Deep learning-based rock texture identification method and system, and medium

Abstract

The invention relates to the technical field of rock texture recognition, in particular to a rock texture recognition method and system based on deep learning and a medium, and the method comprises the steps: capturing a rock image from an image uploading area in response to a triggering action of clicking uploading; in response to an uploading action captured from the picture uploading area, reading a rock image from the picture uploading area; in response to a click action detected for a segmentation start button, calling a rock texture recognition model to perform rock texture semantic segmentation on the rock image, and superposing texture categories and regions recognized from the rock image on the rock image in different colors to obtain a segmented rock image; wherein the rock texture recognition model is obtained by training a deep learning model; displaying the segmented rock image in a result display area; according to the method, the efficiency and accuracy of river sedimentary rock facies classification can be improved.

Description

Technical Field

[0001] The present invention relates to the technical field of deep learning, and particularly relates to a method, system and medium for rock texture recognition based on deep learning. Background Art

[0002] In the technologies related to fluvial outcrops, quantitative research has enhanced the logic and scientific nature of traditional geology mainly based on qualitative analysis. Traditional geological surveys are inefficient, costly, and have a low degree of vectorization of research results, resulting in poor reuse of the results.

[0003] In recent years, deep learning technology has made remarkable progress in the field of image recognition, especially in medical imaging, satellite remote sensing, and natural scene recognition. Deep learning models, such as convolutional neural networks (CNNs), have performed excellently in image recognition tasks due to their powerful feature extraction capabilities and high accuracy.

[0004] Therefore, it is necessary to provide a solution that can apply deep learning-based image recognition technology to outcrop rock texture recognition, so as to quickly and accurately identify the features of outcrop images, realize the automatic recognition of aspects such as outcrop rock facies, and improve the work efficiency of outcrop geological surveys; at the same time, provide reliable basic data and restrictive data for the subsequent fine three-dimensional characterization of rock facies based on the results of automatic recognition. Summary of the Invention

[0005] The purpose of the present invention is to provide a method, system and medium for rock texture recognition based on deep learning, aiming to distinguish and identify various different rock textures, so as to realize the automatic recognition of different rock facies and improve the efficiency and accuracy of fluvial sedimentary rock facies classification.

[0006] To achieve the above purpose, the present invention provides the following technical solutions:

[0007] In the first aspect, an embodiment of the present invention provides a method for rock texture recognition based on deep learning, and the method includes the following steps:

[0008] S101, in response to a trigger action of clicking to upload, capture a rock image from the upload picture area;

[0009] S102, in response to an upload action captured from the upload picture area, read the rock image from the upload picture area;

[0010] S103, in response to a click action detected on the start segmentation button, call a rock texture recognition model to perform semantic segmentation of the rock texture on the rock image, and superimpose the texture categories and regions identified from the rock image on the rock image with different colors to obtain a segmented rock image; wherein, the rock texture recognition model is obtained by training a deep learning model.

[0011] S104, Display the segmented rock image in the result display area.

[0012] Preferably, the method further includes:

[0013] S105, In response to a detected click action on the clear result button, clear the rock image in the upload picture area and the segmented rock image in the result display area.

[0014] Preferably, before invoking the rock texture recognition model to perform rock texture semantic segmentation on the rock image, the method further includes:

[0015] S100, Obtain a rock image containing multiple rock texture features, respectively determine the semantic categories corresponding to the multiple rock texture features, draw the regions of each semantic category on the rock image, and perform semantic annotation to obtain an annotated rock image;

[0016] S200, Perform image preprocessing and image enhancement on the annotated rock image to obtain a sample image, and divide the sample image into a training set and a test set;

[0017] S300, Use the training set to train each deep learning model. During the model training stage, optimize the parameters of the deep learning model through the backpropagation algorithm; after the training is completed, evaluate the performance of each deep learning model;

[0018] S400, Adopt a deep learning model that meets the performance requirements as the rock texture recognition model.

