Granite texture surface layer pattern analysis method based on fractal dimension

Abstract

The invention discloses a granite texture surface layer pattern analysis method based on fractal dimension, and relates to the technical field of image recognition, and the method comprises the steps: obtaining texture images of a granite surface at a plurality of frequency bands, carrying out the multi-modal fusion, and generating an enhanced texture representation image; performing modal decomposition on the enhanced texture representation image to obtain a modal image sequence; the fractal dimension of each modal image in the modal image sequence is calculated, and a fractal heterogeneous feature set of granite texture is constructed; identifying a granite surface texture critical region based on the fractal heterogeneous feature set, and extracting a mineral crystal form feature set of the critical region; and constructing a comprehensive evaluation model based on the fractal isomerism feature set and the mineral crystal morphology feature set, predicting granite performance indexes and generating an analysis result. According to the method, a multi-band imaging and tensor decomposition combined feature fusion method is adopted, effective representation and fusion of cross-band features are realized through information entropy weight, and the integrity of the enhanced texture representation image is improved.

Description

Technical Field

[0001] The invention relates to the technical field of image recognition, in particular to a method for analyzing granite texture surface layer patterns based on fractal dimension. Background Art

[0002] Analysis of granite surface texture features has important applications in building material evaluation and geological engineering. Currently, two main approaches are used for granite texture analysis: one based on traditional image processing, such as grayscale statistics and texture feature extraction; the other based on physical measurements, such as roughness measurement and mineral composition analysis. However, these methods have significant limitations in practical applications and are unable to fully characterize the complexity and multi-scale characteristics of granite textures.

[0003] While fractal theory has been introduced to material surface analysis in recent years, existing fractal analysis methods are limited to single-band image processing and simple fractal dimension calculations. They lack the comprehensive utilization of multi-band information and in-depth research into the correlation between texture characteristics and the actual properties of granite. This makes it difficult to accurately predict the physical and mechanical properties and decorative effects of granite, limiting the scientific selection and quality assessment of granite in engineering applications. Summary of the Invention

[0004] The present invention provides a granite texture surface layer pattern analysis method based on fractal dimension, which is used to solve the technical problems in the prior art of incomplete granite texture feature extraction, insufficient utilization of multi-band information, and inaccurate measurement of the correlation between texture features and performance.

[0005] In view of this, the first aspect of the present invention provides a method for analyzing granite texture surface layer patterns based on fractal dimension, comprising: Acquire texture images of the granite surface in multiple frequency bands, and perform multimodal fusion on the texture images of multiple frequency bands to generate enhanced texture representation images; Performing modal decomposition on the enhanced texture representation image to obtain a modal image sequence; Calculate the fractal dimension of each modal image in the modal image sequence, and construct the fractal heterogeneous feature set of granite texture based on the fractal dimension; Based on the fractal heterogeneous feature set, the critical area of granite surface texture is identified, and the mineral crystal morphology feature set of the critical area is extracted; A comprehensive evaluation model is constructed based on the fractal heterogeneous feature set and the mineral crystal morphology feature set to predict the performance indicators of granite and generate granite texture analysis results.

[0006] Optionally, generating the enhanced texture representation image includes: A multi-band imaging system was used to obtain texture images of the granite surface in the visible and near-infrared bands, and the obtained multi-band texture images were corrected for illumination inhomogeneity. Extract the main texture features of the texture image in each frequency band and establish the feature mapping relationship between frequency bands; Based on the tensor decomposition method, the feature mapping relationship between frequency bands is processed to obtain cross-band shared features and frequency band-specific features; Construct a feature fusion function based on the information entropy weights of cross-band shared features and band-specific features; The fused features are reconstructed into the image space through back-projection transformation to generate an enhanced texture representation image.

[0007] Optionally, performing modal decomposition on the enhanced texture representation image to obtain a modal image sequence includes: Convert the enhanced texture representation image into a grayscale image and perform Gaussian filtering to reduce noise; By calculating the grayscale entropy of the grayscale image, the optimal number of modes for modal decomposition is adaptively determined; The initial center frequency and bandwidth parameters are set, and the variational mode decomposition algorithm is used to decompose the grayscale image into multiple modal functions according to the optimal number of modes; Calculate the energy contribution of the obtained modal functions and eliminate the modal functions whose energy contribution is lower than the energy contribution threshold; The retained modal functions are reconstructed in the spatial domain to generate a modal image sequence.

