Rock type identification method and identification system based on artificial intelligence deep learning

Abstract

The invention provides a rock type identification method and identification system based on artificial intelligence deep learning, and relates to the field of rock identification, and the rock type identification method comprises the steps: carrying out the training of a neural network based on the actual features and actual types of training rock samples, and obtaining a rock type identification model, the actual features comprise surface features, structural features and infrared spectrums, and the infrared spectrums are used for representing the overall rock stratum distribution of the rock sample; acquiring feature data of a rock sample to be identified, inputting the feature data into the rock type identification model, and outputting a rock type by the rock type identification model; and determining mechanical data of the to-be-identified rock sample based on the rock type. According to the rock type identification method, the identification cost can be reduced while the identification precision is improved.

Description

Technical Field

[0001] The present invention relates to the field of rock identification, and in particular to a rock type identification method and identification system based on artificial intelligence deep learning. Background Art

[0002] In the fields of geotechnical engineering, mineral resource development, and geological exploration, the rapid identification of rock types has always been a key technical issue directly related to the progress of the project. The relevant rock type identification methods mainly rely on manual experience or identification through laboratory experiments. Manual identification requires a lot of experience from the identification personnel and the identification accuracy is not high. Experimental identification consumes a lot of time and cost. Summary of the Invention

[0003] The present invention provides a rock type identification method and identification system based on artificial intelligence deep learning, which are used to solve the technical problem of how to improve the accuracy of rock type identification while reducing the identification cost.

[0004] An embodiment of the present invention provides a rock type identification method based on artificial intelligence deep learning. The identification method includes: training a neural network based on actual characteristics and actual types of training rock samples to obtain a rock type identification model, wherein the actual characteristics include: surface characteristics, structural characteristics and infrared spectra, and the infrared spectrum is used to represent the overall rock layer distribution of the rock sample; obtaining characteristic data of the rock sample to be identified, inputting the characteristic data into the rock type identification model, and the rock type identification model outputting the rock type; and determining the mechanical data of the rock sample to be identified based on the rock type.

[0005] In some embodiments, the rock type identification model includes: a surface feature extraction layer, a structural feature extraction layer, a type feature extraction layer and a comprehensive identification layer; the inputting of the feature data into the rock type identification model, and the rock type identification model outputting the rock type includes: inputting the surface feature into the surface feature extraction layer to obtain a surface feature vector, and the surface feature vector is used to represent the surface texture, surface color and particle density of the rock sample to be identified; inputting the structural feature into the structural feature extraction layer to obtain a structural feature vector, and the structural feature vector is used to represent the pore and crack distribution of the rock sample to be identified; inputting the infrared spectrum into the type feature extraction layer to obtain a type distribution feature vector, and the type distribution feature vector is used to represent the rock type distribution of the rock sample to be identified; inputting the surface feature vector, the structural feature vector and the type distribution feature vector into the comprehensive identification layer, and the comprehensive identification layer outputs the rock type.

[0006] In some embodiments, the rock type identification model outputs the rock type, including: the rock type identification model outputs the rock type of different positions of the rock sample to be identified; and determining the mechanical data of the rock sample to be identified based on the rock type includes: determining the mechanical data of the rock sample to be identified based on the rock type of the rock sample to be identified and the position of the rock type.

[0007] In some embodiments, determining the mechanical data of the rock sample to be identified based on the rock type includes: determining the type mechanical data of the rock sample to be identified based on the rock type; determining the particle density of the rock sample to be identified based on the surface feature vector, and determining a loose correction coefficient based on the particle density, wherein the particle density is positively correlated with the loose correction coefficient; determining the number of pores and cracks of the rock sample to be identified based on the structural feature vector, and determining a structural correction coefficient based on the number of pores and cracks, wherein the structural correction coefficient is negatively correlated with the number of pores and cracks; multiplying the type mechanical data by the loose correction coefficient and the structural correction coefficient to obtain the mechanical data of the rock sample to be identified.

[0008] In some embodiments, the rock type identification model outputting the rock type includes: the identification type model outputting the rock type probability of the rock sample to be identified; and determining the rock type corresponding to the maximum probability as the rock type of the rock sample to be identified.

[0009] In some embodiments, the training of the neural network based on the actual characteristics and actual types of the training rock samples to obtain the rock type identification model includes: inputting the actual characteristics into the neural network model, and adjusting the weight parameters of the neural network model according to the difference between the rock type output by the neural network model and the actual type; constructing a confusion matrix based on the actual characteristics of different rock types, inputting the confusion matrix into the neural network model, and determining the prediction accuracy of the neural network model for different rock types based on the rock type output by the neural network model and the actual rock type corresponding to the confusion matrix; iteratively training the neural network model based on multiple groups of actual characteristics, and determining the prediction accuracy of the neural network model for different rock types based on the confusion matrix, until the prediction accuracy of the neural network model for each rock type is greater than the accuracy threshold, thereby obtaining the rock type identification model.

