Geological core image RQD intelligent identification method and system based on electronic grid

Abstract

The application discloses a kind of geological core image RQD intelligent identification method and system based on electronic gridstone.Method includes: collecting core image, using electronic gridstone to automatically mark core box and core, establish geological core image dataset;Geological core image dataset is extracted by Inception-v3 pre-training model after feature, and core and core box are identified, and depth model training is obtained Inception-v3 core identification model;To be identified geological core image input model after training, obtain the pixel size and coordinate information of core and core box after identification, and calculate core actual size;According to the identification result, calculate rock quality index RQD and log core.The application uses the method of automatic marking using electronic gridstone and the model based on image recognition and deep learning, improve efficiency and accuracy, widen the scope of application.

Description

Technical Field

[0001] This invention belongs to the field of geological exploration technology, specifically relating to a method and system for RQD recognition of geological core images. Background Technology

[0002] Rock Quality Determination (RQD) is an important method for evaluating the quality grade of rock masses.

[0003] Traditional core identification and logging methods have the following problems:

[0004] First, the work efficiency is low. Because the length and fracture condition of the rock core need to be manually marked, this process is time-consuming, labor-intensive, and prone to errors.

[0005] Second, the accuracy is low. Existing core RQD identification methods all establish training sets for a certain type of core size, and these training sets are based on manually labeled training sets. The training sets themselves may have biases, resulting in low identification accuracy.

[0006] Third, it lacks universality. Due to the inconsistent size of geological drilling cores and core boxes, the constructed core RQD identification method lacks universality and can only be applied to cores of a certain size.

[0007] Therefore, how to improve the efficiency and accuracy of core RQD identification and broaden its application scope is a problem that urgently needs to be solved by those skilled in the art. Summary of the Invention

[0008] This invention proposes a method and system for intelligent RQD recognition of geological core images based on electronic grids, with the aim of improving the efficiency, accuracy and universality of RQD recognition of geological core images.

[0009] The technical solution of this invention is as follows:

[0010] A method for intelligent RQD recognition of geological core images based on electronic grids includes the following steps:

[0011] S1. Collect core images, automatically label core boxes and cores using electronic grids, and establish a geological core image dataset;

[0012] S2. After extracting features from the geological core image dataset using the Inception-v3 pre-trained model, the Inception-v3 core recognition model is obtained by combining the Softmax function with the core and core box recognition, and deep model training is performed.

[0013] S3. Input the geological core image to be identified into the trained Inception-v3 core recognition model. The output recognition results include the pixel size and coordinate information of the identified core and core box. Then calculate the actual size of the core.

[0014] S4. Calculate the rock quality index RQD based on the identification results of the Inception-v3 core identification model and catalog the core.

[0015] As a further improvement to the aforementioned intelligent identification method for geological core images based on electronic grids using RQD, step S1 specifically includes:

[0016] S11. Construct an electronic grid coordinate system based on the core box dimensions;

[0017] S12. Use the electronic grid coordinate system to obtain the coordinates and lengths of the core box and the core in the geological core image;

[0018] S13. Automatically label core boxes and cores in geological core images to establish a geological core image dataset.

[0019] As a further improvement to the electronic grid-based RQD intelligent recognition method for geological core images: the label objects of the geological core image dataset established in step S1 include core objects and core box objects. The core box objects are set as scales, and then the ratio of pixel size to actual size is calculated. The actual length and number of rows of the core box are stored as fixed values.

[0020] As a further improvement to the aforementioned intelligent RQD recognition method for geological core images based on electronic grids, step S4 specifically includes:

[0021] S41. Obtain the RQD threshold input by the user, and calculate the advance length based on the actual length of the stored core box and the number of rows;

[0022] S42. Calculate the rock quality index RQD for each run;

[0023] S43. Associate the core box objects and core objects identified in the core box images with their pixel size and coordinate information, the actual size of all cores calculated, and the rock quality index (RQD) for each run, and store them in the database.

