Method and device for identifying and classifying geological rock lithology

Abstract

The invention relates to the technical field of rock lithology identification, and discloses a geological rock lithology identification and classification method and device, and the method comprises the steps: obtaining a geological rock lithology image, carrying out the image processing of the geological rock lithology image, and obtaining a sample image data set; based on the sample image data set, performing category discrimination by using an image classification model and a region generation model to obtain a candidate region feature map; and carrying out frame regression detection on the candidate region feature map to obtain an identification and classification result of the geological rock lithology. According to the method, identification and classification of the geological rock lithology are realized, the accuracy and stability of identification and classification of the rock lithology are improved, and development of actual geological research work tasks and improvement of working efficiency are facilitated.

Description

Technical Field

[0001] The invention relates to the technical field of rock lithology identification, and in particular to a method and device for identifying and classifying geological rock lithology. Background Art

[0002] Rock lithology identification is mainly carried out by observing the rock's color, structure, texture, mineral composition and structure. Different types of rocks have different characteristics, and these characteristics can be used to accurately identify the type and nature of the rock. With the increasing shortage of mineral resources, the accuracy requirements for rock identification are getting higher and higher. At the same time, due to the complexity and variability of the geological environment, the rock identification system needs to have good stability and adaptability.

[0003] However, in the relevant geological survey process, it is difficult to identify geological rock properties, and the accuracy and stability of rock property identification and classification are low, which is not conducive to the implementation of actual geological research tasks and reduces work efficiency. Summary of the invention

[0004] In view of this, the present invention provides a method and device for identifying and classifying geological rock properties, so as to solve the problems of low accuracy and poor stability in rock property identification and classification.

[0005] In a first aspect, the present invention provides a method for identifying and classifying geological rock properties, the method comprising:

[0006] Acquire geological rock and lithology images, perform image processing on the geological rock and lithology images, and obtain a sample image data set;

[0007] Based on the sample image dataset, the image classification model and region generation model are used to perform category discrimination and obtain the candidate region feature map;

[0008] The feature map of the candidate area is subjected to bounding box regression detection to obtain the identification and classification results of geological rock properties.

[0009] The method for identifying and classifying geological rock lithology provided in this embodiment obtains geological rock lithology images, performs image processing on the geological rock lithology images, obtains a sample image data set, and then based on the sample image data set, uses an image classification model and a region generation model to perform category discrimination to obtain a candidate region feature map, and finally performs border regression detection on the candidate region feature map to obtain the identification and classification results of the geological rock lithology; in the process of realizing the identification and classification of geological rock lithology, the geological rock lithology images are used as the data source, and the image classification model and the region generation model are used to realize the identification and classification of geological rock lithology, thereby improving the accuracy and stability of rock lithology identification and classification, and facilitating the implementation of actual geological research tasks and improving work efficiency.

[0010] In an optional implementation, the geological rock lithology image is processed to obtain a sample image data set, including:

[0011] Segment and crop the geological rock lithology images to obtain multiple geological category images;

[0012] Labels are added to multiple geological category images respectively, and data preprocessing is performed on the labeled multiple geological category images to obtain a sample image data set.

[0013] The geological rock lithology identification and classification method provided in this embodiment obtains multiple geological category images by segmenting and cropping the geological rock lithology image, and then adds labels to the multiple geological category images respectively, and performs data preprocessing on the multiple geological category images after adding labels, which effectively improves the data quality in the sample image data set, optimizes the model performance and improves the reliability of the analysis results, and lays the foundation for subsequent model training.

[0014] In an optional implementation, based on a sample image dataset, an image classification model and a region generation model are used to perform category discrimination to obtain a candidate region feature map, including:

[0015] Based on the sample images in the sample image dataset, the candidate regions are selected using the region generation model;

[0016] Based on the candidate area, the image classification model is used to extract features to obtain a feature map of the candidate area; wherein the feature map of the candidate area includes a probability vector of each lithology category in the candidate area.

[0017] The method for identifying and classifying geological rock lithology provided in this embodiment selects candidate areas based on sample images in a sample image data set using a region generation model, and then extracts features based on the candidate areas using an image classification model to obtain a feature map of the candidate area, effectively extracts probability vectors of each lithology category in the candidate area, realizes effective identification and classification of rock lithology, and improves the accuracy and effectiveness of lithology identification and classification.

[0018] In an optional implementation, based on the sample image in the sample image data set, the candidate region is selected using the region generation model, including:

[0019] Based on the similarity of the sample image pixels in the sample image data set, the regions are grouped to obtain multiple target regions;

[0020] Acquire regional features of the target area, and calculate the overall similarity of the target area based on the regional features;

[0021] Candidate regions are determined based on the overall similarity of the target region.

[0022] The method for identifying and classifying geological rock lithology provided in this embodiment obtains multiple target areas by performing regional grouping based on the similarity of sample image pixels in a sample image data set, thereby achieving initial division of the sample image, and then obtaining regional features of the target area, calculating the overall similarity of the target area based on the regional features, and finally determining candidate areas based on the overall similarity of the target area. The hierarchical clustering method is used to achieve accurate merging of areas in the sample image, laying a foundation for subsequent classification and identification of rock lithology in candidate areas.

[0023] In an optional implementation, the image classification model includes a convolution layer, a spatial pyramid pooling layer, a fully connected layer, and an activation function layer; based on the candidate region, the image classification model is used to perform feature extraction to obtain a candidate region feature map, including:

[0024] The sample image corresponding to the candidate area is input into the convolution layer for convolution operation to generate ground feature;

[0025] Input the ground feature into the spatial pyramid pooling layer to perform pooling operations at different scales to obtain fixed-dimensional features;

[0026] Input fixed-dimensional features into the fully connected layer for feature concatenation;

[0027] The fixed-dimensional features after feature concatenation are input into the activation function layer for classification and recognition to obtain the candidate region feature map.

