Rock identification system

Abstract

The invention provides a rock identification system, and the system comprises a data acquisition module which is used for collecting a rock image and corresponding lithologic spatial distribution data; the geographical recognition module is used for generating a rock space distribution diagram of the region through the lithology space distribution data, preliminarily judging the type of the rock based on the rock space distribution diagram and generating a first rock feature; the image recognition module is used for recognizing a second rock feature of the rock image; the rock recognition module is used for calculating the matching degree of the first rock feature and the second rock feature, inputting the first rock feature and the second rock feature into a rock recognition model and outputting a rock recognition result if the matching degree is larger than or equal to a preset threshold value, and inputting the second rock feature into the rock recognition model and outputting a rock recognition result if the matching degree is smaller than the preset threshold value; and outputting a rock identification result. The method can be applied to different rock identification environments, and can realize efficient and accurate rock category identification no matter in the field or in the room.

Description

Technical Field

[0001] The present invention relates to the technical field of geological exploration, and particularly to a rock identification system. Background Art

[0002] In the fields of geology and resource exploration, the identification of rock types is a fundamental and crucial task. Traditional rock identification methods mainly rely on manual observation and analysis. For example, rock types are identified by observing features such as the texture and shape of rocks. However, these methods have obvious limitations. For example, the subjectivity of feature selection and the limited identification effect in complex environments. With the development of deep learning technology, especially the emergence of convolutional neural networks (CNNs), the automatic extraction and classification of image features have become more effective. Although deep learning technology has made remarkable progress in the field of image recognition, it still faces challenges in the field of rock identification. Rock sample images have complex texture and shape features. At the same time, the identification of rock samples involves different geological environments and lighting conditions, resulting in the diversity and variability of rock sample images. In addition, the determination of the lithology of field rocks not only needs to consider from the appearance, but also needs to fully consider the geographical environment where the rocks are located. Therefore, simply relying on image recognition to judge the category of rocks leads to certain limitations in the accuracy of the recognition results. Summary of the Invention

[0003] In view of this, the purpose of the present invention is to provide a rock identification system that can achieve efficient and accurate identification of rock categories both in the field and in the laboratory.

[0004] To achieve the above-mentioned invention purpose, the present invention provides a rock identification system, which includes:

[0005] A data acquisition module for collecting rock images and corresponding lithology spatial distribution data;

[0006] A geographical identification module for generating a rock spatial distribution map of the area through the lithology spatial distribution data, and preliminarily determining the category of rocks based on the rock spatial distribution map to generate the first rock feature;

[0007] An image identification module for identifying the second rock feature of the rock image;

[0008] A rock identification module for calculating the matching degree between the first rock feature and the second rock feature. If the matching degree is greater than or equal to a preset threshold, the first rock feature and the second rock feature are input into the rock identification model to output the rock identification result. If the matching degree is less than the preset threshold, the second rock feature is input into the rock identification model to output the rock identification result.

[0009] Further, the data acquisition module is specifically used to perform the following operations:

[0010] S11. Obtain the geographical longitude and latitude information corresponding to the rock image;

[0011] S12. According to the geographical longitude and latitude information, obtain the target area geographical information data corresponding to the rock image;

[0012] S13. Crawl the lithological characteristics and topological characteristics of the target area geographical information data through web crawler technology, and perform feature matching with the geographical location data, lithological characteristics and topological characteristics in the geographical information system database to obtain the lithological spatial distribution data of the corresponding rock image.

[0013] Furthermore, the geographical recognition module is specifically used to perform the following operations:

[0014] S21. Display the collected lithological spatial distribution data geospatially through GIS technology to generate a rock spatial distribution map;

[0015] S22. Perform feature recognition on the rock spatial distribution map through a machine learning model to obtain a preliminary judgment model;

[0016] S23. Input the rock image into the preliminary judgment model to determine whether the rock image belongs to the rock spatial distribution map. If it is, output the first rock feature; if not, output that the rock image does not belong to the rock spatial distribution map.

