Mineral recognition method, training method and device of deep learning semantic segmentation model

Abstract

The present disclosure provides a mineral identification method, a training method and device of a deep learning semantic segmentation model, comprising: first obtaining a backscattered electron image and energy spectrum data of a to-be-identified rock chip; then fusing the backscattered electron image and the energy spectrum data based on super-resolution processing, an affine transformation model and resampling processing to generate a fused image of the to-be-identified rock chip, the affine transformation model representing a spatial mapping relationship from an energy spectrum data coordinate system to a backscattered electron image coordinate system. Then input a target image obtained based on the fused image into a deep learning semantic segmentation model to obtain a mineral identification result of the to-be-identified rock chip, the mineral identification result indicating a mineral type and a mineral distribution area of the to-be-identified rock chip. The method solves the problem of low accuracy when identifying minerals based on a backscattered electron image.

Description

Technical Field

[0001] This disclosure relates to the field of mineral identification technology, and in particular to a mineral identification method, a training method and device for a deep learning semantic segmentation model. Background Technology

[0002] Backscattered electron imaging technology has unique advantages in mineral identification. Backscattered electron images not only provide high-contrast mineral composition information, but also accurately reflect the spatial distribution of minerals. This technology provides important technical support for quantitative mineralogy research.

[0003] In existing technologies, the mineral types in backscattered electron images are typically determined manually, which can easily lead to inconsistent interpretations due to different subjective standards. Furthermore, automated techniques based on backscattered electron images usually utilize the grayscale differences reflected in the image for image segmentation to delineate mineral regions, but this method is prone to misjudgment due to grayscale overlap.

[0004] Therefore, it is evident that existing technologies for mineral identification based on backscattered electron images suffer from low accuracy. Summary of the Invention

[0005] In view of this, this disclosure provides a mineral identification method, a training method and device for a deep learning semantic segmentation model.

[0006] One aspect of this disclosure provides a mineral identification method, comprising: acquiring a backscattered electron image and energy spectrum data of rock fragments to be identified; fusing the backscattered electron image and energy spectrum data based on super-resolution processing, an affine transformation model, and resampling processing to generate a fused image of the rock fragments to be identified, wherein the affine transformation model represents the spatial mapping relationship from the energy spectrum data coordinate system to the backscattered electron image coordinate system; and inputting the target image obtained based on the fused image into a deep learning semantic segmentation model to obtain a mineral identification result of the rock fragments to be identified, wherein the mineral identification result indicates the mineral type and mineral distribution area of ​​the rock fragments to be identified.

[0007] According to embodiments of this disclosure, a fused image of rock debris to be identified is generated by fusing backscattered electron images and energy spectrum data based on super-resolution processing, an affine transformation model, and resampling processing. This includes: performing super-resolution processing on the energy spectrum data to obtain super-resolution energy spectrum data, wherein the spatial resolution of the super-resolution energy spectrum data matches the spatial resolution of the backscattered electron image; mapping the super-resolution energy spectrum data from the energy spectrum data coordinate system to the backscattered electron image coordinate system based on an affine transformation model to obtain an energy spectrum image; and performing resampling processing on the image obtained by superimposing the energy spectrum image and the backscattered electron image to obtain the fused image.

[0008] According to embodiments of this disclosure, the method further includes: obtaining the model input data size; determining a sliding step size for cropping the fused image based on the model input data size, wherein the sliding step size is less than or equal to the model input size; cropping the fused image based on the sliding step size, determining the cropped fused image as the target image, wherein the size of the target image matches the sliding step size; and inputting the target image into a deep learning semantic segmentation model to obtain the mineral identification result of the rock fragments to be identified.

[0009] According to embodiments of this disclosure, the method further includes: stitching together the target image and the mineral identification result corresponding to the target image to obtain a mineral distribution map, wherein the mineral distribution map indicates the mineral distribution of the rock fragments to be identified, and the size of the mineral distribution map matches the size of the fused image; and when the mineral distribution map indicates the presence of pixel isolation noise, performing an opening operation on the mineral distribution map to generate a denoised mineral distribution map.

[0010] Another aspect of this disclosure provides a training method for a deep learning semantic segmentation model, comprising: acquiring backscattered electron images and energy spectrum data of rock debris samples; determining the mineral type and mineral distribution area of ​​the rock debris samples based on the backscattered electron images and energy spectrum data; fusing the backscattered electron images and energy spectrum data samples based on super-resolution processing, an affine transformation model, and resampling processing to generate a fused image of the rock debris samples, wherein the affine transformation model represents the spatial mapping relationship from the energy spectrum data coordinate system to the backscattered electron image coordinate system; obtaining a target image sample based on the fused image, mineral type, and mineral distribution, and training the deep learning semantic segmentation model using the target image sample.

[0011] According to embodiments of this disclosure, determining the mineral types and mineral distribution areas of a rock fragment sample based on backscattered electron images and energy dispersive spectroscopy (EDS) data includes: using a first mapping relationship, determining the basic mineral type of the rock fragment sample based on the grayscale value and mineral morphology of the backscattered electron image, wherein the first mapping relationship includes the mapping relationship between the grayscale value of the backscattered electron image, the mineral morphology, and the mineral type, and the mineral morphology indicates the mineral shape, grain boundary characteristics, and spatial arrangement of the minerals; determining the basic mineral distribution of the rock fragment sample based on the position of the minerals corresponding to the basic mineral type in the backscattered electron image; and determining the mineral type and mineral distribution of the rock fragment sample based on the basic mineral type, the basic mineral distribution, and the EDS data.

[0012] According to embodiments of this disclosure, determining the mineral type and mineral distribution of a rock fragment sample based on the basic mineral type, basic mineral distribution, and energy dispersive spectroscopy (EDS) data includes: correcting the basic mineral type and basic mineral distribution based on the elemental distribution in the EDS data sample to determine the mineral type and mineral distribution of the rock fragment sample.

[0013] According to embodiments of this disclosure, a target image sample is obtained based on a fused image, mineral type, and mineral distribution, including: acquiring a target mineral and the coordinates of the center point of the target mineral in a backscattered electron image coordinate system, wherein the center point coordinates refer to the coordinates of the geometric center point of the target mineral; cropping the fused image sample according to a preset size and the center point coordinates to obtain a target image, wherein the size of the target image matches the preset size; and using the mineral type and mineral distribution area corresponding to the target mineral as labels, and using the labels and the target image to obtain a target image sample.

[0014] According to embodiments of this disclosure, obtaining a target image sample using a label and a target image includes: converting a tagged target image into a mask image, and using the mask image as a target image sample.

