Geological light slice scanning imaging method based on optical imaging and automatic machine platform

Abstract

The invention discloses a geological light slice scanning imaging method based on optical imaging and an automatic machine platform, and the method comprises the steps: collecting a geological light slice multi-view image through carrying optical equipment on the automatic machine platform, and carrying out the feature extraction and preliminary analysis through employing an improved mask region convolutional neural network; the method comprises the following steps: calculating corresponding points by using a multi-view stereoscopic vision algorithm in combination with geological light slice thickness and refractive index parameters, establishing a geometrical relationship model to obtain initial three-dimensional point cloud data, optimizing the point cloud data based on mineral component distribution parameters, and finally performing surface reconstruction and texture mapping to generate a high-precision three-dimensional scanning imaging result. According to the method, the algorithm and geological light slice parameters are deeply fused, the whole-process processing from image acquisition to imaging is realized, the precision and efficiency of geological light slice scanning imaging are effectively improved, and reliable data support is provided for geological analysis and research.

Description

TECHNICAL FIELD

[0001] The present application relates to the field of geological light slice scanning imaging, and particularly to a geological light slice scanning imaging method based on optical imaging and an automatic machine platform. BACKGROUND

[0002] In the field of geological research, a geological light slice is an important carrier for obtaining microstructure and composition information of a rock mineral, and the imaging quality thereof directly affects the accuracy and depth of geological analysis. With the expansion of geological exploration scale and the increasing demand for fine research, the traditional optical imaging and manual analysis methods have been difficult to meet the needs of mass sample processing and complex geological feature analysis, and it is urgent to introduce automatic and high-precision scanning imaging technology to realize rapid and accurate digital characterization of the geological light slice.

[0003] The existing geological light slice scanning imaging technology has two major defects. Firstly, the imaging precision is insufficient. In the collection process of conventional optical imaging equipment, the differences in the thickness and refractive index of the geological light slice cannot be fully considered, resulting in problems such as geometric distortion and detail blur in the image. In the data processing stage, the traditional algorithm is difficult to effectively extract key features such as mineral composition distribution and particle boundary, especially in the multi-mineral complex structure area, which is prone to feature confusion and information loss, resulting in deviation between the three-dimensional reconstruction result and the actual structure.

[0004] Secondly, the degree of intelligence is limited. The traditional method mainly uses a fixed parameter algorithm process, which cannot be adaptively adjusted according to the mineral composition, particle size and other characteristics of different geological light slices. The whole process from image collection to three-dimensional imaging lacks intelligent coordination mechanism, and there are many manual intervention links, which is difficult to realize automatic processing, resulting in low processing efficiency and being difficult to meet the rapid analysis needs of large-scale geological samples, and seriously restricting the efficiency and depth of geological research. SUMMARY

[0005] In order to overcome the shortcomings and deficiencies of the prior art, the present application provides a geological light slice scanning imaging method based on optical imaging and an automatic machine platform.

[0006] The technical scheme adopted by the present application is a geological light slice scanning imaging method based on optical imaging and an automatic machine platform, comprising the following steps:

[0007] Step S1: An optical imaging device carried by the automatic machine platform collects multi-directional images of the geological light slice at different angles and positions, controls the focal length, aperture and exposure time parameters of the optical imaging device during the collection process, and obtains a set of original image data containing different view angle information of the geological light slice;

[0008] Step S2: input the original image data set into the improved mask region convolutional neural network, use the network to perform feature extraction and preliminary analysis on the original image data, identify and strengthen the features of the geological thin section area in the original image through the cooperative action of the multiple convolutional layers, pooling layers and mask generation layers in the network, and obtain a preliminary processing image with feature labels;

[0009] Step S3: using the multi-view stereo vision algorithm, according to the thickness and refractive index parameters of the geological thin section, the preliminary processing images at different viewing angles are matched and corresponding point calculation is performed, a geometric relationship model between the images is established, and preliminary three-dimensional point cloud data is obtained;

[0010] Step S4: based on the mineral composition distribution and particle size parameters of the geological thin section, the preliminary three-dimensional point cloud data is optimized, the effective information and noise information in the point cloud are further identified through the improved mask region convolutional neural network, the noise points are removed, and the effective points reflecting the true structure characteristics of the geological thin section are retained, forming optimized three-dimensional point cloud data.

[0011] Step S5: input the optimized three-dimensional point cloud data into the imaging module of the automatic machine platform, combine the color and transparency parameters of the geological thin section, use the multi-view stereo vision algorithm for surface reconstruction and texture mapping, and generate a three-dimensional scanning imaging result of the geological thin section with real appearance and structure characteristics;

[0012] Step S6: the generated three-dimensional scanning imaging result of the geological thin section is converted into a data format, a standardized data file is generated according to the mineral composition and structure characteristic parameters of the geological thin section, the data is stored using distributed storage technology, an index directory associated with the geological parameters is established, and a standardized data interface is output for external calling;

[0013] Further, the step S2, the improved mask region convolutional neural network uses the following formula for feature extraction and preliminary analysis: Wherein, F out represents the output preliminary processing image with feature labels; σ is the activation function; n is the number of layers of the convolutional neural network; W i is the weight matrix of the i-th layer; f i is the convolution operation function of the i-th layer; I in is the input original image data; b is the bias term; M is a mask matrix generated according to the mineral particle distribution characteristics of the geological thin section, the mask matrix element value is set according to the boundary and distribution of different mineral particles in the geological thin section, and is used to strengthen the feature extraction of the geological thin section area.