[0008] Optionally, the fractal heterogeneous feature set for constructing granite texture based on fractal dimension includes: The structural fractal dimension of each modality image was calculated using the box counting method, and the intensity fractal dimension of each modality image was calculated using the differential box counting method; Based on the structural fractal dimension and the intensity fractal dimension, the global fractal statistical characteristics are calculated; Calculate the multifractal spectrum through q-order moment analysis and extract the geometric characteristics of the fractal spectrum; Based on the global fractal statistical characteristics and fractal spectrum geometric characteristics, a fractal heterogeneous feature set of granite texture is constructed.

[0009] Optionally, identifying the critical region of granite surface texture based on the fractal heterogeneous feature set includes: Calculate the spatial gradient of the fractal heterogeneous feature set and construct a heterogeneity change rate map; Based on the heterogeneity change rate map, the threshold segmentation method is used to determine the critical area, and the morphological optimization and topological structure analysis are performed on the critical area. Perform microscopic imaging of critical areas to obtain mineral crystal distribution maps; Extract mineral crystal morphological feature sets from mineral crystal distribution maps.

[0010] Optionally, building a comprehensive evaluation model includes: The fractal isomerism feature set and the mineral crystal morphology feature set are standardized to construct the granite comprehensive feature vector. A neural network model was constructed, with the granite comprehensive feature vector as input parameter and granite performance index as prediction target; The neural network model was trained using granite samples, and the model parameters and network structure were optimized using cross-validation method. The comprehensive feature vector of the granite sample to be evaluated is input into the neural network model to obtain the prediction results of the performance index; According to the prediction results of the neural network model and combined with the technical requirements of different application scenarios, the granite texture analysis results are generated.

[0011] Optionally, calculating the structural fractal dimension of each modality image using a box counting method includes: Adaptive threshold segmentation is performed on the preprocessed modal image to obtain a binary image of the granite texture boundary; Construct a sequence of decreasing box sizes, divide the binary image into corresponding grids for each box size in the sequence, and count the number of non-empty grids; In the double logarithmic coordinate system, the correspondence between the number of non-empty grids and the inverse of the box size is established. The least squares method is used to fit the linear function, and the absolute value of the slope of the linear function is determined as the structural fractal dimension.

[0012] The beneficial effects of the present invention are as follows: the present invention adopts a feature fusion method combining multi-band imaging and tensor decomposition, realizes effective representation and fusion of cross-band features through information entropy weights, and improves the integrity of the enhanced texture representation image; proposes an adaptive modal decomposition and energy contribution screening mechanism based on grayscale entropy, realizes automatic determination of the number of modes and reasonable screening of key modes, and improves the accuracy of modal decomposition results; combines structural fractal dimension and intensity fractal dimension for multi-fractal analysis, realizes the joint representation of granite surface geometry and grayscale distribution, and improves the comprehensiveness of texture feature extraction; constructs a neural network model based on fractal heterogeneous feature set and mineral crystal morphology feature set, and realizes the prediction of granite performance indicators and evaluation in different application scenarios. BRIEF DESCRIPTION OF THE DRAWINGS

[0013] In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the following briefly introduces the drawings required for use in the embodiments or the description of the prior art. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying any creative work.

[0014] Figure 1 This is a flow chart of the granite texture surface layer pattern analysis method based on fractal dimension.

[0015] Figure 2 A flow chart was generated for the enhanced texture characterization of granite texture surface layer pattern analysis method based on fractal dimension.

[0016] Figure 3 A flow chart is constructed for the fractal heterogeneous feature set of the granite texture surface layer pattern analysis method based on fractal dimension. DETAILED DESCRIPTION

[0017] In order to make the purpose, features, and advantages of the present invention more obvious and easy to understand, the technical solutions in the embodiments of the present invention will be clearly and completely described below in conjunction with the drawings in the embodiments of the present invention. Obviously, the embodiments described below are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without making creative work are within the scope of protection of the present invention.

[0018] Reference Figures 1 to 3 , is an embodiment of the present invention, which provides a method for analyzing granite texture surface layer patterns based on fractal dimension. The flow chart of the method is as follows Figure 1 As shown, the method includes: S1: Acquire texture images of the granite surface in multiple frequency bands, and perform multimodal fusion on the texture images of multiple frequency bands to generate an enhanced texture representation image.

[0019] Specifically, the enhanced texture representation generation flow chart is as follows: Figure 2 As shown, the implementation process includes: S1.1: Use a multi-band imaging system to acquire texture images of the granite surface in the visible and near-infrared bands, and perform illumination non-uniformity correction on the acquired multi-band texture images.