[0019] Preferably, in S200, the performing image preprocessing and image enhancement on the annotated rock image to obtain a sample image and dividing the sample image into a training set and a test set includes:

[0020] S210, Perform at least one of denoising, contrast enhancement, normalization, size adjustment, and color space conversion on the annotated rock image to obtain a preprocessed rock image;

[0021] S220, Perform at least one of rotation, scaling, flipping, cropping, and color jittering on the preprocessed rock image to obtain a sample image;

[0022] S230, Divide multiple sample images into a training set and a test set according to a ratio.

[0023] Preferably, in S300, the using the training set to train each deep learning model includes:

[0024] S310, Use a convolutional neural network to extract features from the input feature map, and learn the local features of the input feature map through multiple convolutional layers and pooling layers, and gradually expand the receptive field to obtain more global context information;

[0025] S320, Extract high-level abstract features from the input feature map through the encoder in the semantic segmentation network, map the abstract features back to the original size of the input feature map through the decoder, and generate pixel-level classification results;

[0026] S330, Connect the feature maps of different levels in the encoder with the corresponding-level feature maps in the decoder through skip connections, and use upsampling operations in the decoder to restore the spatial resolution of the feature map to obtain the output feature map.

[0027] Preferably, in S300, evaluating the performance of each deep learning model includes:

[0028] Evaluate the performance of each deep learning model using pixel accuracy and mean intersection over union to obtain the recognition effect and generalization ability of each deep learning model.

[0029] Preferably, the deep learning models include ENet, DeepLabV3+, and KNet; training each deep learning model using the training set includes:

[0030] Use swin transformer as the backbone network of the KNet to enable the KNet to better learn rock texture features as the training deepens through the swin transformer;

[0031] Through the mechanism of dynamically updating the convolutional kernel of the KNet, the kernel can make conditional responses according to the content on the image to focus on different rock texture features.

[0032] In a second aspect, an embodiment of the present invention provides a deep learning-based rock texture recognition system, and the system includes:

[0033] At least one processor;

[0034] At least one memory for storing at least one program;

[0035] When the at least one program is executed by the at least one processor, the at least one processor implements the method as described in any one of the above.

[0036] In a third aspect, an embodiment of the present invention provides a computer-readable storage medium, in which a processor-executable program is stored, and the processor-executable program is used to execute the method as described in any one of the above when executed by the processor.

[0037] The beneficial effects of the present invention are as follows: By capturing rock images in response to the triggering action of clicking to upload, and distinguishing and identifying various different rock textures through a rock texture recognition model, the automatic recognition of different rock facies is realized. Through automated image recognition and interactive image processing, the efficiency and accuracy of river sediment rock facies classification are improved, and the manual labor intensity is reduced, thereby providing an efficient auxiliary tool for geological workers. BRIEF DESCRIPTION OF THE DRAWINGS

[0038] In order to more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the following will briefly introduce the drawings required for use in the embodiments. Obviously, the drawings described below are only some embodiments of the present invention. For those of ordinary skill in the art, without creative efforts, other drawings can also be obtained based on these drawings.

[0039] Figure 1 It is a schematic flowchart of a method for rock texture recognition based on deep learning in an embodiment of the present invention;

[0040] Figure 2 It is a schematic diagram of an interactive interface in an embodiment of the present invention;

[0041] Figure 3 It is a schematic diagram of the process of automatic rock texture recognition in an embodiment of the present invention;

[0042] Figure 4 It is a schematic diagram of the effect of automatic rock texture recognition in an embodiment of the present invention;

[0043] Figure 5 It is a schematic diagram of semantic segmentation data annotation in an embodiment of the present invention;

[0044] Figure 6 It is a curve graph of the loss function decreasing during the training process in an embodiment of the present invention;

[0045] Figure 7 It is a graph of evaluation indicators on the validation set during the training process in an embodiment of the present invention;

[0046] Figure 8 It is a graph of the effect of segmenting different types of rock textures using 3 segmentation algorithms in an embodiment of the present invention;

[0047] Figure 9 It is a schematic diagram of the structure of a rock texture recognition system based on deep learning in an embodiment of the present invention. DETAILED DESCRIPTION OF THE EMBODIMENTS

[0048] The concept, specific structure, and technical effects of the present invention will be clearly and completely described below in conjunction with the embodiments and the accompanying drawings to fully understand the purpose, solution, and effects of the present invention. It should be noted that, without conflict, the embodiments in the present invention and the features in the embodiments can be combined with each other.