[0014] Further, the step S3, the multi-view stereo vision algorithm establishes a geometric relationship model between the images based on the following formula: wherein D ij represents the geometric distance metric value between the corresponding points in image i and image j; a is a proportional coefficient set according to the thickness of the geological light sheet, the thicker the geological light sheet, the larger the value of a; p i , p j are the coordinates of the corresponding points in image i and image j respectively; d max is a preset maximum distance threshold; β ij is a weight coefficient set according to the refractive index difference of the geological light sheet, used to correct the influence of image deformation on the calculation of geometric relationship due to different refractive indexes.

[0015] Further, in the step S4, when the preliminary three-dimensional point cloud data is optimized by using the improved mask region convolutional neural network, the following calculation process is adopted: according to the mineral composition distribution parameters of the geological light sheet, a mineral composition feature vector set V = {v1, v2, …, v m} is constructed, wherein m is the number of mineral composition types; for each point P(x, y, z) in the preliminary three-dimensional point cloud data, its feature vector f P is extracted by using the improved mask region convolutional neural network, the similarity S P between f i and each feature vector in V is calculated, S P = cos(f i , v i ); a similarity threshold T is set, if there is S i ≥ T, it is determined that the point is a valid point, otherwise it is regarded as a noise point and is removed.

[0016] Further, in the step S5, when the surface reconstruction is performed by using the multi-view stereo vision algorithm, the following model formula is adopted:

[0017]

[0018] wherein S represents the reconstructed geological light sheet surface model; is the surface model to be optimized; l is the number of images participating in the reconstruction; is the pixel value calculated from the kth image on the surface model ; is the pixel value of the kth original image; and are the gradients of the corresponding pixel values respectively; λ is a weight coefficient set according to the transparency of the geological light sheet, the higher the transparency, the smaller the value of λ, used to balance the influence of pixel value difference and gradient difference on the surface reconstruction.

[0019] Further, in the step S5, when the texture mapping is performed, the following formula is adopted: wherein T(x, y, z) is a texture value of a point with coordinates (x, y, z) in the three-dimensional scanning imaging result; q is the number of images used for texture mapping; w s is a weight coefficient set according to the color distribution uniformity of the geological light slice; the more uniform the color distribution, the more similar the values of each w s I s (u s , v s ) is a pixel value corresponding to the texture coordinates (u s , v s ) in the s-th image.

[0020] Further, the step S3 comprises the following sub-steps:

[0021] Step S3-1: According to the thickness parameter of the geological light slice, the interlayer correspondence relationship of the preliminary processed images of different viewing angles is analyzed to determine the overlapping area and the non-overlapping area between adjacent image layers;

[0022] Step S3-2: According to the refractive index parameter of the geological light slice, the geometric deformation correction is performed on the preliminary processed images to eliminate the image distortion caused by the difference in refractive index, so that the structure of the geological light slice in the image is closer to the real morphology;

[0023] Step S3-3: Using the matching module of the multi-view stereo vision algorithm, the corresponding points between different viewing angle images are found through the feature matching algorithm in the preliminary processed images after the geometric deformation correction, and a preliminary point correspondence relationship set is established;

[0024] Step S3-4: Based on the established point correspondence relationship set, the spatial geometric distance and angle relationship between the corresponding points are calculated in combination with the size parameter of the geological light slice, and preliminary three-dimensional point cloud data are generated.

[0025] Further, the step S4 comprises the following sub-steps:

[0026] Step S4-1: The preliminary three-dimensional point cloud data are divided according to the mineral composition distribution area of the geological light slice to form a plurality of sub-point cloud data sets;

[0027] Step S4-2: Each sub-point cloud data set is input into the improved mask region convolutional neural network, and the network is used to extract features of the points in each sub-point cloud data to obtain a feature vector reflecting the mineral characteristics of the region where the points are located;

[0028] Step S4-3: According to the particle size parameter of the geological light slice, a density threshold of the point cloud data is set, and the density of each sub-point cloud data is analyzed to eliminate points with density not meeting the requirements;

[0029] Step S4-4: The processed sub-point cloud data sets are integrated to form optimized three-dimensional point cloud data.

[0030] Further, the step S5 comprises the following sub-steps:

[0031] Step S5-1: Based on the color parameters of the geological light sheet, the optimized three-dimensional point cloud data is color classified, and points with similar color characteristics are divided into the same color region;

[0032] Step S5-2: Using the surface reconstruction module of the multi-view stereo vision algorithm, the three-dimensional point cloud data after color classification is surface meshed according to the transparency parameters of the geological light sheet, and a preliminary surface model is constructed;

[0033] Step S5-3: According to the surface roughness parameters of the geological light sheet, the preliminary surface model is smoothed to remove the uneven surface caused by point cloud data errors, so that the surface model is more consistent with the actual surface morphology of the geological light sheet;

[0034] Step S5-4: According to the texture feature parameters of the geological light sheet, the processed multi-image texture information is mapped to the smoothed surface model to generate a three-dimensional scanning imaging result of the geological light sheet with real appearance and structural characteristics.

[0035] Beneficial effects: The present application proposes a geological light sheet scanning imaging method based on optical imaging and automatic machine platform. In terms of imaging accuracy, in the image acquisition stage, the method controls the multi-parameters of the optical imaging equipment through the automatic machine platform to obtain multi-view original images, laying a foundation for high-precision imaging. In the processing process, the improved mask region convolutional neural network generates a mask matrix combined with the mineral particle distribution of the geological light sheet, strengthens feature extraction, and the multi-view stereo vision algorithm performs geometric relationship calculation and deformation correction according to parameters such as sheet thickness and refractive index. The optimized three-dimensional point cloud data processing and surface reconstruction process are also closely related to the mineral composition and transparency of the sheet, greatly improving the ability to capture and restore complex structures and small details, effectively avoiding feature confusion and information loss, and making the three-dimensional imaging result closer to the real structure. In terms of intelligence, the method realizes the full-process automation and adaptive processing from image acquisition to imaging output. The combination of the automatic machine platform and the improved algorithm can automatically adjust the parameters and processing flow according to the characteristics of different geological light sheets, without the need for excessive human intervention. Not only does it significantly improve processing efficiency, but it also meets the demand for rapid analysis of large-scale geological samples, providing more efficient and accurate data support for geological research and promoting the development of geological analysis towards intelligence and refinement. BRIEF DESCRIPTION OF DRAWINGS

[0036] Figure 1 The method flowchart of the present application;

[0037] Figure 2 To realize the unit graph for the method of the application. DETAILED DESCRIPTION

[0038] It should be noted that the embodiments in the present application and the features in the embodiments can be combined with each other without conflict, and the present application will be further described in detail below in combination with the drawings and specific embodiments.