[0015] According to embodiments of the present disclosure, another aspect of the present disclosure provides an apparatus, comprising: an acquisition module for acquiring a backscattered electron image and energy spectrum data of rock fragments to be identified; a generation module for fusing the backscattered electron image and energy spectrum data based on super-resolution processing, an affine transformation model, and resampling processing to generate a fused image of the rock fragments to be identified, wherein the affine transformation model represents the spatial mapping relationship from the energy spectrum data coordinate system to the backscattered electron image coordinate system; and an output module for inputting a target image obtained based on the fused image into a deep learning semantic segmentation model to obtain a mineral identification result of the rock fragments to be identified, wherein the mineral identification result indicates the mineral type and mineral distribution area of ​​the rock fragments to be identified.

[0016] Another aspect of this disclosure provides an electronic device, including: one or more processors; and a storage device for storing one or more programs, wherein when the one or more programs are executed by the one or more processors, the one or more processors perform the method described above.

[0017] Another aspect of this disclosure provides a computer-readable storage medium having executable instructions stored thereon, which, when executed by a processor, cause the processor to perform the methods described above.

[0018] Another aspect of this disclosure provides a computer program product, including a computer program that, when executed by a processor, implements the above-described method.

[0019] The above one or more embodiments have the following technical effects: (1) For the training data of the deep learning semantic segmentation model, backscattered electron images and energy spectrum data are fused. The former contains grayscale information and morphological information, and the latter contains the elemental composition of minerals, thereby enriching the information of the training data and overcoming the problem of misjudgment caused by grayscale overlap; (2) For the generation of the fused image, with the help of super-resolution processing, affine transformation model and resampling processing, the backscattered electron images and energy spectrum data are fused on the same coordinate system, and the data information contained in the fused image is richer, thereby improving the accuracy of the deep learning semantic segmentation model in identifying minerals and solving the problem of low accuracy when identifying minerals based on backscattered electron images.

[0020] It should be understood that the description in this section is not intended to identify key or essential features of the embodiments of this disclosure, nor is it intended to limit the scope of this disclosure. Other features of this disclosure will become readily apparent from the following description. Attached Figure Description

[0021] The above and other objects, features, and advantages of this disclosure will become clearer from the following description of embodiments of the present disclosure with reference to the accompanying drawings, in which:

[0022] Figure 1 A flowchart illustrating a mineral identification method according to an embodiment of the present disclosure is shown schematically.

[0023] Figure 2 A flowchart illustrating a training method for a deep learning semantic segmentation model according to an embodiment of the present disclosure is shown schematically.

[0024] Figure 3 A schematic diagram of a backscattered electron image according to an embodiment of the present disclosure is shown;

[0025] Figure 4 A schematic diagram of energy spectrum data according to an embodiment of the present disclosure is shown;

[0026] Figure 5 A schematic diagram of a fused image according to an embodiment of the present disclosure is shown;

[0027] Figure 6a This illustration schematically shows the contents of a configuration file according to an embodiment of the present disclosure. Figure 1 ;

[0028] Figure 6b This illustration schematically shows the contents of a configuration file according to an embodiment of the present disclosure. Figure 2 ;

[0029] Figure 7a This illustration schematically depicts a mineral identification process according to an embodiment of the present disclosure. Figure 1 ;

[0030] Figure 7b This illustration schematically depicts a mineral identification process according to an embodiment of the present disclosure. Figure 2 ;

[0031] Figure 7c This illustration schematically depicts a mineral identification process according to an embodiment of the present disclosure. Figure 3 ;

[0032] Figure 7d This illustration schematically depicts a mineral identification process according to an embodiment of the present disclosure. Figure 4 ;

[0033] Figure 7e This illustration schematically depicts a mineral identification process according to an embodiment of the present disclosure. Figure 5 ;

[0034] Figure 7f A schematic diagram of a mineral identification process according to an embodiment of the present disclosure is shown in Figure 6.

[0035] Figure 8 A schematic block diagram of a device according to an embodiment of the present disclosure is shown.

[0036] Figure 9 A schematic block diagram of an electronic device that can be used to implement the methods of embodiments of the present disclosure is shown. Detailed Implementation

[0037] The embodiments of this disclosure will now be described with reference to the accompanying drawings. Various details of the embodiments of this disclosure are included to aid understanding and should be considered merely exemplary. Therefore, those skilled in the art will recognize that various changes and modifications can be made to the embodiments described herein without departing from the scope and spirit of this disclosure. Similarly, for clarity and brevity, descriptions of well-known functions and structures are omitted in the following description.

[0038] In the technical solutions disclosed herein, the collection, storage, use, processing, transmission, provision, disclosure, and application of data (including but not limited to user personal information) comply with the provisions of relevant laws and regulations, necessary confidentiality measures have been taken, and they do not violate public order and good morals.

[0039] First, let's explain the terminology used in the embodiments of this application:

[0040] Backscattered electron image: This refers to the imaging mode acquired by a backscattered electron detector in a scanning electron microscope. For minerals, this image provides both grayscale information and reflects the spatial distribution characteristics of the minerals at the microscopic scale.

[0041] Energy dispersive spectroscopy (EDS) data is a technique that measures and analyzes the energy distribution emitted, absorbed, or scattered by a substance. EDS data for minerals represents their chemical composition characteristics, including the relative abundance or counts of various chemical elements at different spatial locations.

[0042] Super-resolution processing is a technique that improves the resolution of an image through hardware or software methods, enabling the recovery or reconstruction of a high-resolution image from a low-resolution image.

[0043] Resampling is a technique that performs secondary sampling based on the original sampled data.

[0044] Bicubic interpolation is a commonly used interpolation algorithm in two-dimensional space. For the value of a pixel in the target image, it is obtained by weighting the gray values ​​(or component concentration values) of its 16 neighboring pixels (i.e., a 4×4 window) in the original image using a cubic polynomial. This calculation involves two steps: first, four cubic interpolations are performed horizontally to obtain four intermediate values; then, based on these four intermediate values, a third cubic interpolation is performed vertically to finally obtain the value of the target pixel. The weighting function used for interpolation ensures the smoothness of the reconstructed curve. For example, the weighting function can be a Catmull-Rom cubic convolution kernel.

[0045] Bilinear interpolation is a numerical analysis method for interpolating two variables. It uses a weighted average of the values ​​of the four nearest neighbors (2×2 pixels) around the point to be calculated, achieving a good balance between computational complexity and result smoothness.

[0046] Opening is a fundamental operation in morphological image processing. It is defined as performing an erosion operation on the image first, followed by a dilation operation on the erosion result. This operation can effectively eliminate small isolated points, smooth object boundaries, and break narrow connections, while maintaining the area and shape of larger objects in the image essentially unchanged.

[0047] Figure 1 A flowchart illustrating a mineral identification method according to an embodiment of the present disclosure is shown schematically.

[0048] like Figure 1 As shown, the mineral identification method 100 of this embodiment includes operations S110 to S130.

[0049] During operation S110, backscattered electron images and energy spectrum data of the rock cuttings to be identified are acquired.