[0010] In some embodiments, obtaining characteristic data of the rock sample to be identified includes: obtaining image data of the rock sample to be identified at different angles through a high-definition camera, and fusing the image data to form the surface features; scanning the rock sample to be identified through a three-dimensional scanner to obtain a three-dimensional point cloud model of the rock sample to be identified, and obtaining the structural features based on the three-dimensional point cloud model; emitting infrared light in multiple frequency ranges to the rock sample to be identified through an infrared component, and obtaining the infrared spectrum reflected by the rock sample to be identified.

[0011] An embodiment of the present invention also provides a rock type identification system, which includes: a data acquisition component for acquiring characteristic data in the rock type identification method provided in the above embodiment; a data processing device for training a neural network based on the actual characteristics and actual types of training rock samples to obtain a rock type identification model; a device for acquiring characteristic data of a rock sample to be identified, inputting the characteristic data into the rock type identification model, and the rock type identification model outputting the rock type; and a device for determining mechanical data of the rock sample to be identified based on the rock type.

[0012] In some embodiments, the data acquisition component includes: a high-definition camera for acquiring image data of the rock sample to be identified at different angles; a three-dimensional scanner for scanning the rock sample to be identified to obtain a three-dimensional point cloud model of the rock sample to be identified; and an infrared component for emitting infrared light in multiple frequency ranges to the rock sample to be identified and acquiring the infrared spectrum reflected by the rock sample to be identified.

[0013] In some embodiments, the data acquisition component further includes: a turntable for carrying the rock sample to be identified and driving the rock sample to be identified to rotate; wherein, the high-definition camera is located on the side of the turntable in a direction perpendicular to the rotation axis of the turntable.

[0014] An embodiment of the present invention provides a rock type identification method, which includes: training a neural network model based on actual features and actual types of training rock samples to obtain a rock type identification model, obtaining feature data of the rock sample to be identified and inputting the feature data into the rock type identification model, the rock type identification model outputting the rock type and determining mechanical data of the rock sample to be identified based on the rock type, wherein the actual features include: surface features, structural features and infrared spectra, combining visual testing and infrared spectral testing, and performing a multi-dimensional comprehensive analysis of the visual features and infrared spectral features through a neural network to determine the rock type of the rock sample to be identified. The method can more accurately identify the type of the rock sample, and, without destroying the rock sample, can determine not only the rock type of the rock sample but also the distribution of the rock type in the rock sample. BRIEF DESCRIPTION OF THE DRAWINGS

[0015] Figure 1 A schematic flow chart of a first rock type identification method provided in an embodiment of the present invention;

[0016] Figure 2 A schematic flow chart of a second rock type identification method provided in an embodiment of the present invention;

[0017] Figure 3 A schematic flow chart of a third rock type identification method provided in an embodiment of the present invention;

[0018] Figure 4 A schematic flow chart of a fourth rock type identification method provided in an embodiment of the present invention;

[0019] Figure 5 A schematic flow chart of a fifth rock type identification method provided in an embodiment of the present invention;

[0020] Figure 6 A schematic flow chart of a sixth rock type identification method provided in an embodiment of the present invention;

[0021] Figure 7 A schematic flow chart of a seventh rock type identification method provided in an embodiment of the present invention;

[0022] Figure 8 A schematic diagram of the structure of a rock type identification system provided by an embodiment of the present invention;

[0023] Figure 9 A schematic diagram of the structure of a data acquisition component in a rock type identification system provided by an embodiment of the present invention.

[0024] Description of Reference Numerals

[0025] 10. Data acquisition component; 11. High-definition camera; 12. Three-dimensional scanner; 13. Infrared component; 14. Turntable; 20. Data processing device; 30. Display device. DETAILED DESCRIPTION

[0026] In order to make the objectives, technical solutions and advantages of the present invention more clear, the present invention is described in detail below with reference to the accompanying drawings and specific embodiments.

[0027] The various specific technical features in the various embodiments described in the specific implementation methods can be combined in various ways without contradiction. For example, different implementation methods can be formed by combining different specific technical features. In order to avoid unnecessary repetition, the various possible combinations of the specific technical features in the present invention will not be described separately.

[0028] It should also be noted here that, in order to avoid obscuring the present invention due to unnecessary details, only structures and / or processing steps closely related to the solutions of the present invention are shown in the drawings, while other details that are not closely related to the present invention are omitted.