[0039] As shown in the figure, the geological light slice scanning imaging method based on optical imaging and automatic machine platform includes the following steps: Figure 1 Step S1: through the optical imaging equipment carried by the automatic machine platform, multi-directional image acquisition is performed on the geological light slice at different angles and positions, and in the acquisition process, the focal length, aperture and exposure time parameters of the optical imaging equipment are controlled to obtain a set of original image data containing different view angle information of the geological light slice;

[0040] Specifically, in the initial stage of the geological light slice scanning imaging, step S1 performs the multi-directional image acquisition task through the optical imaging equipment carried by the automatic machine platform. The automatic machine platform has precise motion control capability and can realize flexible movement at multiple angles and positions in space to change the relative position relationship between the optical imaging equipment and the geological light slice. In the acquisition process, the key parameters of the optical imaging equipment need to be regulated, in which the focal length determines the clear range of imaging, by adjusting the focal length, it can be ensured that different layers of the geological light slice can be clearly imaged; the aperture controls the amount of light, and reasonable setting of the aperture size is helpful to obtain images with appropriate brightness under different lighting conditions; the exposure time affects the light and dark degree of the image, and according to the light reflection characteristics of the geological light slice, the exposure time is adjusted to avoid over-bright or over-dark images. Through the coordinated adjustment of these parameters, a set of original image data containing rich view angle information of the geological light slice can be obtained, and these data record the surface features and internal structure information of the geological light slice from multiple angles, which provides a comprehensive data basis for subsequent processing.

[0041]

[0042] ​The implementation of this step is to use the automatic control program of the automatic machine platform to sequentially collect images of the geological thin section according to the preset collection path and parameter combination. The collection path can be a circumferential path around the geological thin section, or multiple positions according to a specific grid distribution for shooting, ensuring that all key areas of the geological thin section are covered. The significance of this multi-azimuth collection method is to overcome the limitations of traditional single-view imaging. The multi-angle images obtained can complement each other, fully present the spatial form and structural details of the geological thin section, provide sufficient data support for subsequent algorithm-based three-dimensional model construction, make the final imaging result closer to the real appearance of the geological thin section, and improve the accuracy and integrity of the imaging.

[0043] Step S2: input the original image data set into the improved mask region convolutional neural network, use the network to extract features and preliminarily analyze the original image data, identify and strengthen the features of the geological thin section region in the original image through the synergistic effect of the multiple convolutional layers, pooling layers and mask generation layers in the network, and obtain a preliminarily processed image with feature markers;

[0044] Specifically, step S2 inputs the original image data set obtained in S1 into the improved mask region convolutional neural network. The network is improved based on the traditional mask region convolutional neural network and is optimized for the characteristics of the geological thin section image. The network contains multiple convolutional layers, pooling layers and mask generation layers, which cooperate with each other to extract features and preliminarily analyze the image. The convolutional layer slides on the image through different convolutional kernels to extract different scale and type features such as edges and textures. The pooling layer reduces the dimension of the feature map output by the convolutional layer to retain key features, reduce data volume and improve computational efficiency. The mask generation layer generates a corresponding mask matrix according to the mineral particle distribution and other characteristics of the geological thin section. The matrix can highlight the geological thin section region, suppress background information and strengthen feature extraction of the target region. Through the synergistic effect of these layers, the network can accurately identify the geological thin section region in the original image and enhance its features, and finally output a preliminarily processed image with feature markers.

[0045] The implementation process of this step is to input the original image data into the improved mask region convolutional neural network after format conversion and preprocessing according to the input requirements of the network. The network calculates through forward propagation and sequentially processes each layer to finally output the preliminary processed image. The significance lies in extracting valuable feature information from the original image for subsequent analysis, strengthening the structure and mineral characteristics of the geological thin section, removing irrelevant background interference, and making the geological thin section features in the image more prominent and obvious. This not only provides high-quality input data for subsequent image matching and three-dimensional reconstruction operations, but also improves the accuracy and efficiency of processing, reduces the processing difficulty caused by complex original image information, and lays the foundation for constructing an accurate three-dimensional model of the geological thin section.

[0046] Step S3: For the preliminary processed image, use the multi-view stereo vision algorithm to match and calculate the corresponding points of the preliminary processed images at different angles according to the thickness and refractive index parameters of the geological thin section, establish a geometric relationship model between the images, and obtain preliminary three-dimensional point cloud data.

[0047] Specifically, step S3 uses the multi-view stereo vision algorithm to obtain preliminary three-dimensional point cloud data for the preliminary processed image obtained in S2. The multi-view stereo vision algorithm is based on the principle of geometric optics and analyzes and processes preliminary processed images at different angles according to the thickness and refractive index parameters of the geological thin section. The thickness parameter of the geological thin section affects the relative position relationship of different layers in the image. Through the thickness information, the depth information of each point in the image in the three-dimensional space can be calculated. The refractive index parameter causes the light to refract when passing through the geological thin section, which in turn causes geometric distortion of the image. The algorithm uses the refractive index parameter to correct the image and eliminate errors caused by distortion. The algorithm matches the feature points in the images at different angles, calculates the geometric relationship between these corresponding points, establishes a geometric relationship model between the images, determines the coordinate position of each point in the three-dimensional space, and finally generates preliminary three-dimensional point cloud data. These point cloud data roughly outline the three-dimensional contour and structure of the geological thin section in the form of discrete points.