[0050] Among them, obtaining the backscattered electron image is based on the backscattered electron imaging mode of a scanning electron microscope. In practical applications, the scanning electron microscope observes the polished section on which the cuttings to be identified are placed. Before observation, the cuttings to be identified are cleaned, dried, embedded, polished, etc. to prepare a standard polished section of the cuttings to be identified.

[0051] Among them, obtaining the energy spectrum data is based on the energy spectrum analysis device equipped with the scanning electron microscope. In practical applications, the energy spectrum analysis device performs multi-point, dense, and full-coverage energy spectrum data acquisition on the same polished section of cuttings. When acquiring the energy spectrum data, the scanning step and sampling interval are set according to the size of the polished section and the image resolution, so that the acquired energy spectrum data covers the corresponding area of the entire backscattered electron image in space.

[0052] It should be understood that obtaining the backscattered electron image of the cuttings to be identified can obtain the gray value and morphological characteristics of the cuttings to be identified. And obtaining the energy spectrum data of the cuttings to be identified can obtain the element distribution of the cuttings to be identified. Because it is easy to have the problem of misjudgment caused by gray value overlap only by relying on the gray value and morphological characteristics, the energy spectrum data is further combined for judgment, providing data support for subsequent data fusion and mineral identification.

[0053] In operation S120, based on super-resolution processing, an affine transformation model, and resampling processing, the backscattered electron image and the energy spectrum data are fused to generate a fused image of the cuttings to be identified.

[0054] Among them, the affine transformation model represents the spatial mapping relationship from the energy spectrum data coordinate system to the backscattered electron image coordinate system.

[0055] In some exemplary embodiments, first, super-resolution processing is performed on the energy spectrum data to obtain super-resolved energy spectrum data. Then, based on the affine transformation model, the super-resolved energy spectrum data is mapped from the energy spectrum data coordinate system to the backscattered electron image coordinate system to obtain an energy spectrum image. Then, resampling processing is performed on the image obtained by superimposing the energy spectrum image and the backscattered electron image to obtain a fused image.

[0056] Among them, the spatial resolution of the super-resolved energy spectrum data matches the spatial resolution of the backscattered electron image. The spatial resolution represents the minimum distance between adjacent distinguishable sampling points.

[0057] It should be understood that in order to achieve the fusion of the energy spectrum data and the backscattered electron image, the problem of mismatched spatial resolutions between the two needs to be solved. Therefore, super-resolution processing is performed on the energy spectrum data to obtain super-resolved energy spectrum data with the same spatial resolution as the backscattered electron image.

[0058] For example, the bicubic interpolation algorithm is used to implement super-resolution processing. Specifically, in the elemental surface distribution map of a certain element in the original energy spectrum data, let the point The value at that location is According to the formula:

[0059]

[0060] in, This refers to the interpolated pixel value. It refers to the coordinates of points in the interpolated image. It refers to the coordinates of a point in the image before interpolation, where and It is usually not an integer.

[0061] in, This refers to the integer division operation. This refers to a cubic weighting function. In practical applications, a common form is: when hour, ;when hour, ;when For other values, .in, This is an adjustable parameter, typically set to -0.5 or -0.75.

[0062] It should be understood that by traversing each sampling point in the energy spectrum data through a bicubic interpolation algorithm, a high-density energy spectrum distribution map with the same spatial pixel size as the backscattered electron image is generated, which is an image composed of super-resolution energy spectrum data.

[0063] After obtaining the super-resolution energy spectrum data, for further fusion, the super-resolution energy spectrum data and the backscattered electron image can be placed on the same coordinate system. Specifically, mapping is performed based on an affine transformation model, which is used to describe linear deformations such as translation, rotation, scaling, and shearing (misalignment) between images.

[0064] In some embodiments, the affine transformation model can be expressed as: ,in, These are the coordinates of a point in the super-resolution energy dispersive spectroscopy data. The coordinates of the point mapped to the backscattered electron image coordinate system; Used for rotation, scaling, and cropping. and Used for translation.

[0065] In practical applications, identical physical feature points uniformly distributed within the field of view are selected as control point pairs in both super-resolution energy dispersive spectroscopy (EDS) data and backscattered electron images. To improve registration accuracy, it is recommended to prioritize sites with clear and sharp features, such as mineral grain edges, fracture endpoints, and junctions of different phase boundaries. The number of control point pairs should be appropriate to ensure robust estimation of the affine transformation model parameters.

[0066] To solve for the parameters in the affine transformation model and , will select The coordinates of the control points are substituted into the equations of the affine transformation model, forming an overdetermined linear system of equations. Subsequently, the least squares method is used for parameter estimation. This method aims to find a set of parameters that minimizes the sum of squared errors between the transformed coordinates of all control points and their actual coordinates on the backscattered electron image.

[0067] Specifically, for each pair of control points, two equations can be formulated. Collecting all control point pairs forms a matrix equation. ,in It is a column vector consisting of 6 parameters to be determined. Through calculation... This allows us to obtain the optimal affine transformation parameter solution.

[0068] It should be understood that after obtaining the solution of the affine transformation parameters, the affine transformation model is obtained. For each integer pixel position in the backscattered electron image coordinate system, the inverse affine transformation obtained above is used to calculate its corresponding (usually non-integer) position in the energy spectrum data coordinate system, and then the pixel value at that position is determined by interpolation.

[0069] In practical applications, the bilinear interpolation method is used. Specifically, let the points to be calculated be... The floating-point coordinates in the original image (i.e., the image obtained by superimposing the energy spectrum image and the backscattered electron image) are: The four integer coordinates surrounding it are The corresponding values ​​are respectively .but Point interpolation results The calculation is as follows:

[0070] First, perform two linear interpolations in the horizontal direction:

[0071]

[0072]

[0073] Then, a linear interpolation is performed in the vertical direction:

[0074]

[0075] in, This refers to the pixel values ​​of the interpolated and merged image.

[0076] It should be understood that super-resolution processing of energy spectrum data can make the energy spectrum data and backscattered electron image consistent at the pixel scale. Mapping the super-resolution processed energy spectrum data onto the coordinate system of the backscattered electron image facilitates further image fusion. Resampling processing can simultaneously acquire richer information from the image obtained by superimposing the energy spectrum image and the backscattered electron image. The fused image obtained based on this allows the data input to the deep learning semantic segmentation model to contain both energy spectrum data reflecting elemental composition and backscattered electron image data containing grayscale values ​​and morphological information, thus enriching the data information input to the model.

[0077] In operation S130, the target image obtained based on the fused image is input into the deep learning semantic segmentation model to obtain the mineral identification result of the rock fragments to be identified.

[0078] Among them, the mineral identification results indicate the mineral type and mineral distribution area of ​​the rock fragments to be identified.