[0020] The specific implementation of this step includes: configuring a multi-band imaging system, including a visible light camera, a near-infrared camera, and an ultraviolet camera (optional), where the ultraviolet camera can be selected according to specific detection requirements; conducting a camera spectral response test to ensure signal quality in the frequency band transition region; acquiring multi-band texture images of the same granite sample, ensuring that the optical axes of the cameras are parallel and the fields of view completely overlap; using a standard correction plate to collect grayscale response curves for each frequency band and constructing an illumination non-uniformity compensation function; correcting the original texture images of each frequency band according to the illumination non-uniformity compensation function to eliminate image brightness unevenness caused by lighting conditions; and spatially registering the corrected images of each frequency band using a multi-scale feature registration method based on mutual information combined with affine transformation to achieve precise registration.

[0021] Preferably, the present invention solves the technical problems of limited single-band imaging information and uneven image brightness through multi-band imaging system configuration and standardized correction process; through the illumination non-uniformity compensation function and multi-scale feature alignment based on mutual information, the image brightness unevenness phenomenon is effectively eliminated, and the precise spatial correspondence of images in each band is achieved, providing high-quality multi-band texture images for subsequent feature extraction.

[0022] S1.2: Extract the main texture features of the texture image in each frequency band and establish the feature mapping relationship between frequency bands.

[0023] Furthermore, based on the multi-band texture image after spatial registration, the wavelet transform multi-scale analysis method is used to extract the texture feature set, which includes at least local texture features, global statistical features and inter-band coupling features; the feature importance evaluation method is used to screen the main texture features from the texture feature set; the correlation coefficient of the screened inter-band features is calculated, and the inter-band feature mapping matrix is established based on the correlation coefficient; the inter-band feature mapping matrix is optimized by the least squares regression method to achieve effective mapping of inter-band texture features.

[0024] S1.3: Process the feature mapping relationship between frequency bands based on the tensor decomposition method to obtain cross-band shared features and frequency band-specific features.

[0025] Furthermore, the inter-band feature mapping matrix is reorganized into a third-order or higher-order tensor structure, comprising spatial, feature, and frequency-band dimensions. The constructed high-order tensor is then subjected to the Tucker tensor decomposition algorithm, which is suitable for extracting shared information between modalities, to obtain a core tensor and a factor matrix. By simultaneously analyzing the core tensor and the factor matrices of each dimension obtained through Tucker decomposition, the shared feature subspaces across frequency bands and the unique features of each frequency band are identified. Extracting both shared and unique features through tensor decomposition overcomes the shortcomings of traditional feature fusion methods in analyzing inter-band correlations.

[0026] Among them, the pattern combination corresponding to the high-value elements of the core tensor reflects the shared feature structure, while the feature weight distribution in the frequency band dimension factor matrix represents the unique information contribution of each frequency band.

[0027] S1.4: Construct a feature fusion function based on the information entropy weights of cross-band shared features and band-specific features.

[0028] Furthermore, the information entropy of the obtained cross-band shared features and the unique features of each frequency band is calculated respectively. The higher the entropy value, the richer the effective information contained in the feature. The information entropy weight normalization method is used to assign weight coefficients to the cross-band shared features and the unique features of each frequency band, so that the features with higher entropy values obtain greater weights. The feature fusion function is constructed based on the assigned weight coefficients.

[0029] During the feature fusion process, the weight distribution is dynamically adjusted according to the texture roughness parameters of the granite samples: coarse texture samples enhance the weight of macro-structural features, and fine texture samples enhance the weight of micro-texture features, so that the fused features have both structural integrity and texture discriminability.

[0030] S1.5: Reconstruct the fused features into the image space through back-projection transformation to generate an enhanced texture representation image.

[0031] Among them, the back-projection transformation is the process of mapping the fused features in the feature domain back to the image space. Specifically, it adopts the image reconstruction technology based on wavelet basis function, and restores the multi-scale features to the image hierarchy in sequence through the inverse transformation operation. The local adaptive enhancement algorithm is applied to optimize the texture detail expression and maintain the integrity and distinguishability of the granite surface structural features.

[0032] Through the above steps, the final enhanced texture representation image has higher information richness and contrast, and can fully characterize the texture characteristics of the granite surface.

[0033] S2: Perform modal decomposition on the enhanced texture representation image to obtain a modal image sequence.

[0034] Specifically, the implementation process of step S2 includes: S2.1: Convert the enhanced texture representation image into a grayscale image and perform Gaussian filtering to reduce noise.

[0035] In this step, the enhanced texture representation image is converted to a grayscale image using the RGB weighted averaging method. Specifically, the present invention uses the luminance conversion weights specified in the ITU-R BT.601 standard for RGB-to-grayscale conversion. This weighting scheme takes into account the human eye's high sensitivity to green and maximizes the contrast information of the granite texture.

