Rock debris lithology analysis method based on multi-level fusion image algorithm and X-ray diffraction

Abstract

The invention discloses a rock debris lithology analysis method based on a multilevel fusion image algorithm and X-ray diffraction. The method comprises the following steps: S1, acquiring rock debris image data under different resolutions; s2, generating a multi-level rock debris image feature matrix; s3, preliminarily generating a lithology classification result; s4, generating supplementary data of X-ray diffraction analysis; s5, fusing the supplementary data of the lithology analysis and the preliminarily generated lithology classification result, and finally generating a comprehensive lithology classification result; s6, the comprehensive lithology classification result is input into a dynamic adaptive analysis system, the dynamic adaptive analysis system adaptively adjusts an analysis strategy according to the rock debris sample complexity, and parameters of a dual-path convolutional neural network model and a multi-modal analysis algorithm are optimized through a real-time feedback and learning mechanism; and S7, outputting an analysis report containing mineral distribution, crystal structure and comprehensive lithology classification results. The classification accuracy of different types of rock debris samples is improved by about 15%.

Description

Technical Field

[0001] The present invention relates to the field of rock cuttings technology, and in particular to a rock cuttings lithology analysis method based on a multi-level fusion image algorithm and X-ray diffraction. Background Art

[0002] In the existing technology, rock cuttings lithology analysis mainly relies on a single image processing technology or X-ray diffraction technology. Traditional image processing technology mainly classifies rock cuttings samples through the texture and morphological characteristics of two-dimensional images. Although it can identify the macroscopic characteristics of rock cuttings to a certain extent, it is difficult to accurately distinguish different mineral components and their crystal structures when processing complex multi-component samples. In addition, image processing technology has technical bottlenecks in resolution improvement and noise control, and is easily affected by image quality, resulting in inaccurate classification results.

[0003] On the other hand, X-ray diffraction technology is often used to analyze the crystal structure and mineral composition of rock cuttings. Through the calculation of diffraction intensity data and crystal plane spacing, XRD can identify the mineral composition in the sample and provide relatively accurate crystal structure information. However, XRD technology is limited to providing mineral composition and structural information of rock samples, and cannot generate intuitive results of lithology classification in the absence of image information. In addition, XRD technology has a slow sample processing speed and cannot meet the needs of rapid analysis of large-scale samples. In complex geological exploration projects, traditional XRD analysis takes a long time and it is difficult to provide real-time feedback on analysis results.

[0004] The main shortcomings or problems of existing technologies include the following aspects: First, it is difficult for a single image processing technology to take into account both the macroscopic morphology and microscopic mineral composition of rock cuttings samples, especially the subtle differences in complex rock cuttings samples are easily ignored, resulting in insufficient analysis accuracy; second, although X-ray diffraction technology can provide information on mineral composition and crystal structure, it is disconnected from image processing technology and lacks a complete description of the overall lithological characteristics. In addition, existing technologies are insufficient in the automated processing and real-time analysis of large-scale samples. Traditional technologies are difficult to achieve fast analysis speeds while maintaining high accuracy, which affects the decision-making efficiency of geological exploration projects. Summary of the invention

[0005] One object of the present invention is to propose a rock cuttings lithology analysis method based on a hierarchical fusion image algorithm and X-ray diffraction, which improves the classification accuracy of different types of rock cuttings samples by about 15%.

[0006] A rock fragment lithology analysis method based on a hierarchical fusion image algorithm and X-ray diffraction according to an embodiment of the present invention comprises the following steps:

[0007] S1. Pre-process the rock cuttings samples to remove impurities and pollutants, and perform multispectral image acquisition to obtain rock cuttings image data at different resolutions;

[0008] S2. Use multi-scale image processing technology to process rock cuttings image data of different resolutions, identify the macroscopic lithology characteristics of rock cuttings samples at low resolution, identify the microstructure and texture characteristics at high resolution, and generate a multi-level rock cuttings image feature matrix;

[0009] S3, perform deep feature extraction on the multi-level rock cuttings image feature matrix through a dual-path convolutional neural network model, automatically learn the image features in the rock cuttings samples, and preliminarily generate lithology classification results;

[0010] S4. Use X-ray diffraction technology to perform crystal structure analysis on rock cuttings samples, obtain diffraction patterns of samples, identify mineral composition and crystal structure information of rock cuttings samples, and generate supplementary data for lithology analysis;

[0011] S5. The supplementary data of lithology analysis and the preliminary generated lithology classification results are integrated, and the comprehensive lithology classification results are finally generated through comparison, optimization and refinement of multimodal analysis algorithms;

[0012] S6. Input the comprehensive lithology classification results into the dynamic adaptive analysis system, which adaptively adjusts the analysis strategy according to the complexity of the rock cuttings sample and optimizes the parameters of the dual-path convolutional neural network model and the multimodal analysis algorithm through real-time feedback and learning mechanisms;

[0013] S7. Output an analysis report containing mineral distribution, crystal structure and comprehensive lithology classification results. The analysis report includes a three-dimensional mineral distribution map, XRD spectrum interpretation and image analysis flow chart.