[0048] The implementation of this step is to input the preliminary processed image into a multi-view stereo vision algorithm program. The algorithm first pre-processes the image according to parameters such as the thickness and refractive index of the geological light sheet, and corrects the geometric deformation of the image. Then, a feature matching algorithm such as a feature descriptor-based matching method is used to find corresponding feature points in images at different viewing angles. Then, the three-dimensional coordinates of the corresponding points are calculated through principles such as triangulation, and three-dimensional point cloud data is gradually constructed. The significance of this step is to convert two-dimensional image information into three-dimensional point cloud data, realize the preliminary construction from a plane image to a three-dimensional structure, and make the spatial form of the geological light sheet present in a digital form. The preliminary three-dimensional point cloud data provides a basic framework for subsequent model optimization and surface reconstruction. Although its accuracy may be insufficient, it points the way for further construction of an accurate three-dimensional model and is a key transitional step in realizing three-dimensional imaging of the geological light sheet.

[0049] Step S4: Based on the mineral composition distribution and particle size parameters of the geological light sheet, the preliminary three-dimensional point cloud data is optimized. An improved mask region convolutional neural network is used to further identify effective information and noise information in the point cloud, eliminate noise points, and retain effective points reflecting the true structural characteristics of the geological light sheet, forming optimized three-dimensional point cloud data.

[0050] Specifically, step S4 optimizes the preliminary three-dimensional point cloud data obtained in S3 based on the mineral composition distribution and particle size parameters of the geological light sheet. Different mineral compositions have unique physical and chemical properties and exhibit different characteristics in images. By analyzing the mineral composition distribution parameters of the geological light sheet, the distribution of each mineral in the sheet can be understood. The particle size parameter reflects the size information of the mineral particles, which is crucial for determining effective points and noise points in the point cloud data. The improved mask region convolutional neural network plays a role again. It can identify the mineral characteristics corresponding to each point in the point cloud. According to the pre-set mineral composition characteristics and particle size standards, it determines whether the point is an effective point reflecting the true structural characteristics of the geological light sheet. Points that do not meet the standards, i.e. noise points, are eliminated, and the remaining effective points form the optimized three-dimensional point cloud data, which more accurately reflects the actual structure of the geological light sheet.

[0051] The implementation of this step is to combine the preliminary three-dimensional point cloud data with the parameter information of the mineral composition distribution and particle size of the geological light sheet, and input them into the improved mask region convolutional neural network for analysis and processing. The network extracts and analyzes the features of each point, compares them with the preset mineral characteristics and particle standards, and distinguishes between effective points and noise points. The significance lies in removing noise interference in the preliminary three-dimensional point cloud data and improving the quality and accuracy of the point cloud data. The optimized three-dimensional point cloud data can more truly restore the internal structure and mineral distribution of the geological light sheet, avoid model errors caused by noise points, provide more reliable data for subsequent surface reconstruction and texture mapping, and make the final generated three-dimensional scanning imaging result more accurately reflect the actual morphology and characteristics of the geological light sheet, thereby improving the quality and application value of imaging.

[0052] Step S5: input the optimized three-dimensional point cloud data into the imaging module of the automatic machine platform, combine the color and transparency parameters of the geological light sheet, use the multi-view stereo vision algorithm for surface reconstruction and texture mapping, and generate a three-dimensional scanning imaging result of the geological light sheet with real appearance and structural characteristics;

[0053] Specifically, step S5 inputs the optimized three-dimensional point cloud data of step S4 into the imaging module of the automatic machine platform, combines the color, transparency and other parameters of the geological light sheet, uses the multi-view stereo vision algorithm for surface reconstruction and texture mapping, and generates the final three-dimensional scanning imaging result of the geological light sheet. The color parameters of the geological light sheet reflect the differences in mineral composition and chemical composition, and different colors correspond to different minerals or mineral combinations; the transparency parameter affects the degree of light penetration and determines the visibility of the internal structure of the sheet. In the surface reconstruction process, the multi-view stereo vision algorithm processes the three-dimensional point cloud data according to these parameters to construct a continuous surface model, so that the discrete point cloud data forms a complete three-dimensional object surface; in the texture mapping link, the algorithm maps the texture information extracted from the original image to the surface model according to the color information of the geological light sheet, giving the model a real appearance texture, and finally generating a three-dimensional scanning imaging result of the geological light sheet with real appearance and structural characteristics.

[0054] The implementation of this step is in the imaging module of the automatic machine platform, using the surface reconstruction and texture mapping functions of the multi-view stereo vision algorithm to fuse the optimized three-dimensional point cloud data with the color, transparency and other parameters of the geological light sheet. First, the three-dimensional point cloud data is constructed into a triangular mesh or other form of surface model through a surface reconstruction algorithm, such as a mesh-based reconstruction method; then, according to the color and transparency parameters, the corresponding texture information is extracted from the original image and accurately mapped to the corresponding position of the surface model. The significance of this is to convert abstract three-dimensional point cloud data into intuitive and realistic three-dimensional visualization models, allowing geologists to more clearly and comprehensively observe the structure and characteristics of the geological light sheet. The generated three-dimensional scanning imaging results not only have precise geometric structures, but also have realistic appearance textures, which can provide intuitive and effective data support for geological analysis and research, and help to understand the microstructure and formation process of the geological light sheet.