[0029] In addition, it should be noted that the terms "include", "comprising" or any other variations thereof are intended to cover non-exclusive inclusion, so that a process, method, article or device comprising a series of elements includes not only those elements, but also other elements not explicitly listed, or also includes elements inherent to such process, method, article or device. In the following description, the terms "first\second\..." involved are merely used to distinguish different objects and do not indicate that there is any similarity or connection between the objects. It should be understood that the directions described by the directional nouns such as "above", "below", "inside" and "outside" are all directions in normal use.

[0030] In the following specific embodiments, the rock type identification method based on artificial intelligence deep learning can be applied to different usage scenarios. For example, the rock identification method can be applied to geotechnical engineering, ore exploration, and geological disaster monitoring. The specific steps of the rock type identification method based on artificial intelligence deep learning are exemplified below in combination with various embodiments.

[0031] In some embodiments, as Figure 1 As shown, Figure 1 A schematic flow chart of a first rock type identification method based on artificial intelligence deep learning provided in an embodiment of the present invention includes the following steps:

[0032] Step S101: training a neural network model based on actual characteristics and actual types of training rock samples to obtain a rock type recognition model.

[0033] Among them, the actual characteristics include: surface characteristics, structural characteristics and infrared spectra. The infrared spectrum is used to represent the overall rock layer distribution of the rock sample. It can be understood that the external surface characteristics and structural characteristics of the training rock sample are obtained through visual recognition, and the overall rock layer distribution of the rock is obtained through infrared spectra. The rock data and infrared spectrum data are obtained through visual recognition to perform multi-dimensional training on the neural network model, so that the trained rock type recognition model can more accurately identify the rock type; among them, the training rock sample can be understood as a data sample with a label, and each data sample is an actual feature and an actual type corresponding to the actual feature. The training rock sample can be a sample obtained through experiment, and the training sample can also be a sample obtained through database query.

[0034] Step S102: Acquire characteristic data of the rock sample to be identified.

[0035] Among them, the characteristic data includes the surface characteristics, structural characteristics and infrared spectrum of the rock sample to be identified. The shape and structural characteristics of the rock sample can be obtained through a visual sensor, which can be a camera, for example. The surface characteristics are the appearance characteristics that can be obtained through visual recognition, such as the surface texture distribution of the rock sample to be identified or the color distribution of the rock sample. The structural characteristics are, for example, the structural characteristics that can be obtained through visual recognition, which can be the pores and holes of the rock sample to be identified; the overall rock layer characteristics of the rock sample to be identified can be obtained through the infrared spectrum. It can be understood that different rock types have different infrared spectra. The distribution of different rock types in the rock sample to be identified can be obtained by the infrared spectra reflected back from different parts of the rock sample to be identified. It should be noted that compared with simply determining the rock type of the rock sample to be identified by infrared spectroscopy, this embodiment combines the surface characteristics, structural characteristics and infrared spectrum of the rock sample to be identified, and comprehensively determines the rock type of the rock sample to be identified from multiple dimensions, thereby being able to more accurately identify the rock type of the rock sample to be identified; compared with determining the composition of the rock sample to be identified by X-ray diffraction, the X-ray diffraction experiment requires the rock sample to be ground into powder, which not only requires a long experimental time, but also destroys the overall structure of the rock sample, and can only obtain the composition of the rock type of the rock sample, but cannot obtain the distribution characteristics of the rock type in the rock sample. This embodiment tests the rock sample through infrared spectroscopy, and there is no need to grind the rock sample to be identified, which not only saves experimental time, but also can obtain the composition of the rock sample and the distribution of different rock types in the rock sample without destroying the rock sample.

[0036] Step S103: input the characteristic data into a rock type recognition model, and the rock type recognition model outputs the rock type.

[0037] It can be understood that the surface features, structural features, and infrared spectra of the rock sample to be identified are input into the trained rock type identification model. The rock type identification model can comprehensively analyze the surface features, structural features, and infrared spectra of the rock sample to be identified and output the rock type of the rock sample to be identified in the output layer. The rock sample to be identified can be a neural network model with different structures. For example, the neural network model can be a multi-layer convolutional neural network model. The neural network model can also be a combination of a long short-term memory model and a convolutional neural network model. The neural network model can also be a back-propagation network model. The rock type output by the neural network output layer can be in different forms. For example, the rock type output by the neural network model is a preset rock type number, and the corresponding rock type can be determined by the number. The rock type output by the neural network model can also be a probability distribution corresponding to the preset rock type number, and the corresponding rock type can be determined by the rock type number with the highest probability.

[0038] Step S104: Determine the mechanical data of the rock sample to be identified based on the rock type.