[0014] Optionally, S1 includes the following specific steps:

[0015] S11, performing physical pretreatment on the collected rock cuttings samples, using screening equipment to remove particles with a particle size smaller than d1 and larger than d2, d1 is the minimum allowable diameter of the rock cuttings particles, d2 is the maximum allowable diameter of the rock cuttings particles, and the values ​​of d1 and d2 are set according to the initial particle size distribution of the rock cuttings samples;

[0016] S12, cleaning and drying the pretreated rock cuttings samples to remove impurities and moisture attached to the sample surfaces;

[0017] S13, using multi-spectral imaging equipment to collect images of rock cuttings samples, obtain rock cuttings image data at different resolutions, in the spectral bands λ1 to λ n Corresponding to r1 to r n The rock cuttings image data is collected at the resolution level of

[0018] S14, storing the collected rock cuttings image data in the form of a three-dimensional matrix T (λ, r, x, y):

[0019]

[0020] Among them, I i (λ j ,r k ,x,y) represents the spectral band λ j , resolution k And rock cuttings image data at spatial position (x, y), where x and y represent the spatial coordinates of the rock cuttings image data.

[0021] Optionally, S2 includes the following specific steps:

[0022] S21. Perform multi-scale processing on rock cuttings image data with different resolutions. low Preprocess the rock cuttings image data to reduce the noise of the rock cuttings image data and obtain the macroscopic lithological characteristics of the rock cuttings samples. macro :

[0023]

[0024] Among them, I(x′,y′,r low ) indicates that at low resolution r low The original rock cuttings image pixel value under G(xx′,yy′,σ macro ) is the Gaussian kernel function, and the kernel width is σ macro , represents the scale of the macroscopic feature, and the integration range is the macroscopic area Ω of the cuttings image macro ;

[0025] S22, using wavelet transform based image processing method to analyze high resolution high The cuttings image data is decomposed at multiple scales to extract the microscopic lithological structure characteristics of the cuttings image. micro :

[0026]

[0027] Among them, I(x+m,y+n,r high ) represents the pixel value of the rock chip image at high resolution, is the wavelet basis function, the scale factor is s, M and N are the spatial window sizes of the wavelet transform;

[0028] S23, microscopic lithology structural characteristics F micro and macroscopic lithological characteristics macro Fusion is performed to generate a multi-level rock cuttings image feature matrix T feature (x,y):

[0029]

[0030] Among them, F macro (x, y) represents the macroscopic lithological characteristics at the spatial position (x, y), F micro (x, y) represents the microscopic lithological structural characteristics at the same spatial position.

[0031] Optionally, S3 includes the following specific steps:

[0032] S31, multi-level rock cuttings image feature matrix T feature (x, y) is input into the dual-path convolutional neural network model, which includes two independent feature extraction paths, namely the spatial feature path and the texture feature path for processing different features of rock cuttings samples;

[0033] S32. Use multiple convolutional layers in the spatial feature path to extract features of the overall shape, structure and contour of the rock cuttings sample:

[0034]

[0035] in, Represents the convolution output feature map of the l-th layer spatial feature path, is the convolution kernel weight in the l-th layer spatial feature path, is the bias term, k is the convolution kernel size;

[0036] S33. After each convolution layer in the spatial feature path, the ReLU activation function is used to perform nonlinear processing on the convolution result, and the pooling layer is used to reduce the dimension of the spatial features:

[0037]

[0038] in, is the size of the pooling window, and the overall shape, structure, and contour information of the rock cuttings sample are preserved through the pooling operation;

[0039] S34. In the texture feature path, multiple convolutional layers are used to extract features of the subtle texture and microstructure of the rock cuttings sample:

[0040]

[0041] in, Represents the convolution output feature map of the l-th layer texture feature path, is the convolution kernel weight in the texture feature path of the lth layer, is the bias term;

[0042] S35. After each convolution layer in the texture feature path, the convolution result is processed nonlinearly using the ReLU activation function, and the texture feature is reduced in dimension using the pooling layer:

[0043]

[0044] Preserve subtle texture and microstructural information in rock cuttings samples through pooling operations;

[0045] S36, the feature map F extracted from the spatial feature path and the texture feature path spatial_pool ( x,y ) and F texture_pool ( x,y ) Fusion is performed to generate a comprehensive feature matrix T combined ( x,y ) :

[0046] T combined ( x,y ) =α·F spatial_pool ( x,y ) +β·F texture_pool ( x,y ) ;

[0047] Among them, α and β are fusion coefficients, which represent the weights of spatial features and texture features;

[0048] S37, the integrated feature matrix T combined (x,y) is input to the fully connected layer, and the Softmax classifier is used to classify the rock cuttings samples:

[0049]

[0050] Among them, P(C k ∣T combined ) represents the probability that the sample belongs to the kth lithology, W k and W j is the weight of the classifier, and K is the number of lithology classifications.

[0051] Optionally, S4 includes the following specific steps:

[0052] S41, placing the pretreated rock cuttings sample in an X-ray diffractometer, irradiating the sample with X-rays and obtaining diffraction intensity data I(θ), where θ represents the diffraction angle;

[0053] S42, by analyzing the diffraction intensity data I(θ) and position, identify the type and content of each mineral component in the rock cuttings sample, and generate a mineral composition analysis table C mineral (x,y):

[0054]

[0055] Among them, I i (θ) is the diffraction angle θ of the i-th mineral component i The diffraction intensity under i is the diffraction peak width, Z is the normalization factor used to make the total percentage of the mineral components equal to 1, Indicates that in the diffraction angle range θ min to θ max The total diffraction intensity within, f mineral (x,y,θ i ) is the diffraction angle θ i The characteristic function of the relevant mineral composition describes the distribution characteristics of the mineral at the spatial position (x, y);

[0056] S43. Constructing a mineral crystal structure model based on the changes in diffraction intensity and diffraction angle crystal (x,y), mineral crystal structure model is used to describe the crystal structure characteristics of different mineral components:

[0057]

[0058] Among them, V k is the unit cell volume of the kth mineral, F hkl is the structure factor of the crystal plane family, G hkl is the reciprocal vector of the crystal plane family, describing the lattice periodicity, r(x,y) represents the atomic position vector at the spatial position (x,y), D hkl (T) is the temperature factor of the crystal, which indicates the effect of temperature on the diffraction intensity;

[0059] S44, mineral composition analysis table C mineral (x,y) and mineral crystal structure model S crystal (x,y) are fused to generate supplementary data D for lithology analysis supplement ( x,y ) :

[0060] D supplement ( x,y ) =α1·C mineral ( x,y ) +β1·S crystal ( x,y );

[0061] Among them, α1 and β1 are weight coefficients, which are used to balance the weights of the mineral composition analysis table and the mineral crystal structure model in the supplementary data.