[0024] As a further improvement to the aforementioned intelligent RQD recognition method for geological core images based on electronic grids, the specific calculation method for the rock quality index RQD is as follows:

[0025] For the rock quality index RQD of the nth cycle n The calculation formula is:

[0026]

[0027] Where l is the length of the core box, m n Let n be the number of rows in the nth core box. L is the sum of the core lengths exceeding the RQD threshold in the nth cycle. ni Let be the length of the i-th core in the n-th cycle, where n∈{1,2,…,N} and N is the total number of cycles, i∈{1,2,…,k}. n}, k n Let I be the total number of core samples in the nth cycle. n This refers to the set of core samples that reach the RQD threshold in the nth cycle.

[0028] As a further improvement to the aforementioned intelligent RQD recognition method for geological core images based on electronic grids, step S2 specifically includes:

[0029] S21. Use the Inception-v3 pre-trained model to extract high-dimensional features from core images;

[0030] S22. Freeze all fully connected layers in the Inception-v3 model, train the model using a geological core image dataset, and save the geological core image features extracted by the convolutional layers, pooling layers, and Inception layers in the model as vectors in a cache file for subsequent classification.

[0031] S23. Fuse the geological core image features in the cache file with the high-dimensional features extracted from the core image by the pre-trained model, input the Softmax classification function, and output the pixel length and coordinate information of the identified core and core box;

[0032] S24. Calculate the actual size of all identified cores based on the ratio of the core box pixel size to the actual size.

[0033] As a further improvement to the electronic grid-based RQD intelligent recognition method for geological core images, step S3 also includes processing and correcting incorrectly identified core box images, specifically as follows:

[0034] If an image with an incorrect recognition is detected in the recognition result queue, the recognition result corresponding to the incorrect image is cleared. Then, the redundant background in the image is cropped and added back to the recognition queue for recognition correction. If the reprocessed image is still incorrectly recognized, the image is considered to have been captured incorrectly and is placed in the deletion queue for further processing.

[0035] An image is considered to have an identification error if any of the following conditions are met: the image does not identify the core label or core box label; the length of the identified core box object is less than the length of any core object; the coordinate information of any identified core object is not within the range of the core box object; or the maximum width of all identified core objects is greater than twice the minimum width.

[0036] As a further improvement to the electronic grid-based RQD intelligent recognition method for geological core images, step S3 also includes: saving the core box images, core box row numbers, and pixel size and coordinate information of the core objects and core box objects in the recognition result queue into a new dataset, and periodically using the new dataset to further train the model.

[0037] This invention also proposes an intelligent RQD recognition system for geological core images based on electronic grids. The system is based on the above-mentioned intelligent RQD recognition method for geological core images and includes: a core image acquisition module, an electronic grid annotation module, a model training module, a core recognition module, and a data cataloging module.

[0038] The core image acquisition module is used to acquire core images and geological core images to be identified;

[0039] The electronic grid annotation module is used to automatically annotate core boxes and cores on the acquired core images using electronic grids, and to establish a geological core image dataset;

[0040] The model training module is used to extract features from the geological core image dataset using the Inception-v3 pre-trained model, then combine the Softmax function to identify the core and core box, and train a deep model to obtain the Inception-v3 core recognition model.

[0041] The core recognition module is used to input the geological core image to be recognized into the trained Inception-v3 core recognition model. The output recognition result includes the pixel size and coordinate information of the core and the core box, and calculates the actual size of the core.

[0042] The data cataloging module is used to calculate the rock quality index (RQD) based on the identification results and to catalog the rock core.

[0043] As a further improvement to the aforementioned RQD intelligent recognition system for geological core images based on electronic grids, it also includes a processing and correction module, which is used to detect images with recognition errors in the recognition result queue of the core recognition module, clear the recognition results corresponding to the images with recognition errors, and then crop the redundant background in the image and re-add it to the recognition queue for recognition correction. If the reprocessed image is still recognized incorrectly, it is considered that the image was captured incorrectly and enters the deletion queue for processing.

[0044] Compared with the prior art, the present invention has the following advantages:

[0045] 1. This invention obtains the coordinates and lengths of core boxes and cores in geological core images by constructing an electronic grid coordinate system. Then, it automatically labels the core boxes and cores and establishes a geological core image dataset. Compared with manual labeling methods, the automatic labeling method using electronic grids is time-saving, labor-saving, and has a more accurate labeling range. It can also easily process cores of different types and sizes, improving its universality.