[0028] The method for identifying and classifying geological rock lithology provided in the present embodiment generates ground feature by inputting sample images corresponding to the candidate area into the convolution layer for convolution operation, and then inputting the ground feature features into the spatial pyramid pooling layer for pooling operations of different scales to obtain fixed-dimensional features, and inputting the fixed-dimensional features into the fully connected layer for feature splicing, and finally inputting the fixed-dimensional features after feature splicing into the activation function layer for classification and identification to obtain a candidate area feature map, and normalizes images of different scales into feature vectors of the same size before reaching the fully connected layer through the spatial pyramid pooling layer, so that multi-scale input of image data can be realized, and feature splicing is performed through the fully connected layer to realize classification and identification of different rock lithology probabilities.

[0029] In an optional implementation, border regression detection is performed on the candidate region feature map to obtain the identification and classification results of geological rock lithology, including:

[0030] Determining a marker recognition region based on a sample image data set;

[0031] Based on the marked recognition area and the candidate area in the candidate area feature map, a bounding box regression algorithm is used to determine the position offset of the candidate area;

[0032] Adjusting the position of the feature map of the candidate region based on the position offset of the candidate region;

[0033] The characteristic maps of the candidate regions after position adjustment are classified and counted to obtain the identification and classification results of geological rock lithology.

[0034] The method for identifying and classifying geological rock lithology provided in the present embodiment determines a marked identification area based on a sample image data set, determines the position offset of the candidate area based on the marked identification area and the candidate area in the candidate area feature map by using a bounding box regression algorithm, and then adjusts the position of the candidate area feature map based on the position offset of the candidate area. Finally, the candidate area feature map after the position adjustment is classified and counted to obtain the identification and classification results of geological rock lithology. By adjusting the position of the candidate area feature map, the position of the candidate area feature map is made close to the marked identification area, thereby achieving accurate positioning of the candidate area in the candidate area feature map, and improving the accuracy of rock lithology identification in geological surveys.

[0035] In a second aspect, the present invention provides a device for identifying and classifying geological rock properties, the device comprising:

[0036] An image processing module is used to obtain geological rock lithology images, perform image processing on the geological rock lithology images, and obtain a sample image data set;

[0037] The discrimination module is used to discriminate categories based on the sample image data set using the image classification model and the region generation model to obtain a candidate region feature map;

[0038] The detection module is used to perform border regression detection on the feature map of the candidate area to obtain the identification and classification results of geological rock lithology.

[0039] In a third aspect, the present invention provides a computer device, comprising: a memory and a processor, the memory and the processor being communicatively connected to each other, the memory storing computer instructions, and the processor executing the method for identifying and classifying geological rock lithology according to the first aspect or any corresponding embodiment thereof by executing the computer instructions.

[0040] In a fourth aspect, the present invention provides a computer-readable storage medium having computer instructions stored thereon, the computer instructions being used to enable a computer to execute the method for identifying and classifying geological rock lithology according to the first aspect or any corresponding embodiment thereof.

[0041] In a fifth aspect, the present invention provides a computer program product, comprising computer instructions for causing a computer to execute the method for identifying and classifying geological rock properties according to the first aspect or any corresponding embodiment thereof. BRIEF DESCRIPTION OF THE DRAWINGS

[0042] In order to more clearly illustrate the specific implementation methods of the present invention or the technical solutions in the prior art, the drawings required for use in the specific implementation methods or the description of the prior art will be briefly introduced below. Obviously, the drawings described below are some implementation methods of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying creative work.

[0043] Figure 1 It is a schematic diagram of a flow chart of a method for identifying and classifying geological rock properties according to an embodiment of the present invention;

[0044] Figure 2 is a schematic flow chart of another method for identifying and classifying geological rock properties according to an embodiment of the present invention;

[0045] Figure 3 is a schematic diagram of a process of a rock lithology identification and classification method according to an embodiment of the present invention;

[0046] Figure 4 is a schematic flow chart of another method for identifying and classifying geological rock properties according to an embodiment of the present invention;

[0047] Figure 5 is a principle diagram of a rock lithology classification and identification algorithm based on CNN according to an embodiment of the present invention;

[0048] Figure 6 is a schematic flow chart of another method for identifying and classifying geological rock properties according to an embodiment of the present invention;

[0049] Figure 7 It is a schematic diagram of the principle framework of rock lithology classification and identification based on the Tensorflow platform according to an embodiment of the present invention;

[0050] Figure 8 It is a schematic diagram of a deep learning model framework designed for TORL recognition based on the Tensorflow platform according to an embodiment of the present invention;

[0051] Fig. 9 is a structural block diagram of a method and device for identifying and classifying geological rock properties according to an embodiment of the present invention;

[0052] Fig.10 It is a schematic diagram of the hardware structure of a computer device according to an embodiment of the present invention. DETAILED DESCRIPTION

[0053] In order to make the purpose, technical solution and advantages of the embodiments of the present invention clearer, the technical solution 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 described embodiments are 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 those skilled in the art without creative work are within the scope of protection of the present invention.

[0054] With the continuous advancement of artificial intelligence and computer vision technology, the application of AI (Artificial Intelligence) technology in the identification and classification of rock lithology has shown an upward trend. In terms of technological development, the accuracy of rock lithology identification and classification has been continuously improved by using deep learning technology, and the processing speed is also accelerating. Methods based on deep learning, such as Convolutional Neural Networks (CNN), have shown good results in rock lithology identification; in terms of data fusion, rock lithology identification is developing in the direction of multi-source information fusion. By integrating multiple data sources such as rock images, spectral information, and elemental analysis, the limitations of a single information source can be broken through, and the accuracy and reliability of rock lithology identification can be improved. In addition, with the in-depth development of technology, rock lithology identification will gradually become automated and intelligent. Through machine learning algorithms, AI systems can automatically parse complex information in geological images, automatically identify geological elements such as lithology, structure, and mineral deposits, and generate corresponding structural models, providing strong support for geological disaster prevention and resource exploration.