[0062] Optionally, S5 includes the following specific steps:

[0063] S51. Supplementary data D for lithology analysis supplement ( x,y ) Compared with the preliminary lithology classification results P(C k ∣T combined ) to obtain the initial fused feature matrix T fused ( x,y):

[0064] T fused ( x,y ) =α2∈P ( C k ∣T combined ) +β2·D supplement ( x,y ) ;

[0065] Among them, α2 and β2 are weight coefficients, which represent the weights of the preliminary lithology classification results and the supplementary data of lithology analysis in the fusion process;

[0066] S52, the initial fusion feature matrix T fused (x, y) performs multimodal analysis and calculates the similarity measure S between the preliminary classification results and the supplementary data similarity ( x,y ) :

[0067] S similarity (x,y)=exp(-(P(C k ∣T combined )-D supplement (x,y) 2 );

[0068] The closer the similarity measure is to 1, the higher the consistency between the preliminary classification results and the supplementary data;

[0069] S53, according to the similarity measure S similarity (x,y) pairs of fused feature matrix T fused (x, y) is optimized, and the weight coefficients α2 and β2 are adjusted through the back propagation algorithm to maximize the similarity measure:

[0070]

[0071] Where η is the learning rate, and is the updated weight coefficient;

[0072] S54, the optimized fusion feature matrix T fused (x, y) is input into the classification model, and the Softmax classifier is used to perform the final refinement classification of the lithology category:

[0073]

[0074] Among them, P final (C k ∣T fused_opt ) represents the probability that the sample belongs to the kth lithology under the final fusion feature matrix.

[0075] Optionally, S7 includes the following specific steps:

[0076] S71. Integrate the comprehensive lithology classification results with the mineral composition analysis table and the mineral crystal structure model to generate an analysis report data set;

[0077] S72. Draw a three-dimensional mineral distribution map according to the mineral composition analysis table, where the three-dimensional mineral distribution map represents the distribution of minerals at spatial positions (x, y, z);

[0078] S73. Interpret the diffraction intensity data to generate a spectrum interpretation result including diffraction peak position, peak intensity and mineral composition.

[0079] The beneficial effects of the present invention are:

[0080] (1) The present invention adopts multi-scale image processing technology to capture the macroscopic lithological characteristics of rock cuttings samples at low resolution and identify their microstructure and texture at high resolution. The dual-path convolutional neural network model performs deep feature extraction on feature matrices at different levels to automatically learn the complex rock cuttings image features in the sample and generate preliminary lithological classification results. Compared with traditional single image processing technology, the present invention can identify subtle differences in rock cuttings at a wider scale, especially in complex samples, and shows higher accuracy. Experimental results show that the present invention improves the classification accuracy of different types of rock cuttings samples by about 15%.

[0081] (2) The present invention innovatively combines the mineral composition and crystal structure information obtained by X-ray diffraction technology with the lithology classification results generated by the image processing algorithm. Through optimization by a multimodal analysis algorithm, XRD technology can accurately identify the mineral composition of rock cuttings samples, while the image algorithm can provide detailed macro- and microstructural information. The combination of the two greatly improves the accuracy and robustness of lithology analysis. Through comparative analysis and optimization, the problem of insufficient classification accuracy in a single technology can be effectively solved. The lithology classification results finally generated show obvious advantages in refining mineral composition and structural characteristics.

[0082] (3) The present invention can quickly process a large number of rock cuttings samples through automated image processing and XRD analysis processes, and generate high-precision lithology classification reports in a relatively short period of time. The automatic learning of the dual-path convolutional neural network model and the back-propagation adjustment system of the optimization algorithm can adaptively adjust the analysis parameters to ensure the consistency and efficiency of the classification results. Compared with traditional manual analysis methods, the present invention greatly shortens the analysis time and increases the analysis speed by about 30% in large-scale geological exploration projects while maintaining a high classification accuracy, which can provide more timely decision support for the project. BRIEF DESCRIPTION OF THE DRAWINGS

[0083] The accompanying drawings are used to provide a further understanding of the present invention and constitute a part of the specification. Together with the embodiments of the present invention, they are used to explain the present invention and do not constitute a limitation of the present invention. In the accompanying drawings:

[0084] Figure 1 A flowchart of a rock fragment lithology analysis method based on a multi-level fusion image algorithm and X-ray diffraction proposed by the present invention;

[0085] Figure 2 This is a schematic diagram of deep feature extraction of multi-level image features by a dual-path convolutional neural network model in a cuttings lithology analysis method based on a multi-level fusion image algorithm and X-ray diffraction proposed in the present invention. DETAILED DESCRIPTION

[0086] The present invention will now be described in further detail with reference to the accompanying drawings. These drawings are simplified schematic diagrams, which only illustrate the basic structure of the present invention in a schematic manner, and therefore only show the components related to the present invention.