[0036] S2.2: Adaptively determine the optimal number of modes for modal decomposition by calculating the grayscale entropy of the grayscale image.

[0037] Specifically, a grayscale histogram of the grayscale image is constructed, the occurrence probability distribution of each grayscale level in the grayscale histogram is calculated, and the image information entropy is calculated based on the occurrence probability distribution to quantify the texture complexity; the initial modal number range is determined based on the information entropy value, the initial value is set to 2, and the upper limit is determined based on the image information entropy value. The higher the information entropy, the larger the upper limit; within the initial modal number range, the following evaluation process is performed on each candidate modal number in turn: the parameters of modal decomposition are configured based on the candidate modal number, the reconstruction error under the current configuration is calculated through rapid simulation, and the error change rate between adjacent modal numbers is analyzed; when the error change rate is less than the preset change rate threshold (such as 5%), it is determined that increasing the modal number can no longer significantly improve the decomposition effect, and the corresponding modal number is the optimal modal number.

[0038] Advantageously, an adaptive optimization method based on information entropy and reconstruction error solves the technical problem of difficulty determining the number of modes in traditional modal decomposition. By combining information entropy with reconstruction error, an analytical approach overcomes the drawback of empirical assumptions that can easily lead to inaccurate decomposition. This solution automatically optimizes the number of modes and improves the accuracy and reliability of granite texture modal decomposition.

[0039] S2.3: Set the initial center frequency and bandwidth parameters, and use the variational mode decomposition algorithm to decompose the grayscale image into multiple modal functions according to the optimal number of modes.

[0040] It should be noted that the initial center frequency and bandwidth parameters are adaptively set based on the image spectrum analysis results and texture complexity. Specifically, by performing a Fast Fourier Transform (FFT) on the grayscale image and analyzing its spectral energy distribution, the spectrum range is evenly divided according to the optimal number of modes, and the center of each interval is used as the initial center frequency of the corresponding mode. The bandwidth parameter is determined based on a combination of grayscale entropy and spectral energy concentration. A higher entropy value indicates a more complex texture, and a wider bandwidth is set accordingly to capture more detailed features.

[0041] In addition, the variational mode decomposition algorithm was chosen because of its non-recursive decomposition characteristics and adaptive mode extraction capabilities. Compared with the traditional empirical mode decomposition (EMD) method, it can effectively avoid the modal aliasing problem and has a stronger separation ability for the complex multi-scale texture features of the granite surface.

[0042] S2.4: Calculate the energy contribution of the obtained modal functions, and eliminate modal functions whose energy contribution is lower than the energy contribution threshold.

[0043] Furthermore, the energy value of each modal function is calculated, which is the sum of the squares of the modal functions. The total energy value of all modal functions is calculated. The energy contribution of each modal function is determined based on the ratio of its energy value to the total energy value. An energy contribution threshold is determined, which is adaptively adjusted based on the granite texture complexity. Modal functions with energy contributions below the energy contribution threshold are eliminated, while modal functions with significant texture features are retained. Through an adaptive threshold mechanism based on granite texture complexity and a multi-dimensional energy contribution evaluation system, precise screening of modal functions is achieved, effectively removing noise and redundant modes while maximally retaining the key granite texture features, thereby improving feature extraction accuracy.

[0044] Among them, the energy contribution threshold is dynamically determined based on the information entropy value of the grayscale image. When the information entropy value is high (complex texture), a lower threshold is used to retain more features, and when the information entropy value is low (simple texture), a higher threshold is used to eliminate redundant modes.

[0045] S2.5: Reconstruct the retained modal functions in the spatial domain to generate a modal image sequence.

[0046] Furthermore, the retained modal functions are converted into two-dimensional spatial distribution functions; the two-dimensional spatial distribution functions are grayscale normalized so that their grayscale range is mapped to the standard interval; an adaptive spatial filter is designed according to the center frequency characteristics of each modal function; the two-dimensional spatial distribution functions are enhanced by the adaptive spatial filter; the processed two-dimensional spatial distribution functions are arranged in order of center frequency from high to low to generate a modal image sequence.

[0047] S3: Calculate the fractal dimension of each modal image in the modal image sequence, and construct a fractal heterogeneous feature set of granite texture based on the fractal dimension.

[0048] Specifically, the fractal heterogeneous feature set construction flow chart is as follows: Figure 3 As shown, the implementation process includes: S3.1: Preprocess the modality image sequence.