[0049] Referring to Figures 1 to 4 , the present invention provides a method for rock texture recognition based on deep learning, and the method includes the following steps:

[0050] S101, in response to a trigger action of clicking to upload, capture a rock image from the upload picture area;

[0051] S102, in response to an upload action captured from the upload picture area, read the rock image from the upload picture area;

[0052] S103, in response to a click action detected on the start segmentation button, call the rock texture recognition model to perform semantic segmentation on the rock image, and superimpose the texture categories and regions identified from the rock image on the rock image in different colors to obtain the segmented rock image; wherein, the rock texture recognition model is obtained by training a deep learning model;

[0053] S104, display the segmented rock image in the result display area.

[0054] The present invention captures a rock image in response to a trigger action of clicking to upload, simplifies the operation process, improves the recognition efficiency, and ensures that users can quickly obtain accurate rock texture analysis results without complex operations. Through the above steps, the present invention not only optimizes the user experience but also significantly improves the accuracy and real-time performance of rock texture recognition. Through the efficient recognition of rock textures by the rock texture recognition model, the accuracy and efficiency of geological analysis are significantly improved, providing strong technical support for geological exploration.

[0055] Specifically, in the interface design, it mainly includes some basic components, such as function modules for picture upload, picture display, segmentation result and legend display, and result clearing. The specific interface implementation effect is as Figures 2 to 4 shown:

[0056] The interactive interface includes an upload picture area, a result display area, a category legend area, a start segmentation button, and a clear result button. The category legend area displays the color legends related to the segmentation categories. The lower part of the interface is the button for starting segmentation after the picture upload is completed and the button for clearing the current result. The code for implementing the interface using Gradio is as follows:

[0057]

[0058]

[0059] After clicking the button to upload an image and then clicking the start segmentation button, the segment_image function will be called to complete the semantic segmentation function of the rock texture and display it on the page. The page effect is as shown in the lower interface of the figure. The recognizable texture categories and regions on the rock image are overlaid on the original image in different colors. Among them, red represents trough cross-bedded sandstone St, green represents planar cross-bedded sandstone Sp, orange represents wedge cross-bedded sandstone Slt, blue-violet represents massive (bedded) sandstone Sm, and yellow represents parallel bedding Sh. If the current image has been recognized, you can click to clear the result and then upload a new image. In this way, the construction of the rock texture automatic recognition system is completed.

[0060] In some improved embodiments, in S200, the preprocessing and enhancement of the labeled rock image to obtain a sample image, and dividing the sample image into a training set and a test set include:

[0061] S105, in response to the detected click action on the clear result button, clear the rock image in the upload picture area and the segmented rock image in the result display area.

[0062] In some improved embodiments, before calling the rock texture recognition model to perform semantic segmentation on the rock image, the method further includes:

[0063] S100, obtain a rock image containing multiple rock texture features, respectively determine the semantic categories corresponding to the multiple rock texture features, draw the regions of each semantic category on the rock image, and perform semantic annotation to obtain a labeled rock image;

[0064] S200, perform image preprocessing and enhancement on the labeled rock image to obtain a sample image, and divide the sample image into a training set and a test set;

[0065] S300, use the training set to train each deep learning model. In the model training stage, optimize the parameters of the deep learning model through the backpropagation algorithm; after training is completed, evaluate the performance of each deep learning model;

[0066] S400, adopt the deep learning model that meets the performance requirements as the rock texture recognition model.