[0087] refer to Figure 1-2 A rock fragment lithology analysis method based on hierarchical fusion image algorithm and X-ray diffraction comprises the following steps:

[0088] S1. Pre-process the rock cuttings samples to remove impurities and pollutants, and perform multispectral image acquisition to obtain rock cuttings image data at different resolutions;

[0089] S2. Use multi-scale image processing technology to process rock cuttings image data of different resolutions, identify the macroscopic lithology characteristics of rock cuttings samples at low resolution, identify the microstructure and texture characteristics at high resolution, and generate a multi-level rock cuttings image feature matrix;

[0090] S3, perform deep feature extraction on the multi-level rock cuttings image feature matrix through a dual-path convolutional neural network model, automatically learn the image features in the rock cuttings samples, and preliminarily generate lithology classification results;

[0091] S4. Use X-ray diffraction technology to perform crystal structure analysis on rock cuttings samples, obtain diffraction patterns of samples, identify mineral composition and crystal structure information of rock cuttings samples, and generate supplementary data for lithology analysis;

[0092] S5, integrating the supplementary data of lithology analysis with the preliminary generated lithology classification results, and finally generating the final lithology classification results through comparison, optimization and refinement of multimodal analysis algorithms;

[0093] S6. Input the final lithology classification results into the dynamic adaptive analysis system, which adaptively adjusts the analysis strategy according to the complexity of the rock cuttings sample and optimizes the parameters of the dual-path convolutional neural network model and the multimodal analysis algorithm through real-time feedback and learning mechanisms;

[0094] S7. Output a comprehensive analysis report including mineral distribution, crystal structure and final lithology classification results. The comprehensive analysis report includes a three-dimensional mineral distribution map, XRD spectrum interpretation and image analysis flow chart.

[0095] In this implementation, S1 includes the following specific steps:

[0096] S11, performing physical pretreatment on the collected rock cuttings samples, using screening equipment to remove particles with a particle size smaller than d1 and larger than d2, d1 is the minimum allowable diameter of the rock cuttings particles, d2 is the maximum allowable diameter of the rock cuttings particles, and the values ​​of d1 and d2 are set according to the initial particle size distribution of the rock cuttings samples;

[0097] S12, cleaning and drying the pretreated rock cuttings samples to remove impurities and moisture attached to the sample surfaces;

[0098] S13, using multi-spectral imaging equipment to collect images of rock cuttings samples, obtain rock cuttings image data at different resolutions, in the spectral bands λ1 to λ n Corresponding to r1 to r n The rock cuttings image data is collected at the resolution level of

[0099] S14, storing the collected rock cuttings image data in the form of a three-dimensional matrix T (λ, r, x, y):

[0100]

[0101] Among them, I i (λ j ,r k ,x,y) represents the spectral band λ j , resolution k And rock cuttings image data at spatial position (x, y), where x and y represent the spatial coordinates of the rock cuttings image data.

[0102] In this implementation, S2 includes the following specific steps:

[0103] S21, multi-scale processing of rock cuttings image data with different resolutions, low Preprocess the rock cuttings image data to reduce the noise of the rock cuttings image data and obtain the macroscopic lithological characteristics of the rock cuttings samples. macro :

[0104]

[0105] Among them, I(x′,y′,r low ) indicates that at low resolution r low The original rock cuttings image pixel value under G(xx′,yy′,σ macro ) is the Gaussian kernel function, and the kernel width is σ macro , represents the scale of the macroscopic feature, and the integration range is the macroscopic area Ω of the cuttings image macro ;

[0106] S22, using wavelet transform based image processing method to analyze high resolution high The cuttings image data is decomposed at multiple scales to extract the microscopic lithological structure characteristics of the cuttings image. micro :

[0107]

[0108] Among them, I(x+m,y+n,r high ) represents the pixel value of the rock chip image at high resolution, is the wavelet basis function, the scale factor is s, M and N are the spatial window sizes of the wavelet transform;

[0109] S23, microscopic lithology structural characteristics F micro and macroscopic lithological characteristics macro Fusion is performed to generate a multi-level rock cuttings image feature matrix T feature (x,y):

[0110]

[0111] Among them, F macro (x, y) represents the macroscopic lithological characteristics at the spatial position (x, y), F micro (x, y) represents the microscopic lithological structural characteristics at the same spatial position.

[0112] In this implementation, S3 includes the following specific steps:

[0113] S31, multi-level rock cuttings image feature matrix T feature ( x, y) is input into the dual-path dual-path convolutional neural network model, which includes two independent feature extraction paths, namely the spatial feature path and the texture feature path for processing different features of rock cuttings samples;

[0114] S32. Use multiple convolutional layers in the spatial feature path to extract features of the overall shape, structure and contour of the rock cuttings sample:

[0115]

[0116] in, Represents the convolution output feature map of the l-th layer spatial feature path, is the convolution kernel weight in the l-th layer spatial feature path, is the bias term, k is the convolution kernel size;

[0117] S33. After each convolution layer in the spatial feature path, the ReLU activation function is used to perform nonlinear processing on the convolution result, and the pooling layer is used to reduce the dimension of the spatial features:

[0118]

[0119] in, is the size of the pooling window, and the overall shape, structure, and contour information of the rock cuttings sample are preserved through the pooling operation;

[0120] S34. In the texture feature path, multiple convolutional layers are used to extract features of the subtle texture and microstructure of the rock cuttings sample:

[0121]

[0122] in, Represents the convolution output feature map of the l-th layer texture feature path, is the convolution kernel weight in the texture feature path of the lth layer, is the bias term;

[0123] S35. After each convolution layer in the texture feature path, the convolution result is processed nonlinearly using the ReLU activation function, and the texture feature is reduced in dimension using the pooling layer:

[0124]

[0125] Preserve subtle texture and microstructural information in rock cuttings samples through pooling operations;

[0126] S36, the feature map F extracted from the spatial feature path and the texture feature path spatial_pool ( x,y ) and F texture_pool ( x,y ) Fusion is performed to generate a comprehensive feature matrix T combined ( x,y ) :

[0127] T combined ( x,y ) =α·F spatial_pool ( x,y ) +β·F texture_pool ( x,y ) ;

[0128] Among them, α and β are fusion coefficients, which represent the weights of spatial features and texture features;

[0129] S37, the integrated feature matrix T combined (x,y) is input to the fully connected layer, and the Softmax classifier is used to classify the rock cuttings samples:

[0130]

[0131] Among them, P(C k ∣T combined ) represents the probability that the sample belongs to the kth lithology, W k and W j is the weight of the classifier, and K is the number of lithology classifications.