[0039] Among them, the mechanical data of the rock sample includes compressive strength, elastic modulus and Poisson's ratio. Optionally, after determining the rock type, the mechanical data of the rock sample to be identified can be directly obtained according to the characteristics of the rock type. The correspondence between the rock type and the mechanical data can be achieved by querying the database, or by conducting mechanical experiments on different types of rock samples; optionally, after determining the rock type, a rock model can be constructed according to the rock structure and rock type, and the mechanical data of the rock sample can be obtained through simulation in the simulation program.

[0040] An embodiment of the present invention provides a rock type identification method, which includes: training a neural network model based on actual features and actual types of training rock samples to obtain a rock type identification model, obtaining feature data of the rock sample to be identified and inputting the feature data into the rock type identification model, the rock type identification model outputting the rock type and determining mechanical data of the rock sample to be identified based on the rock type, wherein the actual features include: surface features, structural features and infrared spectra, combining visual testing and infrared spectral testing, and performing a multi-dimensional comprehensive analysis of the visual features and infrared spectral features through a neural network to determine the rock type of the rock sample to be identified. The method can more accurately identify the type of the rock sample, and, without destroying the rock sample, can determine not only the rock type of the rock sample but also the distribution of the rock type in the rock sample.

[0041] In some embodiments, the rock type recognition model includes: a surface feature extraction layer, a structural feature extraction layer, a type feature extraction layer and a comprehensive recognition layer, such as Figure 2As shown, Figure 2 A schematic flow chart of a second rock type identification method provided in an embodiment of the present invention, based on Figure 1 , Figure 1 Step S103 in the embodiment includes:

[0042] Step S201: Input the surface features into the surface feature extraction layer to obtain a surface feature vector.

[0043] Specifically, the surface feature extraction layer extracts characteristic values of the surface features, which are represented by surface feature vectors. The surface feature vectors are used to represent the surface texture, surface color, and granularity of the rock sample to be identified. Optionally, the surface feature extraction layer is a one-dimensional convolutional neural network, which convolves the surface features with a one-dimensional convolution kernel that slides in the surface feature space, thereby obtaining a surface feature vector.

[0044] Step S202: Input the structural features into the structural feature extraction layer to obtain a structural feature vector.

[0045] That is, the structural feature extraction layer extracts the characteristic value of the structural feature, which is represented by a structural feature vector. The surface feature vector is used to represent the pore and fracture distribution of the rock sample to be identified. Optionally, the structural feature extraction layer is a three-dimensional convolutional neural network, which slides a three-dimensional convolution kernel in the structural feature space to perform convolution calculations with the structural feature, thereby obtaining a structural feature vector.

[0046] Optionally, the structural feature extraction layer includes multiple three-dimensional convolution layers, and each three-dimensional convolution layer is connected in series through a pooling layer. Through the extraction of multiple layers of three-dimensional convolution layers and pooling layers, redundant features in the structural features can be gradually removed, thereby retaining the key features in the structural features. In this way, not only low-level spatial features (low-level spatial features such as holes and cracks) can be extracted, but also more advanced spatial features can be gradually extracted. After multiple convolutions and pooling, the obtained three-dimensional feature map will be flattened into a one-dimensional vector, which is the structural feature vector.

[0047] Step S203: Input the infrared spectrum into the type feature extraction layer to obtain a type distribution feature vector.

[0048] Specifically, the type feature extraction layer extracts type distribution feature values, which are represented by type feature vectors. The type feature vectors are used to represent the rock type distribution of the rock sample to be identified. Optionally, the type feature extraction layer is a one-dimensional convolutional neural network that convolves the infrared spectral data with a one-dimensional convolution kernel that slides across the infrared spectral space, thereby obtaining the type distribution feature vector.

[0049] It can be understood that three different convolutional layers are used to extract feature vectors of three dimensions based on surface features, structural features and infrared spectra, respectively, to improve the accuracy of rock type identification. The order of steps S201 to S203 can be interchanged arbitrarily or executed in parallel.

[0050] Step S204: input the surface feature vector, the structural feature vector and the type distribution feature vector into the comprehensive identification layer, and the comprehensive identification layer outputs the rock type.

[0051] It can be understood that the surface feature vector, structural feature vector and type distribution vector are connected through the comprehensive recognition layer to comprehensively analyze the rock type from multiple dimensions and output the rock type; optionally, the comprehensive recognition layer is a fully connected layer.

[0052] In some embodiments, as Figure 3 As shown, Figure 3 A schematic flow chart of a third rock type identification method provided in an embodiment of the present invention, based on Figure 1 , Figure 1 Step S103 in the embodiment includes:

[0053] Step S301: input feature data into a rock type recognition model, and the rock type recognition model outputs rock types at different locations of a rock sample to be identified.

[0054] It can be understood that, for rock samples formed by different rock types, the rock type identification model can not only output the rock type of the rock sample to be identified, but also output the rock type at different positions of the rock sample, thereby obtaining the distribution of different rock types of the rock sample.