[0055] Step S6: The generated three-dimensional scanning imaging results of the geological light sheet are converted into data formats, and standardized data files are generated according to the mineral composition and structural characteristic parameters of the geological light sheet. Distributed storage technology is used to store the data, and an index directory associated with the geological parameters is established, and a standardized data interface is output for external calling.

[0056] Specifically, step S6 is the last link of the entire scanning imaging method, responsible for storing and outputting the three-dimensional scanning imaging results of the geological light sheet generated in S5. In terms of storage, to ensure the safety and accessibility of data, a professional data storage system is used to store the imaging results. The storage system will perform format conversion and compression processing on the data to make it comply with specific storage standards and specifications, reduce the storage space occupied by the data, and facilitate subsequent data management and retrieval. During storage, the data will also be backed up to prevent damage or loss of the imaging results due to hardware failure, data loss, etc. In terms of output, according to different application requirements and usage scenarios, the three-dimensional scanning imaging results are output in various forms, such as three-dimensional model files, image files, data reports, etc. The output results can be directly used in geological analysis software for further research and processing, or can be visualized through display devices for observation and analysis by geologists.

[0057] The implementation of this step is to process the three-dimensional scanning imaging results by the data management program, store the data into the designated storage device or storage system according to the preset storage format and backup strategy; at the same time, according to the output requirements of the user, the corresponding output interface and conversion tool are called to convert the imaging results into the required format for output. The significance lies in properly saving and effectively utilizing the generated three-dimensional scanning imaging results, so that the results of the complex scanning and processing work in the early stage can be continued and applied. The stored data provides reliable historical data for subsequent geological research, which facilitates researchers to check and analyze at any time; the diversified output form meets the needs of different users and application scenarios, promotes the exchange and sharing of geological research results, and helps to promote the development and progress of geological science research.

[0058] Preferably, in the step S2, the improved mask region convolutional neural network adopts the following formula for feature extraction and preliminary analysis: Wherein, F out represents the output preliminary processing image with feature labels; σ is an activation function; n is the number of layers of the convolutional neural network; W i is the weight matrix of the i-th layer; f i is the convolution operation function of the i-th layer; I in is the input original image data; b is a bias term; and M is a mask matrix generated according to the mineral particle distribution characteristics of the geological thin section, the element values of which are set according to the boundaries and distribution of different mineral particles in the geological thin section, and is used to strengthen the feature extraction of the region of the geological thin section.

[0059] Specifically, in step S2, the improved mask region convolutional neural network realizes deep feature extraction of the original image data by constructing a specific model. In this model, the network multi-layer structure cooperates to perform convolution operation and feature transformation on the input image, and the feature mapping is obtained after the activation function processing. The key lies in the mask matrix generated according to the mineral particle distribution of the geological thin section, which sets the element values based on the mineral particle boundaries and distribution, strengthens the feature response of the region of the geological thin section in the network operation, suppresses the background interference, and makes the output preliminary processing image highlight the target region features, providing a high-quality feature enhancement image for subsequent processing, effectively improving the pertinence and accuracy of feature extraction.

[0060] Preferably, in the step S3, the multi-view stereo vision algorithm establishes a geometric relationship model between images based on the following formula: Wherein, D ij represents the geometric distance measurement value between corresponding points in image i and image j; α is a proportional coefficient set according to the thickness of the geological thin section, the thicker the geological thin section, the larger the value of α; p i , p jare the corresponding point coordinates in image i and image j respectively; d max is a preset maximum distance threshold; β ij is a weight coefficient set according to the refractive index difference of the geological light slice, used to correct the influence of image deformation caused by the refractive index difference on the calculation of geometric relationship.

[0061] Specifically, for step S3, the multi-view stereo vision algorithm accurately processes the geometric relationship between the images of different angles by establishing a geometric relationship model formula. In the formula, a proportionality coefficient is set according to the thickness of the geological light slice, and the greater the thickness, the stronger the adjustment of the coefficient; a weight coefficient is set according to the refractive index difference, used to compensate for the image deformation error caused by the refractive index. In implementation, the geometric distance measurement value of the corresponding points of the images is calculated by the formula, so as to determine the matching relationship between the images and provide a quantitative basis for establishing a reliable geometric model, realizing accurate conversion from two-dimensional images to three-dimensional space relationship, and ensuring the accuracy of the preliminary three-dimensional point cloud data acquisition.

[0062] Preferably, in step S4, when the improved mask region convolutional neural network is used to optimize the preliminary three-dimensional point cloud data, the following calculation process is adopted: first, according to the mineral composition distribution parameters of the geological light slice, a mineral composition feature vector set V = {v1, v2, …, vm} is constructed, where m is the number of mineral composition types; then, for each point P(x, y, z) in the preliminary three-dimensional point cloud data, its feature vector f m is extracted by the improved mask region convolutional neural network, the similarity S P between f P and each feature vector in V is calculated, S i = cos(f P , v i ); finally, a similarity threshold T is set, if there is S i ≥ T, it is determined that the point is a valid point, otherwise it is regarded as a noise point and is removed.

[0063] Specifically, in the implementation of the three-dimensional point cloud data optimization of step S4, a feature vector set is first constructed according to the mineral composition distribution of the geological light slice, and then a feature vector is extracted for each point in the preliminary three-dimensional point cloud data by using the improved mask region convolutional neural network. The validity of the point is determined by calculating the similarity between the point feature vector and the mineral composition feature vector and comparing it with the set threshold. This process closely surrounds the mineral composition and particle size parameters, realizes intelligent screening of the preliminary point cloud data, removes noise points, retains valid points that truly reflect the structure of the slice, and greatly improves the quality of the three-dimensional point cloud data, laying a solid data foundation for subsequent high-precision modeling.

