A method for intelligent identification of heavy minerals based on multi-element image information

Abstract

This invention discloses an intelligent identification method for heavy minerals based on multi-source image information. The method involves separating heavy minerals to obtain single-polarized light images, orthogonal light images, and reflected light images; aligning these images to obtain single-polarized light panoramic images, orthogonal light panoramic images, and reflected light panoramic images; segmenting these panoramic images into particles to obtain segmented particle images; and using the segmented particle images and corresponding real labels as training data to obtain a multi-channel recognition model for identifying the particle images of the heavy minerals to be identified, thus obtaining the recognition results. This invention utilizes image recognition technology and automatic learning capabilities to establish a complex relationship between the image features of different heavy minerals under different optical conditions (single-polarized light, orthogonal light, reflected light, etc.) and the types of heavy minerals. Furthermore, by segmenting and recognizing multi-source images, it achieves the identification and automatic measurement of different types of heavy minerals, improving work efficiency and accuracy, reducing the time personnel are exposed to toxic reagents, and lowering safety risks.

Description

Technical Field

[0001] This invention belongs to the field of oil and gas exploration and development technology, and in particular relates to a method for intelligent identification of heavy minerals based on multi-source image information. Background Technology

[0002] Currently, the oil and gas exploration industry commonly uses heavy mineral identification methods to identify heavy minerals. These methods are mainly based on manual identification and counting under a microscope. Manual identification of heavy minerals under a microscope is currently recognized as the most effective method in the industry, but it also has the following problems.

[0003] When identifying heavy minerals manually under a microscope, the identification is mainly based on the morphological and optical characteristics exhibited by the heavy minerals under single-polarized light, crossed light, and reflected light. The quantity of each heavy mineral still needs to be counted manually, and the counting requirement is that the total number should not be less than 400. If the total number is less than 400, all minerals are counted, which results in low identification efficiency and is prone to omissions and duplicate counting.

[0004] In addition, the immersion oil used for microscopic identification contains α-bromonaphthalene, a toxic substance, and prolonged exposure to it can harm the health of laboratory personnel. Summary of the Invention

[0005] The problem to be solved by this invention is to provide a method for identifying heavy minerals; in particular, a smart method for identifying heavy minerals based on multi-source image information that has high identification efficiency, high accuracy and high security.

[0006] To solve the above-mentioned technical problems, the technical solution adopted by the present invention is: a method for intelligent identification of heavy minerals based on multi-source image information, comprising the following steps:

[0007] S1: Separate heavy minerals and obtain single-polarized light images, cross-polarized light images, and reflected light images;

[0008] S2: Align the single-polarized light image, the orthogonal light image, and the reflected light image to obtain a single-polarized light panoramic image, an orthogonal light panoramic image, and a reflected light panoramic image;

[0009] S3: Perform particle segmentation on the single-polarized light overall image, the orthogonal light overall image, and the reflected light overall image to obtain a single-polarized light overall particle image, an orthogonal light overall particle image, and a reflected light overall particle image;

[0010] S4: Use the single-polarized light overall particle map, the orthogonal light overall particle map, the reflected light overall particle map and the corresponding real labels as training data to train the multi-channel recognition model and obtain the trained multi-channel recognition model.

[0011] S5: Perform identification using the multi-channel recognition model to obtain the recognition result.

[0012] Furthermore, S1 includes the following steps:

[0013] S11: The sedimentary rock is crushed and ground with dilute hydrochloric acid to remove carbonates and muddy cement to obtain the clastic particle composition;

[0014] S12: The debris particles are placed in a tribromomethane heavy liquid to obtain a density greater than 2.86 g / cm³. 3 Heavy minerals;

[0015] S13: Place the heavy mineral on a thin film and drop in a mixture of α-bromonaphthalene and liquid paraffin;

[0016] S14: Using a microscope imaging system, images of the heavy mineral are captured under single-polarized light, cross-polarized light, and reflected light from an external light source, respectively, to obtain the single-polarized light image, cross-polarized light image, and reflected light image of the heavy mineral.

[0017] Furthermore, S2 includes the following steps:

[0018] S21: Obtain the heavy mineral feature points in the single-polarized light image, the orthogonal light image, and the reflected light image respectively using the FAST algorithm;

[0019] S22: Obtain the comparison results of N point pairs surrounding the heavy mineral feature point using the BRIFE algorithm, and use the comparison results as the descriptor of the feature point. The comparison formula for the point pair is as follows:

[0020]

[0021] In the formula, T(P(A,B)) represents the comparison result of a set of point pairs (A,B) around feature point P, and I A I represents the gray level of point A. B This represents the gray level of point B;

[0022] S23: When the descriptor similarity between two points in the image to be aligned is greater than 80%, then the two points are used as matching feature points;

[0023] S24: Based on the matched feature points, calculate the homography matrix using the LM algorithm, rotate the image using the homography matrix, and align the single-polarized image, the orthogonal light image, and the reflected light image to obtain the aligned overall image of the single-polarized light, the orthogonal light, and the reflected light.

[0024] Furthermore, S3 includes the following steps:

[0025] S31: For the single-polarized panoramic image, according to the formula, obtain the correlation ρ(x,y) between pixel x and any surrounding pixel y in the single-polarized panoramic image in terms of RGB values, as follows:

[0026]

[0027] Where E[·] represents the expected value; X is the vector composed of the values ​​of the three RGB channels at pixel x; Y is the vector composed of the values ​​of the three RGB channels at pixel y; μ X Let μ be the mean of vector X. Y Let X be the mean of vector Y; i Y is a vector formed by the values ​​of the i-th channel in vector X; i Let Y be a vector composed of the values ​​of the i-th channel.