[0046] 2. This invention constructs and uses a trained core identification model based on the Inception-v3 transfer learning neural network to identify the cores in the core box. It also calculates the rock quality index for each run based on the RQD threshold and the drilling length. The identification information and RQD results are stored and cataloged separately, which improves the efficiency of core identification and cataloging and reduces the identification cost.

[0047] 3. This invention can detect images in the recognition result queue, clear the recognition results corresponding to images with incorrect recognition, and reprocess and re-recognize the incorrect images, thereby reducing the occurrence of recognition errors and improving the reliability of core identification. Attached Figure Description

[0048] Figure 1 This is a schematic flowchart of the method of the present invention;

[0049] Figure 2 A schematic diagram of the training process of the Inception-v3 core identification model in this invention;

[0050] Figure 3 This is a schematic diagram of the core logging process of the present invention. Detailed Implementation

[0051] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0052] like Figure 1 A method for intelligent RQD recognition of geological core images based on electronic grids, comprising the following steps:

[0053] S1. Collect core images, use electronic grids to automatically label core boxes and cores, and establish a geological core image dataset.

[0054] Step S1 specifically includes:

[0055] S11. Construct an electronic grid coordinate system based on the core box dimensions.

[0056] S12. Use the electronic grid coordinate system to obtain the coordinates and lengths of the core box and the core in the geological core image.

[0057] S13. Automatically label core boxes and cores in geological core images to establish a geological core image dataset.

[0058] Furthermore, the pre-established geological core image dataset includes core objects and core box objects as labeled objects. The core box objects are set as rulers, and then the ratio of pixel size to actual size is calculated. The actual length and number of rows of the core box are stored as fixed values.

[0059] S2. For example Figure 2 After feature extraction using the Inception-v3 pre-trained model on the geological core image dataset, the Inception-v3 core recognition model was obtained by combining the Softmax function with the core and core box recognition, and deep model training was performed.

[0060] Step S2 specifically includes:

[0061] S21. Use the Inception-v3 pre-trained model to extract high-dimensional features from the core image. In this embodiment, to further improve the accuracy of core feature extraction, three convolutional layers are added consecutively before the fully connected layer of the original model to enhance the model's feature extraction capability.

[0062] S22. Freeze all fully connected layers in the Inception-v3 model, train the model using a geological core image dataset, and save the geological core image features extracted by the convolutional layers, pooling layers, and Inception layers in the model as vectors in a cache file for subsequent classification.

[0063] S23. Fuse the geological core image features in the cache file with the high-dimensional features extracted from the core image by the pre-trained model, input the Softmax classification function, and output the pixel length and coordinate information of the identified core and core box.

[0064] S24. Calculate the actual size of all identified cores based on the ratio of the core box pixel size to the actual size.

[0065] S3. Input the geological core image to be identified into the trained Inception-v3 core recognition model. The output recognition results include the pixel size and coordinate information of the identified core and core box. Then calculate the actual size of the core.

[0066] Furthermore, step S3 also includes processing and correcting the incorrectly identified core box images, specifically as follows:

[0067] If an image with an incorrect recognition is detected in the recognition result queue, the recognition result corresponding to the incorrect image is cleared. Then, the redundant background in the image is cropped and added back to the recognition queue for recognition correction. If the reprocessed image is still incorrectly recognized, the image is considered to have been captured incorrectly and is placed in the deletion queue for further processing.

[0068] An image is considered to have an identification error if any of the following conditions are met: the image does not identify the core label or core box label; the length of the identified core box object is less than the length of any core object; the coordinate information of any identified core object is not within the range of the core box object; or the maximum width of all identified core objects is greater than twice the minimum width.

[0069] Furthermore, step S3 also includes: saving the core box images, core box row counts, and pixel size and coordinate information of the core objects and core box objects in the recognition result queue into a new dataset, and periodically using the new dataset to further train the model.

[0070] S4. For example Figure 3 Based on the identification results of the Inception-v3 core identification model, the rock quality index RQD was calculated and the core was cataloged.