[0067] The automatic rock texture recognition method based on deep learning of the present invention uses advanced convolutional neural networks (CNNs) and Transformer architectures, such as DeepLabv3, ENet, and KNet, to perform semantic segmentation and recognition on rock texture images. The specific process includes: First, collect and preprocess rock texture images; Second, use the Labelme tool to perform pixel-level annotation on the images to create the annotation data required for training; Then, divide the annotated data into a training set and a test set for training and evaluating the model. In the model training stage, the present invention uses the annotated image dataset to optimize the model parameters through the backpropagation algorithm to improve the recognition accuracy. After training, the present invention evaluates the performance of different models on an independent validation set, and comprehensively measures the effects and generalization capabilities of different types of models through metrics such as accuracy, IOU, model size, and inference speed. Finally, the present invention deploys the verified model into practical applications to achieve automatic recognition and classification of new rock texture images.

[0068] The processing of rock images is described below:

[0069] As Figure 5 shown, data processing is a crucial step in machine learning and deep learning model training, which directly affects the performance and accuracy of the model. In the present invention, the processing of rock images includes the following key steps:

[0070] Data collection and annotation: The present invention collects rock images with rock textures from multiple sources, including rock photos collected in the field, rock images of rock samples taken in the laboratory, and public geological databases, and a total of 1108 rock images are obtained. These rock images cover various different rock textures, including: trough cross-bedded sandstone St, planar cross-bedded sandstone - Sp, wedge cross-bedded - Slt, massive (bedded) sandstone - Sm, and parallel bedding - Sh. Multiple rock texture features can ensure that the model can learn the features of various rocks.

[0071] The collected rock images need to be accurately annotated. According to the rock texture characteristics of the rocks, the rock images are divided into different semantic categories to facilitate the training of deep learning models. During the annotation process, the present invention uses the pixel-level annotation tool labelme. By using the polygon tool to place points, the areas of each semantic category are drawn on the rock images to ensure that each pixel point is correctly classified, and sample images are obtained.

[0072] In some improved embodiments, in S200, the preprocessing and image enhancement of the annotated rock images to obtain sample images, and dividing the sample images into a training set and a test set include:

[0073] S210, perform at least one of denoising, contrast enhancement, normalization, size adjustment, and color space conversion on the marked rock image to obtain a preprocessed rock image;

[0074] S220, perform at least one of rotation, scaling, flipping, cropping, and color jittering on the preprocessed rock image to obtain a sample image;

[0075] S230, divide multiple sample images into a training set and a test set according to a ratio.

[0076] The steps of image preprocessing include the following aspects:

[0077] (1) Denoising: Use a filter to remove noise in the image and improve image quality.

[0078] (2) Contrast enhancement: Adjust the contrast of the image to make the rock texture features more obvious.

[0079] (3) Normalization: Scale the pixel values of the image to the range of [0, 1] or [-1, 1] to meet the input requirements of the deep learning model.

[0080] (4) Size adjustment: Adjust all images to a unified size for easy model processing.

[0081] (5) Color space conversion: In some cases, convert the image from the RGB color space to other color spaces (such as HSV or Lab) to better highlight the rock texture features.

[0082] The steps of image enhancement include the following aspects:

[0083] Since the currently collected rock image data is limited, in order to improve the generalization ability of the model and reduce overfitting, the present invention uses data augmentation technology to expand the training set. Data augmentation includes:

[0084] (1) Rotation: Randomly rotate the image by a certain angle.

[0085] (2) Scaling: Randomly scale the image.

[0086] (3) Flipping: Horizontally or vertically flip the image.

[0087] (4) Cropping: Randomly crop a small area from the original image.

[0088] (5) Color jittering: Randomly change the brightness, contrast, and saturation of the image

[0089] Through the above detailed image data processing steps, the present invention can ensure that the deep learning model can receive high-quality and diverse image data during the training process, thereby improving the recognition ability and accuracy of the model.