[0028] S32: Obtain the average value of the correlation between the pixel x and the eight surrounding pixels, and use the average value as the feature value of the pixel x, thereby obtaining the feature values ​​of all pixels in the single-polarized panoramic image.

[0029] S33: Pixels with feature values ​​less than 0.8 are used as boundary points of heavy minerals. Particle segmentation is performed based on the boundary points of heavy minerals to obtain the single-polarized overall particle map.

[0030] S34: According to the coordinates of the boundary points in the single-polarized light panorama, perform particle segmentation in the orthogonal light panorama and the reflected light panorama to obtain the orthogonal light panorama particle map and the reflected light panorama particle map, respectively.

[0031] Furthermore, the multi-channel recognition model includes a first image feature extraction module, a second image feature extraction module, a third image feature extraction module, a first single image feature fusion module, a second single image feature fusion module, a third single image feature fusion module, a multi-image feature fusion module, and a classification and recognition module. The input terminal of the first image feature extraction module corresponds to the input of the single-polarized light overall particle image, and the output terminal of the first image feature extraction module is connected to the input terminal of the first single image feature fusion module. The input terminal of the second image feature extraction module corresponds to the input of the orthogonal light overall particle image, and the output terminal of the second image feature extraction module is connected to the input terminal of the second single image feature fusion module. The input terminal of the third image feature extraction module corresponds to the input of the reflected light overall particle image, and the output terminal of the third image feature extraction module is connected to the input terminal of the third single image feature fusion module. The output terminals of the first, second, and third single image feature fusion modules are respectively connected to the input terminal of the multi-image feature fusion module. The output terminal of the multi-image feature fusion module is connected to the input terminal of the classification and recognition module, and the output terminal of the classification and recognition module is the output terminal of the multi-channel recognition model.

[0032] Furthermore, the first image feature extraction module, the second image feature extraction module, and the third image feature extraction module each include a first convolutional layer, a second convolutional layer, a third convolutional layer, and a fourth convolutional layer connected in sequence, and the input end of each first convolutional layer is the input end of the multi-image feature fusion module;

[0033] The first single-image feature fusion module, the second single-image feature fusion module, and the third single-image feature fusion module all include a fifth convolutional layer, a sixth convolutional layer, a seventh convolutional layer, an eighth convolutional layer, and a first single-image feature fusion unit, a second single-image feature fusion unit, and a third single-image feature fusion unit connected in series.

[0034] The output of each of the first convolutional layers is connected to the corresponding fifth convolutional layer; the output of each of the second convolutional layers is connected to the corresponding sixth convolutional layer; the output of each of the third convolutional layers is connected to the corresponding seventh convolutional layer; and the output of each of the fourth convolutional layers is connected to the corresponding eighth convolutional layer.

[0035] The first single-image feature fusion unit, the second single-image feature fusion unit, and the third single-image feature fusion unit all include a ninth convolutional layer, a tenth convolutional layer, an eleventh convolutional layer, a twelfth convolutional layer, a thirteenth convolutional layer, and a fourteenth convolutional layer;

[0036] The input of the fourteenth convolutional layer of the first single image feature fusion unit is connected to the output of the corresponding eighth convolutional layer;

[0037] The input of the thirteenth convolutional layer of the first single image feature fusion unit is connected to the output of the corresponding seventh convolutional layer;

[0038] The input of the twelfth convolutional layer of the first single image feature fusion unit is connected to the output of the corresponding seventh convolutional layer;

[0039] The input of the eleventh convolutional layer of the first single image feature fusion unit is connected to the output of the corresponding sixth convolutional layer;

[0040] The input of the tenth convolutional layer of the first single image feature fusion unit is connected to the output of the corresponding sixth convolutional layer;

[0041] The input of the ninth convolutional layer of the first single image feature fusion unit is connected to the output of the corresponding fifth convolutional layer;

[0042] The connection rules between the second single-image feature fusion unit and the third single-image feature fusion unit are as follows:

[0043] The output of the fourteenth convolutional layer of the nth single image feature fusion unit is connected to the input of the fourteenth convolutional layer of the (n+1)th single image feature fusion unit and the input of the twelfth convolutional layer of the (n+1)th single image feature fusion unit, respectively.

[0044] The output of the thirteenth convolutional layer of the nth single image feature fusion unit is connected to the input of the fourteenth convolutional layer of the nth single image feature fusion unit, the input of the twelfth convolutional layer of the (n+1)th single image feature fusion unit, and the input of the thirteenth convolutional layer of the (n+1)th single image feature fusion unit, respectively.

[0045] The output of the twelfth convolutional layer of the nth single image feature fusion unit is connected to the input of the tenth convolutional layer of the nth single image feature fusion unit and the input of the thirteenth convolutional layer of the nth single image feature fusion unit, respectively.

[0046] The output of the eleventh convolutional layer of the nth single image feature fusion unit is connected to the input of the thirteenth convolutional layer of the nth single image feature fusion unit, the input of the tenth convolutional layer of the (n+1)th single image feature fusion unit, and the input of the eleventh convolutional layer of the (n+1)th single image feature fusion unit, respectively.

[0047] The output of the tenth convolutional layer of the nth single image feature fusion unit is connected to the input of the ninth convolutional layer of the nth single image feature fusion unit and the input of the eleventh convolutional layer of the nth single image feature fusion unit, respectively.

[0048] The output of the ninth convolutional layer of the nth single image feature fusion unit is connected to the input of the ninth convolutional layer of the (n+1)th single image feature fusion unit.