[0055] In the relevant geological survey process, the identification of geological rock lithology is difficult and not accurate enough, which is not conducive to the implementation of actual work tasks and reduces work efficiency. Therefore, the automatic, fast and accurate identification of rock lithology through AI technology can better serve actual production and work, meet application needs, and improve quality and efficiency. With the increasing tension of mineral resources, the accuracy requirements for rock identification are getting higher and higher. At the same time, due to the complexity and variability of the geological environment, the rock identification system needs to have good stability and adaptability. In addition, in the field of geological exploration, the processing speed requirements for rock identification are also high.

[0056] Therefore, how to improve the operating efficiency of the algorithm to meet the needs of real-time rapid response and accurate identification of rock lithology has become the main task of geological research.

[0057] The embodiment of the present invention provides a method for identifying and classifying geological rock lithology. The method uses geological rock lithology images as data sources, identifies TORL (Typical Objects of Rock Lithology) through an improved CNN model algorithm, and integrates an image classification network and a region generation network to extract features of TORL and quickly generate candidate region windows. It also combines border regression and classification statistics to improve the recognition accuracy of TORL.

[0058] According to an embodiment of the present invention, an embodiment of a method for identifying and classifying geological rock lithology is provided. It should be noted that the steps shown in the flowchart of the accompanying drawings can be executed in a computer system such as a set of computer executable instructions, and although a logical order is shown in the flowchart, in some cases, the steps shown or described can be executed in an order different from that shown here.

[0059] In this embodiment, a method for identifying and classifying geological rock properties is provided, which can be used in the above-mentioned electronic device. Figure 1 FIG. 1 is a flow chart of a method for identifying and classifying geological rock properties according to an embodiment of the present invention. Figure 1 As shown, the process includes the following steps:

[0060] Step S101, obtaining geological rock lithology images, performing image processing on the geological rock lithology images, and obtaining a sample image data set.

[0061] Specifically, geological rock and lithology images are images that contain typical geological categories, including stratigraphic sections, igneous rock sections, metamorphic rock phase sections, sedimentary rock phase sections, structural features, paleontology, minerals and mineral deposits, landforms and landscapes, and water landscapes.

[0062] Step S102, based on the sample image data set, classify the categories using the image classification model and the region generation model to obtain a candidate region feature map.

[0063] Step S103, performing border regression detection on the candidate region feature map to obtain the identification and classification results of geological rock properties.

[0064] Specifically, the recognition and classification results of the geological rock lithology include the position of the geological rock lithology in the image and the category of the geological rock lithology.

[0065] The method for identifying and classifying geological rock lithology provided in this embodiment obtains geological rock lithology images, performs image processing on the geological rock lithology images, obtains a sample image data set, and then based on the sample image data set, uses an image classification model and a region generation model to perform category discrimination to obtain a candidate region feature map, and finally performs border regression detection on the candidate region feature map to obtain the identification and classification results of the geological rock lithology; in the process of realizing the identification and classification of geological rock lithology, the geological rock lithology images are used as the data source, and the image classification model and the region generation model are used to realize the identification and classification of geological rock lithology, thereby improving the accuracy and stability of rock lithology identification and classification, and facilitating the implementation of actual geological research tasks and improving work efficiency.

[0066] In this embodiment, a method for identifying and classifying geological rock properties is provided, which can be used in the above-mentioned electronic device. Figure 2 FIG. 1 is a flow chart of a method for identifying and classifying geological rock properties according to an embodiment of the present invention. Figure 2 As shown, the process includes the following steps:

[0067] Step S201, obtaining geological rock lithology images, performing image processing on the geological rock lithology images, and obtaining a sample image data set.

[0068] Specifically, the above step S201 includes:

[0069] Step S2011, segment and crop the geological rock lithology image to obtain a plurality of geological category images.

[0070] Specifically, the geological rock and lithology images are segmented and cropped according to typical geological categories.

[0071] Step S2012: adding labels to the plurality of geological category images respectively, and performing data preprocessing on the labeled plurality of geological category images to obtain a sample image data set.

[0072] Specifically, the label includes a mark identification area, and the mark identification area includes the position of the geological rock lithology in the sample image and the category of the geological rock lithology.

[0073] Furthermore, if Figure 3 As shown, noise filtering and image enhancement are performed on multiple geological category images after adding labels to obtain a sample image dataset, which is then divided into a training set and a test set according to a preset ratio. The training set is used to train the image classification model and the region generation model to obtain a category discrimination network, and the test set is used to test the recognition effect of the category discrimination network.

[0074] Step S202: Based on the sample image data set, the image classification model and the region generation model are used to perform category discrimination to obtain a candidate region feature map. Figure 1 Step S102 of the illustrated embodiment will not be described in detail here.

[0075] Step S203, perform border regression detection on the candidate region feature map to obtain the identification and classification results of geological rock lithology. Figure 1 Step S103 of the illustrated embodiment will not be described in detail here.

[0076] The geological rock lithology identification and classification method provided in this embodiment obtains multiple geological category images by segmenting and cropping the geological rock lithology image; then labels are added to the multiple geological category images respectively, and data preprocessing is performed on the multiple geological category images after adding labels to obtain a sample image data set, which improves image quality, enhances geological rock lithology information, reduces noise and interference, and lays a foundation for the subsequent deep learning model training.

[0077] In this embodiment, a method for identifying and classifying geological rock properties is provided, which can be used in the above-mentioned electronic device. Figure 4 FIG. 1 is a flow chart of a method for identifying and classifying geological rock properties according to an embodiment of the present invention. Figure 4 As shown, the process includes the following steps:

[0078] Step S401, obtain geological rock lithology images, perform image processing on the geological rock lithology images, and obtain a sample image data set. Figure 2 Step S201 of the illustrated embodiment will not be described in detail here.

[0079] Step S402: Based on the sample image data set, the image classification model and the region generation model are used to perform category discrimination to obtain a candidate region feature map.