[0090] In some improved embodiments, in S300, the training of each deep learning model using the training set includes:

[0091] S310, using a convolutional neural network to extract features from the input feature map, learning the local features of the input feature map through multiple convolutional layers and pooling layers, and gradually expanding the receptive field to obtain more global context information;

[0092] S320, extracting high-level abstract features from the input feature map through the encoder in the semantic segmentation network, mapping the abstract features back to the original size of the input feature map through the decoder, and generating pixel-level classification results;

[0093] S330, connecting the feature maps of different levels in the encoder with the corresponding-level feature maps in the decoder through skip connections, and using upsampling operations in the decoder to restore the spatial resolution of the feature map to obtain the output feature map.

[0094] Semantic segmentation is a highly challenging and widely applied technology in the field of computer vision. Its core task is to segment a given image into several semantically similar regions, and identify the semantic information of each region in the image through pixel-level classification. This technology can not only identify the objects in the image, but also accurately depict the contours of the objects, thereby achieving an in-depth understanding of the image structure.

[0095] In some improved embodiments, in S300, the evaluation of the performance of each deep learning model includes:

[0096] Evaluating the performance of each deep learning model using pixel accuracy and mean intersection over union to obtain the recognition effect and generalization ability of each deep learning model.

[0097] With the rapid development of deep learning technology, the architecture of semantic segmentation networks has evolved from traditional convolutional neural networks (CNNs) to Transformer-based architectures, and the performance has been continuously improved. This invention comprehensively considers the development process of semantic segmentation networks and selects three representative network architectures: ENet, DeepLabV3+, and KNet. ENet is suitable for real-time semantic segmentation tasks due to its efficient computational performance. DeepLabV3+ effectively balances the recognition of context information and object boundaries through an encoder-decoder structure, while KNet demonstrates flexibility and efficiency in multi-task segmentation with its dynamic kernel update mechanism. This invention will conduct a comprehensive comparison of these three networks to evaluate their performance in the rock texture recognition task.

[0098] Model Training:

[0099] This invention uses the pre-trained model of Swin Transformer as the backbone network, which has been pre-trained on a large-scale dataset to improve the model's initial feature extraction ability. The input images are resized to 512x512 pixels and normalized to match the expected input of the pre-trained model. The AdamW optimizer is used with an initial learning rate of 6*10-6 and configured with a learning rate warm-up and multi-step decay strategy. The batch size is batch size = 2, the maximum number of training iterations is max_iters = 50000, and the validation set completes the evaluation of the current model every 1000 times and retains the best model weights. The specific training process and the decrease of the loss function are as Figure 6 shown, and the evaluation metrics of the validation test set are as Figure 7 shown. It can be seen that the loss function is decreasing slowly, and at the same time, the various metrics on the validation set are also increasing slowly, proving that the model training is effective and there is no obvious overfitting phenomenon.

[0100] Model Comparison and Analysis:

[0101] Commonly used semantic segmentation evaluation metrics include:

[0102] ACC - Pixel Accuracy: Calculate the matching degree between all predicted pixels and true label pixels. It simply calculates the proportion of correctly classified pixels to the total number of pixels.

[0103] mIOU (Mean Intersection over Union): Intersection over union (IoU) is a commonly used evaluation metric. For each category, IoU calculates the ratio of the intersection area to the union area between the predicted region and the true label region. The mean intersection over union calculates the average of the IoUs of all categories and can better measure the accuracy of segmentation.

[0104] A is the ground truth label, occupying a certain area of the image; B is the prediction result, occupying a certain area of the image. The TP part in the middle is the intersection of the ground truth label and the prediction value, and the entire colored part of the image is the union of the ground truth label and the prediction value. And mIOU is the average of the intersection over union of each class in the dataset.