[0071] Step S4 specifically includes:

[0072] S41. Obtain the RQD threshold input by the user, and calculate the advance length based on the actual length of the stored core box and the number of rows;

[0073] S42. Calculate the rock quality index RQD for each run;

[0074] S43. Associate the core box objects and core objects identified in the core box images with their pixel size and coordinate information, the actual size of all cores calculated, and the rock quality index (RQD) for each run, and store them in the database.

[0075] The specific calculation method for the rock quality index RQD is as follows:

[0076] For the rock quality index RQD of the nth cycle n The calculation formula is:

[0077]

[0078] Where l is the length of the core box, m n Let n be the number of rows in the nth core box. L is the sum of the core lengths exceeding the RQD threshold in the nth cycle. ni Let be the length of the i-th core in the n-th cycle, where n∈{1,2,…,N} and N is the total number of cycles, i∈{1,2,…,k}. n}, k n Let I be the total number of core samples in the nth cycle. n This refers to the set of core samples that reach the RQD threshold in the nth cycle.

[0079] This embodiment also includes a geological core image RQD intelligent recognition system based on an electronic grid, which is based on the above-mentioned geological core image RQD intelligent recognition method.

[0080] The system includes: a core image acquisition module, an electronic grid annotation module, a model training module, a core recognition module, and a data logging module. Among them:

[0081] The core image acquisition module is used to acquire core images and geological core images to be identified.

[0082] The electronic grid annotation module is used to automatically annotate core boxes and cores on acquired core images using electronic grids, thereby establishing a geological core image dataset.

[0083] The model training module is used to extract features from the geological core image dataset using the Inception-v3 pre-trained model, then use the Softmax function to identify the core and core box, and finally train a deep model to obtain the Inception-v3 core recognition model.

[0084] The core recognition module is used to input the geological core image to be recognized into the trained Inception-v3 core recognition model. The output recognition results include the pixel size and coordinate information of the core and core box, and calculate the actual size of the core.

[0085] The data cataloging module is used to calculate the rock quality index (RQD) based on the identification results and to catalog the rock core.

[0086] Furthermore, the system also includes a processing and correction module, which is used to detect images with recognition errors in the recognition result queue of the core identification module, clear the recognition results corresponding to the images with recognition errors, and then crop the redundant background in the image and re-add it to the recognition queue for recognition correction. If the reprocessed image is still recognized incorrectly, it is considered that the image was taken incorrectly and enters the deletion queue for processing.

[0087] The various embodiments in this specification are described in a progressive manner, with each embodiment focusing on its differences from other embodiments. Similar or identical parts between embodiments can be referred to interchangeably. For the apparatus disclosed in the embodiments, since they correspond to the methods disclosed in the embodiments, the description is relatively simple; relevant parts can be referred to the method section.

[0088] The above description of the disclosed embodiments enables those skilled in the art to make or use the invention. Various modifications to these embodiments will be readily apparent to those skilled in the art, and the general principles defined herein may be implemented in other embodiments without departing from the spirit or scope of the invention. Therefore, the invention is not to be limited to the embodiments shown herein, but is to be accorded the widest scope consistent with the principles and novel features disclosed herein.

Claims

1. A method for intelligent RQD recognition of geological core images based on electronic grids, characterized in that... Includes the following steps: S1. Collect core images, automatically label core boxes and cores using electronic grids, and establish a geological core image dataset; Step S1 specifically includes: S11. Construct an electronic grid coordinate system based on the core box dimensions; S12. Use the electronic grid coordinate system to obtain the coordinates and lengths of the core box and the core in the geological core image; S13. Automatically label core boxes and cores in geological core images and establish a geological core image dataset; S2. After extracting features from the geological core image dataset using the Inception-v3 pre-trained model, the Inception-v3 core recognition model is obtained by combining the Softmax function with the core and core box recognition, and deep model training is performed. Step S2 specifically includes: S21. Use the Inception-v3 pre-trained model to extract high-dimensional features from core images; S22. Freeze all fully connected layers in the Inception-v3 model, train the model using a geological core image dataset, and save the geological core image features extracted by the convolutional layers, pooling layers, and Inception layers in the model as vectors in a cache file for subsequent classification. S23. Fuse the geological core image features in the cache file with the high-dimensional features extracted from the core image by the pre-trained model, input the Softmax classification function, and output the pixel length and coordinate information of the identified core and core box; S24. Calculate the actual size of all identified cores based on the ratio of the core box pixel size to the actual size; S3. Input the geological core image to be identified into the trained Inception-v3 core recognition model. The output recognition results include the pixel size and coordinate information of the identified core and core box. Then calculate the actual size of the core. S4. Calculate the rock quality index RQD based on the identification results of the Inception-v3 core identification model and catalog the core.