[0049] Preprocessing includes image normalization and contrast enhancement. Image normalization eliminates differences in brightness and grayscale distribution between images of different modalities, ensuring that fractal dimension calculations are performed under a unified standard. Contrast enhancement highlights the structural characteristics and boundary information of granite textures, improving the ability of fractal dimension calculations to capture texture complexity and self-similarity.

[0050] S3.2: Calculate the structural fractal dimension of each modality image using the box counting method, and calculate the intensity fractal dimension of each modality image using the differential box counting method.

[0051] Specifically, the box counting method is used to calculate the structural fractal dimension of each modal image, including: performing adaptive threshold segmentation on the preprocessed modal image to obtain a binary image of the granite texture boundary; constructing a decreasing box size sequence, dividing the binary image into corresponding grids for each box size in the sequence, and counting the number of non-empty grids, where the number of non-empty grids represents the number of grids containing granite texture boundary pixels; establishing a correspondence between the number of non-empty grids and the inverse of the box size in a double logarithmic coordinate system, and using the least squares method to fit a linear function. The absolute value of the slope of the linear function is determined as the structural fractal dimension.

[0052] Furthermore, the intensity fractal dimension of each modal image is calculated using the differential box counting method, which includes: constructing a three-dimensional surface for the preprocessed modal image, using the pixel position as the plane coordinate and the pixel grayscale value as the height coordinate; using a multi-scale box system to cover the three-dimensional surface, and calculating the minimum number of boxes required for coverage at each scale; establishing a linear relationship between the number of boxes and the inverse of the scale in a double logarithmic coordinate system, and obtaining a linear function by least squares fitting; and determining the absolute value of the slope of the linear function as the intensity fractal dimension.

[0053] S3.3: Calculate global fractal statistical characteristics based on the structural fractal dimension and the intensity fractal dimension.

[0054] Among them, the global fractal statistical features include weighted fractal dimension, maximum coefficient of variation and composite entropy. The weighted fractal dimension is determined by combining the arithmetic mean of the structural fractal dimension sequence with the arithmetic mean of the intensity fractal dimension sequence with a preset weight; the maximum coefficient of variation is determined by comparing the coefficient of variation of the structural fractal dimension sequence with the coefficient of variation of the intensity fractal dimension sequence and taking the maximum value of the two, where the coefficient of variation is the ratio of the standard deviation of the sequence to the arithmetic mean; the composite entropy is determined by combining the Shannon entropy of the structural fractal dimension sequence with the Shannon entropy of the intensity fractal dimension sequence with a preset weight; In this embodiment, the following parameter settings are adopted: the structural mean weight is 0.6, the intensity mean weight is 0.4; the structural entropy weight is 0.8, and the intensity entropy weight is 0.2.

[0055] S3.4: Calculate the multifractal spectrum through q-order moment analysis and extract the geometric characteristics of the fractal spectrum.

[0056] Among them, the fractal spectrum geometric characteristics include spectrum width and spectrum symmetry indicators.

[0057] Specifically, a multi-resolution measure was constructed for the preprocessed modal images. The images were divided into grids of different scales, and the probability distribution of the pixel grayscale values within each grid was calculated. The range of moment analysis parameters was set, and the generalized fractal dimension at different orders was calculated by varying the moment analysis parameters. The moment analysis parameter range was set to [-10, 10] based on the complexity of granite texture and computational stability, with a step size of 0.5, to balance computational accuracy and efficiency. Based on the generalized fractal dimension, the correspondence between the multifractal index and the multifractal spectrum was constructed to obtain the multifractal spectrum curve. The spectral width was extracted from the multifractal spectrum curve. The spectral width was defined as the difference between the maximum and minimum multifractal indices, which represents the degree of heterogeneity of the granite texture. The spectral symmetry index was calculated. The spectral symmetry index was determined by measuring the deviation of the shapes on the left and right sides of the fractal spectrum, which represents the structural complexity distribution of the granite texture.

[0058] S3.5: Construct a fractal heterogeneous feature set of granite texture based on global fractal statistical features and fractal spectrum geometric features.

[0059] By combining the calculation methods of structural and strength fractal dimensions, this approach overcomes the inability of traditional single fractal dimensions to fully characterize the complex textures of granite surfaces. By synergistically analyzing global fractal statistics and fractal spectrum geometry, this approach overcomes the limitations of conventional fractal analysis methods in multi-scale feature extraction. This approach not only improves the completeness of granite texture characterization but also provides a more reliable quantitative indicator for granite material performance evaluation, effectively addressing the technical issue of insufficient feature characterization in granite texture analysis.

[0060] S4: Identify the critical area of granite surface texture based on the fractal heterogeneous feature set, and extract the mineral crystal morphology feature set in the critical area.