[0049] The outputs of the ninth, eleventh, thirteenth, and fourteenth convolutional layers of the third single-image feature fusion unit together serve as the output of the single-image feature fusion module to which it belongs.

[0050] Furthermore, the multi-image feature fusion module includes a fifteenth convolutional layer, a sixteenth convolutional layer, a seventeenth convolutional layer, and a first multi-image feature fusion unit, a second multi-image feature fusion unit, a third multi-image feature fusion unit, and a stitching unit connected in sequence.

[0051] The first multi-image feature fusion unit, the second multi-image feature fusion unit, and the third multi-image feature fusion unit all include an eighteenth convolutional layer, a nineteenth convolutional layer, a twentieth convolutional layer, and a twenty-first convolutional layer;

[0052] The input of the eighteenth convolutional layer of the first multi-image feature fusion unit is connected to the output of the corresponding fifteenth convolutional layer;

[0053] The input of the nineteenth convolutional layer of the first multi-image feature fusion unit is connected to the output of the sixteenth convolutional layer and the output of the seventeenth convolutional layer, respectively.

[0054] The input of the twentieth convolutional layer of the first multi-image feature fusion unit is connected to the output of the corresponding sixteenth convolutional layer;

[0055] The input of the twenty-first convolutional layer of the first multi-image feature fusion unit is connected to the output of the corresponding seventeenth convolutional layer;

[0056] The connection rules between the second multi-image feature fusion unit and the third multi-image feature fusion unit are as follows:

[0057] The output of the eighteenth convolutional layer of the m-th multi-image feature fusion unit is connected to the input of the twentieth convolutional layer of the m-th multi-image feature fusion unit and the input of the eighteenth convolutional layer of the (m+1)-th multi-image feature fusion unit, respectively.

[0058] The output of the nineteenth convolutional layer of the m-th multi-image feature fusion unit is connected to the input of the eighteenth convolutional layer of the m-th multi-image feature fusion unit and the input of the twentieth convolutional layer of the m-th multi-image feature fusion unit, respectively.

[0059] The output of the twentieth convolutional layer of the m-th multi-image feature fusion unit is connected to the input of the twentieth convolutional layer of the m-th multi-image feature fusion unit, the input of the nineteenth convolutional layer of the (m+1)-th multi-image feature fusion unit, and the input of the twentieth convolutional layer of the (m+1)-th multi-image feature fusion unit, respectively.

[0060] The output of the 21st convolutional layer of the m-th multi-image feature fusion unit is connected to the input of the 19th convolutional layer of the (m+1)-th multi-image feature fusion unit and the input of the 21st convolutional layer of the (m+1)-th multi-image feature fusion unit, respectively.

[0061] The outputs of the eighteenth, twentieth, and twenty-first convolutional layers of the third multi-image feature fusion unit are respectively connected to the input of the stitching unit.

[0062] Furthermore, the present invention also provides an apparatus for performing the above-described data processing method.

[0063] Furthermore, the present invention also provides an apparatus including a memory, a processor, and an algorithm stored in the memory and executable on the processor, wherein the processor implements the data processing method when executing the computer program.

[0064] Furthermore, the present invention also provides a computer-readable storage medium storing a computer algorithm, which, when executed by a processor, implements the data processing method described above.

[0065] The advantages and positive effects of this invention are:

[0066] This invention applies multi-source image intelligent recognition technology to traditional heavy mineral identification experiments. Utilizing image recognition technology and automatic learning capabilities, it establishes a complex relationship between the image features of different heavy minerals under varying optical conditions (single-polarized light, orthogonal light, reflected light, etc.) and the types of heavy minerals. Furthermore, through the segmentation and recognition of multi-source images, it achieves the identification and automatic measurement of different types of heavy minerals. Compared to traditional heavy mineral identification methods, this significantly improves work efficiency and accuracy, reduces the time personnel are exposed to toxic reagents, and lowers safety risks. Attached Figure Description

[0067] Figure 1 This is an overall flowchart of an embodiment of the present invention.

[0068] Figure 2 This is a schematic diagram of the multi-channel recognition model structure according to an embodiment of the present invention.

[0069] Figure 3 This is a schematic diagram of three identical single-image feature fusion units connected in series according to an embodiment of the present invention.

[0070] Figure 4 This is a schematic diagram of three identical and sequentially connected multi-image feature fusion units according to an embodiment of the present invention.

[0071] Figure 5 This is a schematic diagram illustrating the effect of particle segmentation in an embodiment of the present invention. Detailed Implementation

[0072] The technical solution of the present invention will now be clearly and completely described with reference to the accompanying drawings. Obviously, the described embodiments are only some, not all, of the embodiments of the present invention. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0073] In the description of this invention, it should be noted that the terms "center," "upper," "lower," "left," "right," "vertical," "horizontal," "inner," and "outer," etc., indicate the orientation or positional relationship based on the orientation or positional relationship shown in the accompanying drawings. They are used only for the convenience of describing the invention and for simplifying the description, and do not indicate or imply that the device or element referred to must have a specific orientation, or be constructed and operated in a specific orientation. Therefore, they should not be construed as limitations on the invention. Furthermore, the terms "first," "second," and "third" are used for descriptive purposes only and should not be construed as indicating or implying relative importance.

[0074] In the description of this invention, it should be noted that, unless otherwise explicitly specified and limited, the terms "installation", "connection" and "linking" should be interpreted broadly, and those skilled in the art can understand the specific meaning of the above terms in this invention according to the specific circumstances.

[0075] The embodiments of the present invention will be further described below with reference to the accompanying drawings:

[0076] like Figure 1 As shown, a method for intelligent identification of heavy minerals based on multi-source image information includes the following steps:

[0077] S1: Separate the heavy minerals and obtain images of the heavy minerals under single-polarized light, under cross-polarized light, and under reflected light using an external light source.

