Intelligent rock image lithology identification method and system based on Inception-v3 transfer learning model

Abstract

The invention discloses an intelligent rock image lithology identification method and system based on an Inception-v3 transfer learning model, and the method comprises the steps: collecting a rock image, carrying out the lithology marking of the rock image, and building a rock image data set; an Inception-v3 lithology identification model is constructed, the Inception-v3 lithology identification model comprises an Inception-v3 transfer learning model and a classifier, and the output of the Inception-v3 transfer learning model is used as the input of the classifier; the rock image data set is utilized to carry out pre-training on the Inception-v3 lithology identification model, and a trained Inception-v3 lithology identification model is obtained; the Inception-v3 transfer learning model is used for carrying out rock feature extraction, and the classifier is used for carrying out lithology identification according to rock features; and inputting a rock image to be identified into the trained Inception-v3 lithology identification model, and outputting a lithology identification result. According to the method, the modeling speed is greatly improved on the whole, the requirement for the data size of the rock image is low, and the modeling difficulty is reduced.

Description

Technical Field

[0001] The present invention relates to the technical field of engineering geological exploration, and in particular to an intelligent rock image lithology recognition method and system based on an Inception-v3 migration learning model. Background Art

[0002] Rocks are the working objects of geology, resource exploration, tunnel and underground engineering, geological disaster identification and prevention, etc., and are composed of minerals, structural structures, etc.

[0003] Manual identification and physical testing are often used to identify the mineral composition, structure, particle size, color and other lithology of rocks. However, manual identification requires the identification personnel to be highly professional and have rich practical experience, and the identification results are greatly affected by subjective factors. The physical testing method has high requirements on the accuracy of the detection instrument and is limited by the accuracy of the experimental equipment and the detection environment.

[0004] Deep learning mainly refers to an algorithm that processes information through neural networks to achieve feature extraction. In the field of lithology identification, traditional machine learning methods require manual extraction of rock features such as color, shape, and texture. Traditional machine learning methods not only require a lot of repetitive work and slow modeling, but also require a large amount of rock image data, making modeling more difficult.

[0005] Therefore, how to provide a method and system for intelligently identifying the lithology of rock images that improves the efficiency and reliability of lithology identification while reducing the cost of lithology identification is a technical problem that needs to be urgently solved by those skilled in the art. Summary of the invention

[0006] In view of the above research status and existing problems, the present invention provides an intelligent rock image lithology recognition method and system based on the Inception-v3 transfer learning model.

[0007] The present invention first provides an intelligent rock image lithology recognition method based on the Inception-v3 transfer learning model, comprising the following steps:

[0008] S1: Collect rock images, annotate the rock properties of the rock images, and establish a rock image dataset;

[0009] S2: constructing an Inception-v3 lithology recognition model, including an Inception-v3 transfer learning model and a classifier, wherein the output of the Inception-v3 transfer learning model is used as the input of the classifier; pre-training the Inception-v3 lithology recognition model using the rock image dataset to obtain a trained Inception-v3 lithology recognition model; the Inception-v3 transfer learning model is used to extract rock features, and the classifier is used to identify lithology according to the rock features;

[0010] S3: The rock image to be identified is input into the trained Inception-v3 lithology identification model, and the lithology identification result is output.

[0011] Preferably, the S1 further comprises a step of preprocessing the collected rock images, including:

[0012] The rock image data is subjected to data elimination, data cleaning and data enhancement processing to obtain standardized rock image data, and the standardized rock image data is used to establish a rock image data set.

[0013] Preferably, the data removal process of the rock image in S1 is to remove damaged rock images and blurred images:

[0014] The data cleaning process is to proportionally intercept rock images with long imaging distance, small target rocks or off-center rock images, so that the characteristics of the target rock are sufficiently obvious. For rock images with cluttered backgrounds or multiple rock samples, background removal is performed so that one image contains only one rock sample.

[0015] The data enhancement process is to randomly select half of the rock images and perform random flipping, mirror reversal, random translation, color change, brightness change and contrast change.

[0016] Preferably, the rock image in S1 includes one or more of the following rock image combinations:

[0017] Images of gold ore, copper ore, iron ore, lead and zinc ore, granite, gneiss, marble and shale.

[0018] Preferably, S2 comprises the following steps:

[0019] S21. Extract high-dimensional features of rock images;

[0020] S32. Freeze the convolution layer, pooling layer and Inception layer in the Inception-v3 transfer learning model, input the rock image data set into the Inception-v3 transfer learning model, and save the rock image features extracted by the convolution layer, pooling layer, Inception layer and fully connected layer in sequence;

[0021] S33. After fusing the rock image features and the high-dimensional features, the features are input into the classifier. The output result of the classifier is combined with the annotation to train the Inception-v3 lithology recognition model to obtain the trained Inception-v3 lithology recognition model.