[0061] Specifically, the implementation process of step S4 includes: S4.1: Calculate the spatial gradient of the fractal heterogeneous feature set and construct a heterogeneity change rate map.

[0062] Specifically, the fractal heterogeneous feature set is organized into feature vectors according to spatial position, and the feature vectors are normalized; a second-order difference operator is designed to calculate the gradients of the feature vectors in the horizontal and vertical directions respectively; feature smoothing at multiple scales is achieved through Gaussian filters, and the feature gradients at each scale are calculated and fused through adaptive weights to obtain a multi-scale gradient representation; based on the multi-scale gradient representation, the gradient amplitude of each feature component is calculated, and the feature weight is determined based on information gain or Fisher discriminant ratio; the comprehensive heterogeneity change rate is generated by weighted fusion of the gradient amplitudes of each feature component; the comprehensive heterogeneity change rate is subjected to tensor-guided filtering and local contrast enhancement to construct the final heterogeneity change rate map.

[0063] Preferably, the heterogeneity change rate map is constructed through multi-scale gradient analysis and feature fusion, which achieves the quantitative characterization of the spatial changes of fractal heterogeneous features and provides a reliable data basis for subsequent critical area identification.

[0064] S4.2: Based on the heterogeneity change rate map, the threshold segmentation method is used to determine the critical area, and the critical area is subjected to morphological optimization and topological structure analysis.

[0065] Furthermore, an iterative optimal segmentation algorithm is applied to the heterogeneous change rate map, and the optimal segmentation parameters are automatically determined based on the image statistical characteristics and the multimodal distribution of the histogram, and the region boundaries are optimized in combination with the region growing strategy; a morphological opening and closing joint operator is implemented on the initially identified critical regions to suppress high-frequency noise and maintain structural integrity; a regional topology analysis method is used to construct a connected domain feature description, and geometric feature vectors such as the area, boundary complexity, and compactness of the region are extracted; based on feature space clustering analysis, the morphological feature distribution law of typical heterogeneous regions is determined, and a discriminant function is constructed to automatically eliminate pseudo-critical regions; a boundary vector optimization algorithm is applied to accurately extract the contours of the retained regions to enhance the degree of fit with the actual crystal boundary, and finally a critical region feature descriptor is generated.

[0066] S4.3: Perform microscopic imaging of the critical area to obtain a map of the mineral crystal distribution.

[0067] S4.4: Extract a mineral crystal morphology feature set from the mineral crystal distribution map.

[0068] Specifically, the extraction process includes: segmenting the mineral crystal distribution map to divide each crystal area; obtaining the crystal size distribution through statistics of the major axis length, minor axis length and their ratio; generating the crystal orientation distribution through main direction measurement and statistics; evaluating the integrity of the crystal boundary based on edge characteristics, and the evaluation process includes edge continuity analysis, curvature change measurement and cross-boundary grayscale gradient calculation; analyzing the intercrystalline bonding strength through grain boundary structural characteristics, and the analysis content includes grain boundary width measurement, grayscale contrast evaluation and network connectivity analysis; integrating the crystal size distribution, crystal orientation distribution, crystal boundary integrity and intercrystalline bonding strength to form a mineral crystal morphological feature set.

[0069] The present invention achieves accurate and automatic identification of critical regions by constructing a heterogeneity change rate map based on a fractal heterogeneous feature set, overcoming the subjectivity and uncertainty of traditional manual judgment. The accuracy of critical region boundaries is improved through morphological optimization and topological analysis. Furthermore, a systematic crystal feature extraction method is used to achieve comprehensive quantitative characterization of the crystal structure in the critical region, resolving the technical problems of inaccurate critical region identification and incomplete crystal feature characterization in the prior art, and providing a more reliable analytical basis for the performance evaluation of granite materials.

[0070] S5: A comprehensive evaluation model is constructed based on the fractal heterogeneity feature set and the mineral crystal morphology feature set to predict the performance indicators of granite and generate granite texture analysis results.

[0071] Specifically, the implementation process of step S5 includes: S5.1: Normalize the fractal isomerism feature set and the mineral crystal morphology feature set to construct a comprehensive feature vector for granite.

[0072] S5.2: Construct a neural network model with the granite comprehensive feature vector as input parameter and granite performance index as prediction target.