[0105] A comprehensive comparison is made on the segmentation quality and performance metrics of three algorithms. Among them, DeepLabV3+ selects the structure with a resnet101-layer backbone network, ENet is a lightweight network, and KNet selects the structure with a swin transformer large backbone network for training under the same conditions. The same test set is evaluated and calculated, and the results are shown in the following table:

[0106] Comparison results of the three model metrics:

[0107]

[0108]

[0109] The segmentation effects of 3 segmentation algorithms for different types of rock textures are as Figure 8 shown:

[0110] As can be seen from the table, under the condition of the same number of training iterations for the three algorithms of DeepLabV3+, ENet, and KNet, the results of the KNet model in performance metrics such as Acc, mIoU, and Recall are much higher than those of DeepLabV3+ and ENet. It can be known that compared with the network based on the traditional CNN architecture, the Transformer-based KNet can better learn the texture features as the training deepens. Due to the small number of network parameters, ENet is difficult to learn the difficult-to-separate samples of complex textures, resulting in a poor overall effect; Although DeepLabV3+ was a very advanced network before and should also have very strong feature extraction ability with the support of the resnet101 deep backbone network, the structure of DeepLabV3+ is more suitable for extracting target regions of different scales, and the effect of distinguishing the details of rock textures is not ideal. From Figure 8It can also be seen from the segmentation recognition effects of the 3 algorithms that DeepLabV3+ and ENet have relatively serious errors in the segmentation of Slt, Sp, and Sm classes, indicating that the learning of these 2 models for these 3 classes is very limited and prone to misclassification. It can also be seen from the segmentation effect diagrams that the segmentation effect of KNet is very similar and correct compared to the true annotation GT. Even the segmentation details of the trained model on 2 samples of Slt and Sp are more delicate than the original annotation, indicating that KNet can truly learn the texture expressions between different classes and correctly divide different regions of the rock. Although the currently collected training samples are very limited, KNet already has a preliminary ability to segment and recognize complex rock textures, with certain research value and application potential. The disadvantage of KNet compared to the other two algorithms is mainly reflected in that the large number of model parameters is caused by using swin-transformer large in the backbone network, and the inference speed is also slower, requiring stronger graphics card hardware to meet the needs of actual applications.

[0111] Corresponding to the method of Figure 1 , referring to Figure 9 , an embodiment of the present invention provides a rock texture recognition system based on deep learning, including:

[0112] At least one processor;

[0113] At least one memory for storing at least one program;

[0114] When the at least one program is executed by the at least one processor, the at least one processor implements the above method.

[0115] It can be seen that the content in the above method embodiments is applicable to the system embodiments of the present invention. The functions specifically implemented by the system embodiments of the present invention are the same as those of the above method embodiments, and the beneficial effects achieved are also the same as those of the above method embodiments.

[0116] In addition, an embodiment of the present invention also discloses a computer program product or a computer program. The computer program product or the computer program is stored in a computer-readable storage medium. The processor of the computer device can read the computer program from the computer-readable storage medium, and the processor executes the computer program, so that the computer device executes the above method. Similarly, the content in the above method embodiments is applicable to the storage medium embodiments of the present invention. The functions specifically implemented by the storage medium embodiments of the present invention are the same as those of the above method embodiments, and the beneficial effects achieved are also the same as those of the above method embodiments.

[0117] Those of ordinary skill in the art will appreciate that all or some of the methods and systems disclosed above can be implemented as software, firmware, hardware, and appropriate combinations thereof. Some or all of the physical components can be implemented as software executed by a processor, such as a central processing unit, a digital signal processor, or a microprocessor, or as hardware, or as an integrated circuit, such as an application specific integrated circuit. Such software can be distributed on a computer-readable medium, which can include a computer storage medium (or non-transitory medium) and a communication medium (or transitory medium). As is well known to those of ordinary skill in the art, the term computer storage medium includes volatile and non-volatile, removable and non-removable media implemented in any method or technology for storing information, such as computer-readable instructions, data structures, program modules, or other data. Computer storage media includes, but is not limited to, RAM, ROM, EEPROM, flash memory or other memory technologies, CD-ROM, digital versatile disks (DVD) or other optical disk storage, magnetic cassettes, magnetic tape, magnetic disk storage or other magnetic storage devices, or any other medium that can be used to store the desired information and that can be accessed by a computer. In addition, it is well known to those of ordinary skill in the art that communication media typically embodies computer-readable instructions, data structures, program modules, or other data in a modulated data signal such as a carrier wave or other transport mechanism, and can include any information delivery media.