[0078] Specifically, S1 includes the following steps:

[0079] S11: Crushed blocky sedimentary rock is ground with dilute hydrochloric acid to remove carbonates and muddy cement in the rock and obtain the detrital particle composition of the sedimentary rock.

[0080] S12: The debris particles were placed in a tribromomethane heavy liquid to obtain a density greater than 2.86 g / cm³. 3 Heavy minerals.

[0081] S13: Place the heavy mineral on a thin film and add a mixture of α-bromonaphthalene and liquid paraffin.

[0082] S14: Using a microscope imaging system, images of heavy minerals are captured under single-polarized light, cross-polarized light, and reflected light from an external light source, respectively, resulting in single-polarized light images, cross-polarized light images, and reflected light images of the heavy minerals.

[0083] S2: Align the single-polarized light image, cross-beam light image, and reflected light image of the heavy mineral to obtain the aligned image, namely the single-polarized light panoramic image, the cross-beam light panoramic image, and the reflected light panoramic image.

[0084] Specifically, S2 includes the following steps:

[0085] S21: Based on the similarity between each pixel and its surrounding pixels, the FAST algorithm is used to obtain the heavy mineral feature points in the single-polarized light image, the orthogonal light image, and the reflected light image, respectively.

[0086] S22: Obtain the comparison results of N point pairs around the heavy mineral feature point using the BRIFE algorithm, and use the comparison results as the descriptor of the feature point. The comparison formula for the point pair is:

[0087]

[0088] In the formula, T(P(A,B)) represents the comparison result of a set of point pairs (A,B) around feature point P, and I A I represents the gray level of point A. B This represents the gray level of point B.

[0089] S23: Obtain matching feature points. When the descriptor similarity between two points in the image to be aligned is greater than 80%, then the two points are used as matching feature points.

[0090] S24: Based on the matched feature points, calculate the homography matrix using the LM algorithm, rotate the image using the homography matrix, and complete the alignment of the single-polarized light image, the orthogonal light image, and the reflected light image to obtain the aligned single-polarized light panoramic image, the orthogonal light panoramic image, and the reflected light panoramic image.

[0091] S3: Perform particle segmentation on the single-polarized light overall image, the orthogonal light overall image, and the reflected light overall image to obtain the single-polarized light overall particle image, the orthogonal light overall particle image, and the reflected light overall particle image.

[0092] Specifically, S3 includes the following steps:

[0093] S31: For a single-polarized panoramic image, according to the formula, obtain the correlation ρ(x,y) between pixel x and any surrounding pixel y in the single-polarized panoramic image in terms of RGB values. The formula is as follows:

[0094]

[0095] Where E[·] represents the expected value; X is the vector composed of the values ​​of the three RGB channels at pixel x; Y is the vector composed of the values ​​of the three RGB channels at pixel y; μ X Let μ be the mean of vector X. Y Let X be the mean of vector Y; i Y is a vector formed by the values ​​of the i-th channel in vector X; i Let Y be a vector composed of the values ​​of the i-th channel.

[0096] S32: Obtain the mean value of the correlation between pixel x and its eight surrounding pixels, and use the mean value as the feature value of pixel x, thereby obtaining the feature values ​​of all pixels in the single-polarized panoramic image.

[0097] S33: Pixels with feature values ​​less than 0.8 are used as heavy mineral boundary points. Particle segmentation is performed based on the heavy mineral boundary points to obtain a single-polarized overall particle map.

[0098] S34: Based on the coordinates of the boundary points in the single-polarized panoramic image, perform particle segmentation in the orthogonal panoramic image and the reflected panoramic image to obtain the particle image corresponding to the orthogonal panoramic image and the particle image corresponding to the reflected panoramic image, namely the orthogonal panoramic particle image and the reflected panoramic particle image.

[0099] S4: Use the single-polarized light overall particle image, the orthogonal light overall particle image, the reflected light overall particle image, and the corresponding real labels as training data to train the multi-channel recognition model and obtain the trained multi-channel recognition model.

[0100] The multi-channel recognition model includes a first image feature extraction module, a second image feature extraction module, a third image feature extraction module, a first single image feature fusion module, a second single image feature fusion module, a third single image feature fusion module, a multi-image feature fusion module, and a classification and recognition module. The first, second, and third image feature extraction modules have identical structures; similarly, the first, second, and third single image feature fusion modules also have identical structures.

[0101] Specifically, the input of the first image feature extraction module corresponds to the input of a single-polarized light overall grain image, and the output of the first image feature extraction module is connected to the input of the first single-image feature fusion module; the input of the second image feature extraction module corresponds to the input of an orthogonal light overall grain image, and the output of the second image feature extraction module is connected to the input of the second single-image feature fusion module; the input of the third image feature extraction module corresponds to the input of a reflected light overall grain image, and the output of the third image feature extraction module is connected to the input of the third single-image feature fusion module. The outputs of the first, second, and third single-image feature fusion modules are respectively connected to the input of the multi-image feature fusion module; the output of the multi-image feature fusion module is connected to the input of the classification and recognition module; the output of the classification and recognition module is the output of the multi-channel recognition model.

[0102] Specifically, the first image feature extraction module, the second image feature extraction module, and the third image feature extraction module all include a first convolutional layer, a second convolutional layer, a third convolutional layer, and a fourth convolutional layer connected in sequence, and the input end of each first convolutional layer is the input end of the multi-image feature fusion module.