[0080] Specifically, the above step S402 includes:

[0081] Step S4021, based on the sample images in the sample image dataset, a candidate region is selected using a region generation model.

[0082] Specifically, since the candidate region detection method takes a long time to generate a detection frame, such as the adaptive enhancement algorithm in OpenCV (a cross-platform computer vision and machine learning software library released under a licensed open source license) uses a sliding window combined with an image pyramid to generate a detection frame and an edge detection (EB) method used in deep learning to generate a detection frame; therefore, considering the defects of the above-mentioned candidate region detection methods, a selective search (SS) algorithm is used to replace the sliding window and EB methods to improve the generation speed of the detection frame; among them, SS is an algorithm that applies the idea of ​​hierarchical clustering to region merging. For the entire image, SS first groups regions according to the similarity of pixels, and then combines the regions according to features such as texture, color and area between regions.

[0083] In some optional implementations, the above step S4021 includes:

[0084] Step a1: grouping regions based on the similarity of sample image pixels in the sample image data set to obtain multiple target regions.

[0085] Specifically, assuming that there are n pre-segmented regions in the sample image data set, represented by R = {R1, R2, ..., Rn}, the corresponding similarity calculation operation is performed between each region in the sample image and its adjacent regions to obtain an n*n similarity matrix, where the similarity between the same region and between two non-adjacent regions in the sample image can be set to an empty set Then obtain two areas with the maximum similarity values ​​from the similarity matrix and integrate the two areas together; then obtain two areas with the maximum similarity values ​​from the similarity matrix and integrate the two areas together. At this time, there are n-1 areas remaining in the sample image; repeat the above calculation process to calculate the new similarities between the remaining new areas in the sample image and their adjacent areas. The areas that have participated in the calculation do not need to be calculated repeatedly. Each time the calculation is repeated, the total number of areas to be grouped in the sample image is reduced by 1 until the final segmented areas are merged into the target area.

[0086] Among them, the computer languages ​​corresponding to the regional groups are as follows:

[0087]

[0088] Extract target area boxes from all regions R: L

[0089] Step a2, obtaining the regional features of the target area, and calculating the overall similarity of the target area based on the regional features.

[0090] Specifically, each target region is combined according to its regional features such as texture, color and area, that is, the similarity between different target regions is calculated using color similarity, texture similarity, size similarity and block matching similarity, that is, the overall similarity of the target regions.

[0091] Furthermore, the color similarity, texture similarity, size similarity and block matching similarity are weighted and integrated by coefficients to obtain the overall similarity of the two target areas. The calculation formula of the overall similarity of the target area is as follows:

[0092] s(r i , r j )=α1s color (r i , r j )+α2s stripes (r i , r j )+α3s size (r i , r j )+α4s adapt (r i , r j ) (1)

[0093] Among them, s(r i , r j ) represents the similarity of adjacent target regions, r i represents the target area corresponding to pixel i, r j represents the target area corresponding to pixel j, s color (r i , r j ) represents the color similarity of adjacent target areas, s stripes (r i , r j ) represents the texture similarity of adjacent target regions, s size (r i , r j ) represents the size similarity of adjacent target regions, s adapt (r i , r j ) represents the block matching similarity of adjacent target areas, α1 represents the weight of color similarity, α2 represents the weight of texture similarity, α3 represents the weight of size similarity, and α4 represents the weight of matching similarity.

[0094] Step a3: determine the candidate region based on the overall similarity of the target region.

[0095] Specifically, the overall similarities of the target regions are sorted from large to small, and the target region corresponding to the maximum overall similarity is selected as the candidate region.

[0096] Step S4022: Based on the candidate region, feature extraction is performed using an image classification model to obtain a feature map of the candidate region; wherein the feature map of the candidate region includes a probability vector of each lithology category in the candidate region.

[0097] Specifically, Figure 5 As shown, the image classification model includes convolution ( co n v ) layer, Spatial Pyramid Pooling Network (SSP Net) layer, fully connected layer and activation function (softmax) layer; among them, since the network layers such as conv (convolution), pooling (pooling), ReLU (Rectified Linear Unit, a commonly used nonlinear activation function, mainly used in convolutional neural networks, which can effectively extract the nonlinear features of data) do not require fixed-scale input, the different sizes of the input images result in different scales of the obtained feature maps, which in turn leads to the inability to directly access the same fully connected layer for classification of multiple feature maps; therefore, after adding the SPP layer to the image classification model, fixed-dimensional feature representations are extracted for different blocks, and then classification and recognition operations can be performed through softmax (activation function layer), and finally the probabilities of different rock lithologies are output.

[0098] In some optional implementations, the above step S4022 includes:

[0099] Step b1: input the sample image corresponding to the candidate area into the convolution layer for convolution operation to generate ground feature.

[0100] Step b2: input the ground feature into the spatial pyramid pooling layer to perform pooling operations of different scales to obtain fixed-dimensional features.

[0101] Specifically, Figure 5 As shown in the figure, the SPP layer (pyramid pooling layer) is regarded as ROIPooling (region of interest pooling) and added between the convolution layer and the fully connected layer. The pyramid pooling layer divides the input images of different sizes into 4*4, 2*2 and 1*1 regions, and then collects different blocks with maximum pooling, which can map inputs of different sizes to a feature vector of fixed dimension.

[0102] Step b3: Input the fixed-dimensional features into the fully connected layer for feature concatenation.

[0103] Step b4, input the fixed-dimensional features after feature splicing into the activation function layer for classification and recognition to obtain the candidate region feature map.

[0104] Specifically, the candidate region feature map includes the candidate region and probability vectors of different rock lithologies in the candidate region.

[0105] Step S403, perform border regression detection on the candidate region feature map to obtain the identification and classification results of geological rock lithology. Figure 2 Step S203 of the illustrated embodiment will not be described in detail here.