[0017] Furthermore, the image recognition module includes:

[0018] An image processing unit, used to perform binarization processing on the rock image to obtain the structural characteristics of the rock image;

[0019] An image block unit, used to perform block processing on the rock image;

[0020] An edge processing unit, used to perform edge processing on the block-divided rock image;

[0021] An image recognition unit, used to perform feature recognition on the edge-processed block-divided rock image, and perform feature fusion on the recognized features and the structural characteristics to determine the second rock feature.

[0022] Furthermore, the image processing unit is specifically used to perform the following operations:

[0023] S31. Convert the rock image into a grayscale image, perform pixel grayscale statistics on the grayscale image to obtain a grayscale statistical histogram;

[0024] S32. Take a grayscale value as the grayscale image segmentation threshold, and divide the grayscale statistical histogram into a dark part and a bright part based on the grayscale image segmentation threshold;

[0025] S33. Based on the result of step S32, perform preliminary binarization on the grayscale image through the OTSU algorithm;

[0026] S34. Perform local binarization on the dark parts of the grayscale image after preliminary binarization through the Sauvola algorithm to obtain a binarized image;

[0027] S35. Detect the edges of the binarized image through an edge detection algorithm to obtain the structural features of the rock image.

[0028] Furthermore, the image block unit is specifically used to perform the following operations:

[0029] S41. Perform superpixel segmentation on the rock image through the superpixel algorithm to generate superpixel blocks;

[0030] S42. Coarsely merge the superpixel blocks using a merging strategy to obtain the regions to be recognized.

[0031] Furthermore, the edge processing unit is specifically used to perform the following operations:

[0032] S51. Label the categories of the regions to be recognized and perform normalization processing;

[0033] S52. Extract color features by calculating the average value of all pixels in the LAB color space within each region to be recognized, and extract texture features through the gray-level co-occurrence matrix within each region to be recognized;

[0034] S53. Perform preliminary edge recognition on the edges by combining the edge detection algorithm with the texture features within each region to be recognized to obtain an initial region edge map;

[0035] S54. For the boundary lines of each initial region edge map, combine upsampling using bilinear interpolation with an attention mechanism to adaptively increase the weight of edge information and enhance the detailed features of the edge contours of the initial region edge map, obtaining the region edge map of the regions to be recognized in the rock image.

[0036] Furthermore, the image recognition unit is specifically used to perform the following operations:

[0037] S61. After performing 3*3 convolution, 5*5 convolution, and 1*1 convolution operations on the region edge map through a neural network respectively, perform average pooling operation, and splice the output global features and local features in the channel dimension to obtain the first fusion feature of the region edge map;

[0038] S62. After obtaining the channel attention weights for the first fusion feature through the channel attention module, then obtain the spatial attention weights through the spatial attention module;

[0039] S63. Multiply the channel attention weights along the channel dimension with the first fused feature, and multiply the spatial attention weights along the spatial dimension with the first fused feature to obtain the enhanced feature of the regional edge map.

[0040] S64. After capturing the different scale information of the enhanced feature and the structural feature at different network levels through multi-modal feature fusion, perform feature splicing to obtain the second rock feature.

[0041] Further, the rock recognition model is specifically used to perform the following operations:

[0042] When the first rock feature and the second rock feature are input simultaneously, perform feature fusion after assigning weights to the first rock feature and the second rock feature through the attention mechanism, and output the rock recognition result by passing the fused feature through the classifier.

[0043] When the second rock feature is input alone, output the rock recognition result by passing it through the classifier.

[0044] Compared with the prior art, the beneficial effects of the present invention are as follows:

[0045] A rock recognition system provided by the present invention combines the geographical information (the first rock feature) and the image feature (the second rock feature) of the rock. When the matching degree between the geographical information and the image feature reaches the preset threshold, the joint information of the two is used for recognition, which can improve the reliability of recognition in the case of a complex environment in the wild. In the case where the matching degree between the geographical information and the image feature is lower than the preset threshold, it automatically switches to using only the image feature for recognition, so that the system can still maintain a high recognition rate when facing rock images in different recognition environments. In summary, through the intelligent recognition process, the present invention can adapt to different recognition environments, and can provide stable recognition performance whether in the wild or indoors, which not only improves the accuracy and efficiency of rock recognition, but also provides strong technical support for geological exploration and research. BRIEF DESCRIPTION OF THE DRAWINGS

[0046] In order to more clearly illustrate the technical solutions in the embodiments of the present application or the prior art, the following will briefly introduce the drawings required for use in the description of the embodiments or the prior art. Obviously, the drawings in the following description are only the embodiments of the present application, and for those of ordinary skill in the art, other drawings can be obtained according to the provided drawings without creative efforts.