[0103] like Figure 2 As shown, Figure 2 In part A, the gradually shrinking rectangles represent feature maps of different scales, and the arrowed straight lines from bottom to top represent convolutional layers (first convolutional layer to fourth convolutional layer); Figure 2 In parts B and C, the arrows primarily indicate the data flow, and the circles represent the various convolutional layers. Figure 2 Taking a single-polarized image (512*512) as an example, the single-polarized image is successively passed through a convolutional layer with a kernel size of 3*3 and a stride of 2, resulting in feature maps p1, p2, p3, and p4 with sizes of 256*256, 128*128, 64*64, and 32*32, respectively.

[0104] The first single image feature fusion module, the second single image feature fusion module, and the third single image feature fusion module all include a fifth convolutional layer, a sixth convolutional layer, a seventh convolutional layer, an eighth convolutional layer, and a first single image feature fusion unit, a second single image feature fusion unit, and a third single image feature fusion unit connected in series. The first single image feature fusion unit, the second single image feature fusion unit, and the third single image feature fusion unit have the same structure.

[0105] like Figure 3 As shown in the diagram, the numbers inside the circles represent the convolutional layer numbers, and the arrows indicate the data flow direction. The output of each first convolutional layer is connected to the corresponding fifth convolutional layer; the output of each second convolutional layer is connected to the corresponding sixth convolutional layer; the output of each third convolutional layer is connected to the corresponding seventh convolutional layer; and the output of each fourth convolutional layer is connected to the corresponding eighth convolutional layer.

[0106] The first single image feature fusion unit, the second single image feature fusion unit, and the third single image feature fusion unit all include a ninth convolutional layer, a tenth convolutional layer, an eleventh convolutional layer, a twelfth convolutional layer, a thirteenth convolutional layer, and a fourteenth convolutional layer.

[0107] Specifically, the input of the fourteenth convolutional layer of the first single image feature fusion unit is connected to the output of the corresponding eighth convolutional layer; the input of the thirteenth convolutional layer of the first single image feature fusion unit is connected to the output of the corresponding seventh convolutional layer; the input of the twelfth convolutional layer of the first single image feature fusion unit is connected to the output of the corresponding seventh convolutional layer; the input of the eleventh convolutional layer of the first single image feature fusion unit is connected to the output of the corresponding sixth convolutional layer; the input of the tenth convolutional layer of the first single image feature fusion unit is connected to the output of the corresponding sixth convolutional layer; and the input of the ninth convolutional layer of the first single image feature fusion unit is connected to the output of the corresponding fifth convolutional layer.

[0108] The connection rules between the second and third single-image feature fusion units are as follows:

[0109] The output of the fourteenth convolutional layer of the nth single-image feature fusion unit is connected to the input of the fourteenth convolutional layer of the (n+1)th single-image feature fusion unit and the input of the twelfth convolutional layer of the (n+1)th single-image feature fusion unit, respectively; the output of the thirteenth convolutional layer of the nth single-image feature fusion unit is connected to the input of the fourteenth convolutional layer of the nth single-image feature fusion unit, the input of the twelfth convolutional layer of the (n+1)th single-image feature fusion unit, and the input of the thirteenth convolutional layer of the (n+1)th single-image feature fusion unit, respectively; the output of the twelfth convolutional layer of the nth single-image feature fusion unit is connected to the input of the tenth convolutional layer of the nth single-image feature fusion unit and the input of the thirteenth convolutional layer of the nth single-image feature fusion unit, respectively; the output of the eleventh convolutional layer of the nth single-image feature fusion unit is connected to the input of the thirteenth convolutional layer of the nth single-image feature fusion unit, the input of the tenth convolutional layer of the (n+1)th single-image feature fusion unit, and the input of the thirteenth convolutional layer of the (n+1)th single-image feature fusion unit, respectively. The input of the eleventh convolutional layer of the single image feature fusion unit is connected; the output of the tenth convolutional layer of the nth single image feature fusion unit is connected to the input of the ninth convolutional layer of the nth single image feature fusion unit and the input of the eleventh convolutional layer of the nth single image feature fusion unit; the output of the ninth convolutional layer of the nth single image feature fusion unit is connected to the input of the ninth convolutional layer of the (n+1)th single image feature fusion unit.

[0110] The outputs of the ninth, eleventh, thirteenth, and fourteenth convolutional layers of the third single-image feature fusion unit serve as the outputs of its respective single-image feature fusion module.

[0111] Specifically, the calculation formula for the twelfth convolutional layer in the first single image feature fusion unit is as follows:

[0112]

[0113] in, This is the output of the twelfth convolutional layer in the first single image feature fusion unit; This is the output of the seventh convolutional layer; This is the output of the eighth convolutional layer; Up(.) represents the upsampling operation; w1 and w2 are both weight coefficients; ∈ is a constant, which can take the value 0.0001; conv(.) represents the convolution operation.

[0114] The calculation formula for the fourteenth convolutional layer in the first single image feature fusion unit is:

[0115]

[0116] in, is the output of the fourteenth convolutional layer in the first single image feature fusion unit; w″1 is the weight coefficient.

[0117] The calculation formula for the thirteenth convolutional layer in the first single image feature fusion unit is:

[0118]

[0119] in, This is the output of the thirteenth convolutional layer in the first single image feature fusion unit; Down(.) indicates a downsampling operation; w′1, w′2, and w′3 are all weight coefficients; This is the output of the eleventh convolutional layer in the first single image feature fusion unit.

[0120] The calculation formula for the eleventh convolutional layer in the first single image feature fusion unit is:

[0121]

[0122] in, This is the output of the eleventh convolutional layer in the first single image feature fusion unit; This is the output of the sixth convolutional layer; This is the output of the tenth convolutional layer in the first single image feature fusion unit; This is the output of the ninth convolutional layer in the first single image feature fusion unit; w″′1, w″2 and w″3 are all weight coefficients.