[0079] The mineral distribution area refers to the region where the identified minerals are located. For example, the mineral types of the rock fragments to be identified include fir olivine, pyroxene, and quartz. The corresponding mineral distribution areas are: fir olivine at the geometric center of the section containing the rock fragments, pyroxene on the right side of the section containing the rock fragments, and quartz on the left side of the section containing the rock fragments.

[0080] In some exemplary embodiments, the model input data size is first obtained. Then, based on the model input data size, a sliding step size for cropping the fused image is determined, wherein the sliding step size is less than or equal to the model input size. The fused image is then cropped according to the sliding step size, and the cropped fused image is determined as the target image, with the target image's size matching the sliding step size. Finally, the target image is input into a deep learning semantic segmentation model to obtain the mineral identification result of the rock fragments to be identified.

[0081] Here, the model input data size refers to the size of the data input to the deep learning semantic segmentation model. Because the backscattered electron image is large enough that it cannot be input into the deep learning semantic segmentation model for inference all at once, the model is enabled to recognize the entire image using a sliding window approach. The sliding step size is the size of the sliding window. The target image is the fused image cropped using the sliding step size.

[0082] In some embodiments, the operation of a simulated scanner is used to slide across the fused image with a sliding step size, each time capturing an image patch that matches the size of the model input data, i.e., the target image, and then feeding the target image into a deep learning semantic segmentation model for recognition. The model outputs the mineral recognition result of the mineral type to which each pixel in the target image belongs.

[0083] By cropping the fused image, the large image size can be avoided from affecting the recognition of the deep learning semantic segmentation model. This approach accelerates the model's recognition efficiency and further improves the efficiency of mineral type recognition.

[0084] In some exemplary embodiments, the target image and the corresponding mineral identification result are first stitched together to obtain a mineral distribution map. If the mineral distribution map indicates the presence of isolated pixel noise, an opening operation is performed on the mineral distribution map to generate a denoised mineral distribution map.

[0085] The mineral distribution map indicates the mineral distribution of the rock fragments to be identified, and the size of the mineral distribution map matches the size of the fused image.

[0086] In practical applications, to avoid obvious boundaries or gaps during the stitching process, a certain overlap area needs to be set between sliding windows. For example, the overlap ratio can be set to 25%, meaning that adjacent windows have 25% of their pixel areas overlapping each other in both the horizontal and vertical directions. For pixels within the overlap area, since they are predicted multiple times by multiple windows, the final value is the average of the multiple prediction results, thereby smoothing the transition and reducing local inconsistencies that may be caused by a single prediction.

[0087] In addition, after stitching, the resulting mineral distribution map may contain several scattered, isolated noise points consisting of only a few pixels. These noise points usually do not conform to the actual morphological characteristics of the minerals and may originate from errors in model prediction or noise in the image itself. To remove these invalid areas and retain mineral regions with reasonable morphological structures, the "opening operation" from mathematical morphology can be used for processing.

[0088] Specifically, for a binary image (where the target mineral region is the foreground and the rest is the background), the mathematical expression of the opening operation is as follows: Let the input image be... The structural element is Then the opening operation can be expressed as .

[0089] Among them, corrosion operation Defined as: That is, all structural elements Translation It is still completely contained within points within A set. The expansion operation. Defined as: That is, all structural elements Reflection translation about the origin Later and Intersecting points The set. In practice, a binary mask can be generated for each type of mineral region, and an opening operation can be performed independently on each mask by selecting a structural element of appropriate shape and size (such as a circle or square with a radius of 3 pixels), thereby removing isolated pixel groups with too small an area and retaining mineral regions with continuous shapes.

[0090] Generating mineral distribution maps clarifies the distribution of minerals on the rock fragments to be identified, providing data support for subsequent mineral processing. Furthermore, the morphological processing of opening operations makes the mineral distribution map reflect the mineral distribution more clearly, improving the accuracy of mineral identification.

[0091] In some embodiments, the pixel area ratio of each mineral phase can be further calculated. Specifically, the total number of foreground pixels in the binary mask corresponding to each mineral category is calculated, and this value represents the pixel area of ​​the mineral in the image; the ratio of the pixel area of ​​each mineral to the total number of pixels in the entire image is the area ratio of the mineral, which can be used to quantitatively analyze the distribution and abundance of minerals in the sample.

[0092] According to embodiments of this application, backscattered electron images and energy spectrum data are fused based on super-resolution processing, affine transformation model, and resampling processing, so that the fused image provides input data rich in mineral information for a deep learning semantic segmentation model. This mineral information not only reflects the mineral elemental composition, but also reflects the mineral grayscale value and morphology, thereby improving the accuracy and efficiency of mineral identification.

[0093] Figure 2 A flowchart illustrating a training method for a deep learning semantic segmentation model according to an embodiment of the present disclosure is shown.

[0094] like Figure 2 As shown, the training method 200 of the deep learning semantic segmentation model in this embodiment includes operations S210 to S240.

[0095] During operation S210, backscattered electron images and energy spectrum data of rock cutting samples were acquired.

[0096] In operation S220, the mineral type and mineral distribution area of ​​the rock fragment sample are determined based on backscattered electron images and energy spectrum data.

[0097] In some exemplary embodiments, the basic mineral type of the rock fragment sample is first determined using a first mapping relationship based on the grayscale values ​​and mineral morphology of the backscattered electron image. Next, the basic mineral distribution of the rock fragment sample is determined based on the location of the minerals corresponding to the basic mineral type in the backscattered electron image. Finally, the mineral type and mineral distribution of the rock fragment sample are determined based on the basic mineral type, basic mineral distribution, and energy dispersive spectroscopy (EDS) data.

[0098] The first mapping relationship includes the mapping relationship between the gray value of the backscattered electron image, the mineral morphology and the mineral type. The mineral morphology indicates the mineral shape, grain boundary features and the spatial arrangement of the minerals.

[0099] Among them, basic mineral type refers to the mineral type initially determined based on backscattered electron images, and basic mineral distribution refers to the mineral distribution initially determined based on backscattered electron images.

[0100] It should be understood that the brightness of a pixel in a backscattered electron image directly corresponds to the average atomic number of the imaging region; therefore, different minerals will exhibit significant differences in grayscale. Bright areas in the image indicate minerals with high average atomic numbers, such as metal oxides or certain accessory minerals; while darker areas reflect minerals with low average atomic numbers, such as quartz or feldspar. Therefore, by analyzing the magnitude of grayscale values ​​to reflect the contrast between light and dark areas, different potential mineral regions can be identified.

[0101] Simultaneously, the geometric morphology of each region in the backscattered electron image, such as whether mineral grains exhibit specific crystal habits (minerals showing platy or columnar outlines may be feldspar, while platy morphology is usually mica), the clarity and smoothness of grain boundaries, and whether minerals appear as aggregates with specific structures are all important factors. By combining grayscale values, mineral shapes, grain boundary characteristics, and the spatial arrangement of minerals, a preliminary prediction can be made about the major mineral categories and their approximate distribution in rock fragments, serving as a basis for subsequent judgments.