[0132] In this implementation, S4 includes the following specific steps:

[0133] S41, placing the pretreated rock cuttings sample in an X-ray diffractometer, irradiating the sample with X-rays and obtaining diffraction intensity data I(θ), where θ represents the diffraction angle;

[0134] S42, by analyzing the diffraction intensity data I(θ) and position, identify the type and content of each mineral component in the rock cuttings sample, and generate a mineral composition analysis table C mineral (x,y):

[0135]

[0136] Among them, I i (θ) is the diffraction angle θ of the i-th mineral component i The diffraction intensity under i is the diffraction peak width, Z is the normalization factor used to make the total percentage of the mineral components equal to 1, Indicates that in the diffraction angle range θ min to θ max The total diffraction intensity within, f mineral (x,y,θ i ) is the diffraction angle θ i The characteristic function of the relevant mineral composition describes the distribution characteristics of the mineral at the spatial position (x, y);

[0137] S43. Constructing a mineral crystal structure model based on the changes in diffraction intensity and diffraction angle crystal (x,y), mineral crystal structure model is used to describe the crystal structure characteristics of different mineral components:

[0138]

[0139] Among them, V k is the unit cell volume of the kth mineral, F hkl is the structure factor of the crystal plane family, G hkl is the reciprocal vector of the crystal plane family, describing the lattice periodicity, r(x,y) represents the atomic position vector at the spatial position (x,y), D hkl (T) is the temperature factor of the crystal, which indicates the effect of temperature on the diffraction intensity;

[0140] S44, mineral composition analysis table C mineral (x,y) and mineral crystal structure model S crystal (x,y) are fused to generate supplementary data D for lithology analysis supplement ( x,y ) :

[0141] D supplement ( x,y ) =α1·C mineral ( x,y ) +β1·S crystal ( x,y );

[0142] Among them, α1 and β1 are weight coefficients, which are used to balance the weights of the mineral composition analysis table and the mineral crystal structure model in the supplementary data.

[0143] In this implementation, S5 includes the following specific steps:

[0144] S51. Supplementary data D for lithology analysis supplement ( x,y ) Compared with the preliminary lithology classification results P(C k ∣T combined ) to obtain the initial fused feature matrix T fused (x,y):

[0145] T fused ( x,y ) =α2·P ( C k ∣T combined ) +β2·D supplement ( x,y ) ;

[0146] Among them, α2 and β2 are weight coefficients, which represent the weights of the preliminary lithology classification results and the supplementary data of lithology analysis in the fusion process;

[0147] S52, the initial fusion feature matrix T fused (x, y) performs multimodal analysis and calculates the similarity measure S between the preliminary classification results and the supplementary data similarity (x,y):

[0148] S similarity (x,y)=exp(-(P(C k ∣T combined )-D supplement (x,y) 2 );

[0149] The closer the similarity measure is to 1, the higher the consistency between the preliminary classification results and the supplementary data;

[0150] S53, according to the similarity measure S similarity (x,y) pairs of fused feature matrix T fused (x, y) is optimized, and the weight coefficients α2 and β2 are adjusted through the back propagation algorithm to maximize the similarity measure:

[0151]

[0152] Where η is the learning rate, and is the updated weight coefficient;

[0153] S54, the optimized fusion feature matrix T fused (x, y) is input into the classification model, and the Softmax classifier is used to perform the final refinement classification of the lithology category:

[0154]

[0155] Among them, P final (C k ∣T fused_opt ) represents the probability that the sample belongs to the kth lithology under the final fusion feature matrix.

[0156] In this implementation, S7 includes the following specific steps:

[0157] S71. Integrate the final lithology classification results with the mineral composition analysis table and the mineral crystal structure model to generate a comprehensive analysis report data set;

[0158] S72. Draw a three-dimensional mineral distribution map according to the mineral composition analysis table, where the three-dimensional mineral distribution map represents the distribution of minerals at spatial positions (x, y, z);

[0159] S73. Interpret the diffraction intensity data to generate a spectrum interpretation result including diffraction peak position, peak intensity and mineral composition.

[0160] Embodiment 1:

[0161] In August 2023, a geological exploration company was faced with the problem of analyzing complex rock chip samples while conducting oil and gas exploration in the Tarim Basin in Xinjiang. Due to the complex geological structure of the Tarim Basin, the exploration depth reached 4,000 to 4,500 meters, the rock types were diverse, and the sample composition was complex. Traditional rock chip lithology analysis methods were difficult to cope with complex situations. Therefore, the exploration team decided to use the rock chip lithology analysis method based on hierarchical fusion image algorithm and X-ray diffraction technology of the present invention to improve the accuracy and efficiency of lithology classification.