[0055] Figure 1 Step S104 in the embodiment includes:

[0056] Step S302: Determine the mechanical data of the rock based on the rock type and the location of the rock type of the rock sample to be identified.

[0057] It can be understood that not only is the mechanical data of a rock sample determined by its rock type, but for rock samples composed of different rock types, the mechanical data of the rock sample is comprehensively determined based on the rock type and the distribution of the different rock types, thereby making the obtained mechanical data more accurate. Optionally, mechanical data of different rock types can be separately queried by querying a database or experimental data, and then theoretical mechanical analysis can be performed based on the distribution of each rock type in the rock sample to obtain the mechanical data of the rock sample to be identified. Optionally, a model of the rock sample to be identified can be constructed in mechanical simulation software based on the rock type distribution obtained by identification, and the mechanical data of the rock sample to be identified can be obtained through simulation.

[0058] In some embodiments, as Figure 4 As shown, Figure 4 A schematic flow chart of a fourth rock type identification method provided in an embodiment of the present invention, based on Figure 1 , Figure 1 Step S104 in the embodiment includes:

[0059] Step S401: Determine the type mechanical data of the rock sample to be identified based on the rock type.

[0060] Among them, type mechanical data refers to the mechanical data when only the rock type and rock type distribution are considered. However, since the structure of the rock will also affect the mechanical data of the rock, only considering the type distribution is not enough to accurately reflect the mechanical data of the rock sample. The type mechanical data also needs to be corrected according to the rock structure.

[0061] Step S402: Determine the particle density of the rock sample to be identified based on the surface feature vector, and determine the porosity correction coefficient based on the particle density.

[0062] It can be understood that the surface feature vector obtained by the surface feature extraction layer in the rock type identification model is not only used to identify the rock type, but also can extract the particle density of the rock sample. The particle density is a parameter in the surface feature vector. The comprehensive identification layer not only outputs the rock type but also the particle density. Based on the output particle density, the looseness correction coefficient can be determined. The looseness correction coefficient is positively correlated with the particle density. It can be understood that the smaller the particle density of the rock sample to be identified, the looser its structure, and the mechanical data of the rock sample needs to be corrected downward accordingly.

[0063] Step S403: Determine the number of pores and cracks in the rock sample to be identified based on the structural characteristic vector, and determine a structural correction coefficient based on the number of pores and cracks.

[0064] It can be understood that the structural feature vector extracted by the structural feature extraction layer in the rock type identification model is not only used to identify the type of rock sample, but also can extract the number of pores and cracks in the rock sample. The number of pores and cracks is a parameter in the structural feature vector. The structural correction coefficient is determined based on the number of pores and cracks. The structural correction coefficient is negatively correlated with the number of pores and cracks. It can be understood that the more pores and cracks a rock sample has, the more likely it is that stress concentration will occur in the rock sample, and the mechanical data of the rock sample needs to be corrected downward accordingly.

[0065] Step S404: multiply the type mechanical data by the porosity correction coefficient and the structure correction coefficient to obtain the mechanical data of the rock sample to be identified.

[0066] It can be understood that, on the basis of obtaining the type mechanical data, the type mechanical data is corrected according to the appearance characteristics and structural characteristics of the rock sample to be identified, so as to obtain more accurate mechanical data.

[0067] In some embodiments, as Figure 5 As shown in FIG5 , a flow chart of the fifth rock type identification method provided by an embodiment of the present invention is shown in FIG5 . Figure 1 , Figure 1 Step S103 in the embodiment includes:

[0068] Step S501: input the characteristic data of the rock sample to be identified into a rock type identification model, and the rock type identification model outputs the rock type probability of the rock sample to be identified.

[0069] It can be understood that the rock sample to be identified is selected from a variety of preset rock types, and the rock type identification model outputs the probability of each rock type, that is, the probability distribution of the rock type. For example, the preset rock types include granite, limestone and sandstone, and the rock type probabilities output by the rock type identification model are the probability that the rock type of the rock sample to be identified is granite, the probability that the rock type of the rock sample to be identified is limestone, and the probability that the rock type of the rock sample to be identified is sandstone.

[0070] Step S502: Determine the rock type corresponding to the maximum probability as the rock type of the sample to be identified.

[0071] That is, the rock type corresponding to the maximum probability in the probability distribution is determined as the rock type of the rock sample to be identified.

[0072] In some embodiments, as Figure 6 As shown, Figure 6 A schematic flow chart of the sixth rock type identification method provided in an embodiment of the present invention, based on Figure 1 , Figure 1 Step S101 in the embodiment includes:

[0073] Step S601: The actual characteristic neural network model is used to adjust the weight parameters of the neural network model according to the difference between the rock type output by the neural network model and the actual type.