[0106] The method for identifying and classifying geological rock lithology provided in this embodiment selects candidate areas based on sample images in a sample image data set using a region generation model, and then extracts features based on the candidate areas using an image classification model to obtain a feature map of the candidate area, effectively extracts probability vectors of each lithology category in the candidate area, realizes effective identification and classification of rock lithology, and improves the accuracy and effectiveness of lithology identification and classification.

[0107] In this embodiment, a method for identifying and classifying geological rock properties is provided, which can be used in the above-mentioned electronic device. Figure 6 FIG. 1 is a flow chart of a method for identifying and classifying geological rock properties according to an embodiment of the present invention. Figure 6 As shown, the process includes the following steps:

[0108] Step S601, obtain geological rock lithology images, perform image processing on the geological rock lithology images, and obtain a sample image data set. Figure 4 Step S401 of the illustrated embodiment will not be described in detail here.

[0109] Step S602: Based on the sample image data set, the image classification model and the region generation model are used to perform category discrimination to obtain a candidate region feature map. Figure 4 Step S402 of the illustrated embodiment will not be described in detail here.

[0110] Step S603, performing border regression detection on the candidate region feature map to obtain the identification and classification results of geological rock properties.

[0111] Specifically, the above step S603 includes:

[0112] Step S6031, determining a marker recognition area based on a sample image data set.

[0113] Step S6032, based on the marked recognition area and the candidate area in the candidate area feature map, a bounding box regression algorithm is used to determine the position offset of the candidate area.

[0114] Specifically, it is assumed that the target recognition box D is the target recognition box that correctly marks the location of the geological rock lithology, and the rectangular box R is the box corresponding to the candidate area in the candidate area feature map; even if the rock lithology in the rectangular box R is identified as TORL by the classifier, the TORL is not correctly detected in the sample image due to the inaccurate positioning of the rectangular box R. Therefore, the rectangular box R is fine-tuned to make the rectangular box R close to the rectangular box D corresponding to the marked recognition area;

[0115] Furthermore, it is assumed that the target recognition box D is a four-dimensional vector (x, y, w, h), where x represents the horizontal coordinate of the center point of the detection window, y represents the vertical coordinate of the center point of the detection window, w represents the width of the detection window, and h represents the height of the detection window; the candidate region R in the candidate region feature map = (R x , R y , R w , Rh), where R x Represents the horizontal coordinate of the center point of the candidate region, R y The vertical coordinate of the center point of the candidate region, R w represents the width of the candidate region, and Rh represents the height of the candidate region.

[0116] Furthermore, the candidate region R is mapped to obtain a regression window D′ that is closer to the true recognition window D, where the mapping relationship f can be expressed as:

[0117] f(R x , R y , R w , R h )=(D′ x , D′ y , D′ w , D′ h )≈(D x , D y , D w , D h ) (2)

[0118] Among them, D′ x represents the horizontal coordinate of the center point of the regression window, D′ y Indicates the vertical coordinate of the center point of the regression window, D w Represents the width of the regression window, D′ h Indicates the height of the regression window, D x Indicates the horizontal coordinate of the center point of the mark recognition area, D y Indicates the vertical coordinate of the center point of the mark recognition area, D w Indicates the width of the mark recognition area, D h Indicates the height of the marker recognition area.

[0119] Furthermore, the rectangular box R is first translated and then scaled using the mapping f to obtain the regression window D′. The specific formula is as follows:

[0120]

[0121] Among them, d x (R) represents the position offset of the horizontal coordinate of the center point of the candidate region, d y (R) represents the position offset of the vertical coordinate of the center point of the candidate region, d w (R) represents the position offset of the wide candidate region, and dh(R) represents the position offset of the high candidate region.

[0122] Furthermore, when the difference between the rectangular frame R and the target recognition frame D is small, the above formula (3) is a linear transformation, and the linear regression model can be used to fine-tune the rectangular frame R, where the translation amount (p x , p y ) and scale factor (p w , p h ) is calculated as follows:

[0123]

[0124] Among them, p x represents the horizontal coordinate translation, p y represents the vertical coordinate translation, p w represents the width scale factor, p h Represents the height scale factor.

[0125] Furthermore, after linear regression processing using the above formula (4), the rectangular box R further approaches the target recognition box D. The loss function can be constructed to adjust the rectangular box R. The specific steps are as follows: the feature vector obtained after the convolution operation is defined as Φ, and the target recognition box D that correctly detects the ground object (i.e., rock lithology) is input, that is, (p x , p v , p w , p h ), and get the output d x (R), d y (R), d w (R) and d h (R) Four transformations form the objective function based on the input data. The expression of the objective function is as follows:

[0126]

[0127] Among them, Φ(R) represents the feature vector in the candidate region feature vector map, represents the learning parameter, d(R) represents the predicted value, that is, (p x , py , p w , p h ), d * (R) represents the objective function, and * represents x, y, w, and h.

[0128] Furthermore, by calculating the loss function, (p x , p y , p w , p h ) has the smallest difference with the true value. The specific calculation formula of the loss function is as follows:

[0129]

[0130] Where N represents the number of eigenvectors, represents the true target area feature vector, and λ represents the error correction parameter.

[0131] Furthermore, in order to solve the minimum value of the position offset, the optimization objective function is determined based on the loss function. The specific calculation formula of the optimization objective function is as follows:

[0132]

[0133] Among them, w * Indicates the minimum value of the position offset.

[0134] Step S6033: adjust the position of the feature map of the candidate region based on the position offset of the candidate region.

[0135] Specifically, the gradient descent method or the least squares method is used to solve the optimization objective function to obtain the position offset of the candidate region, and then the position of the candidate region feature map is adjusted according to the position offset of the candidate region, so that the candidate region feature map can correctly locate the candidate region.

[0136] Step S6034, classify and count the candidate region feature maps after position adjustment to obtain the identification and classification results of geological rock properties.