[0123] The calculation formula for the tenth convolutional layer in the first single image feature fusion unit is:

[0124]

[0125] in, This is the output of the tenth convolutional layer in the first single image feature fusion unit; w″″1 and w″′2 are both weight parameters.

[0126] The calculation formula for the ninth convolutional layer in the first single image feature fusion unit is:

[0127]

[0128] in, This is the output of the ninth convolutional layer in the first single image feature fusion unit; w″″′1 and w″″2 are both weight parameters.

[0129] The multi-image feature fusion module includes a fifteenth convolutional layer, a sixteenth convolutional layer, a seventeenth convolutional layer, and a first multi-image feature fusion unit, a second multi-image feature fusion unit, a third multi-image feature fusion unit, and a stitching unit connected in sequence; the first multi-image feature fusion unit, the second multi-image feature fusion unit, and the third multi-image feature fusion unit have the same structure.

[0130] The fifteenth convolutional layer is used to unify and stitch the outputs of the first single-image feature fusion module, and adjust the size of the stitched feature map; the sixteenth convolutional layer is used to unify and stitch the outputs of the second single-image feature fusion module, and adjust the size of the stitched feature map; the seventeenth convolutional layer is used to unify and stitch the outputs of the third single-image feature fusion module, and adjust the size of the stitched feature map.

[0131] like Figure 4 As shown, Figure 4 The numbers inside the circles indicate the convolutional layer numbers, and the arrows indicate the data flow direction. The first, second, and third multi-image feature fusion units all include the eighteenth, nineteenth, twentieth, and twenty-first convolutional layers.

[0132] Specifically, the input of the eighteenth convolutional layer of the first multi-image feature fusion unit is connected to the output of the corresponding fifteenth convolutional layer; the input of the nineteenth convolutional layer of the first multi-image feature fusion unit is connected to the outputs of the corresponding sixteenth and seventeenth convolutional layers, respectively; the input of the twentieth convolutional layer of the first multi-image feature fusion unit is connected to the output of the corresponding sixteenth convolutional layer; and the input of the twenty-first convolutional layer of the first multi-image feature fusion unit is connected to the output of the corresponding seventeenth convolutional layer.

[0133] The connection rules between the second and third multi-image feature fusion units are as follows:

[0134] The output of the 18th convolutional layer of the m-th multi-image feature fusion unit is connected to the input of the 20th convolutional layer of the m-th multi-image feature fusion unit and the input of the 18th convolutional layer of the (m+1)-th multi-image feature fusion unit, respectively; the output of the 19th convolutional layer of the m-th multi-image feature fusion unit is connected to the input of the 18th convolutional layer of the m-th multi-image feature fusion unit and the input of the 20th convolutional layer of the m-th multi-image feature fusion unit, respectively; the output of the 20th convolutional layer of the m-th multi-image feature fusion unit is connected to the input of the 21st convolutional layer of the m-th multi-image feature fusion unit, the input of the 19th convolutional layer of the (m+1)-th multi-image feature fusion unit, respectively; and the output of the 21st convolutional layer of the m-th multi-image feature fusion unit is connected to the input of the 19th convolutional layer of the (m+1)-th multi-image feature fusion unit and the input of the 21st convolutional layer of the (m+1)-th multi-image feature fusion unit, respectively.

[0135] The outputs of the eighteenth, twentieth, and twenty-first convolutional layers of the third multi-image feature fusion unit are respectively connected to the input of the splicing unit.

[0136] The stitching unit is used to stitch together the outputs of the eighteenth convolutional layer, the twentieth convolutional layer, and the twenty-first convolutional layer of the third multi-image feature fusion unit along the channel dimension; the output of the stitching unit is the output of the multi-image feature fusion module.

[0137] The classification and recognition module consists of a fully connected layer and a softmax layer connected in sequence. The fully connected layer adjusts the output vector size of the splicing unit to 1*F, where F is the number of heavy mineral types. Finally, the softmax layer is used to calculate the output vector, and the index of the highest probability is taken as the recognition result.

[0138] S5: Obtain particle images of the heavy mineral to be identified using the same methods as S1-S3, such as... Figure 5 As shown, the trained multi-channel recognition model is used to identify the particle images of the heavy minerals to be identified, and the recognition results are obtained.

[0139] In summary, this invention applies multi-source image intelligent recognition technology to traditional heavy mineral identification experiments. Utilizing image recognition technology and automatic learning capabilities, it establishes a complex relationship between the image features of different heavy minerals under varying optical conditions (single-polarized light, orthogonal light, reflected light, etc.) and the types of heavy minerals. Furthermore, through the segmentation and recognition of multi-source images, it achieves the identification and automatic measurement of different types of heavy minerals. Compared to traditional heavy mineral identification methods, this significantly improves work efficiency and accuracy, reduces the time personnel are exposed to toxic reagents, and lowers safety risks.

[0140] The foregoing has provided a detailed description of one embodiment of the present invention, but this description is merely a preferred embodiment and should not be construed as limiting the scope of the invention. All equivalent variations and modifications made within the scope of the claims of this invention should still fall within the patent coverage of this invention.