[0047] Figure 1 It is a schematic structural diagram of a rock recognition system provided by an embodiment of the present invention. DETAILED DESCRIPTION OF THE EMBODIMENTS

[0048] The present invention will be further described in detail below with reference to the accompanying drawings and embodiments. It can be understood that the specific embodiments described herein are only used to explain the present invention, rather than limiting the present invention. Additionally, it should be noted that for the convenience of description, only the parts related to the present invention rather than all the structures are shown in the drawings.

[0049] Referring Figure 1 , this embodiment provides a rock identification system, and the system includes:

[0050] A data acquisition module, configured to collect rock images and corresponding lithologic spatial distribution data. The data acquisition module is specifically configured to perform the following operations:

[0051] S11. Obtain the geographical longitude and latitude information corresponding to the rock image.

[0052] S12. According to the geographical longitude and latitude information, obtain the target area geographical information data corresponding to the rock image.

[0053] S13. Crawl the lithologic characteristics and topological characteristics of the target area geographical information data through web crawler technology, and perform feature matching with the geographical location data, lithologic characteristics, and topological characteristics in the geographical information system database to obtain the lithologic spatial distribution data corresponding to the rock image.

[0054] In this embodiment, when the user inputs the rock image to be identified into the rock identification system, the system automatically obtains the geographical longitude and latitude information corresponding to the rock image through positioning technology. According to the geographical longitude and latitude information, the system can locate the geographical location where the rock image to be identified is located, and obtain the geographical information data belonging to this area through the geographical location, that is, the target area geographical information data. The target area geographical information data often contains hidden information about the target geographical area. By using web crawler technology to crawl the lithologic characteristics and topological characteristics of the required target area geographical information data, useful information for identifying the rock image can be obtained.

[0055] The data obtained by crawling is then subjected to feature matching with the geographical location data, lithologic characteristics, and topological characteristics in the geographical information system database to determine the validity of the information, and at the same time, the lithologic spatial distribution data corresponding to the rock image is obtained. The lithologic spatial distribution data provides the distribution characteristics of rocks in geographical space, which is crucial for the correct identification of field rocks. It can more accurately identify rock categories. For example, methods based on deep learning can process the lithologic spatial distribution data to automatically identify and classify the lithology of rocks in the lithologic spatial distribution data of the target geographical location.

[0056] A geographical recognition module, which is used to generate a rock spatial distribution map of the area through lithological spatial distribution data, and preliminarily determine the category of the rock based on the rock spatial distribution map to generate the first rock feature. Specifically, the geographical recognition module is used to perform the following operations:

[0057] S21. Display the collected lithological spatial distribution data geospatially through GIS technology to generate a rock spatial distribution map.

[0058] S22. Perform feature recognition on the rock spatial distribution map through a machine learning model to obtain a preliminary judgment model.

[0059] S23. Input the rock image into the preliminary judgment model to determine whether the rock image belongs to the rock spatial distribution map. If so, output the first rock feature; if not, output that the rock image does not belong to the rock spatial distribution map.

[0060] In this embodiment, GIS technology can convert abstract lithological spatial distribution data into an intuitive rock spatial distribution map, making complex geological information clear at a glance, providing intuitive data input for the subsequent machine learning model, and helping the machine learning model better understand and learn the rock spatial distribution characteristics of the target area. For example, accurately extract rock features such as texture, color, and shape. Perform feature recognition on the rock spatial distribution map of the target geographical area through a machine learning model to obtain a preliminary judgment model for the target area. Through the preliminary judgment model, it is possible to preliminarily determine whether the rock input by the user belongs to this area. If the rock does not belong to this area, it is prompted that the rock image taken or input by the user does not belong to the target geographical area, and at the same time, the system automatically switches to only using image features to identify the category of the rock; if the rock belongs to this area, the preliminary judgment model preliminarily extracts the features of the rock image to obtain the first rock feature, that is, the system can use prior knowledge to assist the rock recognition module to quickly locate and identify the category of the rock.