[0074] Optionally, a loss value is constructed based on the difference between the probability corresponding to the actual type output by the neural network model and 1, so that the weight parameter is adjusted according to the loss value. The weight parameter is adjusted by the gradient descent method until the loss value reaches the minimum value. At this time, it can be considered that the neural network has been trained once, and this training is the initial generation training.

[0075] Step S602: construct a confusion matrix based on the actual characteristics of different rock types, input the confusion matrix into the neural network model, and determine the prediction accuracy of the neural network model for different rock types based on the rock types output by the neural network model and the actual rock types corresponding to the confusion matrix.

[0076] It can be understood that after each training is completed, the confusion matrix is used to verify whether the neural network model meets the convergence conditions, that is, the actual characteristics of different rock types are input into the neural network model to determine the prediction accuracy of the neural network model for different types of rocks. Optionally, the prediction accuracy can be expressed by the area under the curve of the receiver operating characteristic curve (for the sake of simplicity, the receiver operating characteristic curve is referred to as the ROC curve below) (for the sake of simplicity, the area under the curve is represented by the AUC value below). That is, according to the probability distribution of the output of the neural network model for different types of rocks, multiple ROC curves are drawn respectively and the AUC value of each ROC curve is calculated. If the AUC values of the ROC curves obtained for different rock types are all greater than the preset threshold, it is considered that the neural network model has reached the convergence conditions and the training of the neural network model is completed. Otherwise, the neural network model needs to be further iteratively trained with more training data.

[0077] Step S603: Iteratively train the neural network model based on multiple sets of known features, and determine the prediction accuracy of the neural network model for different rock types based on the confusion matrix, until the prediction accuracy of the neural network model for each rock type is greater than the accuracy threshold, thereby obtaining a rock type recognition model.

[0078] It can be understood that, based on the initial generation neural network model, the neural network model is iteratively trained through a large amount of training data, and the convergence conditions of the trained neural network model are verified through the confusion matrix after each training, until the AUC values of each ROC curve obtained for different rock types are greater than the accuracy threshold. At this time, it can be considered that the neural network model training is completed, and the neural network model is a rock type recognition model, and the rock type recognition model can identify different rock types within the preset rock type range with high accuracy.

[0079] In some embodiments, as Figure 7 As shown, Figure 7 A flow chart of the seventh rock type identification method provided in an embodiment of the present invention, based on Figure 1 , Figure 1 Step S102 in the embodiment includes:

[0080] Step S701: Obtain image data of the rock sample to be identified from different angles through a high-definition camera, and fuse the image data to form surface features.

[0081] It can be understood that the image data of the rock sample at different angles is obtained by shooting with a high-definition camera, and then the images are fused and spliced according to the angle and edge of the image acquisition to obtain the surface image of the rock. The high-definition camera is a color camera. Each pixel in the surface image has not only grayscale information but also color information. After obtaining the surface image, the image needs to be preprocessed to obtain the surface features. The image preprocessing process is exemplified below.

[0082] The collected data is denoised using Gaussian filtering and bilateral filtering. The filtered and denoised image data is then normalized to eliminate the influence of the absolute grayscale on subsequent recognition. Gaussian filtering is suitable for image denoising. By adjusting the filter window size (e.g., 3×3 or 5×5), the image is smoothed to reduce the influence of random noise. The formula is as follows:

[0083]

[0084] Where G(x,y) is the weight of the two-dimensional Gaussian function, which represents the weighting coefficient at the point (x,y) in the image; σ is the standard deviation, which determines the expansion range of the Gaussian filter and controls the degree of image smoothing; x and y are the offsets between the current pixel and the center of the filter.

[0085] Bilateral filtering can retain edge information while removing noise, making it suitable for processing complex rock surface textures. This algorithm is suitable for rock images with relatively complex textures. Its formula is as follows:

[0086]

[0087] Where I'(p) is the filtered pixel value, which represents the output value at position p; I(q) is the original pixel value at position q in image I; Ω is the neighborhood area of the filtering operation; f r (||I(p)-I(q)||) is a weight function based on the difference in pixel values. A Gaussian function is usually used to calculate the similarity between pixel values. It is defined as:

[0088]

[0089] Where, σ T Is the standard deviation that controls the effect of pixel value differences, and a smaller σ T This will give pixels of similar colors higher weights.

[0090] f d (||pq|| is a weight function based on spatial distance, usually a Gaussian function, which calculates the similarity of pixels in a spatial neighborhood. It is defined as:

[0091]

[0092] Where, σ d is the standard deviation of the control space distance, the smaller σ d This will give closer pixels higher weights.