Claims

1. A method for intelligent identification of heavy minerals based on multi-element image information, characterized in that: Includes the following steps, S1: Separate heavy minerals and obtain single-polarized light images, cross-polarized light images, and reflected light images; S2: Align the single-polarized light image, the orthogonal light image, and the reflected light image to obtain a single-polarized light overall image, an orthogonal light overall image, and a reflected light overall image; S3: Perform particle segmentation on the single-polarized light overall image, the orthogonal light overall image, and the reflected light overall image to obtain a single-polarized light overall particle image, an orthogonal light overall particle image, and a reflected light overall particle image. S3 includes the following steps. S31: for the single-polarization full-view image, according to the formula, the correlation of the RGB values of the pixel point x and any pixel point y around the pixel point x in the single-polarization full-view image is obtained , and the formula is as follows: in Represents the mathematical expectation; X is a vector composed of the values ​​of the three RGB channels at pixel x; Y is a vector composed of the values ​​of the three RGB channels at pixel y. Let X be the mean of vector X. Let Y be the mean of vector Y; Let X be a vector composed of the values ​​of the i-th channel. Let Y be a vector composed of the values ​​of the i-th channel. S32: Obtain the average value of the correlation between the pixel x and the eight surrounding pixels, and use the average value as the feature value of the pixel x, thereby obtaining the feature values ​​of all pixels in the single-polarized panoramic image. S33: Pixels with feature values ​​less than 0.8 are used as boundary points of heavy minerals. Particle segmentation is performed based on the boundary points of heavy minerals to obtain the single-polarized overall particle map. S34: According to the coordinates of the boundary points in the single-polarized light panoramic image, perform particle segmentation in the orthogonal light panoramic image and the reflected light panoramic image to obtain the orthogonal light panoramic particle image and the reflected light panoramic particle image, respectively. S4: Use the single-polarized light overall particle map, the orthogonal light overall particle map, the reflected light overall particle map and the corresponding real labels as training data to train the multi-channel recognition model and obtain the trained multi-channel recognition model; S5: Recognition is performed using the multi-channel recognition model to obtain the recognition result. The multi-channel recognition model includes a first image feature extraction module, a second image feature extraction module, a third image feature extraction module, a first single image feature fusion module, a second single image feature fusion module, a third single image feature fusion module, a multi-image feature fusion module, and a classification and recognition module. The input terminal of the first image feature extraction module corresponds to the input of the single-polarized panoramic grain image. The output terminal of the first image feature extraction module is connected to the input terminal of the first single image feature fusion module. The input terminal of the second image feature extraction module corresponds to the input of the orthogonal light panoramic grain image. The output of the image feature extraction module is connected to the input of the second single-image feature fusion module. The input of the third image feature extraction module corresponds to the input of the overall particle image of reflected light. The output of the third image feature extraction module is connected to the input of the third single-image feature fusion module. The outputs of the first single-image feature fusion module, the second single-image feature fusion module, and the third single-image feature fusion module are respectively connected to the input of the multi-image feature fusion module. The output of the multi-image feature fusion module is connected to the input of the classification and recognition module. The output of the classification and recognition module is the output of the multi-channel recognition model.

2. The method for intelligent identification of heavy minerals based on multi-source image information according to claim 1, characterized in that: S1 includes the following steps: S11: The sedimentary rock is crushed and ground with dilute hydrochloric acid to remove carbonates and muddy cement to obtain the clastic particle composition; S12: The detrital particles are placed in tribromomethane heavy liquid to obtain heavy minerals with a density greater than 2.86 g / cm³; S13: Place the heavy mineral on a thin film and drop in a mixture of α-bromonaphthalene and liquid paraffin; S14: Using a microscope imaging system, images of the heavy mineral are captured under single-polarized light, cross-polarized light, and reflected light from an external light source, respectively, to obtain the single-polarized light image, cross-polarized light image, and reflected light image of the heavy mineral.

3. The method for intelligent identification of heavy minerals based on multi-source image information according to claim 1 or 2, characterized in that: S2 includes the following steps: S21: Obtain the heavy mineral feature points in the single-polarized light image, the orthogonal light image, and the reflected light image respectively using the FAST algorithm; S22: Obtain the comparison results of N point pairs around the heavy mineral feature point using the BRIFE algorithm, and use the comparison results as the descriptor of the feature point. The comparison formula for the point pair is: In the formula, Represents a set of point pairs around feature point P. The comparison results This represents the gray level of point A. This represents the gray level of point B; S23: When the descriptor similarity between two points in the image to be aligned is greater than 80%, then the two points are used as matching feature points; S24: Based on the matched feature points, calculate the homography matrix using the LM algorithm, rotate the image using the homography matrix, and align the single-polarized image, the orthogonal image, and the reflected image to obtain the aligned overall image of the single-polarized light, the orthogonal image, and the reflected image.