[0061] An image recognition module, which is used to identify the second rock feature of the rock image. The image recognition module includes:

[0062] An image processing unit, which is used to perform binarization processing on the rock image to obtain the structural feature of the rock image.

[0063] An image block unit, which is used to perform block processing on the rock image.

[0064] An edge processing unit, which is used to perform edge processing on the block rock image.

[0065] An image recognition unit, which is used to perform feature recognition on the edge-processed block rock image and perform feature fusion on the recognized features and the structural features to determine the second rock feature.

[0066] In this embodiment, the features of the rock image are extracted to identify the category of the rock. By binarizing the rock image, the key structural features in the rock image are highlighted; the block processing can reduce the complexity of the image, lower the computational complexity of image processing, improve the processing speed, and make the edge detection and feature recognition more focused on the important parts of the image; the edge information in the rock image is identified through edge processing. Fusing the structural features with the features of the block-processed and edge-processed rock image can provide a more comprehensive feature description for the rock image, enhance the ability to extract rock features, and determine the second rock feature of the rock image. That is, processing the image step by step can optimize the image recognition process, making each step focus on a specific image processing task.

[0067] The image processing unit is specifically used to perform the following operations:

[0068] S31. Convert the rock image into a grayscale image, perform pixel grayscale statistics on the grayscale image, and obtain a grayscale statistical histogram.

[0069] S32. Take a grayscale value as the grayscale image segmentation threshold, and divide the grayscale statistical histogram into a dark part and a bright part based on the grayscale image segmentation threshold.

[0070] S33. Perform preliminary binarization processing on the grayscale image based on the OTSU algorithm in step S32.

[0071] S34. Perform local binarization processing on the dark part of the grayscale image after preliminary binarization processing through the Sauvola algorithm to obtain a binarized image.

[0072] S35. Detect the edges of the binarized image through an edge detection algorithm to obtain the structural features of the rock image.

[0073] In this embodiment, the color rock image is simplified to a grayscale image through grayscale conversion to reduce the complexity of data, and at the same time provides a basis for the generation of the grayscale statistical histogram. The grayscale histogram can enhance the contrast of the image to highlight the texture and structural features of the rock. Based on the grayscale image segmentation threshold, the grayscale statistical histogram is divided into a dark part and a bright part to adapt to rock images under different lighting conditions. The divided dark part and bright part can respectively represent different features in the rock image. For example, the dark part represents the shadow or crack of the rock, and the bright part represents the high-reflection area of the rock. The adaptive characteristic of the OTSU algorithm enables it to adapt to rock images under different lighting conditions. By traversing the grayscale values and calculating the between-class variance under each threshold, the threshold that maximizes the variance is found to determine the optimal threshold and find the segmentation point, thereby realizing the preliminary binary automatic segmentation of the image. The Sauvola algorithm dynamically adjusts the threshold by calculating the local mean and standard deviation of each pixel point in the dark part of the grayscale image after preliminary binary processing, thereby realizing the local binary processing of the image. By using the edge detection algorithm, the edges of the rock in the image are automatically identified to highlight the key structural features in the rock image, such as cracks, bedding, and joints, etc.

[0074] The image block unit is specifically used to perform the following operations:

[0075] S41. Perform superpixel segmentation on the rock image through the superpixel algorithm to generate superpixel blocks.

[0076] S42. Coarsely merge the superpixel blocks using the merging strategy to obtain the area to be recognized.

[0077] In this embodiment, superpixel segmentation is performed on the rock image through the superpixel algorithm to generate superpixel blocks, which can divide the rock image into regions with similar colors or textures while maintaining the structural features of the rock image. The merging strategy is used to coarsely merge the superpixel blocks to generate the area to be recognized. The generated superpixel blocks and the area to be recognized provide a recognition basis for subsequent edge detection, feature extraction, and rock classification.