[0093] W p is a normalization coefficient used to ensure that the total weight of the filtered pixel values is 1, and is usually defined as:

[0094]

[0095] Optionally, after obtaining the denoised and normalized image data, the sample size of the image data can be expanded by segmenting, rotating, and scaling the image data, thereby improving the generalization ability of the rock sample recognition model obtained after training.

[0096] Step S702: Scan the rock sample to be identified with a three-dimensional scanner to obtain a three-dimensional point cloud model of the rock sample to be identified, and obtain structural features based on the three-dimensional point cloud model.

[0097] This can be understood as discretizing the rock sample to be identified into a point cloud model consisting of a finite number of points through 3D scanning. The coordinates of each point cloud in 3D space are used to represent the structural characteristics of the rock sample to be identified. Optionally, outlier removal methods are used to remove invalid isolated points in the point cloud model to achieve denoising of the point cloud model.

[0098] Step S703: emitting infrared light of multiple frequency ranges to the rock sample to be identified through the infrared component, and acquiring the infrared spectrum reflected by the rock sample to be identified.

[0099] The infrared component can be understood as consisting of an infrared light emitting device and a multi-band infrared sensor. The infrared light emitting device emits infrared light of different frequency ranges toward the rock sample to be identified. The multi-band infrared sensor captures infrared light reflected from the rock sample within the corresponding frequency range, thereby obtaining an infrared spectrum within the predetermined frequency range. Optionally, whiteboard normalization can be used to normalize the multispectral data, eliminating reflectivity differences between different devices and achieving standardization of the infrared spectrum. Optionally, principal component analysis can be used to extract information about the main spectral bands of the infrared spectrum, reducing redundant infrared spectrum data and reducing the workload of training and identification.

[0100] The embodiment of the present invention also provides a rock type identification system, which is used to achieve the following Figures 1 to 7 The rock type identification method shown in any one of the figures is described below. The structure and function of the identification system are exemplarily described in combination with various embodiments.

[0101] like Figure 8As shown, the identification system includes a data acquisition component 10 and a data processing device 20. The data acquisition component 10 is used to acquire characteristic data of the rock sample to be identified, so that the data processing device 20 can identify the rock type based on the characteristic data. The data processing device 20 can be a personal computer or a data processor. The data processing device 20 is used to train a neural network model based on the actual characteristics and actual type characteristics of the training rock samples to obtain a rock type identification model. The data acquisition component 10 is also used to acquire the characteristic data of the rock sample to be identified, input the characteristic data into the rock type identification model, and cause the rock type identification model to output the rock type of the rock sample to be identified. The data processing device 20 is also used to determine the mechanical data of the rock sample to be identified based on the rock type.

[0102] Optionally, the identification system further includes a display device 30, which is used to display the rock type and mechanical data of the rock sample to be identified.

[0103] In some embodiments, as Figure 9 As shown, the data acquisition component 10 includes: a high-definition camera 11, a three-dimensional scanner 12 and an infrared component 13. The high-definition camera 11 is used to obtain image data of the rock sample to be identified at different angles. The three-dimensional scanner 12 is used to scan the rock sample to be identified to obtain a three-dimensional point cloud model of the rock sample to be identified. The infrared component 13 is used to emit infrared light in multiple frequency ranges to the rock sample to be identified and obtain the infrared spectrum reflected by the rock sample to be identified.

[0104] Optional, such as Figure 9 As shown, the data acquisition component 10 also includes a turntable 14, which is used to carry the rock sample to be identified and drive the rock sample to be identified to rotate, wherein the high-definition camera 11 is located on one side of the turntable 14 in a direction perpendicular to the rotation axis of the turntable 14, so that the high-definition camera 11 can capture images of the rock sample to be identified at different angles through the rotation of the turntable 14.

[0105] The above embodiments are only used to illustrate the technical solutions of the present invention and are not intended to limit the present invention. Although the present invention has been described in detail with reference to the preferred embodiments, it should be understood by those skilled in the art that the technical solutions of the present invention may be modified or replaced by equivalents without departing from the spirit and scope of the technical solutions of the present invention.

Claims

1. A rock type identification method based on artificial intelligence deep learning, characterized in that: The rock type identification method comprises: Training a neural network based on actual features and actual types of training rock samples to obtain a rock type recognition model, wherein the actual features include: surface features, structural features, and infrared spectra, and the infrared spectra are used to represent the overall rock layer distribution of the rock samples; Acquire characteristic data of a rock sample to be identified, input the characteristic data into the rock type identification model, and the rock type identification model outputs the rock type; Mechanical data of the rock sample to be identified is determined based on the rock type.