4. A method for intelligent identification of heavy minerals based on multi-source image information according to claim 1 or 2, characterized in that: The first image feature extraction module, the second image feature extraction module, and the third image feature extraction module each include a first convolutional layer, a second convolutional layer, a third convolutional layer, and a fourth convolutional layer connected in sequence, and the input end of each first convolutional layer is the input end of the multi-image feature fusion module; The first single-image feature fusion module, the second single-image feature fusion module, and the third single-image feature fusion module all include a fifth convolutional layer, a sixth convolutional layer, a seventh convolutional layer, an eighth convolutional layer, and a first single-image feature fusion unit, a second single-image feature fusion unit, and a third single-image feature fusion unit connected in series. The output of each of the first convolutional layers is connected to the corresponding fifth convolutional layer; the output of each of the second convolutional layers is connected to the corresponding sixth convolutional layer; the output of each of the third convolutional layers is connected to the corresponding seventh convolutional layer; and the output of each of the fourth convolutional layers is connected to the corresponding eighth convolutional layer. The first single-image feature fusion unit, the second single-image feature fusion unit, and the third single-image feature fusion unit all include a ninth convolutional layer, a tenth convolutional layer, an eleventh convolutional layer, a twelfth convolutional layer, a thirteenth convolutional layer, and a fourteenth convolutional layer; The input of the fourteenth convolutional layer of the first single image feature fusion unit is connected to the output of the corresponding eighth convolutional layer; The input of the thirteenth convolutional layer of the first single image feature fusion unit is connected to the output of the corresponding seventh convolutional layer; The input of the twelfth convolutional layer of the first single image feature fusion unit is connected to the output of the corresponding seventh convolutional layer; The input of the eleventh convolutional layer of the first single image feature fusion unit is connected to the output of the corresponding sixth convolutional layer; The input of the tenth convolutional layer of the first single image feature fusion unit is connected to the output of the corresponding sixth convolutional layer; The input of the ninth convolutional layer of the first single image feature fusion unit is connected to the output of the corresponding fifth convolutional layer; The connection rules between the second single-image feature fusion unit and the third single-image feature fusion unit are as follows: The output of the fourteenth convolutional layer of the nth single image feature fusion unit is connected to the input of the fourteenth convolutional layer of the (n+1)th single image feature fusion unit and the input of the twelfth convolutional layer of the (n+1)th single image feature fusion unit, respectively. The output of the thirteenth convolutional layer of the nth single image feature fusion unit is connected to the input of the fourteenth convolutional layer of the nth single image feature fusion unit, the input of the twelfth convolutional layer of the (n+1)th single image feature fusion unit, and the input of the thirteenth convolutional layer of the (n+1)th single image feature fusion unit, respectively. The output of the twelfth convolutional layer of the nth single image feature fusion unit is connected to the input of the tenth convolutional layer of the nth single image feature fusion unit and the input of the thirteenth convolutional layer of the nth single image feature fusion unit, respectively. The output of the eleventh convolutional layer of the nth single image feature fusion unit is connected to the input of the thirteenth convolutional layer of the nth single image feature fusion unit, the input of the tenth convolutional layer of the (n+1)th single image feature fusion unit, and the input of the eleventh convolutional layer of the (n+1)th single image feature fusion unit, respectively. The output of the tenth convolutional layer of the nth single image feature fusion unit is connected to the input of the ninth convolutional layer of the nth single image feature fusion unit and the input of the eleventh convolutional layer of the nth single image feature fusion unit, respectively. The output of the ninth convolutional layer of the nth single image feature fusion unit is connected to the input of the ninth convolutional layer of the (n+1)th single image feature fusion unit. The outputs of the ninth, eleventh, thirteenth, and fourteenth convolutional layers of the third single-image feature fusion unit together serve as the output of the single-image feature fusion module to which it belongs.

5. The method for intelligent identification of heavy minerals based on multi-source image information according to claim 4, characterized in that: The multi-image feature fusion module includes a fifteenth convolutional layer, a sixteenth convolutional layer, a seventeenth convolutional layer, and a first multi-image feature fusion unit, a second multi-image feature fusion unit, a third multi-image feature fusion unit, and a stitching unit connected in sequence. The first multi-image feature fusion unit, the second multi-image feature fusion unit, and the third multi-image feature fusion unit all include an eighteenth convolutional layer, a nineteenth convolutional layer, a twentieth convolutional layer, and a twenty-first convolutional layer; The input of the eighteenth convolutional layer of the first multi-image feature fusion unit is connected to the output of the corresponding fifteenth convolutional layer; The input of the nineteenth convolutional layer of the first multi-image feature fusion unit is connected to the output of the sixteenth convolutional layer and the output of the seventeenth convolutional layer, respectively. The input of the twentieth convolutional layer of the first multi-image feature fusion unit is connected to the output of the corresponding sixteenth convolutional layer; The input of the twenty-first convolutional layer of the first multi-image feature fusion unit is connected to the output of the corresponding seventeenth convolutional layer; The connection rules between the second multi-image feature fusion unit and the third multi-image feature fusion unit are as follows: The output of the eighteenth convolutional layer of the m-th multi-image feature fusion unit is connected to the input of the twentieth convolutional layer of the m-th multi-image feature fusion unit and the input of the eighteenth convolutional layer of the (m+1)-th multi-image feature fusion unit, respectively. The output of the nineteenth convolutional layer of the m-th multi-image feature fusion unit is connected to the input of the eighteenth convolutional layer of the m-th multi-image feature fusion unit and the input of the twentieth convolutional layer of the m-th multi-image feature fusion unit, respectively. The output of the twentieth convolutional layer of the m-th multi-image feature fusion unit is connected to the input of the twentieth convolutional layer of the m-th multi-image feature fusion unit, the input of the nineteenth convolutional layer of the (m+1)-th multi-image feature fusion unit, and the input of the twentieth convolutional layer of the (m+1)-th multi-image feature fusion unit, respectively. The output of the 21st convolutional layer of the m-th multi-image feature fusion unit is connected to the input of the 19th convolutional layer of the (m+1)-th multi-image feature fusion unit and the input of the 21st convolutional layer of the (m+1)-th multi-image feature fusion unit, respectively. The outputs of the eighteenth, twentieth, and twenty-first convolutional layers of the third multi-image feature fusion unit are respectively connected to the input of the stitching unit.

6. A heavy mineral intelligent identification device based on multi-source image information, characterized in that: The identification method described in any one of claims 1 to 5 is executed.

7. A computer device comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, characterized in that: When the processor executes the computer program, it implements the identification method as described in any one of claims 1 to 5.

8. A computer-readable storage medium storing a computer program, characterized in that, When the computer program is executed by the processor, it implements the identification method as described in any one of claims 1 to 5.

Images