[0137] Specifically, the probability vectors of different rock lithologies contained in the candidate area are obtained according to the feature map of the candidate area after position adjustment. The probability vectors of different rock lithologies are statistically analyzed to accurately determine whether the candidate area includes TORL, and then the number and position information of TORL in each sample image are obtained, that is, the identification and classification results of geological rock lithology.

[0138] The method for identifying and classifying geological rock lithology provided in the present embodiment determines a marked identification area based on a sample image data set, determines the position offset of the candidate area based on the marked identification area and the candidate area in the candidate area feature map by using a bounding box regression algorithm, and then adjusts the position of the candidate area feature map based on the position offset of the candidate area. Finally, the candidate area feature map after the position adjustment is classified and counted to obtain the identification and classification results of geological rock lithology. By adjusting the position of the candidate area feature map, the position of the candidate area feature map is made close to the marked identification area, thereby achieving accurate positioning of the candidate area in the candidate area feature map, and improving the accuracy of rock lithology identification in geological surveys.

[0139] The following is a specific example to illustrate the specific steps of a runoff forecasting method based on data statistical analysis.

[0140] Embodiment 1:

[0141] like Figure 7 As shown in Figure 1, the specific steps of the runoff forecasting method based on statistical analysis of data include:

[0142] First, a deep learning model is built based on Tensorflow (an open source machine learning framework developed by a Google team). The essence of deep learning is to build a multi-hidden layer neural network model and use massive training data for feature learning to improve the accuracy of training model recognition and prediction; Convolutional Neural Networks (CNN) is a multi-layer neural network structure with shared weights. CNN reduces the number of parameters in the neural network structure by sharing weights, reducing the data size and computational complexity; it solves the problems of too many parameters, high computational complexity, slow speed and overfitting in the image processing process of fully connected neural networks.

[0143] The training model obtained by improving the CNN model can reduce the number of training network parameters, reduce model complexity, reduce the training time of object recognition, and improve the efficiency of object recognition by sharing weights. Based on the improved CNN model, a deep learning model framework for TORL recognition is designed through the Tensorflow platform. Figure 8As shown in the figure, the model is divided into two parts: sample training and object recognition, which correspond to the construction graph and execution graph in the Tensorflow training and recognition process; the construction graph creates a session (node) and defines the operation method between nodes; the execution graph executes the corresponding operations between nodes and updates the variable values, and finally saves the final data; the node structure and node operation of the sample training and object recognition parts in the deep learning model are the same, and the initial assignment of weight variables in the network structure of sample training and object recognition is different; the initial weight of the neural network in the sample training part is a fixed value, and the weights of each layer of the convolutional neural network structure are adjusted through the back propagation algorithm, and then the cross entropy is used to reduce the error between the actual output value and the true value; the weight of the object recognition part uses the weights of each layer of the convolutional neural network structure obtained by sample training; the test sample image data of TORL is digitized into a multidimensional matrix, and then mapped through each layer of the network, and finally the object recognition result is obtained.

[0144] Secondly, the steps of designing the overall algorithm for TORL recognition based on CNN include: segmenting and cropping the image to obtain multiple geological category images, then adding labels to the objects to be identified in the geological category images, data preprocessing and image enhancement, and finally dividing the processed sample image data set into a training set and a test set; using the training set data to train the image classification model and the region generation model, and the test set is used to test the recognition effect of the model; in the testing stage, the target candidate area mentioned in each TORL image is extracted, and then the candidate area is input into the trained category discrimination network, and finally, the classification probability statistics and bounding box regression detection are combined to determine whether the candidate area includes TORL and determine the number and location information of TORL in each image.

[0145] During the training phase, spatial pyramid pooling is used to train the image classification network: a pyramid pooling layer is added between the convolution layer and the fully connected layer. The pyramid pooling layer divides the input images of different sizes into 4*4, 2*2 and 1*1 areas, and then collects the different blocks using maximum pooling; finally, the inputs of different sizes are mapped to a feature vector of a fixed scale; the convolution layer, pooling layer and activation layer do not require fixed-scale inputs. After the input image passes through these neural layers, feature maps of different scales are obtained, and feature maps of different scales cannot be connected to a fully connected layer for classification; after adding the pyramid pooling layer, a fixed-dimensional feature representation is extracted for each block to obtain multiple feature vectors of fixed scales, and multiple feature vectors of fixed scales can be connected to a fully connected layer for classification.

[0146] During the training phase, the selective search algorithm is used to build a region generation model. Related detection methods consume a lot of time to generate detection frames. Related detection methods, such as the adaptive enhancement algorithm in OpenCV, use sliding windows combined with image pyramids to generate detection frames. In deep learning, the EB method is used to generate detection frames. Using selective search SS instead of sliding windows and EB methods can increase the speed of generating detection frames.

[0147] In the recognition stage, bounding box regression and classification statistics are used to obtain the recognition accuracy of TORL. Based on the candidate region feature map, the fully connected layer and the activation function layer are used to calculate whether each feature vector in the candidate region feature map belongs to TORL, and then the probability vector of classification statistics is output. Finally, bounding box regression is used to obtain the position offset of each feature vector, and a more accurate target detection frame is determined based on the position offset.

[0148] In the above-mentioned embodiment 1, the complex information in the geological rock lithology image is parsed by a deep learning algorithm, and geological elements such as lithology, structure and mineral deposits are automatically identified, thereby improving the accuracy of lithology identification in geological surveys; the relevant rock lithology identification methods require a lot of manpower and time, while AI recognition and classification technology can automatically perform image analysis, greatly improving exploration efficiency; at the same time, AI recognition and classification technology makes decisions based on a large amount of data and algorithms, reducing the influence of human factors in the identification process and reducing human errors; in addition, in mine exploration and mining operations, AI technology can replace manpower to work in dangerous environments, greatly reducing personal risks and property losses, and improving the safety of operations.

[0149] In this embodiment, a device for identifying and classifying geological rock properties is also provided, which is used to implement the above-mentioned embodiments and preferred embodiments, and the descriptions that have been made will not be repeated. As used below, the term "module" can be a combination of software and / or hardware that implements a predetermined function. Although the device described in the following embodiments is preferably implemented in software, the implementation of hardware, or a combination of software and hardware, is also possible and conceivable.