[0022] Preferably, in S21, the high-dimensional features of the rock image include color features, and the extraction steps are as follows:

[0023] Divide the rock image into discrete intervals in color space;

[0024] Count the number of pixels in each discrete interval to obtain the color characteristics of the image.

[0025] Preferably, the high-dimensional features of the rock image include texture features, and the extraction steps are as follows:

[0026] Divide the rock image into several image sub-blocks and calculate the LBP value of each pixel in each image sub-block;

[0027] Perform histogram statistics on each image sub-block to obtain the histogram of the image sub-block;

[0028] After normalizing the histograms of all image sub-blocks, the normalized histograms of all image sub-blocks are connected to obtain the texture features of the image.

[0029] Preferably, S2 includes a step of processing the rock images with identification errors, including:

[0030] According to the category label, detect the images with recognition errors in the lithology recognition result queue, clear the lithology recognition results corresponding to the images with recognition errors, crop the background of the images and re-add them to the current rock image recognition queue for recognition correction. If the images after background cropping are still recognized incorrectly, they enter the queue to be deleted and wait for processing;

[0031] Misidentified images include: images in which the rock lithology category is not identified, images in which multiple different rock lithology categories are identified, images in which the wrong rock lithology category is identified, or images in which the rock lithology category is identified with a confidence level lower than a threshold.

[0032] Preferably, the rock images in the recognition result queue and the corresponding recognition and classification results are saved in a new dataset for regular addition to the rock image dataset to participate in the subsequent training of the Inception-v3 lithology recognition model.

[0033] The present invention further provides an intelligent rock image lithology recognition system based on the Inception-v3 transfer learning model, comprising: a rock image acquisition module, a rock image annotation module, a model training module and a lithology recognition module; wherein,

[0034] A rock image acquisition module, used for acquiring rock images and rock images to be identified;

[0035] A rock image annotation module, used to annotate the rock image with lithology and establish a rock image data set;

[0036] A model training module is used to construct an Inception-v3 lithology recognition model, including an Inception-v3 transfer learning model and a classifier, wherein the output of the Inception-v3 transfer learning model is used as the input of the classifier; the Inception-v3 lithology recognition model is pre-trained using the rock image data set to obtain a trained Inception-v3 lithology recognition model; the Inception-v3 transfer learning model is used to extract rock features, and the classifier is used to identify lithology according to the rock features;

[0037] The lithology recognition module is used to input the rock image to be identified into the trained Inception-v3 lithology recognition model and output the lithology recognition result.

[0038] Preferably, it also includes a processing and correction module, which is used to detect images with recognition errors in the lithology recognition result queue according to the category label, clear the lithology recognition results corresponding to the images with recognition errors, and add the images to the current rock image recognition queue for recognition correction after background cropping. If the images with background cropping are still recognized incorrectly, they enter the queue to be deleted and wait for processing;

[0039] Misidentified images include: images in which the rock lithology category is not identified, images in which multiple different rock lithology categories are identified, images in which the wrong rock lithology category is identified, or images in which the rock lithology category is identified with a confidence level lower than a threshold.

[0040] Compared with the prior art, it has the following beneficial effects:

[0041] The present invention adopts a transfer learning model to learn and extract rock features, adjusts the obtained model structure, parameter weights and other knowledge and transfers them to the task of lithology identification, which can reduce the amount of training parameters, greatly improve the model training speed, and has a low requirement for the amount of rock image data, thus reducing the difficulty of model training.

[0042] The present invention uses the Inception-v3 model, the color extraction layer and the texture extraction layer to extract information of different scales of the target image, and finally fuses them to obtain the fusion features of the image. The model extracts the features of the rock image through automatic search without manual extraction.

[0043] The present invention detects the pictures in the recognition result queue, clears the recognition results corresponding to the pictures with recognition errors, crops the redundant background in the pictures and re-adds them to the recognition queue for recognition correction; if the re-processed pictures are still recognized incorrectly, it is considered that the pictures were taken incorrectly and enter the queue to be deleted to wait for processing, which can reduce the occurrence of recognition errors and improve the reliability of lithology recognition. BRIEF DESCRIPTION OF THE DRAWINGS

[0044] In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the drawings required for use in the embodiments or the prior art descriptions are briefly introduced below. Obviously, the drawings described below are only embodiments of the present invention, and for ordinary technicians in this field, other drawings can be obtained based on the provided drawings without creative work.