[0073] Specifically, a dual-branch network structure is designed, in which the first branch receives the input of a fractal heterogeneous feature set, which contains multiple fully connected layers and nonlinear activation functions, and the second branch receives the input of a mineral crystal morphology feature set, adopting the same hierarchical structure; weight coefficients are set at the output ends of the two branches to achieve weighted fusion of features; an attention mechanism layer is added to enhance the feature fusion effect; a shared fully connected layer network is constructed, including multiple hidden layers and activation functions, and residual connections are introduced to establish a mapping relationship between fusion features and multiple performance indicators; prediction units for mechanical properties, physical properties and decorative properties are set in the output layer respectively, and each prediction unit outputs a corresponding confidence score.

[0074] Optimally, a dual-branch network structure and a weighted feature fusion mechanism address the technical challenge of unifying the heterogeneous fractal and crystal morphology feature sets. A multi-output structure with a shared fully connected layer overcomes the inadequate prediction accuracy of multiple performance indicators. This approach effectively integrates granite's comprehensive characteristics and enables the coordinated prediction of multiple performance indicators, improving the accuracy of performance evaluation.

[0075] S5.3: Use granite samples with pre-labeled performance indicators to train the neural network model, and use cross-validation method to optimize model parameters and network structure.

[0076] The granite samples with pre-labeled performance indicators included granite image samples with known mechanical, physical, and decorative performance test data. The sample dataset was divided into training and test sets in an 8:2 ratio. Stratified sampling was used to ensure a balanced distribution of samples across different performance indicator ranges.

[0077] In addition, the cross-validation method is used to optimize the model parameters and network structure, including: randomly dividing the training set into K subsets; using K-1 subsets in turn to train the model, and the remaining subset is used for validation; calculating the prediction error on the validation set, including the average relative error of various performance indicators, the accuracy of the prediction confidence score, and the stability of the feature fusion weight; by comparing the prediction errors of different cross-validation rounds, the optimal model parameter configuration is determined to obtain the optimized granite performance prediction model.

[0078] S5.4: Input the comprehensive feature vector of the granite sample to be evaluated into the neural network model to obtain the prediction results of the performance index.

[0079] S5.5: Generate granite texture analysis results based on the prediction results of the neural network model and the technical requirements of different application scenarios.

[0080] Furthermore, based on the predicted mechanical, physical, and decorative performance indicators and their confidence scores, the correspondence between granite texture characteristics and performance indicators was analyzed, specifically including: calculating the correlation coefficient matrix between texture characteristics and various performance indicators to identify key influencing factors; applying feature importance analysis techniques (such as permutation importance method and partial dependence plot analysis) to quantify the contribution of each texture parameter to performance; and determining the reliability interval of each performance prediction result based on the confidence score, and identifying high-uncertainty predictions.

[0081] Furthermore, the applicability of the granite samples to be tested in different application scenarios was evaluated by comparing the performance indicator requirements under different application scenarios. Specifically, this included: establishing a performance indicator requirement matrix for each application scenario, including the minimum threshold and ideal value range; calculating the matching score between the sample prediction performance and the scenario requirements, and using a weighted scoring method to comprehensively consider various indicators; and generating a suitability grade (preferred, applicable, conditionally applicable, not applicable) with an accompanying description of the restrictive conditions.

[0082] Furthermore, a granite texture analysis report is generated, including texture feature analysis results, performance prediction results and application scenario suggestions.

[0083] In summary, the present invention adopts a feature fusion method that combines multi-band imaging with tensor decomposition, realizes the effective representation and fusion of cross-band features through information entropy weights, and improves the integrity of the enhanced texture representation image; proposes an adaptive modal decomposition and energy contribution screening mechanism based on grayscale entropy, realizes the automatic determination of the number of modes and the reasonable screening of key modes, and improves the accuracy of the modal decomposition results; combines structural fractal dimension and intensity fractal dimension for multifractal analysis, realizes the joint representation of the geometric structure and grayscale distribution of the granite surface, and improves the comprehensiveness of texture feature extraction; constructs a neural network model based on fractal heterogeneous feature sets and mineral crystal morphology feature sets, and realizes the prediction of granite performance indicators and evaluation in different application scenarios.

[0084] The above embodiments are only used to illustrate the technical solutions of the present invention, rather than to limit the same. Although the present invention has been described in detail with reference to the aforementioned embodiments, those skilled in the art should understand that they can still modify the technical solutions described in the aforementioned embodiments, or make equivalent replacements for some of the technical features therein. However, these modifications or replacements do not deviate the essence of the corresponding technical solutions from the spirit and scope of the technical solutions of the various embodiments of the present invention.