2. The rock type identification method according to claim 1, characterized in that: The rock type identification model includes: a surface feature extraction layer, a structural feature extraction layer, a type feature extraction layer and a comprehensive identification layer; Inputting the characteristic data into the rock type identification model, and the rock type identification model outputting the rock type comprises: Inputting the surface features into the surface feature extraction layer to obtain a surface feature vector, wherein the surface feature vector is used to represent the surface texture, surface color and particle density of the rock sample to be identified; Inputting the structural features into a structural feature extraction layer to obtain a structural feature vector, wherein the structural feature vector is used to represent the pore and crack distribution of the rock sample to be identified; Inputting the infrared spectrum into the type feature extraction layer to obtain a type distribution feature vector, wherein the type distribution feature vector is used to represent the rock type distribution of the rock sample to be identified; The surface feature vector, the structural feature vector and the type distribution feature vector are input into the comprehensive identification layer, and the comprehensive identification layer outputs the rock type.

3. The rock type identification method according to claim 1 or 2, characterized in that: The rock type identification model outputs rock types including: The rock type identification model outputs the rock types at different positions of the rock sample to be identified; The determining of the mechanical data of the rock sample to be identified based on the rock type includes: Based on the rock type and the location of the rock type of the rock sample to be identified, mechanical data of the rock sample to be identified is determined.

4. The rock type identification method according to claim 2, characterized in that: The determining of the mechanical data of the rock sample to be identified based on the rock type includes: Determining type mechanical data of the rock sample to be identified based on the rock type; Determining the particle density of the rock sample to be identified based on the surface feature vector, and determining a porosity correction coefficient based on the particle density, wherein the particle density and the porosity correction coefficient are positively correlated; Determining the number of pores and cracks in the rock sample to be identified based on the structural feature vector, and determining a structural correction coefficient based on the number of pores and cracks, wherein the structural correction coefficient is negatively correlated with the number of pores and cracks; The type mechanical data is multiplied by the porosity correction coefficient and the structure correction coefficient to obtain the mechanical data of the rock sample to be identified.

5. The rock type identification method according to claim 1 or 2, characterized in that: The rock type identification model outputs rock types including: The identification type model outputs the rock type probability of the rock sample to be identified; The rock type corresponding to the maximum probability is determined as the rock type of the rock sample to be identified.

6. The rock type identification method according to claim 1, characterized in that: The training of the neural network based on the actual characteristics and actual types of the training rock samples to obtain the rock type recognition model includes: Inputting the actual characteristics into the neural network model, and adjusting the weight parameters of the neural network model according to the difference between the rock type output by the neural network model and the actual type; Constructing a confusion matrix based on actual characteristics of different rock types, inputting the confusion matrix into the neural network model, and determining the prediction accuracy of the neural network model for different rock types based on the rock types output by the neural network model and the actual rock types corresponding to the confusion matrix; The neural network model is iteratively trained based on multiple groups of actual features, and the prediction accuracy of the neural network model for different rock types is determined based on a confusion matrix, until the prediction accuracy of the neural network model for each rock type is greater than an accuracy threshold, thereby obtaining the rock type recognition model.

7. The rock type identification method according to claim 1, characterized in that: The step of obtaining characteristic data of the rock sample to be identified includes: Acquire image data of the rock sample to be identified at different angles through a high-definition camera, and fuse the image data to form the surface features; Scanning the rock sample to be identified by a three-dimensional scanner to obtain a three-dimensional point cloud model of the rock sample to be identified, and obtaining the structural features based on the three-dimensional point cloud model; Infrared light of multiple frequency ranges is emitted to the rock sample to be identified through an infrared component, and an infrared spectrum reflected by the rock sample to be identified is obtained.

8. A rock type identification system, characterized in that: The identification system comprises: A data acquisition component, configured to acquire characteristic data of a rock sample to be identified in the rock type identification method according to any one of claims 1 to 7; The data processing device is used to train a neural network based on the actual characteristics and actual types of training rock samples to obtain a rock type recognition model; is also used to obtain characteristic data of the rock sample to be identified, input the characteristic data into the rock type recognition model, and the rock type recognition model outputs the rock type; and is further used to determine the mechanical data of the rock sample to be identified based on the rock type.

9. The identification system according to claim 8, characterized in that The data acquisition component includes: A high-definition camera, used to obtain image data of the rock sample to be identified at different angles; A three-dimensional scanner, used for scanning the rock sample to be identified to obtain a three-dimensional point cloud model of the rock sample to be identified; The infrared component is used to emit infrared light of multiple frequency ranges to the rock sample to be identified and obtain the infrared spectrum reflected by the rock sample to be identified.

10. The identification system according to claim 9, characterized in that The data acquisition component also includes: A turntable, used for carrying the rock sample to be identified and driving the rock sample to be identified to rotate; Wherein, in a direction perpendicular to the rotation axis of the turntable, the high-definition camera is located on the side of the turntable.

Images