[0045] Figure 1 A flow chart of an intelligent rock image lithology recognition method based on the Inception-v3 transfer learning model provided in an embodiment of the present invention;

[0046] Figure 2 Schematic diagram of the intelligent rock image lithology recognition method based on the Inception-v3 transfer learning model provided in an embodiment of the present invention;

[0047] Figure 3 A schematic diagram of the Inception-v3 lithology recognition model training provided in an embodiment of the present invention. DETAILED DESCRIPTION

[0048] The following will be combined with the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0049] The first aspect of the present invention discloses an intelligent rock image lithology recognition method based on the Inception-v3 transfer learning model. Using transfer learning, after considering the specific association between tasks, the model structure, parameter weights and other knowledge obtained in the previous task are adjusted and transferred to the lithology recognition task. It includes the following steps:

[0050] S1: Collect rock images, annotate the rock properties of the rock images, and establish a rock image dataset;

[0051] S2: Construct an Inception-v3 lithology recognition model, including an Inception-v3 transfer learning model and a classifier. The output of the Inception-v3 transfer learning model is used as the input of the classifier. The Inception-v3 lithology recognition model is pre-trained using a rock image dataset to obtain a trained Inception-v3 lithology recognition model. The Inception-v3 transfer learning model is used to extract rock features, and the classifier is used to identify lithology based on rock features.

[0052] S3: The rock image to be identified is input into the trained Inception-v3 lithology recognition model, and the lithology recognition result is output.

[0053] In one embodiment, S1 further includes a step of preprocessing the collected rock images, including:

[0054] The rock image data are processed by data elimination, data cleaning and data enhancement to obtain standardized rock image data, and the standardized rock image data are used to establish a rock image data set.

[0055] In one embodiment, the data removal process of the rock image in S1 is to remove damaged rock images and blurred images, and at least includes the following execution methods:

[0056] The data cleaning process is to proportionally intercept rock images with long imaging distances, small target rocks or rocks that are off-center, so that the characteristics of the target rocks are sufficiently obvious.

[0057] Background removal is performed on rock images with cluttered backgrounds or multiple rock samples so that one image contains only one rock sample.

[0058] The data enhancement process is to randomly select half of the rock images and perform random flipping, mirror reversal, random translation, color change, brightness change and contrast change.

[0059] In one embodiment, the rock image in S1 includes one or more of the following rock image combinations:

[0060] Images of gold ore, copper ore, iron ore, lead and zinc ore, granite, gneiss, marble and shale.

[0061] In one embodiment, the rock image data set is divided into a training set and a test set for training and testing of the Inception-v3 lithology recognition model, wherein the total amount of data of the training set and the test set is divided according to a preset ratio of 8:2.

[0062] In one embodiment, S2 includes the following steps:

[0063] S21. Extract high-dimensional features of rock images;

[0064] S32. Freeze the convolutional layer, pooling layer, and Inception layer in the Inception-v3 transfer learning model, input the rock image dataset into the Inception-v3 transfer learning model, and save the rock image features extracted by the convolutional layer, pooling layer, Inception layer, and fully connected layer in sequence;

[0065] S33. After the rock image features and high-dimensional features are fused, they are input into the classifier. The output results of the classifier are combined with the annotations to train the Inception-v3 lithology recognition model to obtain the trained Inception-v3 lithology recognition model.

[0066] In this embodiment, the rock image features extracted in S32 are saved in a cache file in the form of vectors for subsequent classification.

[0067] In one embodiment, the classifier adopts a Softmax classification function.

[0068] In one embodiment, in S21, the high-dimensional features of the rock image include color features, and the extraction steps are as follows:

[0069] Divide the rock image into discrete intervals in color space;

[0070] Count the number of pixels in each discrete interval to obtain the color characteristics of the image.

[0071] In one embodiment, the high-dimensional features of the rock image include texture features, and the extraction steps are as follows:

[0072] Divide the rock image into several image sub-blocks and calculate the LBP value of each pixel in each image sub-block;

[0073] Perform histogram statistics on each image sub-block to obtain the histogram of the image sub-block;

[0074] After normalizing the histograms of all image sub-blocks, the normalized histograms of all image sub-blocks are connected to obtain the texture features of the image.

[0075] In one embodiment, S2 includes steps for processing the rock images with identification errors, including:

[0076] According to the category labeling, the images with recognition errors are detected in the lithology recognition result queue, the lithology recognition results corresponding to the images with recognition errors are cleared, the images are background-cropped and then re-added to the current rock image recognition queue for recognition correction. If the re-processed images are still recognized incorrectly, it is considered that the images were taken incorrectly and are placed in the queue to be deleted for processing;

[0077] Misidentified images include: images in which the rock lithology category is not identified, images in which multiple different rock lithology categories are identified, images in which the wrong rock lithology category is identified, or images in which the rock lithology category is identified with a confidence level lower than a threshold.