[0078] The use of the edge detection algorithm is obvious at the positions where the color of the rock image changes drastically, but the recognition result for similar targets is not good. There will be some points with large gradient changes inside the target to be segmented in the edge detection algorithm, which will cause them to be misjudged as edge points. These problems are disadvantageous for an algorithm like the edge detection algorithm whose goal is to output the final segmentation result. For the superpixel algorithm, its ultimate goal is not to output the final segmentation result, but to over-segment the target to be segmented, that is, to generate a large number of pixel blocks. For data such as rock images with complex components to be segmented and high similarity, using the superpixel algorithm to output the image as a large number of pixel blocks and then using the merging algorithm to merge the same rock components will obtain better results than directly using the segmentation algorithm.

[0079] The edge processing unit is specifically used to perform the following operations:

[0080] S51. Label the category of the area to be recognized and perform normalization processing.

[0081] S52. Extract color features by calculating the average value of the LAB color space of all pixels in each area to be recognized, and extract texture features in each area to be recognized through a gray-level co-occurrence matrix.

[0082] S53. Perform a preliminary recognition of the edge by combining an edge detection algorithm with the texture features in each area to be recognized, and obtain an initial area edge map.

[0083] S54. For the boundary line of each initial area edge map, combine the upsampling of the bilinear interpolation method with the attention mechanism, adaptively increase the weight of the edge information, and strengthen the detailed features of the edge contour of the initial area edge map to obtain the area edge map of the area to be recognized in the rock image.

[0084] In this embodiment, first, labeling and normalizing the area to be recognized can reduce the complexity of subsequent data processing. Each area to be recognized has different color features, and the LAB color space can better reflect the human eye's perception of color. By calculating the LAB average value of all pixels in each area to be recognized, the color information of a certain area can be obtained, which helps to distinguish different rock features. The gray-level co-occurrence matrix is used to describe the texture features of the rock image, and texture information is extracted by analyzing the gray relationship between pixels to distinguish the texture changes in each area to be recognized. By using the texture features in the recognition area, it can help to distinguish the real edge from the false edge caused by noise or other factors, and reduce misdetection.

[0085] Edge details are an important part of rock image feature extraction. Strengthening the edge details can better identify the features of rocks in the feature extraction task. When upsampling and magnifying back to the original image size, it is easy to cause edge blurring. The upsampling of the bilinear interpolation method considers the correlation influence of the 4 nearest neighbor points around the point to be sampled on the sampled point, which improves the continuity of the image edge, thereby reducing edge blurring. After sampling, more detailed features in the image edge are retained. Combining with the attention mechanism, the feature channels related to the rock image information are retained or enhanced, and the feature channels related to the noise information are removed or weakened, so as to enhance the aggregation ability of the edge boundary speckles and the anti-background interference ability, thereby strengthening the detailed features of the edge contour of the initial area edge map.

[0086] The image recognition unit is specifically used to perform the following operations:

[0087] S61. After performing 3×3 convolution, 5×5 convolution, and 1×1 convolution operations on the regional edge map through a neural network respectively, perform average pooling operation, and splice the output global features and local features in the channel dimension to obtain the first fusion feature of the regional edge map;

[0088] S62. After obtaining the channel attention weights from the first fusion feature through the channel attention module, obtain the spatial attention weights through the spatial attention module;

[0089] S63. Multiply the channel attention weights along the channel dimension with the first fusion feature, and multiply the spatial attention weights along the spatial dimension with the first fusion feature to obtain the enhanced feature of the regional edge map;

[0090] S64. After capturing the different scale information of the enhanced feature and the structural feature at different network levels through multi-modal feature fusion, perform feature splicing to obtain the second rock feature.

[0091] In this embodiment, by using convolution kernels of different scales, the neural network can extract feature information of different scales from the regional edge map, including global features and local features. Among them, the global features can obtain more extensive background knowledge from the regional edge map, which helps to grasp the overall features and helps to reduce the ambiguity that may be brought by local features, while the local features can effectively capture details and improve the perception of the subtle features of the regional edge map. By splicing the output global features and local features in the channel dimension, the input regional edge map can be understood more comprehensively, and it can better adapt to input data of different scales and complexities, thereby improving the feature extraction and recognition of the regional edge map.