[0150] This embodiment provides a device for identifying and classifying geological rock properties. Fig. 9 As shown, including:

[0151] The image processing module 901 is used to obtain geological rock lithology images, perform image processing on the geological rock lithology images, and obtain a sample image data set.

[0152] The discrimination module 902 is used to perform category discrimination based on the sample image data set using the image classification model and the region generation model to obtain a candidate region feature map.

[0153] The detection module 903 is used to perform border regression detection on the candidate region feature map to obtain the identification and classification results of geological rock properties.

[0154] In some optional implementations, the image processing module 901 includes:

[0155] The segmentation and cropping unit is used to segment and crop the geological rock lithology image to obtain multiple geological category images.

[0156] The preprocessing unit is used to add labels to multiple geological category images respectively, and perform data preprocessing on the multiple geological category images after adding labels to obtain a sample image data set.

[0157] In some optional implementations, the determination module 902 includes:

[0158] The generation unit is used to select a candidate region by using a region generation model based on a sample image in a sample image data set.

[0159] The feature extraction unit is used to extract features based on the candidate area using an image classification model to obtain a feature map of the candidate area; wherein the feature map of the candidate area includes a probability vector of each lithology category in the candidate area.

[0160] In some optional implementations, the generating unit includes:

[0161] The grouping subunit is used to group regions based on the similarity of sample image pixels in the sample image data set to obtain multiple target regions.

[0162] The calculation subunit is used to obtain the regional features of the target area and calculate the overall similarity of the target area based on the regional features.

[0163] A determination subunit is used to determine a candidate region based on the overall similarity of the target region.

[0164] In some optional implementations, the feature extraction unit includes:

[0165] The grouping subunit is used to input the sample images corresponding to the candidate area into the convolution layer for convolution operation to generate ground feature.

[0166] The pooling subunit is used to input the ground feature into the spatial pyramid pooling layer to perform pooling operations of different scales to obtain fixed-dimensional features.

[0167] The concatenation subunit is used to input fixed-dimensional features into the fully connected layer for feature concatenation.

[0168] The classification and recognition subunit is used to input the fixed-dimensional features after feature splicing into the activation function layer for classification and recognition to obtain the candidate region feature map.

[0169] In some optional implementations, the detection module 903 includes:

[0170] The marking unit is used to determine a marking recognition area based on a sample image data set.

[0171] The determination unit is used to determine the position offset of the candidate region by using a border regression algorithm based on the marked recognition region and the candidate region in the candidate region feature map.

[0172] The position adjustment unit is used to adjust the position of the candidate region feature map based on the position offset of the candidate region.

[0173] The classification and statistics unit is used to perform classification and statistics on the candidate area feature map after position adjustment to obtain the identification and classification results of geological rock lithology.

[0174] The further functional description of each of the above modules and units is the same as that of the above corresponding embodiments and will not be repeated here.

[0175] In this embodiment, a geological rock lithology identification and classification device is presented in the form of a functional unit, where the unit refers to an ASIC (Application Specific Integrated Circuit) circuit, a processor and memory that executes one or more software or fixed programs, and / or other devices that can provide the above functions.

[0176] The embodiment of the present invention also provides a computer device having the above Fig. 9 A device for identifying and classifying geological rock properties is shown.

[0177] See also Fig.10 , Fig.10 is a schematic diagram of the structure of a computer device provided by an optional embodiment of the present invention, such as Fig.10 As shown, the computer device includes: one or more processors 10, a memory 20, and interfaces for connecting various components, including high-speed interfaces and low-speed interfaces. Various components are connected to each other using different buses for communication, and can be installed on a common mainboard or installed in other ways as needed. The processor can process instructions executed in the computer device, including instructions stored in or on the memory to display the graphical information of the GUI on an external input / output device (such as a display device coupled to the interface). In some optional embodiments, if necessary, multiple processors and / or multiple buses can be used together with multiple memories. Similarly, multiple computer devices can be connected, and each device provides some necessary operations (for example, as a server array, a group of blade servers, or a multi-processor system). Fig.10A processor 10 is taken as an example.

[0178] The processor 10 may be a central processing unit, a network processor or a combination thereof. The processor 10 may further include a hardware chip. The hardware chip may be a dedicated integrated circuit, a programmable logic device or a combination thereof. The programmable logic device may be a complex programmable logic device, a field programmable gate array, a general purpose array logic or any combination thereof.

[0179] The memory 20 stores instructions executable by at least one processor 10, so that at least one processor 10 executes the method shown in the above embodiment.

[0180] The memory 20 may include a program storage area and a data storage area, wherein the program storage area may store an operating system, an application required for at least one function; the data storage area may store data created according to the use of the computer device, etc. In addition, the memory 20 may include a high-speed random access memory, and may also include a non-transient memory, such as at least one disk storage device, a flash memory device, or other non-transient solid-state storage device. In some optional embodiments, the memory 20 may optionally include a memory remotely arranged relative to the processor 10, and these remote memories may be connected to the computer device via a network. Examples of the above-mentioned network include, but are not limited to, the Internet, an intranet, a local area network, a mobile communication network, and combinations thereof.

[0181] The memory 20 may include a volatile memory, such as a random access memory; the memory may also include a non-volatile memory, such as a flash memory, a hard disk or a solid state drive; the memory 20 may also include a combination of the above types of memory.

[0182] The computer device also includes an input device 30 and an output device 40. The processor 10, the memory 20, the input device 30 and the output device 40 may be connected via a bus or other means. Fig.10 The example of connecting through bus is taken in the following.