[0102] In some exemplary embodiments, the basic mineral type and basic mineral distribution are corrected based on the elemental distribution in the energy dispersive spectroscopy data sample to determine the mineral type and mineral distribution of the rock fragment sample.

[0103] It should be understood that grayscale overlap can lead to misjudgment, so further judgment can be made by combining energy dispersive spectral data that reflects elemental distribution. In practical applications, the elemental distribution information is precisely correlated with the grayscale regions in the backscattered electron image. Special attention is paid to the distribution of major rock-forming elements such as silicon, aluminum, iron, calcium, magnesium, sodium, and potassium. By examining or comparing the distribution maps of different elements one by one, spatial enrichment regions of specific elements can be observed.

[0104] For example, if a bright area appears highly bright in a backscattered electron image and shows a strong signal in the iron element distribution map of the energy dispersive spectroscopy (EDS), then that area is very likely an iron mineral. Conversely, if a dark area in the image is rich in silicon and contains almost no other metallic elements, it strongly suggests that it is quartz. For the differentiation of feldspar minerals, it is necessary to observe the distribution patterns of elements such as silicon, aluminum, potassium, sodium, and calcium.

[0105] By cross-referencing and comprehensively interpreting image grayscale features and element spatial distribution data, it is possible to effectively distinguish various common minerals such as quartz, potassium feldspar, plagioclase, carbonate minerals, mica, and iron-titanium oxides, and ultimately accurately delineate the mineral distribution of various minerals in rock fragment samples, completing the analysis process from preliminary identification to final confirmation.

[0106] In operation S230, based on super-resolution processing, affine transformation model and resampling processing, backscattered electron images and energy spectrum data samples are fused to generate a fused image of rock debris samples.

[0107] The affine transformation model represents the spatial mapping relationship from the energy spectrum data coordinate system to the backscattered electron image coordinate system.

[0108] In operation S240, target image samples are obtained based on the fused image, mineral type, and mineral distribution, and the deep learning semantic segmentation model is trained using the target image samples.

[0109] In some exemplary embodiments, the target mineral and its center point coordinates in the backscattered electron image coordinate system are first obtained. Then, the fused image sample is cropped according to a preset size and center point coordinates to obtain the target image. Next, the mineral type and distribution area corresponding to the target mineral are used as labels, and the target image sample is obtained using the labels and the target image.

[0110] The center point coordinates refer to the coordinates of the geometric center point of the target mineral.

[0111] The target image size matches a preset size. The preset size is used to limit the image size input to the deep learning semantic segmentation model.

[0112] For example, a mineral of a certain type is selected as the target mineral, and the coordinates of the center point of the identified target mineral are used as the basis for identification. Based on this, the preset size is set to 512 pixels × 512 pixels. Calculate the coordinates of the starting point at the top left corner of the center point. This allows for the expansion and cropping of a rectangular region of interest with a strictly fixed size of 512 pixels × 512 pixels.

[0113] When cropping, ensure that the target mineral is located in the center or near the center of the rectangular area, and that it is surrounded by a sufficient number of symbiotic minerals as background information. If the target mineral is located at the edge of the image, preventing the capture of the entire area, the center point position should be adjusted or the sample should be recorded as a special case.

[0114] In practical applications, to prevent the model from identifying all regions as the target mineral during training, several center points should be randomly selected from areas of the image without the target mineral, and cropped to the same size to serve as negative samples for comparison. All cropped sample images are named and saved using a sequence number, which is a four-digit number (e.g., "0001"). The corresponding annotation files will use the exact same main filename to ensure accurate data pair association.

[0115] Meanwhile, the target images in the target image samples are labeled with mineral type and mineral distribution area. The annotation process is as follows: the target image display mode is adjusted to a suitable channel combination for clear boundary identification, and then points are plotted sequentially along the actual edge of the target mineral at an appropriate density to form a closed polygonal region. During the annotation process, it should be ensured that the polygon closely follows the mineral edge, avoiding including too much background or omitting some minerals. Then, the polygon is categorized, for example, the target mineral is labeled as "Class_1", and the background or other associated minerals can be labeled as "Class_0 (Background)", "Class_2", etc. All annotation results will be automatically saved as a JSON file with the same name.

[0116] In some exemplary embodiments, the labeled target image is converted into a mask image, and the mask image is used as a sample of the target image.

[0117] It should be understood that using the masked image as a target image sample is to facilitate the reading by the deep learning semantic segmentation model.

[0118] In practical applications, JSON-formatted annotations are uniformly converted into single-channel PNG mask images. This conversion process ensures that the category identifier for each pixel is correctly mapped. In the generated mask image, different pixel values ​​correspond to different categories. Typically, 0 represents the background, and 1, 2, 3… represent different categories of minerals, and the pixel value definitions for all images must be completely consistent. The original image (.png), the JSON annotation file, and the converted mask image (_mask.png) can be stored in three separate directories: "images", "json", and "mask" to maintain a clear and orderly file system.

[0119] According to embodiments of this application, the mineral type and distribution area determined based on backscattered electron images and energy dispersive spectroscopy (EDS) data are used as training labels for a deep learning semantic segmentation model. A fused image, integrating both backscattered electron images and EDS data, is used as training data for the deep learning semantic segmentation model. This improves the training accuracy of the deep learning semantic segmentation model by combining training labels and training data, thereby enhancing the model's accuracy in mineral identification.

[0120] In practical applications, to ensure the objectivity of deep learning semantic segmentation model evaluation, all data pairs consisting of samples and their masks must be completely and randomly shuffled. Then, all data is divided into three independent parts according to a preset ratio: a training set (approximately 70%), i.e., the rock debris samples in this embodiment, used for learning model parameters; a validation set (approximately 20%), used to evaluate model performance and adjust hyperparameters during training; and a test set (approximately 10%), i.e., the rock debris to be identified in this embodiment, used for final testing of the model's generalization performance after training. During the division, it should be ensured that samples of each category are evenly distributed across the datasets, and the resulting lists should be saved as text files (e.g., train.txt, val.txt, test.txt), recording the filenames to ensure experimental reproducibility. The entire dataset construction process is now complete, forming a standardized data resource with a clear structure, accurate annotations, and direct access by deep learning models.

[0121] For deep learning semantic segmentation models, this application uses the DeepLabv3+ semantic segmentation model for training. Its core lies in fusing multi-scale information through an encoder-decoder structure and capturing context using dilated convolution and atrous spatial pyramid pooling (ASPP) layers, thereby improving segmentation accuracy and robustness. The specific technical solution is as follows.