[0118] The above is a specific description of the preferred embodiments of the present disclosure, but the present disclosure is not limited to the above embodiments. Those skilled in the art can make various equivalent deformations or substitutions without departing from the spirit of the present disclosure, and these equivalent deformations or substitutions are all included within the scope defined by the claims of the present disclosure.

Claims

1. A rock texture recognition method based on deep learning, characterized in that: The method comprises the following steps: S101, in response to a trigger action of clicking upload, capturing a rock image from an upload image area; S102, in response to an upload action captured from the upload image area, reading a rock image from the upload image area; S103, in response to a click action detected on a start segmentation button, calling a rock texture recognition model to perform rock texture semantic segmentation on the rock image, and superimposing texture categories and regions identified from the rock image on the rock image in different colors to obtain a segmented rock image; wherein the rock texture recognition model is obtained based on training a deep learning model; S104, displaying the segmented rock image in a result display area.

2. The method according to claim 1, characterized in that The method further comprises: S105, in response to a click action detected on the clear result button, clearing the rock image in the uploaded image area and the segmented rock image in the result display area.

3. The method according to claim 1, characterized in that Before calling the rock texture recognition model to perform rock texture semantic segmentation on the rock image, the method further includes: S100, obtaining a rock image including a plurality of rock texture features, respectively determining semantic categories corresponding to the plurality of rock texture features, drawing regions of each semantic category on the rock image, and performing semantic annotation to obtain an annotated rock image; S200, performing image preprocessing and image enhancement on the annotated rock image to obtain sample images, and dividing the sample images into a training set and a test set; S300, using the training set to train each deep learning model, in the model training stage, optimizing the parameters of the deep learning model by a back propagation algorithm; after the training is completed, evaluating the performance of each deep learning model; S400, uses a deep learning model that meets the performance requirements as the rock texture recognition model.

4. The method according to claim 3, characterized in that In S200, the image preprocessing and image enhancement are performed on the annotated rock image to obtain a sample image, and the sample image is divided into a training set and a test set, including: S210, performing at least one of denoising, contrast enhancement, normalization, size adjustment and color space conversion on the labeled rock image to obtain a preprocessed rock image; S220, performing at least one of rotation, scaling, flipping, cropping and color dithering on the preprocessed rock image to obtain a sample image; S230, dividing the plurality of sample images into a training set and a test set according to a ratio.

5. The method according to claim 3, characterized in that: In S300, the use of the training set to train each deep learning model includes: S310 uses a convolutional neural network to extract features from the input feature map, learns the local features of the input feature map through multiple convolutional layers and pooling layers, and gradually expands the receptive field to obtain more global context information; S320, extracting high-level abstract features from the input feature map through the encoder in the semantic segmentation network, mapping the abstract features back to the original size of the input feature map through the decoder, and generating pixel-level classification results; S330, connecting feature maps of different levels in the encoder with feature maps of corresponding levels in the decoder through skip connections, and restoring the spatial resolution of the feature map by using an upsampling operation in the decoder to obtain an output feature map.

6. The method according to claim 5, characterized in that In S300, the evaluating the performance of each deep learning model includes: The pixel accuracy and average intersection-over-union ratio were used to evaluate the performance of each deep learning model, and the recognition effect and generalization ability of each deep learning model were obtained.

7. The method according to claim 3, characterized in that The deep learning models include ENet, DeepLabV3+ and KNet; the training of each deep learning model using the training set includes: Using swin transformer as the backbone network of the KNet, so that KNet can better learn rock texture features as the training deepens through swin transformer; The mechanism of dynamically updating the convolution kernel through KNet enables the kernel to respond conditionally based on the content on the image to focus on different rock texture features.

8. A rock texture recognition system based on deep learning, characterized in that: The system comprises: at least one processor; at least one memory for storing at least one program; When the at least one program is executed by the at least one processor, the at least one processor implements the method according to any one of claims 1 to 7.

9. A computer-readable storage medium storing a program executable by a processor, characterized in that: The processor-executable program is used to perform the method according to any one of claims 1 to 7 when executed by the processor.

Images