[0162] On August 15, 2023, the exploration team collected a batch of rock chip samples from a depth of 4,200 meters, numbered "XJ4200-15". The samples were analyzed by traditional methods. The results showed that the main components of the samples were quartz and feldspar. The analysis time was about 45 minutes. However, in terms of microstructure, the traditional method failed to identify other possible mineral components. The classification result was "quartz sandstone". When compared with historical data, analysts found that the rock types in the nearby area were mostly complex mixed rocks containing multiple minerals, which made the accuracy of the analysis results questionable.

[0163] Subsequently, the research team used the method of the present invention to further analyze the sample numbered "XJ4200-15". First, a multispectral imaging device was used to collect image data of the sample and obtain rock chip images at different resolutions. After preliminary processing at low resolution, the sample showed obvious macroscopic features of quartz and feldspar, but the researchers noticed that in the high-resolution image, there were still some tiny crystalline textures on the surface of the sample. Through deep learning and feature extraction of the image data by convolutional neural network, the system automatically generated preliminary classification results, and the results showed that in addition to quartz and feldspar, the sample may also contain a small amount of calcite and dolomite.

[0164] To verify the preliminary results, the research team further used an X-ray diffractometer to analyze the "XJ4200-15" sample. After calculating the diffraction intensity data, the system identified the presence of calcite and dolomite crystal structures in the sample, with a mineral content of approximately 13% calcite and 5% dolomite. Comprehensive analysis results showed that the sample should be classified as "calcite-containing quartz sandstone", a classification result that is more accurate than traditional methods.

[0165] In the analysis of another batch of rock chip samples numbered "XJ4300-25" on August 18, 2023, the traditional method took 50 minutes, and the analysis results showed that the sample was ordinary feldspar sandstone, but no trace minerals in the sample were identified. Through the method of the present invention, the sample was also subjected to image acquisition, convolutional neural network feature extraction and X-ray diffraction analysis, and it was finally determined that the sample contained a small amount of pyrite in addition to feldspar and quartz. The detection of pyrite was not identified in the traditional method, but under the method of the present invention, the microstructure of pyrite was clearly visible through the refinement of high-resolution images, and was further confirmed by X-ray diffraction data.

[0166] The analysis of the two groups of samples showed that the method of the present invention has significant advantages over the traditional method. On September 10, 2023, the exploration team processed 250 samples of the 500 samples collected using traditional methods. The average analysis time for each sample was 45 minutes, and the total time was about 18,750 minutes (about 312.5 hours). The classification accuracy of the samples was 78%, especially in the identification of trace minerals.

[0167] For the remaining 250 samples, the method of the present invention was used for analysis, with an average of 25 minutes per sample and a total time of about 6250 minutes (about 104.2 hours), which is 66% shorter than the total time of the traditional method. The classification accuracy was improved to 93%, especially in the identification of trace mineral components and detailed classification. The system accurately detected the microscopic mineral structure in a variety of samples, including the components of calcite, pyrite and other trace minerals. The accurate identification of trace components provides more accurate data support for the stratigraphic structure analysis of oil and gas exploration.

[0168] In the sample numbered "XJ4500-35", the traditional method failed to identify the dolomite in the sample. However, through the method of the present invention, after high-resolution image processing, the convolutional neural network model automatically extracted the microstructural characteristics of the dolomite. Subsequently, X-ray diffraction technology was used to further confirm that the sample contained about 12% dolomite. The researchers were able to accurately identify the sample as "quartz sandstone containing dolomite". This result greatly improved the accuracy of sample classification and provided important mineral distribution information for subsequent exploration.

[0169] Table 1 Sample analysis comparison data table

[0170]

[0171] From the data comparison in Table 1 above, it can be seen that the traditional method has a long analysis time and low classification accuracy when processing complex rock cuttings samples, especially in the identification of microscopic mineral components, the performance is not ideal. The method of the present invention not only significantly shortens the analysis time and improves the classification accuracy, but also shows a higher recognition rate in the detection of subtle mineral components.

[0172] Through the actual application in the oil and gas exploration in the Tarim Basin, the rock fragments lithology analysis method of the present invention has shown excellent performance in a variety of complex samples. It combines multi-scale image processing, deep learning algorithms and X-ray diffraction technology to achieve efficient and accurate lithology classification, especially in microstructure identification and multi-mineral composition analysis. Compared with traditional methods, the present invention shortens the analysis time by 66%, improves the classification accuracy by 15%, and improves the micro-mineral identification rate by 33%. The improvement significantly improves the efficiency and data accuracy of geological exploration and has broad application prospects.

[0173] The present invention adopts multi-scale image processing technology to capture the macroscopic lithological characteristics of rock cuttings samples at low resolution, and identify their microstructure and texture at high resolution. Through the dual-path convolutional neural network model, deep feature extraction is performed on feature matrices at different levels to automatically learn the complex rock cuttings image features in the samples and generate preliminary lithological classification results. Compared with traditional single image processing technology, it can identify subtle differences in rock cuttings on a wider scale, especially in complex samples, and shows higher accuracy. Experimental results show that the present invention improves the classification accuracy of different types of rock cuttings samples by about 15%.

[0174] The present invention innovatively integrates the mineral composition and crystal structure information obtained by X-ray diffraction technology with the lithology classification results generated by image processing algorithms, and optimizes them through a multimodal analysis algorithm. XRD technology can accurately identify the mineral composition of rock cuttings samples, while the image algorithm can provide detailed macro- and microstructural information. The combination of the two greatly improves the accuracy and robustness of lithology analysis. Comparative analysis and optimization can effectively solve the problem of insufficient classification accuracy in a single technology. The lithology classification results finally generated show obvious advantages in refining mineral composition and structural characteristics.