[0122] To improve model performance and training efficiency, this embodiment selects the Xception network as the backbone feature extraction network for DeepLabv3+. The Xception network employs depthwise separable convolutions, maintaining strong feature extraction capabilities while achieving high computational efficiency. At the start of training, instead of randomly initializing weights, pre-trained Xception weights on a large, general-purpose dataset (such as the COCO dataset) are loaded. This transfer learning strategy effectively utilizes the general image features (such as edges and textures) already learned by the pre-trained model, accelerating the model's convergence speed on specific mineral datasets and mitigating the potential problem of insufficient feature extraction capabilities due to limited training data.

[0123] Figure 3 A schematic diagram of a backscattered electron image according to an embodiment of the present disclosure is shown.

[0124] like Figure 3As shown, the different gray levels and morphologies of each mineral can be used to distinguish their approximate range and boundaries. Specifically, zrc represents zircon, Fa represents fayalite, pyx represents pyroxene, ilm represents ilmenite, kf represents potassium feldspar, qz represents quartz, MERR represents sodium phosphate rock, and REE represents rare earth elements.

[0125] Figure 4 A schematic diagram of energy spectrum data according to an embodiment of the present disclosure is shown.

[0126] like Figure 4 As shown, the image uses a backscattered electron image as the background, and the white dots represent the dot matrix of phosphorus.

[0127] Figure 5 A schematic diagram of a fused image according to an embodiment of the present disclosure is shown.

[0128] like Figure 5 As shown, the energy spectrum of this image uses the green band to represent the distribution of phosphorus. When the green band is bright, it indicates that the region is rich in phosphorus. At the same time, this image retains the grayscale and morphological information of the backscattered electron image.

[0129] Figure 6a This illustration schematically shows the contents of a configuration file according to an embodiment of the present disclosure. Figure 1 .

[0130] like Figure 6a As shown in the figure, this diagram contains some of the core content of the configuration file for mineral segmentation, including the parameter settings and descriptions for the dataset and model modules.

[0131] Figure 6b This illustration schematically shows the contents of a configuration file according to an embodiment of the present disclosure. Figure 2 .

[0132] like Figure 6b As shown in the figure, this diagram contains some of the core content of the configuration file for mineral segmentation, including the parameter settings and explanations for the training and optimization, and pre-training and saving modules.

[0133] In practical applications, the training process requires systematic configuration of dataset paths, model structures, training hyperparameters, and optimization strategies. The core monitoring metric for training is the Mean Intersection over Union (mIoU) calculated on the independent validation set. To prevent overfitting, a strategy should be configured to save only the model weights with the highest mIoU on the validation set (Best mIoU), rather than the final weights after each training epoch. Initial training can be set to around 100 epochs to observe the trend of loss reduction and accuracy increase. If the validation set accuracy fluctuates or decreases continuously in the later stages, overfitting may occur, requiring consideration of introducing early stopping or adjusting data augmentation strategies. If the accuracy continues to steadily increase, the total number of training epochs can be increased to 500 or even 1000 epochs to allow the model to fully converge. After training, the saved best model is used for final evaluation on the test set to obtain an objective indicator of the model's generalization performance. Furthermore, a visualization script can be used to visually view the model's segmentation effect on specific samples; the prediction results will be saved for analysis. After multiple rounds of iterative optimization, a high-precision deep learning semantic segmentation model with stable and good recognition performance for mineral images can finally be trained.

[0134] Figure 7a This illustration schematically depicts a mineral identification process according to an embodiment of the present disclosure. Figure 1 .

[0135] like Figure 7a As shown in the figure, this figure represents the data acquisition stage for mineral identification, corresponding to the data acquisition for mineral identification in embodiment 700 of this disclosure. The embodiment disclosed in this figure includes operations S701 to S703.

[0136] In operation S701, standard rock cuttings were prepared.

[0137] The S702 is used to acquire backscattered electron images.

[0138] Using the S703, acquire energy spectrum data.

[0139] Figure 7b This illustration schematically depicts a mineral identification process according to an embodiment of the present disclosure. Figure 2 .

[0140] like Figure 7b As shown in the figure, this figure represents the mineral type determination stage of mineral identification, corresponding to the mineral type determination stage of mineral identification in embodiment 700 of this disclosure. The embodiment disclosed in this figure includes operations S704 to S706.

[0141] When operating the S704, grayscale and morphological analysis are performed based on backscattered electron images.

[0142] When operating the S705, elemental analysis is performed based on energy spectrum data.

[0143] The S706 function is used to determine the mineral type.

[0144] Figure 7c This illustration schematically depicts a mineral identification process according to an embodiment of the present disclosure. Figure 3 .

[0145] like Figure 7c As shown in the figure, this figure represents the data preprocessing stage for mineral identification, corresponding to the data preprocessing for mineral identification in Embodiment 700 of this disclosure. The embodiment disclosed in this figure includes operations S707 to S709.

[0146] When operating the S707, super-resolution processing is performed on the energy spectrum data.

[0147] In operation of S708, the super-resolution processed data is mapped onto the backscattered electron image coordinate system and resampled.

[0148] In operation S709, the resampled data is fused to obtain a fused image.

[0149] Figure 7d This illustration schematically depicts a mineral identification process according to an embodiment of the present disclosure. Figure 4 .

[0150] like Figure 7d As shown in the figure, this figure represents the training set construction stage for mineral identification, corresponding to the training set construction for mineral identification in embodiment 700 of this disclosure. The embodiment disclosed in this figure includes operations S710 to S712.

[0151] The S710 is used to crop the merged image.

[0152] In operation S711, annotations are made based on mineral type and mineral distribution area.

[0153] When operating the S712, the dataset is divided into a training set, a validation set, and a test set.

[0154] Figure 7e This illustration schematically depicts a mineral identification process according to an embodiment of the present disclosure. Figure 5 .

[0155] like Figure 7e As shown in the figure, this figure represents the model training stage for mineral identification, corresponding to the mineral identification model training in embodiment 700 of this disclosure. The embodiment disclosed in this figure includes operations S713 to S715.

[0156] Using S713, build a DeepLabv3+ network.

[0157] When operating the S714, configure training parameters and perform iterative optimization.

[0158] When operating S715, save the optimal weight model.

[0159] Figure 7f A schematic diagram of a mineral identification process according to an embodiment of the present disclosure is shown in Figure 6.

[0160] like Figure 7f As shown in the figure, this figure represents the identification and analysis stage of mineral identification, corresponding to the identification and analysis of mineral identification in Embodiment 700 of this disclosure. The embodiment disclosed in this figure includes operations S716 to S718.

[0161] In operation S716, the target image and the corresponding mineral identification results are stitched together to obtain a mineral distribution map.

[0162] When operating S717, if the mineral distribution map indicates the presence of isolated pixel noise, an opening operation is performed on the mineral distribution map.

[0163] Using S718, a noise-reduced mineral distribution map is generated.