[0064] Preferably, in step S5, when the multi-view stereo vision algorithm performs surface reconstruction, the following model formula is adopted:

[0065]

[0066] wherein S represents a reconstructed geological light slice surface model; is a surface model to be optimized; and l is the number of images participating in reconstruction; is a pixel value calculated from the kth image on the surface model S; is a pixel value of the kth original image; is a pixel value of the kth original image; and respectively represent gradients of the corresponding pixel values; and λ is a weight coefficient set according to the transparency of the geological light slice, and the higher the transparency, the smaller the value of λ, which is used to balance the influence of the pixel value difference and the gradient difference on the surface reconstruction.

[0067] Specifically, in the surface reconstruction link of step S5, the model formula of weight 5 optimizes the geological light slice surface model by balancing the pixel value difference and the gradient difference. In the formula, the weight coefficient is set according to the transparency of the geological light slice, and the gradient difference weight is reduced when the transparency is high, emphasizing the consistency of the pixel values. In implementation, by continuously adjusting the surface model to be optimized, the total sum of differences calculated by the formula is minimized, so that the reconstructed surface model is highly consistent with the original image in terms of pixel value and structural gradient, and a preliminary surface model conforming to the actual morphology is constructed, providing an accurate geometric basis for texture mapping and final three-dimensional imaging.

[0068] Preferably, in step S5, when texture mapping is performed, the following formula is used: wherein T(x, y, z) is the texture value of a point with coordinates (x, y, z) in the three-dimensional scanning imaging result; q is the number of images used for texture mapping; w s is a weight coefficient set according to the color distribution uniformity of the geological light slice, and the more uniform the color distribution, the closer the values of each w s ; I s (u s , v s ) is the pixel value corresponding to the texture coordinates (u s , v s ) in the s th image.

[0069] Specifically, in the texture mapping of step S5, the weight coefficient is set according to the color distribution uniformity of the geological light slice by the formula, and when the color distribution is uniform, the contribution weight of each image texture is similar. In implementation, for a point in the three-dimensional scanning imaging result, the pixel values of the corresponding texture coordinates of multiple images are integrated according to the formula to calculate the texture value of the point, and the texture information of multiple-view images is reasonably fused and mapped to the surface of the three-dimensional model, so that the generated three-dimensional imaging result not only has accurate geometric structure, but also presents realistic appearance texture, enhancing the visualization effect and the value of geological analysis.

[0070] Preferably, the step S3 comprises the following sub-steps:

[0071] Step S3-1: According to the thickness parameter of the geological light section, the interlayer correspondence relationship of the preliminary processed images at different viewing angles is analyzed to determine the overlapping and non-overlapping areas between adjacent image layers, thereby providing a basis for subsequent corresponding point calculation;

[0072] Step S3-2: According to the refractive index parameter of the geological light section, the preliminary processed images are geometrically deformed and corrected to eliminate the image distortion caused by the difference in refractive index, so that the structure of the geological light section in the image is closer to the real morphology;

[0073] Step S3-3: Using the matching module of the multi-view stereo vision algorithm, the corresponding points between the images at different viewing angles are found in the preliminary processed images after geometric deformation correction through a feature matching algorithm, and a preliminary point correspondence relationship set is established;

[0074] Step S3-4: Based on the established point correspondence relationship set, the spatial geometric distance and angle relationship between the corresponding points are calculated in combination with the size parameter of the geological light section, and preliminary three-dimensional point cloud data is generated.

[0075] Preferably, the step S4 comprises the following sub-steps:

[0076] Step S4-1: The preliminary three-dimensional point cloud data is divided according to the mineral composition distribution area of the geological light section to form a plurality of sub-point cloud data sets, thereby facilitating the corresponding processing of different mineral composition areas;

[0077] Step S4-2: Each sub-point cloud data set is input into the improved mask region convolutional neural network, and the network is used to extract features of the points in each sub-point cloud data to obtain a feature vector reflecting the mineral characteristics of the region where the points are located;

[0078] Step S4-3: According to the particle size parameter of the geological light section, a density threshold of the point cloud data is set, and the density of each sub-point cloud data is analyzed to eliminate points with density not meeting the requirements and retain effective structure points;

[0079] Step S4-4: The processed sub-point cloud data sets are integrated to form optimized three-dimensional point cloud data to accurately reflect the real structure characteristics of the geological light section.

[0080] Preferably, the step S5 comprises the following sub-steps:

[0081] Step S5-1: Based on the color parameter of the geological light section, the optimized three-dimensional point cloud data is color classified, and points with similar color characteristics are divided into the same color region to provide a basis for texture mapping;

[0082] Step S5-2: using the surface reconstruction module of the multi-view stereo vision algorithm, according to the transparency parameter of the geological light sheet, the color- classified three-dimensional point cloud data is subjected to surface meshing processing, and a preliminary surface model is constructed;

[0083] Step S5-3: according to the surface roughness parameter of the geological light sheet, the preliminary surface model is subjected to smoothing processing, the uneven surface caused by point cloud data error is removed, and the surface model is more in line with the actual surface morphology of the geological light sheet;

[0084] Step S5-4: according to the texture feature parameter of the geological light sheet, the texture information of the processed multiple images is mapped to the smoothed surface model, and a three-dimensional scanning imaging result of the geological light sheet with real appearance and structural characteristics is generated.

[0085] As shown in Figure 2 The geological light sheet scanning imaging method based on optical imaging and automatic machine platform is implemented through the following units, which include:

[0086] An image acquisition control unit is configured to control the optical imaging device carried by the automatic machine platform to acquire multi-directional images of the geological light sheet at different angles and positions, and adjust the focal length, aperture and exposure time parameters of the optical imaging device.

[0087] An image feature processing unit is connected with the image acquisition control unit and configured to input the collected raw image data set into the improved mask region convolutional neural network for feature extraction and preliminary analysis, so as to obtain preliminary processed images with feature labels.