[0092] The channel attention module can determine the weight relationship between different channels of the first fusion feature, enhance the weights of key channels and suppress channels with less effect. The spatial attention module is to determine the weight relationship between different pixels in the spatial domain of the first fusion feature, enhance the weights of pixels in key regions, and reduce the weights of unimportant regions. The input first fusion feature first passes through the channel attention module, then through the spatial attention module, and finally outputs to enhance the region of interest from both the channel and spatial aspects. After obtaining the channel and spatial attention weights, the channel attention weights are multiplied with the first fusion feature along the channel dimension, while the spatial attention weights are multiplied with the first fusion feature along the spatial dimension to simultaneously focus on important channels and important spatial positions and obtain the enhanced feature of the regional edge map.

[0093] The enhanced features and structural features are respectively input into their respective sub-networks for extracting different-scale information at different network levels, that is, feature extraction is performed at different levels of the neural network. The extracted features are adjusted in the number of channels through 1×1 convolution to splice the features to obtain the second rock feature.

[0094] A rock recognition module is used to calculate the matching degree between the first rock feature and the second rock feature. If the matching degree is greater than or equal to a preset threshold, the first rock feature and the second rock feature are input into the rock recognition model to output the rock recognition result. If the matching degree is less than the preset threshold, the second rock feature is input into the rock recognition model to output the rock recognition result.

[0095] In this embodiment, when the matching degree between the first rock feature and the second rock feature reaches the preset threshold, the joint information of the two is used to identify the rock, which can improve the reliability of identification in the case of a complex field environment. In the case where the matching degree between the first rock feature and the second rock feature is lower than the preset threshold, it automatically switches to using only the second rock feature for identification, so as to still maintain a high recognition rate when facing rock images in different recognition environments.

[0096] The rock recognition model is specifically used to perform the following operations:

[0097] When the first rock feature and the second rock feature are input simultaneously, weights are assigned to the first rock feature and the second rock feature through the attention mechanism and then feature fusion is performed. The fused features are output through the classifier as the rock recognition result;

[0098] When the second rock feature is input alone, the rock recognition result is output through the classifier.

[0099] In this embodiment, through the attention mechanism, the rock recognition model can distinguish the importance of the first rock feature and the second rock feature, assign higher weights to the key features to enhance the feature expression, perform feature fusion on the first rock feature and the second rock feature with assigned weights to provide more comprehensive rock information, and analyze the fused features from multiple angles through the classifier to output the rock recognition result. When the geographical information is not sufficient to provide useful recognition information, the classifier analyzes the second rock feature alone to output the rock recognition result.

[0100] The above are only the preferred embodiments of the present invention and are not intended to limit the present invention. Any modifications, equivalent replacements, improvements, etc. made within the spirit and principle of the present invention shall be included within the protection scope of the present invention.

Claims

1. A rock identification system, characterized in that: The system comprises: A data acquisition module, used to collect rock images and corresponding lithology spatial distribution data; A geographic identification module is used to generate a rock spatial distribution map of the area through the lithology spatial distribution data, and to make a preliminary determination of the rock category based on the rock spatial distribution map to generate a first rock feature; An image recognition module, used for recognizing a second rock feature of the rock image; The rock identification module is used to calculate the matching degree between the first rock feature and the second rock feature. If the matching degree is greater than or equal to a preset threshold, the first rock feature and the second rock feature are input into the rock identification model and the rock identification result is output. If the matching degree is less than the preset threshold, the second rock feature is input into the rock identification model and the rock identification result is output.

2. The rock identification system according to claim 1, characterized in that: The data acquisition module is specifically used to perform the following operations: S11, obtaining geographic longitude and latitude information corresponding to the rock image; S12, acquiring geographic information data of a target area corresponding to the rock image according to the geographic longitude and latitude information; S13. The lithologic characteristics and topological characteristics of the geographic information data of the target area are captured by crawler technology, and feature matching is performed with the geographic location data, lithologic characteristics and topological characteristics in the geographic information system database to obtain the lithologic spatial distribution data of the corresponding rock image.