[0164] Figure 8 A schematic block diagram of a device according to an embodiment of the present disclosure is shown.

[0165] like Figure 8 As shown, the apparatus 800 of this embodiment includes...

[0166] The acquisition module 810 is used to acquire backscattered electron images and energy spectrum data of the rock cuttings to be identified. In one embodiment, the acquisition module 810 can be used to perform the operation S110 described above, which will not be repeated here.

[0167] The generation module 820 is used to fuse backscattered electron images and energy spectrum data based on super-resolution processing, an affine transformation model, and resampling processing to generate a fused image of the rock debris to be identified. The affine transformation model represents the spatial mapping relationship from the energy spectrum data coordinate system to the backscattered electron image coordinate system. In one embodiment, the generation module 820 can be used to perform the operation S120 described above, which will not be repeated here.

[0168] The output module 830 is used to input the target image obtained based on the fused image into the deep learning semantic segmentation model to obtain the mineral identification result of the rock fragments to be identified. The mineral identification result indicates the mineral type and mineral distribution area of ​​the rock fragments to be identified. In one embodiment, the output module 830 can be used to perform the operation S130 described above, which will not be repeated here.

[0169] Figure 9A schematic block diagram of an electronic device that can be used to implement the methods of embodiments of the present disclosure is shown.

[0170] like Figure 9 As shown, an electronic device 900 according to an embodiment of the present disclosure includes a processor 901, which can perform various appropriate actions and processes according to a program stored in a read-only memory (ROM) 902 or a program loaded from a storage portion 908 into a random access memory (RAM) 903. The processor 901 may include, for example, a general-purpose microprocessor (e.g., a CPU), an instruction set processor and / or an associated chipset and / or a special-purpose microprocessor (e.g., an application-specific integrated circuit (ASIC)), etc. The processor 901 may also include onboard memory for caching purposes. The processor 901 may include a single processing unit or multiple processing units for performing different actions of the method flow according to an embodiment of the present disclosure.

[0171] RAM 903 stores various programs and data required for the operation of electronic device 900. Processor 901, ROM 902, and RAM 903 are interconnected via bus 904. Processor 901 performs various operations of the method flow according to embodiments of this disclosure by executing programs in ROM 902 and / or RAM 903. It should be noted that programs may also be stored in one or more memories other than ROM 902 and RAM 903. Processor 901 may also implement the methods provided in embodiments of this disclosure by executing programs stored in one or more memories.

[0172] According to embodiments of this disclosure, the electronic device 900 may further include an input / output (I / O) interface 905, which is also connected to a bus 904. The electronic device 900 may also include one or more of the following components connected to the I / O interface 905: an input section 906 including a keyboard, mouse, etc.; an output section 907 including a cathode ray tube (CRT), liquid crystal display (LCD), etc., and a speaker, etc.; a storage section 908 including a hard disk, etc.; and a communication section 909 including a network interface card such as a LAN card, modem, etc. The communication section 909 performs communication processing via a network such as the Internet. A drive 910 is also connected to the I / O interface 905 as needed. A removable medium 911, such as a disk, optical disk, magneto-optical disk, semiconductor memory, etc., is installed on the drive 910 as needed so that computer programs read from it can be installed into the storage section 908 as needed.

[0173] This disclosure also provides a computer-readable storage medium, which may be included in the device / apparatus / system described in the above embodiments; or it may exist independently and not assembled into the device / apparatus / system. The computer-readable storage medium carries one or more programs that, when executed, implement the method according to the embodiments of this disclosure.

[0174] According to embodiments of this disclosure, the computer-readable storage medium may be a non-volatile computer-readable storage medium, such as including, but not limited to: portable computer disks, hard disks, random access memory (RAM), read-only memory (ROM), erasable programmable read-only memory (EPROM or flash memory), portable compact disk read-only memory (CD-ROM), optical storage devices, magnetic storage devices, or any suitable combination thereof. In this disclosure, the computer-readable storage medium may be any tangible medium that contains or stores a program that can be used by or in conjunction with an instruction execution system, apparatus, or device. For example, according to embodiments of this disclosure, the computer-readable storage medium may include ROM 902 and / or RAM 903 and / or one or more memories other than ROM 902 and RAM 903 described above.

[0175] Embodiments of this disclosure also include a computer program product comprising a computer program containing program code for performing the methods shown in the flowchart. When the computer program product is run on a computer system, the program code is used to cause the computer system to implement the methods provided in the embodiments of this disclosure.

[0176] When the computer program is executed by the processor 901, it performs the functions defined in the system / apparatus of this disclosure embodiments. According to embodiments of this disclosure, the systems, apparatuses, modules, units, etc., described above can be implemented by computer program modules.

[0177] In one embodiment, the computer program may rely on a tangible storage medium such as an optical storage device or a magnetic storage device. In another embodiment, the computer program may also be transmitted and distributed in the form of signals over a network medium, and downloaded and installed via the communication section 909, and / or installed from a removable medium 911. The program code contained in the computer program can be transmitted using any suitable network medium, including but not limited to: wireless, wired, etc., or any suitable combination thereof.

[0178] In such an embodiment, the computer program can be downloaded and installed from a network via the communication section 909, and / or installed from the removable medium 911. When the computer program is executed by the processor 901, it performs the functions defined in the system of this disclosure embodiment. According to embodiments of this disclosure, the systems, devices, apparatuses, modules, units, etc., described above can be implemented by computer program modules.

[0179] It should be noted that the collection, storage, use, processing, transmission, provision, disclosure, and application of user personal information in this disclosed technical solution comply with relevant laws and regulations, necessary confidentiality measures have been taken, and it does not violate public order and good morals. In this disclosed technical solution, user authorization or consent has been obtained before acquiring or collecting user personal information.

[0180] According to embodiments of this disclosure, program code for executing the computer programs provided in embodiments of this disclosure can be written in any combination of one or more programming languages. Specifically, these computational programs can be implemented using high-level procedural and / or object-oriented programming languages, and / or assembly / machine languages. Programming languages ​​include, but are not limited to, languages ​​such as Java, C++, Python, "C", or similar programming languages. The program code can execute entirely on a user's computing device, partially on a user's device, partially on a remote computing device, or entirely on a remote computing device or server. In cases involving remote computing devices, the remote computing device can be connected to the user's computing device via any type of network, including a local area network (LAN) or a wide area network (WAN), or it can be connected to an external computing device (e.g., via the Internet using an Internet service provider).

[0181] The flowcharts and block diagrams in the accompanying drawings illustrate the architecture, functionality, and operation of possible implementations of systems, methods, and computer program products according to various embodiments of this disclosure. In this regard, each block in a flowchart or block diagram may represent a module, segment, or portion of code containing one or more executable instructions for implementing a specified logical function. It should also be noted that in some alternative implementations, the functions indicated in the blocks may occur in a different order than those indicated in the drawings. For example, two consecutively indicated blocks may actually be executed substantially in parallel, and they may sometimes be executed in reverse order, depending on the functions involved. It should also be noted that each block in a block diagram or flowchart, and combinations of blocks in a block diagram or flowchart, may be implemented using a dedicated hardware-based system that performs the specified function or operation, or using a combination of dedicated hardware and computer instructions.