[0078] In one embodiment, the rock lithology information in the recognition result queue is saved in a new dataset, including rock images and corresponding recognition and classification results, for regular addition to the rock image dataset to participate in the subsequent training of the Inception-v3 lithology recognition model.

[0079] A second aspect of an embodiment of the present invention provides an intelligent rock image lithology recognition system according to the first aspect of the embodiment, including: a rock image acquisition module, a rock image annotation module, a model training module and a lithology recognition module; the rock image acquisition module is used to acquire rock images and rock images to be identified; the rock image annotation module is used to perform lithology annotation on rock images and establish a rock image data set; the model training module is used to construct an Inception-v3 lithology recognition model, including an Inception-v3 transfer learning model and a classifier, and the output of the Inception-v3 transfer learning model is used as the input of the classifier; the Inception-v3 lithology recognition model is pre-trained using the rock image data set to obtain a trained Inception-v3 lithology recognition model; the Inception-v3 transfer learning model is used to extract rock features, and the classifier is used to identify lithology according to rock features; the lithology recognition module is used to input the rock image to be identified into the trained Inception-v3 lithology recognition model, and output the lithology recognition result.

[0080] In one embodiment, a preprocessing module is also included for preprocessing the collected rock images, including: performing data elimination, data cleaning and data enhancement processing on the rock image data to obtain standardized rock image data, and using the standardized rock image data to establish a rock image data set.

[0081] In this embodiment, the execution steps of the preprocessing module include: the data cleaning process is to proportionally intercept rock images with a long imaging distance, a small target rock or an off-center rock, so that the characteristics of the target rock are sufficiently obvious. The background of the rock image with a cluttered background or multiple rock samples is removed so that one image contains only one rock sample.

[0082] The data enhancement process is to randomly select half of the rock images and perform random flipping, mirror reversal, random translation, color change, brightness change and contrast change.

[0083] In one embodiment, a processing and correction module is further included, which is used to detect images with recognition errors in the lithology recognition result queue according to the category label, clear the lithology recognition results corresponding to the images with recognition errors, crop the background of the images and re-add them to the current rock image recognition queue for recognition correction; if the images after the background cropping are still recognized incorrectly, they are put into the queue to be deleted and wait for processing;

[0084] Misidentified images include: images in which the rock lithology category is not identified, images in which multiple different rock lithology categories are identified, images in which the wrong rock lithology category is identified, or images in which the rock lithology category is identified with a confidence level lower than a threshold.

[0085] The second aspect of the embodiment of the present invention is applicable to executing all the steps in the first aspect of the embodiment.

[0086] The above is a detailed introduction to an intelligent rock image lithology recognition method and system based on the Inception-v3 transfer learning model provided by the present invention. Specific examples are used in this article to illustrate the principles and implementation methods of the present invention. The description of the above embodiments is only used to help understand the method of the present invention and its core idea; at the same time, for those skilled in the art, according to the idea of ​​the present invention, there will be changes in the specific implementation method and application scope. In summary, the content of this specification should not be understood as a limitation on the present invention.

[0087] In this article, relational terms such as first and second, etc. are used only to distinguish one entity or operation from another entity or operation, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Moreover, the terms "comprise", "include" or any other variants thereof are intended to cover non-exclusive inclusion, so that a process, method, article or device including a series of elements includes not only those elements, but also other elements not explicitly listed, or also includes elements inherent to such process, method, article or device. In the absence of further restrictions, the elements defined by the statement "comprise a ..." do not exclude the presence of other identical elements in the process, method, article or device including the elements.

Claims

1. An intelligent rock image lithology recognition method based on the Inception-v3 transfer learning model, characterized in that: The steps include: S1: Collect rock images, annotate the rock properties of the rock images, and establish a rock image dataset; S2: constructing an Inception-v3 lithology recognition model, including an Inception-v3 transfer learning model and a classifier, wherein the output of the Inception-v3 transfer learning model is used as the input of the classifier; pre-training the Inception-v3 lithology recognition model using the rock image dataset to obtain a trained Inception-v3 lithology recognition model; the Inception-v3 transfer learning model is used to extract rock features, and the classifier is used to identify lithology according to the rock features; S3: The rock image to be identified is input into the trained Inception-v3 lithology identification model, and the lithology identification result is output.