[0088] A three-dimensional point cloud generation unit is connected with the image feature processing unit and configured to use a multi-view stereo vision algorithm to calculate corresponding points of the preliminary processed images according to the thickness and refractive index parameters of the geological light sheet, establish a geometric relationship model between the images, and obtain preliminary three-dimensional point cloud data.

[0089] A point cloud optimization unit is connected with the three-dimensional point cloud generation unit and configured to use the improved mask region convolutional neural network to optimize the preliminary three-dimensional point cloud data based on the mineral composition distribution and particle size parameters of the geological light sheet, so as to form optimized three-dimensional point cloud data.

[0090] A three-dimensional imaging unit is connected with the point cloud optimization unit and configured to use the multi-view stereo vision algorithm to perform surface reconstruction and texture mapping on the optimized three-dimensional point cloud data in combination with the color and transparency parameters of the geological light sheet, so as to generate a three-dimensional scanning imaging result of the geological light sheet.

[0091] The storage output unit is connected with the three-dimensional imaging unit, and is used for storing and outputting the generated three-dimensional scanning imaging result of the geological light thin section.

[0092] The present application realizes breakthrough through multi-link cooperative optimization. In the image acquisition link, the automatic machine platform accurately regulates and controls the focal length, aperture and other parameters of the optical equipment, and collects original images from multiple perspectives, providing a rich data basis for high-precision imaging. In the processing stage, the improved mask region convolutional neural network generates a mask matrix according to the mineral particle distribution of the geological light thin section, strengthens feature extraction, and avoids missing key information; the multi-view stereo vision algorithm combines parameters such as thin section thickness and refractive index to correct the geometric deformation of the image, accurately calculate the corresponding point relationship, and establish a reliable geometric model. In the process of three-dimensional point cloud optimization and surface reconstruction, further combined with parameters such as mineral composition and transparency, noise points are removed, and the surface morphology is optimized, so that the final imaging highly restores the real structure, solving the problem of blurred details and confused features in the traditional method.

[0093] In terms of intelligence and processing efficiency, the disadvantages of fixed traditional technical process and frequent manual intervention are effectively improved. The present method realizes full-process automation and self-adaptive processing, and the automatic machine platform and the improved algorithm cooperate deeply, which can automatically adjust the image acquisition parameters and data processing flow according to the characteristics such as mineral composition and particle size of different geological light thin sections. From image acquisition, feature extraction, to three-dimensional point cloud processing and imaging output, each step is automatically optimized based on the parameters of the geological light thin section, without frequent manual adjustment. At the same time, through the establishment of multi-level index directory and the generation of multi-resolution data copies, efficient storage and rapid calling of data are realized, the processing efficiency is greatly improved, the rapid analysis demand of large-scale geological samples is met, and the development of geological research towards intelligence and high efficiency is promoted.

[0094] In the description of the present application, it should be pointed out that, unless otherwise explicitly specified and limited, the terms "arrangement", "installation", "connection", "link", "fixing" should be understood broadly, for example, it can be fixed connection, or detachable connection, or integrally connected; it can be mechanical connection, or electrical connection; it can be directly connected, or indirectly connected through intermediate medium, or the communication inside two elements. For ordinary skilled in the art, the specific meaning of the above terms in the present application can be understood through specific circumstances.

[0095] Although the embodiments of the present application have been shown and described, it can be understood by those skilled in the art that various equivalent changes, modifications, replacements and variations can be made to these embodiments without departing from the principles and spirits of the present application, and the scope of the present application is defined by the appended claims and their equivalent scope.

Claims

1. A geological light-sheet scanning imaging method based on optical imaging and an automatic machine platform, characterized in that: The following steps are involved: Step S1: Using the optical imaging device carried by the automatic machine platform, multi-directional images of the geological light slice are collected at different angles and positions. During the collection process, the focal length, aperture and exposure time parameters of the optical imaging device are controlled to obtain a set of original image data containing information of the geological light slice at different viewing angles; Step S2: The original image data set is input into the improved mask region convolutional neural network, and the network is used to extract features and perform preliminary analysis on the original image data. Through the synergistic effect of multiple convolutional layers, pooling layers, and mask generation layers in the network, the geological light slice areas in the original image are identified and their features are enhanced, and a preliminary processed image with feature labels is obtained; Step S3: Using a multi-view stereo vision algorithm, the preliminarily processed images from different perspectives are matched and corresponding points are calculated based on the thickness and refractive index parameters of the geological light slices. A geometric relationship model between the images is established to obtain preliminary 3D point cloud data. Step S4: Based on the mineral composition distribution and particle size parameters of the geological thin section, the preliminary three-dimensional point cloud data is optimized and processed. The effective information and noise information in the point cloud are further identified through the improved mask region convolutional neural network, the noise points are eliminated, and the effective points reflecting the true structural characteristics of the geological thin section are retained to form the optimized three-dimensional point cloud data.

2. The geological light slice scanning imaging method based on optical imaging and an automatic machine platform according to claim 1 is characterized in that: The method further includes: Step S5: Input the optimized 3D point cloud data into the imaging module of the automatic machine platform, combine the color and transparency parameters of the geological light slice, and use the multi-view stereo vision algorithm to perform surface reconstruction and texture mapping to generate a 3D scanning imaging result of the geological light slice with realistic appearance and structural characteristics; Step S6: Convert the generated geological thin-section 3D scanning imaging results into a data format, generate standardized data files based on the mineral composition and structural characteristic parameters of the geological thin section, use distributed storage technology to store the data, establish an index directory associated with the geological parameters, and output a standardized data interface for external calls; In step S2, the improved mask region convolutional neural network uses the following formula to perform feature extraction and preliminary analysis: Among them, F out represents the output of the preliminary processed image with feature labels; σ is the activation function; n is the number of layers of the convolutional neural network; W i is the weight matrix of the i-th layer; f i is the convolution operation function of the i-th layer; I in is the input original image data; b is the bias term; M is the mask matrix generated according to the mineral grain distribution characteristics of the geological thin section. The element values ​​of the mask matrix are set according to the boundaries and distribution of different mineral grains in the geological thin section, and are used to enhance the feature extraction of the geological thin section area.

