale image, and converting the grayscale image into a binary image by using a preset threshold; The binary image is segmented to obtain particle images corresponding to each of the lunar soil particles.

3. The method for three-dimensional morphology recognition and reconstruction of lunar soil particles according to claim 1, characterized in that: The step of obtaining a training data set includes: Acquire a preprocessed training lunar soil particle set; wherein the preprocessing includes impurity removal, screening, and drying; the training lunar soil particle set includes a plurality of training lunar soil particles; The training lunar soil particle set is photographed to obtain a two-dimensional image data set; wherein the two-dimensional image data set includes particle images corresponding to each of the training lunar soil particles; The training lunar soil particle set is placed in a fixed container for three-dimensional scanning to obtain a three-dimensional morphology aggregate; Performing particle segmentation on the three-dimensional morphology aggregate to obtain a three-dimensional morphology data set; wherein the three-dimensional morphology data set includes real three-dimensional morphology data corresponding to each of the training lunar soil particles; The training data set is obtained according to the two-dimensional image data set and the three-dimensional shape data set.

4. The method for three-dimensional morphology recognition and reconstruction of lunar soil particles according to claim 1, characterized in that: The method further comprises: The reconstructed three-dimensional morphological data of the lunar soil particles are visualized, wherein the reconstructed three-dimensional morphological data can be rotated and the background can be rendered.

5. A three-dimensional morphology recognition and reconstruction system for lunar soil particles, characterized in that: The system comprises: A particle acquisition module, used to acquire a lunar soil particle set, wherein the lunar soil particle set includes a plurality of lunar soil particles; An image acquisition module, used for photographing the lunar soil particle set to obtain a two-dimensional image corresponding to the lunar soil particle set; An image segmentation module is used to segment the two-dimensional image to obtain particle images corresponding to each of the lunar soil particles; A three-dimensional reconstruction module is used for inputting the particle image of each lunar soil particle into the trained up-dimensional reconstruction model to obtain the reconstructed three-dimensional shape data corresponding to the lunar soil particle; wherein the up-dimensional reconstruction model includes an encoder and a decoder; the encoder is used to extract a plurality of feature maps of different dimensions according to the input particle image; the decoder is used to perform three-dimensional deconvolution on each of the feature maps to obtain the reconstructed three-dimensional shape data; The encoder is specifically used for: The input particle image is subjected to feature extraction through a two-dimensional convolutional layer and a maximum pooling layer, so that the size of the particle image gradually decreases, while the feature dimension gradually increases, and several feature maps of different dimensions are obtained; The decoder is specifically used for: Generate a number of input voxels according to each of the feature maps, perform three-dimensional deconvolution on each of the input voxels through a three-dimensional deconvolution layer so that the voxel size gradually increases, and obtain a number of output voxels; Generate reconstructed three-dimensional shape data according to each of the output voxels; The method for generating the trained dimensionality-raising reconstruction model is as follows: Acquire a training data set, wherein the training data set includes a plurality of training samples and real labels corresponding to the training samples; the training samples are particle images, and the real labels are real three-dimensional morphology data corresponding to the particle images; Inputting one of the training samples into the dimension-raising reconstruction model to obtain reconstructed three-dimensional shape data corresponding to the training sample; Calculate a binary cross entropy loss value and a mean square error loss value according to the true label corresponding to the training sample and the reconstructed three-dimensional shape data; Determining a loss function value of the dimensionality-raising reconstruction model according to the binary cross entropy loss value and the mean square error loss value; Determine whether the loss function value meets the preset training requirements, and if not, update the dimension-raising reconstruction model according to the loss function value; Continue to execute the step of inputting one of the training samples into the dimension-raising reconstruction model until the loss function value meets the training requirement, thereby obtaining the trained dimension-raising reconstruction model.

6. A terminal, characterized in that: The terminal includes a memory and one or more processors; the memory stores one or more programs; the program contains instructions for executing the three-dimensional morphology recognition and reconstruction method of lunar soil particles as described in any one of claims 1-4; and the processor is used to execute the program.

7. A computer-readable storage medium having a plurality of instructions stored thereon, characterized in that: The instructions are suitable for being loaded and executed by a processor to implement the steps of the three-dimensional morphology recognition and reconstruction method of lunar soil particles as described in any of claims 1-4 above.

Images