2. The intelligent rock image lithology recognition method based on the Inception-v3 transfer learning model according to claim 1 is characterized in that: The S1 also includes a step of preprocessing the collected rock images, including: The rock image data is subjected to data elimination, data cleaning and data enhancement processing to obtain standardized rock image data, and the standardized rock image data is used to establish a rock image data set.

3. The intelligent rock image lithology recognition method based on the Inception-v3 transfer learning model according to claim 1 is characterized in that: The rock image in S1 includes one or more of the following rock image combinations: Images of gold ore, copper ore, iron ore, lead and zinc ore, granite, gneiss, marble and shale.

4. The intelligent rock image lithology recognition method based on the Inception-v3 transfer learning model according to claim 1 is characterized in that: The S2 comprises the following steps: S21. Extract high-dimensional features of rock images; S32. Freeze the convolution layer, pooling layer and Inception layer in the Inception-v3 transfer learning model, input the rock image data set into the Inception-v3 transfer learning model, and save the rock image features extracted by the convolution layer, pooling layer, Inception layer and fully connected layer in sequence; S33. After fusing the rock image features and the high-dimensional features, the features are input into the classifier. The output result of the classifier is combined with the annotation to train the Inception-v3 lithology recognition model to obtain the trained Inception-v3 lithology recognition model.

5. The intelligent rock image lithology recognition method based on the Inception-v3 transfer learning model according to claim 4 is characterized in that: In S21, the high-dimensional features of the rock image include color features, and the extraction steps are as follows: Divide the rock image into discrete intervals in color space; Count the number of pixels in each discrete interval to obtain the color characteristics of the image.

6. The intelligent rock image lithology recognition method based on the Inception-v3 transfer learning model according to claim 1 is characterized in that: The high-dimensional features of the rock image include texture features, and the extraction steps are as follows: Divide the rock image into several image sub-blocks and calculate the LBP value of each pixel in each image sub-block; Perform histogram statistics on each image sub-block to obtain the histogram of the image sub-block; After normalizing the histograms of all image sub-blocks, the normalized histograms of all image sub-blocks are connected to obtain the texture features of the image.

7. The intelligent rock image lithology recognition method based on the Inception-v3 transfer learning model according to claim 1 is characterized in that: S2 includes the steps of processing the wrongly identified rock images, including: According to the category label, detect the images with recognition errors in the lithology recognition result queue, clear the lithology recognition results corresponding to the images with recognition errors, crop the background of the images and re-add them to the current rock image recognition queue for recognition correction. If the images after background cropping are still recognized incorrectly, they enter the queue to be deleted and wait for processing; Misidentified images include: images in which the rock lithology category is not identified, images in which multiple different rock lithology categories are identified, images in which the wrong rock lithology category is identified, or images in which the rock lithology category is identified with a confidence level lower than a threshold.

8. The intelligent rock image lithology recognition method based on the Inception-v3 transfer learning model according to claim 1 is characterized in that: The rock images and corresponding recognition and classification results in the recognition result queue are saved in a new dataset for regular addition to the rock image dataset to participate in the subsequent training of the Inception-v3 lithology recognition model.

9. An intelligent rock image lithology recognition system according to any one of claims 1 to 8, characterized in that: include: Rock image acquisition module, rock image annotation module, model training module and lithology identification module; among them, A rock image acquisition module, used for acquiring rock images and rock images to be identified; A rock image annotation module, used to annotate the rock image with lithology and establish a rock image data set; A model training module is used to construct an Inception-v3 lithology recognition model, including an Inception-v3 transfer learning model and a classifier, wherein the output of the Inception-v3 transfer learning model is used as the input of the classifier; the Inception-v3 lithology recognition model is pre-trained using the rock image data set to obtain a trained Inception-v3 lithology recognition model; the Inception-v3 transfer learning model is used to extract rock features, and the classifier is used to identify lithology according to the rock features; The lithology recognition module is used to input the rock image to be identified into the trained Inception-v3 lithology recognition model and output the lithology recognition result.

10. The intelligent rock image lithology identification system according to claim 9, characterized in that: It also includes a processing and correction module, which is used to detect images with recognition errors in the lithology recognition result queue according to the category label, clear the lithology recognition results corresponding to the images with recognition errors, and add the images to the current rock image recognition queue for recognition correction after background cropping. If the images after background cropping are still recognized incorrectly, they enter the queue to be deleted and wait for processing; Misidentified images include: images in which the rock lithology category is not identified, images in which multiple different rock lithology categories are identified, images in which the wrong rock lithology category is identified, or images in which the rock lithology category is identified with a confidence level lower than a threshold.

Images