3. The geological light slice scanning imaging method based on optical imaging and an automatic machine platform according to claim 1, characterized in that: In step S3, the multi-view stereo vision algorithm establishes a geometric relationship model between images based on the following formula: Among them, D ij represents the geometric distance measurement between the corresponding points in image i and image j; α is the proportional coefficient set according to the thickness of the geological light slice. The thicker the geological light slice, the larger the value of α; p i 、p j are the coordinates of the corresponding points in image i and image j respectively; d max is the preset maximum distance threshold; β ij It is a weight coefficient set according to the refractive index difference of geological light slices, which is used to correct the influence of image deformation caused by different refractive indices on the calculation of geometric relationships.

4. The geological light slice scanning imaging method based on optical imaging and an automatic machine platform according to claim 1, characterized in that: In step S4, when optimizing the preliminary three-dimensional point cloud data using the improved mask region convolutional neural network, the following calculation process is adopted: according to the mineral composition distribution parameters of the geological light slice, a mineral composition feature vector set V = {v1, v2, ..., v m }, where m is the number of mineral components; For each point P(x, y, z) in the preliminary 3D point cloud data, its feature vector f is extracted by the improved mask region convolutional neural network P , calculate f P Similarity S with each eigenvector in V i =cos(f P , v i ); Set the similarity threshold T, if there is S i ≥T, the point is considered a valid point, otherwise it is considered a noise point and removed.

5. The geological light slice scanning imaging method based on optical imaging and an automatic machine platform according to claim 2, characterized in that: In step S5, when the multi-view stereo vision algorithm performs surface reconstruction, the following model formula is used: Where S represents the reconstructed geological light slice surface model; is the surface model to be optimized; l is the number of images involved in reconstruction; For the surface model The pixel value calculated from the k-th image; is the pixel value of the kth original image; and are the gradients of the corresponding pixel values; λ is the weight coefficient set according to the transparency of the geological light slice. The higher the transparency, the smaller the value of λ, which is used to balance the influence of pixel value difference and gradient difference on surface reconstruction.

6. The geological light slice scanning imaging method based on optical imaging and an automatic machine platform according to claim 2, characterized in that: In step S5, when performing texture mapping, the following formula is used: Where T(x, y, z) is the texture value of the point with coordinates (x, y, z) in the three-dimensional scanning imaging result; q is the number of images used for texture mapping; w s is the weight coefficient set according to the uniformity of color distribution of geological light slices. The more uniform the color distribution is, the better the w s The closer the values ​​are, the s (u s , v s ) is the corresponding texture coordinate (u s , v s )’s pixel value.

7. The geological light slice scanning imaging method based on optical imaging and an automatic machine platform according to claim 1 is characterized in that: The step S3 includes the following sub-steps: Step S3-1: Based on the thickness parameters of the geological light slice, perform inter-layer correspondence analysis on the preliminary processed images of different viewing angles to determine the overlapping areas and non-overlapping areas between adjacent image layers; Step S3-2: Based on the refractive index parameters of the geological light slice, geometric distortion correction is performed on the preliminarily processed image to eliminate the image distortion caused by the refractive index difference, so that the geological light slice structure in the image is closer to the real form; Step S3-3: Using the matching module of the multi-view stereo vision algorithm, in the preliminary processed image that has been corrected for geometric distortion, a feature matching algorithm is used to find corresponding points between images from different perspectives, and a preliminary set of point correspondence relationships is established; Step S3-4: Based on the established point correspondence set and combined with the size parameters of the geological light slice, the spatial geometric distance and angle relationship between the corresponding points are calculated to generate preliminary three-dimensional point cloud data.

8. The geological light slice scanning imaging method based on optical imaging and an automatic machine platform according to claim 7, characterized in that: The step S4 includes the following sub-steps: Step S4-1: Divide the preliminary 3D point cloud data according to the mineral composition distribution areas of the geological thin sections to form multiple sub-point cloud data sets; Step S4-2: Input each sub-point cloud data set into the improved mask region convolutional neural network, use the network to extract features of the points in each sub-point cloud data, and obtain feature vectors reflecting the mineral characteristics of the area where the points are located; Step S4-3: according to the particle size parameters of the geological light slice, the density threshold of the point cloud data is set, and the density of each sub-point cloud data is analyzed to eliminate points that do not meet the density requirements; Step S4-4: Integrate the processed sub-point cloud data sets to form optimized three-dimensional point cloud data.

9. The geological light slice scanning imaging method based on optical imaging and an automatic machine platform according to claim 2, characterized in that: The step S5 includes the following sub-steps: Step S5-1: Based on the color parameters of the geological light slice, the optimized three-dimensional point cloud data is color-classified, and points with similar color characteristics are divided into the same color area; Step S5-2: Using the surface reconstruction module of the multi-view stereo vision algorithm, the color-classified 3D point cloud data is subjected to surface meshing processing according to the transparency parameters of the geological light slice to construct a preliminary surface model; Step S5-3: Smoothing the preliminary surface model based on the surface roughness parameters of the geological slice to remove surface unevenness caused by point cloud data errors, so that the surface model is more consistent with the actual surface morphology of the geological slice; Step S5-4: Mapping the processed texture information of the multiple images onto the smoothed surface model based on the texture feature parameters of the geological light slice to generate a three-dimensional scanning imaging result of the geological light slice with realistic appearance and structural features.

Images