[0175] The present invention can quickly process a large number of rock cuttings samples through automated image processing and XRD analysis processes, and generate high-precision lithology classification reports in a relatively short time. The automatic learning of the dual-path convolutional neural network model and the back-propagation adjustment system of the optimization algorithm can adaptively adjust the analysis parameters to ensure the consistency and efficiency of the classification results. Compared with traditional manual analysis methods, the present invention greatly shortens the analysis time, and increases the analysis speed by about 30% in large-scale geological exploration projects, while maintaining a high classification accuracy, which can provide more timely decision support for the project.

[0176] The above description is only a preferred specific implementation manner of the present invention, but the protection scope of the present invention is not limited thereto. Any technician familiar with the technical field can make equivalent replacements or changes according to the technical scheme and inventive concept of the present invention within the technical scope disclosed by the present invention, which should be covered by the protection scope of the present invention.

Claims

1. A method for analyzing rock debris lithology based on multi-level fusion image algorithm and X-ray diffraction, characterized in that: The steps include: S1. Pre-process the rock cuttings samples to remove impurities and contaminants, and perform multispectral image acquisition to obtain rock cuttings image data at different resolutions; S2. Use multi-scale image processing technology to process rock cuttings image data of different resolutions, identify the macroscopic lithology characteristics of rock cuttings samples at low resolution, identify the microstructure and texture characteristics at high resolution, and generate a multi-level rock cuttings image feature matrix; S3, perform deep feature extraction on the multi-level rock cuttings image feature matrix through a dual-path convolutional neural network model, automatically learn the image features in the rock cuttings samples, and preliminarily generate lithology classification results; S4. Use X-ray diffraction technology to perform crystal structure analysis on rock cuttings samples, obtain diffraction patterns of samples, identify mineral composition and crystal structure information of rock cuttings samples, and generate supplementary data for lithology analysis; S5. The supplementary data of lithology analysis and the preliminary generated lithology classification results are integrated, and the comprehensive lithology classification results are finally generated through comparison, optimization and refinement of multimodal analysis algorithms; S6. Input the comprehensive lithology classification results into the dynamic adaptive analysis system, which adaptively adjusts the analysis strategy according to the complexity of the rock cuttings sample and optimizes the parameters of the dual-path convolutional neural network model and the multimodal analysis algorithm through real-time feedback and learning mechanisms; S7. Output an analysis report containing mineral distribution, crystal structure and comprehensive lithology classification results. The analysis report includes a three-dimensional mineral distribution map, XRD spectrum interpretation and image analysis flow chart.

2. The method for analyzing rock debris lithology based on multi-level fusion image algorithm and X-ray diffraction according to claim 1, characterized in that: The S1 comprises the following specific steps: S11, performing physical pretreatment on the collected rock cuttings samples, using screening equipment to remove particles with a particle size smaller than d1 and larger than d2, d1 is the minimum allowable diameter of the rock cuttings particles, d2 is the maximum allowable diameter of the rock cuttings particles, and the values ​​of d1 and d2 are set according to the initial particle size distribution of the rock cuttings samples; S12, cleaning and drying the pretreated rock cuttings samples to remove impurities and moisture attached to the sample surfaces; S13, using multi-spectral imaging equipment to collect images of rock cuttings samples, obtain rock cuttings image data at different resolutions, in the spectral bands λ1 to λ n Corresponding to r1 to r n The rock cuttings image data is collected at the resolution level of S14, storing the collected rock cuttings image data in the form of a three-dimensional matrix T (λ, r, x, y): Among them, I i (λ j ,r k ,x,y) represents the spectral band λ j , resolution k And rock cuttings image data at spatial position (x, y), where x and y represent the spatial coordinates of the rock cuttings image data.

3. The method for analyzing rock debris lithology based on hierarchical fusion image algorithm and X-ray diffraction according to claim 1, characterized in that: The S2 comprises the following specific steps: S21. Perform multi-scale processing on rock cuttings image data with different resolutions. low Preprocess the rock cuttings image data to reduce the noise of the rock cuttings image data and obtain the macroscopic lithological characteristics of the rock cuttings samples. macro : Among them, I(x′,y′,r low ) indicates that at low resolution r low The original rock cuttings image pixel value under G(xx′,yy′,σ macro ) is the Gaussian kernel function, and the kernel width is σ macro , represents the scale of the macroscopic feature, and the integration range is the macroscopic area Ω of the cuttings image macro ; S22, using wavelet transform based image processing method to analyze high resolution high The cuttings image data is decomposed at multiple scales to extract the microscopic lithological structure characteristics of the cuttings image. micro : Among them, I(x+m,y+n,r high ) represents the pixel value of the rock chip image at high resolution, is the wavelet basis function, the scale factor is s, M and N are the spatial window sizes of the wavelet transform; S23, the microscopic lithology structural characteristics F micro and macroscopic lithological characteristics macro Fusion is performed to generate a multi-level rock cuttings image feature matrix T feature (x,y): Among them, F macro (x, y) represents the macroscopic lithological characteristics at the spatial position (x, y), F micro (x, y) represents the microscopic lithological structural characteristics at the same spatial position.