[0182] Those skilled in the art will understand that the features described in the various embodiments and / or claims of this disclosure can be combined and / or combined in various ways, even if such combinations or combinations are not explicitly described in this disclosure. In particular, the features described in the various embodiments and / or claims of this disclosure can be combined and / or combined in various ways without departing from the spirit and teachings of this disclosure. All such combinations and / or combinations fall within the scope of this disclosure. Embodiments of this disclosure have been described above. However, these embodiments are for illustrative purposes only and are not intended to limit the scope of this disclosure. Although various embodiments have been described above, this does not mean that the measures in the various embodiments cannot be used advantageously in combination. The scope of this disclosure is defined by the appended claims and their equivalents. Various substitutions and modifications can be made by those skilled in the art without departing from the scope of this disclosure, and all such substitutions and modifications should fall within the scope of this disclosure.

Claims

1. A mineral identification method, characterized in that, The method includes: Acquire backscattered electron images and energy spectrum data of the rock cuttings to be identified; Based on super-resolution processing, affine transformation model, and resampling processing, the backscattered electron image and the energy spectrum data are fused to generate a fused image of the rock debris to be identified. The affine transformation model represents the spatial mapping relationship from the energy spectrum data coordinate system to the backscattered electron image coordinate system. The target image obtained based on the fused image is input into a deep learning semantic segmentation model to obtain the mineral identification result of the rock fragment to be identified. The mineral identification result indicates the mineral type and mineral distribution area of ​​the rock fragment to be identified. The step of fusing the backscattered electron image and the energy spectrum data based on super-resolution processing, affine transformation model, and resampling processing to generate the fused image of the rock debris to be identified includes: The energy spectrum data is subjected to super-resolution processing to obtain super-resolution energy spectrum data, and the spatial resolution of the super-resolution energy spectrum data matches the spatial resolution of the backscattered electron image. Based on the affine transformation model, the super-resolution energy spectrum data is mapped from the energy spectrum data coordinate system to the backscattered electron image coordinate system to obtain the energy spectrum image; The image obtained by superimposing the energy spectrum image and the backscattered electron image is resampled to obtain the fused image.

2. The method according to claim 1, characterized in that, The method further includes: Obtain the model input data dimensions; Based on the model input data size, a sliding step size for cropping the fused image is determined, wherein the sliding step size is less than or equal to the model input size; The fused image is cropped according to the sliding step size, and the cropped fused image is determined as the target image. The size of the target image matches the sliding step size. The target image is input into a deep learning semantic segmentation model to obtain the mineral identification results of the rock fragments to be identified.

3. The method according to claim 2, characterized in that, The method further includes: The target image and the corresponding mineral identification results are stitched together to obtain a mineral distribution map. The mineral distribution map indicates the mineral distribution of the rock fragments to be identified, and the size of the mineral distribution map matches the size of the fused image. If the mineral distribution map indicates the presence of isolated pixel noise, an opening operation is performed on the mineral distribution map to generate a denoised mineral distribution map.

4. A training method for a deep learning semantic segmentation model, characterized in that, The method includes: Acquire backscattered electron images and energy dispersive spectroscopy data of rock cuttings samples; Based on the backscattered electron image and the energy dispersive spectroscopy data, the mineral type and mineral distribution area of ​​the rock fragment sample are determined. Based on super-resolution processing, affine transformation model, and resampling processing, the backscattered electron image and the energy spectrum data are fused to generate a fused image of the rock debris sample. The affine transformation model represents the spatial mapping relationship from the energy spectrum data coordinate system to the backscattered electron image coordinate system. Based on the fused image, the mineral type, and the mineral distribution, a target image sample is obtained, and the target image sample is used to train a deep learning semantic segmentation model. The step of fusing the backscattered electron image and the energy spectrum data based on super-resolution processing, affine transformation model, and resampling processing to generate a fused image of the rock debris sample includes: The energy spectrum data is subjected to super-resolution processing to obtain super-resolution energy spectrum data, and the spatial resolution of the super-resolution energy spectrum data matches the spatial resolution of the backscattered electron image. Based on the affine transformation model, the super-resolution energy spectrum data is mapped from the energy spectrum data coordinate system to the backscattered electron image coordinate system to obtain the energy spectrum image; The image obtained by superimposing the energy spectrum image and the backscattered electron image is resampled to obtain the fused image.

5. The method according to claim 4, characterized in that, The step of determining the mineral types contained in the rock fragment sample and the mineral distribution area of ​​the rock fragment sample based on the backscattered electron image and the energy spectrum data includes: Using a first mapping relationship, the basic mineral type of the rock fragment sample is determined based on the grayscale value and mineral morphology of the backscattered electron image. The first mapping relationship includes the mapping relationship between the grayscale value of the backscattered electron image, the mineral morphology, and the mineral type. The mineral morphology indicates the mineral shape, grain boundary features, and spatial arrangement of the minerals. The basic mineral distribution of the rock fragment sample is determined based on the position of the minerals corresponding to the basic mineral types in the backscattered electron image. Based on the basic mineral type, the basic mineral distribution, and the energy dispersive spectroscopy data, the mineral type and mineral distribution of the rock fragment sample are determined.

6. The method according to claim 5, characterized in that, The step of determining the mineral types and mineral distribution of the rock fragment sample based on the basic mineral type, the basic mineral distribution, and the energy dispersive spectroscopy data includes: Based on the elemental distribution in the energy dispersive spectroscopy data, the basic mineral type and the basic mineral distribution are corrected to determine the mineral type and mineral distribution of the rock fragment sample.

7. The method according to claim 4, characterized in that, The step of obtaining the target image sample based on the fused image, the mineral type, and the mineral distribution includes: Obtain the target mineral and the coordinates of the center point of the target mineral in the backscattered electron image coordinate system, wherein the center point coordinates refer to the coordinates of the geometric center point of the target mineral; The fused image sample is cropped according to the preset size and the center point coordinates to obtain the target image, and the size of the target image matches the preset size; The mineral type and distribution area corresponding to the target mineral are used as labels, and the target image sample is obtained using the labels and the target image.

8. The method according to claim 7, characterized in that, The step of obtaining the target image sample using the label and the target image includes: The target image with the added label is converted into a mask image, and the mask image is used as a sample of the target image.

9. An electronic device, comprising: One or more processors; Storage device for storing one or more programs. Wherein, when the one or more programs are executed by the one or more processors, the one or more processors perform the method according to any one of claims 1 to 8.

Images