2. The intelligent RQD recognition method for geological core images based on electronic grids as described in claim 1, characterized in that: The label objects of the geological core image dataset established in step S1 include core objects and core box objects. The core box objects are set as rulers, and then the ratio of pixel size to actual size is calculated. The actual length and number of rows of the core box are stored as fixed values.

3. The intelligent RQD recognition method for geological core images based on electronic grids as described in claim 2, characterized in that, Step S4 specifically includes: S41. Obtain the RQD threshold input by the user, and calculate the advance length based on the actual length of the stored core box and the number of rows; S42. Calculate the rock quality index RQD for each run; S43. Associate the core box objects and core objects identified in the core box images with their pixel size and coordinate information, the actual size of all cores calculated, and the rock quality index (RQD) for each run, and store them in the database.

4. The intelligent RQD recognition method for geological core images based on electronic grids as described in claim 3, characterized in that, The specific calculation method for the rock quality index RQD is as follows: For the Rock quality indicators for each cycle The calculation formula is: ； in, The length of the core box, For the first The number of rows in a core box per cycle For the first The sum of core lengths exceeding the RQD threshold in each round. For the first The length of the i-th core in each iteration , The total number of rounds, , For the first The total number of core samples in each run. For the first Core collections that reach the RQD threshold in each round .

5. The intelligent RQD recognition method for geological core images based on electronic grids as described in claim 1, characterized in that, Step S3 also includes processing and correcting the incorrectly identified core box images, specifically as follows: If an image with an incorrect recognition is detected in the recognition result queue, the recognition result corresponding to the incorrect image is cleared. Then, the redundant background in the image is cropped and added back to the recognition queue for recognition correction. If the reprocessed image is still incorrectly recognized, the image is considered to have been captured incorrectly and is placed in the deletion queue for further processing. An image is considered to have an identification error if any of the following conditions are met: the image does not identify the core label or core box label; the length of the identified core box object is less than the length of any core object; the coordinate information of any identified core object is not within the range of the core box object; or the maximum width of all identified core objects is greater than twice the minimum width.

6. The intelligent RQD recognition method for geological core images based on electronic grids as described in claim 1, characterized in that, Step S3 also includes: saving the core box images, core box row counts, and pixel size and coordinate information of the core objects and core box objects in the recognition result queue to a new dataset, and periodically using the new dataset to further train the model.

7. A geological core image RQD intelligent recognition system based on electronic grid, characterized in that: The system is based on the RQD intelligent recognition method for geological core images based on electronic grids as described in any one of claims 1 to 6. The system includes: a core image acquisition module, an electronic grid annotation module, a model training module, a core recognition module, and a data cataloging module. The core image acquisition module is used to acquire core images and geological core images to be identified; The electronic grid annotation module is used to automatically annotate core boxes and cores on the acquired core images using electronic grids, and to establish a geological core image dataset; The model training module is used to extract features from the geological core image dataset using the Inception-v3 pre-trained model, then combine the Softmax function to identify the core and core box, and train a deep model to obtain the Inception-v3 core recognition model. The core recognition module is used to input the geological core image to be recognized into the trained Inception-v3 core recognition model. The output recognition result includes the pixel size and coordinate information of the core and the core box, and calculates the actual size of the core. The data cataloging module is used to calculate the rock quality index (RQD) based on the identification results and to catalog the rock core.

8. The RQD intelligent recognition system for geological core images based on electronic grids as described in claim 7, characterized in that: It also includes a processing and correction module, which is used to detect images with recognition errors in the recognition result queue of the core recognition module, clear the recognition results corresponding to the images with recognition errors, and then crop the redundant background in the image and add it back to the recognition queue for recognition correction. If the reprocessed image is still recognized incorrectly, it is considered that the image was taken incorrectly and enters the deletion queue for processing.

Images