3. The rock identification system according to claim 1, characterized in that: The geographic identification module is specifically used to perform the following operations: S21. Display the collected lithology spatial distribution data in geographic space through GIS technology to generate a rock spatial distribution map; S22, performing feature recognition on the rock spatial distribution map through a machine learning model to obtain a preliminary judgment model; S23, inputting the rock image into the preliminary judgment model to judge whether the rock image belongs to the rock spatial distribution map, if yes, outputting the first rock feature, if no, outputting that the rock image does not belong to the rock spatial distribution map.

4. The rock identification system according to claim 1, characterized in that: The image recognition module includes: An image processing unit, used for performing binarization processing on the rock image to obtain the structural features of the rock image; An image segmentation unit, used for segmenting the rock image; An edge processing unit, used for performing edge processing on the divided rock images; The image recognition unit is used to perform feature recognition on the edge-processed block rock image, and to fuse the recognized features with the structural features to determine the second rock feature.

5. The rock identification system according to claim 4, characterized in that: The image processing unit is specifically used to perform the following operations: S31, converting the rock image into a grayscale image, performing grayscale statistics of pixels in the grayscale image, and obtaining a grayscale statistical histogram; S32, taking a grayscale value as a grayscale image segmentation threshold, and dividing the grayscale statistical histogram into a dark part and a bright part based on the grayscale image segmentation threshold; S33, performing preliminary binarization processing on the grayscale image based on step S32 by using the OTSU algorithm; S34, performing local binarization processing on the dark part of the grayscale image after preliminary binarization processing by using Sauvola algorithm to obtain a binarized image; S35. Detect the edge of the binary image using an edge detection algorithm to obtain structural features of the rock image.

6. The rock identification system according to claim 4, characterized in that: The image segmentation unit is specifically used to perform the following operations: S41, performing super-pixel segmentation on the rock image by using a super-pixel algorithm to generate super-pixel blocks; S42: Use a merging strategy to roughly merge the superpixel blocks to obtain a region to be identified.

7. The rock identification system according to claim 6, characterized in that: The edge processing unit is specifically used to perform the following operations: S51, marking the category of the area to be identified and performing normalization processing; S52, extracting color features by calculating the average value of the LAB color space of all pixels in each area to be identified, and extracting texture features in each area to be identified by using a gray level co-occurrence matrix; S53, performing preliminary identification of edges by combining an edge detection algorithm with texture features in each area to be identified, and obtaining an initial area edge map; S54. For the boundary line of each initial regional edge map, the upsampling of the bilinear interpolation method is combined with the attention mechanism to adaptively increase the edge information weight, strengthen the detailed features of the edge contour of the initial regional edge map, and obtain the regional edge map of the rock image to be identified.

8. The rock identification system according to claim 7, characterized in that: The image recognition unit is specifically used to perform the following operations: S61, after performing a 3*3 convolution, a 5*5 convolution, and a 1*1 convolution operation on the region edge map through a neural network, an average pooling operation is performed, and the output global features and local features are spliced ​​in the channel dimension to obtain a first fusion feature of the region edge map; S62, obtaining a channel attention weight by passing the first fusion feature through a channel attention module, and then obtaining a spatial attention weight by passing the first fusion feature through a spatial attention module; S63, multiplying the channel attention weight by the first fusion feature along the channel dimension, and multiplying the spatial attention weight by the first fusion feature along the spatial dimension to obtain an enhanced feature of the regional edge map; S64. After capturing the different scale information of enhanced features and structural features at different network levels through multimodal feature fusion, feature splicing is performed to obtain the second rock feature.

9. The rock identification system according to claim 1, characterized in that: The rock identification model is specifically used to perform the following operations: When the first rock feature and the second rock feature are input at the same time, the first rock feature and the second rock feature are assigned weights through the attention mechanism and then feature fusion is performed, and the fused features are output as rock recognition results through the classifier; When the second rock feature is input alone, the rock identification result is output through the classifier.

Images