[0183] The input device 30 can receive input digital or character information, and generate key signal input related to the user settings and function control of the computer device, such as a touch screen, a keypad, a mouse, a track pad, a touch pad, an indicator bar, one or more mouse buttons, a trackball, a joystick, etc. The output device 40 may include a display device, an auxiliary lighting device (e.g., an LED) and a tactile feedback device (e.g., a vibration motor), etc. The above-mentioned display device includes but is not limited to a liquid crystal display, a light emitting diode, a display and a plasma display. In some optional embodiments, the display device can be a touch screen.

[0184] The embodiment of the present invention also provides a computer-readable storage medium. The method according to the embodiment of the present invention can be implemented in hardware, firmware, or can be implemented as a computer code that can be recorded in a storage medium, or can be implemented as a computer code that is originally stored in a remote storage medium or a non-temporary machine-readable storage medium and will be stored in a local storage medium through a network download, so that the method described herein can be stored in such software processing on a storage medium using a general-purpose computer, a dedicated processor, or programmable or dedicated hardware. Among them, the storage medium can be a magnetic disk, an optical disk, a read-only storage memory, a random access memory, a flash memory, a hard disk or a solid-state hard disk, etc.; further, the storage medium can also include a combination of the above types of memories. It can be understood that a computer, a processor, a microprocessor controller, or programmable hardware includes a storage component that can store or receive software or computer code. When the software or computer code is accessed and executed by a computer, a processor, or hardware, the method shown in the above embodiment is implemented.

[0185] A part of the present invention may be applied as a computer program product, such as a computer program instruction, which, when executed by a computer, can call or provide the method and / or technical solution according to the present invention through the operation of the computer. Those skilled in the art should understand that the existence of the computer program instruction in a computer-readable medium includes, but is not limited to, a source file, an executable file, an installation package file, etc., and accordingly, the way in which the computer program instruction is executed by the computer includes, but is not limited to: the computer directly executes the instruction, or the computer compiles the instruction and then executes the corresponding compiled program, or the computer reads and executes the instruction, or the computer reads and installs the instruction and then executes the corresponding installed program. Here, the computer-readable medium may be any available computer-readable storage medium or communication medium accessible to the computer.

[0186] Although the embodiments of the present invention have been described in conjunction with the accompanying drawings, those skilled in the art may make various modifications and variations without departing from the spirit and scope of the present invention, and such modifications and variations are all within the scope defined by the appended claims.

Claims

1. A method for identifying and classifying geological rock properties, characterized in that: The method comprises: Acquire geological rock lithology images, perform image processing on the geological rock lithology images, and obtain a sample image data set; Based on the sample image data set, classify the images using an image classification model and a region generation model to obtain a candidate region feature map; The feature map of the candidate region is subjected to bounding box regression detection to obtain the identification and classification results of the geological rock properties.

2. The method according to claim 1, characterized in that The step of performing image processing on the geological rock lithology image to obtain a sample image data set includes: Segmenting and cropping the geological rock lithology image to obtain multiple geological category images; Labels are added to the multiple geological category images respectively, and data preprocessing is performed on the multiple geological category images after the labels are added to obtain the sample image data set.

3. The method according to claim 1, characterized in that The method of performing category discrimination based on the sample image data set using an image classification model and a region generation model to obtain a candidate region feature map includes: Based on the sample images in the sample image data set, selecting candidate regions using the region generation model; Based on the candidate area, the image classification model is used to perform feature extraction to obtain a feature map of the candidate area; wherein the feature map of the candidate area includes a probability vector of each lithology category in the candidate area.

4. The method according to claim 3, characterized in that The selecting a candidate region by using the region generation model based on the sample image in the sample image data set includes: Performing region grouping based on the similarity of sample image pixels in the sample image data set to obtain a plurality of target regions; Acquire regional features of the target area, and calculate the overall similarity of the target area based on the regional features; A candidate region is determined based on the overall similarity of the target region.

5. The method according to claim 3, characterized in that: The image classification model includes a convolution layer, a spatial pyramid pooling layer, a fully connected layer and an activation function layer; the feature extraction based on the candidate region using the image classification model to obtain the feature map of the candidate region includes: Inputting the sample image corresponding to the candidate area into the convolution layer for convolution operation to generate ground feature; Inputting the feature of the ground object into the spatial pyramid pooling layer to perform pooling operations of different scales to obtain fixed-dimensional features; Inputting the fixed dimension features into the fully connected layer for feature concatenation; The fixed-dimensional features after feature concatenation are input into the activation function layer for classification and recognition to obtain the candidate region feature map.

6. The method according to claim 1, characterized in that The step of performing border regression detection on the candidate region feature map to obtain the identification and classification results of geological rock properties includes: Determining a marker recognition area based on the sample image data set; Based on the marked recognition area and the candidate area in the candidate area feature map, a frame regression algorithm is used to determine the position offset of the candidate area; Adjusting the position of the feature map of the candidate region based on the position offset of the candidate region; The position-adjusted candidate region feature map is classified and counted to obtain the identification and classification results of the geological rock lithology.

7. A device for identifying and classifying geological rock properties, characterized in that: The device comprises: An image processing module is used to obtain geological rock lithology images, perform image processing on the geological rock lithology images, and obtain a sample image data set; A discrimination module, used to perform category discrimination based on the sample image data set using an image classification model and a region generation model to obtain a candidate region feature map; The detection module is used to perform border regression detection on the feature map of the candidate area to obtain the identification and classification results of the geological rock lithology.

8. A computer device, characterized in that: include: A memory and a processor, wherein the memory and the processor are communicatively connected to each other, the memory stores computer instructions, and the processor executes the method for identifying and classifying geological rock properties as described in any one of claims 1 to 6 by executing the computer instructions.

9. A computer-readable storage medium, characterized in that: The computer-readable storage medium stores computer instructions, and the computer instructions are used to enable a computer to execute the method for identifying and classifying geological rock properties according to any one of claims 1 to 6.

10. A computer program product, characterized in that It comprises computer instructions, and the computer instructions are used to make a computer execute the identification and classification method of geological rock lithology according to any one of claims 1 to 6.

Images