4. The method for analyzing rock debris lithology based on hierarchical fusion image algorithm and X-ray diffraction according to claim 1, characterized in that: The S3 includes the following specific steps: S31, multi-level rock cuttings image feature matrix T feature (x, y) is input into the dual-path convolutional neural network model, which includes two independent feature extraction paths, namely the spatial feature path and the texture feature path for processing different features of rock cuttings samples; S32. Use multiple convolutional layers in the spatial feature path to extract features of the overall shape, structure and contour of the rock cuttings sample: in, Represents the convolution output feature map of the l-th layer spatial feature path, is the convolution kernel weight in the l-th layer spatial feature path, is the bias term, k is the convolution kernel size; S33. After each convolution layer in the spatial feature path, the ReLU activation function is used to perform nonlinear processing on the convolution result, and the pooling layer is used to reduce the dimension of the spatial features: in, is the size of the pooling window, and the overall shape, structure, and contour information of the rock cuttings sample are preserved through the pooling operation; S34. In the texture feature path, multiple convolutional layers are used to extract features of the subtle texture and microstructure of the rock cuttings sample: in, Represents the convolution output feature map of the l-th layer texture feature path, is the convolution kernel weight in the texture feature path of the lth layer, is the bias term; S35. After each convolution layer in the texture feature path, the convolution result is processed nonlinearly using the ReLU activation function, and the texture feature is reduced in dimension using the pooling layer: Preserve subtle texture and microstructural information in rock cuttings samples through pooling operations; S36, the feature map F extracted from the spatial feature path and the texture feature path spatial_pool (x,y) and F texture_pool (x, y) to generate a comprehensive feature matrix T combined (x,y): T combined (x,y)=α·F spatial_pool (x,y)+β·F texture_pool (x,y); Among them, α and β are fusion coefficients, which represent the weights of spatial features and texture features; S37, the fused comprehensive feature matrix T combined (x,y) is input to the fully connected layer, and the Softmax classifier is used to classify the rock cuttings samples: Among them, P(C k ∣T combined ) represents the probability that the sample belongs to the kth lithology, W k and W j is the weight of the classifier, and K is the number of lithology classifications.

5. The method for analyzing rock debris lithology based on hierarchical fusion image algorithm and X-ray diffraction according to claim 1, characterized in that: The S4 comprises the following specific steps: S41, placing the pretreated rock cuttings sample in an X-ray diffractometer, irradiating the sample with X-rays and obtaining diffraction intensity data I(θ), where θ represents the diffraction angle; S42, by analyzing the diffraction intensity data I(θ) and position, identify the type and content of each mineral component in the rock cuttings sample, and generate a mineral composition analysis table C mineral (x,y): Among them, I i (θ) is the diffraction angle θ of the i-th mineral component i The diffraction intensity under i is the diffraction peak width, Z is the normalization factor used to make the total percentage of the mineral components equal to 1, Indicates that in the diffraction angle range θ min to θ max The total diffraction intensity within, f mineral (x,y,θ i ) is the diffraction angle θ i The characteristic function of the relevant mineral composition describes the distribution characteristics of the mineral at the spatial position (x, y); S43. Constructing a mineral crystal structure model based on the changes in diffraction intensity and diffraction angle crystal (x,y), the mineral crystal structure model is used to describe the crystal structure characteristics of different mineral components: Among them, V k is the unit cell volume of the kth mineral, F hkl is the structure factor of the crystal plane family, G hkl is the reciprocal vector of the crystal plane family, describing the lattice periodicity, r(x,y) represents the atomic position vector at the spatial position (x,y), D hkl (T) is the temperature factor of the crystal, which indicates the effect of temperature on the diffraction intensity; S44, mineral composition analysis table C mineral (x,y) and mineral crystal structure model S crystal (x,y) are fused to generate supplementary data D for lithology analysis supplement (x,y): D supplement (x,y)=α1·C mineral (x,y)+β1·S crystal (x,y); Among them, α1 and β1 are weight coefficients, which are used to balance the weights of the mineral composition analysis table and the mineral crystal structure model in the supplementary data.

6. The method for analyzing rock debris lithology based on multi-level fusion image algorithm and X-ray diffraction according to claim 1, characterized in that: The S5 comprises the following specific steps: S51. Supplementary data D for lithology analysis supplement (x,y) and the preliminary lithology classification result P(C k ∣T combined ) to obtain the initial fused feature matrix T fused (x,y): T fused (x,y)=α2·P(C k ∣T combined )+β2·D supplement (x,y); Among them, α2 and β2 are weight coefficients, which represent the weights of the preliminary lithology classification results and the supplementary data of lithology analysis in the fusion process; S52, the initial fusion feature matrix T fused (x, y) performs multimodal analysis and calculates the similarity measure S between the preliminary classification results and the supplementary data similarity (x,y): S similarity (x,y)=exp(-(P(C k ∣T combined )-D supplement (x,y)) 2 ); The closer the similarity measure is to 1, the higher the consistency between the preliminary classification results and the supplementary data; S53, according to the similarity measure S similarity (x,y) pairs of fused feature matrix T fused (x, y) is optimized, and the weight coefficients α2 and β2 are adjusted through the back propagation algorithm to maximize the similarity measure: Where η is the learning rate, and is the updated weight coefficient; S54, the optimized fusion feature matrix T fused (x, y) is input into the classification model, and the Softmax classifier is used to perform the final refinement classification of the lithology category: Among them, P final (C k ∣T fused_opt ) represents the probability that the sample belongs to the kth lithology under the final fusion feature matrix.

7. The method for analyzing rock debris lithology based on multi-level fusion image algorithm and X-ray diffraction according to claim 1, characterized in that: The S7 comprises the following specific steps: S71. Integrate the comprehensive lithology classification results with the mineral composition analysis table and the mineral crystal structure model to generate an analysis report data set; S72. Draw a three-dimensional mineral distribution map according to the mineral composition analysis table, where the three-dimensional mineral distribution map represents the distribution of minerals at spatial positions (x, y, z); S73. Interpret the diffraction intensity data to generate a spectrum interpretation result including diffraction peak position, peak intensity and mineral composition.

Images