Claims

1. A method for analyzing granite texture surface patterns based on fractal dimension, characterized in that: include: Acquire texture images of the granite surface in multiple frequency bands, and perform multimodal fusion on the texture images of the multiple frequency bands to generate an enhanced texture representation image; Performing modal decomposition on the enhanced texture representation image to obtain a modal image sequence; calculating the fractal dimension of each modal image in the modal image sequence, and constructing a fractal heterogeneous feature set of granite texture based on the fractal dimension; Identifying a critical region of granite surface texture based on the fractal heterogeneous feature set, and extracting a mineral crystal morphology feature set in the critical region; A comprehensive evaluation model is constructed based on the fractal isomerization feature set and the mineral crystal morphology feature set to predict the performance index of granite and generate granite texture analysis results.

2. The method for analyzing granite texture surface layer patterns based on fractal dimension according to claim 1, characterized in that: Generating the enhanced texture representation image comprises: A multi-band imaging system was used to obtain texture images of the granite surface in the visible and near-infrared bands, and the obtained multi-band texture images were corrected for illumination inhomogeneity. Extract the main texture features of the texture image in each frequency band and establish the feature mapping relationship between frequency bands; Processing the feature mapping relationship between the frequency bands based on a tensor decomposition method to obtain cross-band shared features and frequency band-specific features; Constructing a feature fusion function according to the information entropy weights of the cross-band shared features and the band-specific features; The fused features are reconstructed into the image space through back-projection transformation to generate an enhanced texture representation image.

3. The method for analyzing granite texture surface layer patterns based on fractal dimension according to claim 1, characterized in that: The step of performing modal decomposition on the enhanced texture representation image to obtain a modal image sequence includes: Converting the enhanced texture representation image into a grayscale image and performing Gaussian filtering to reduce noise; Adaptively determining the optimal number of modes for modal decomposition by calculating the grayscale entropy of the grayscale image; Setting initial center frequency and bandwidth parameters, and using a variational mode decomposition algorithm to decompose the grayscale image into multiple mode functions according to the optimal mode number; Calculate the energy contribution of the obtained modal functions and eliminate the modal functions whose energy contribution is lower than the energy contribution threshold; The retained modal functions are reconstructed in the spatial domain to generate a modal image sequence.

4. The method for analyzing granite texture surface patterns based on fractal dimension according to claim 1, characterized in that: The fractal heterogeneous feature set of granite texture constructed based on the fractal dimension includes: The structural fractal dimension of each modality image was calculated using the box counting method, and the intensity fractal dimension of each modality image was calculated using the differential box counting method; Calculating global fractal statistical features based on the structural fractal dimension and the intensity fractal dimension; Calculate the multifractal spectrum through q-order moment analysis and extract the geometric characteristics of the fractal spectrum; Based on the global fractal statistical characteristics and fractal spectrum geometric characteristics, a fractal heterogeneous feature set of granite texture is constructed.

5. The method for analyzing granite texture surface patterns based on fractal dimension according to claim 1, characterized in that: Identifying critical areas of granite surface texture based on the fractal heterogeneous feature set includes: Calculate the spatial gradient of the fractal heterogeneous feature set and construct a heterogeneity change rate map; Based on the heterogeneity change rate map, a threshold segmentation method is used to determine the critical area, and morphological optimization and topological structure analysis are performed on the critical area; Performing microscopic imaging on the critical area to obtain a mineral crystal distribution map; A mineral crystal morphology feature set is extracted from the mineral crystal distribution map.

6. The method for analyzing granite texture surface layer patterns based on fractal dimension according to claim 1, characterized in that: The construction of the comprehensive evaluation model includes: The fractal isomerism feature set and the mineral crystal morphology feature set are standardized to construct the granite comprehensive feature vector. Constructing a neural network model, taking the granite comprehensive feature vector as an input parameter and the granite performance index as a prediction target; The neural network model is trained using granite samples, and the model parameters and network structure are optimized using a cross-validation method; Inputting the comprehensive feature vector of the granite sample to be evaluated into the neural network model to obtain the prediction result of the performance index; According to the prediction results of the neural network model and combined with the technical requirements of different application scenarios, granite texture analysis results are generated.

7. The method for analyzing granite texture surface layer patterns based on fractal dimension according to claim 4, characterized in that: The method of calculating the structural fractal dimension of each modality image using the box counting method includes: Adaptive threshold segmentation is performed on the preprocessed modal image to obtain a binary image of the granite texture boundary; Constructing a sequence of decreasing box sizes, dividing the binary image into corresponding grids for each box size in the sequence, and counting the number of non-empty grids; In a double logarithmic coordinate system, a corresponding relationship between the number of non-empty grids and the inverse of the box size is established, and a linear function is obtained by least squares fitting. The absolute value of the slope of the linear function is determined as the structural fractal dimension.

Images