Microscopic image data acquisition and analysis system for profile survey

Abstract

The invention relates to the technical field of image processing, in particular to a microscopic image data acquisition and analysis system for profile investigation, which comprises a soil and stone detection unit, an image enhancement unit, an expert detection unit, a feature extraction unit, a stratum recognition unit and a block chain storage unit, and further comprises an image acquisition unit and a data processing unit, the image acquisition unit is used for performing image acquisition on a target geological section through a digital camera so as to obtain a microscopic image of the target geological section, and transmitting the microscopic image of the target geological section to the image denoising unit. According to the invention, the image acquisition unit uses a digital camera to acquire a microscopic image of a geological section, initial data is provided for subsequent data processing and analysis, high-quality images are acquired, the image denoising unit is responsible for removing noise and interference in the images, the images are ensured to be clearer and more stable in quality, and the image quality is improved. Therefore, the accuracy of subsequent image processing and analysis can be improved, and the reliability of data results is ensured.

Description

Technical Field

[0001] The present invention relates to the technical field of image processing, and particularly relates to a microscopic image data acquisition and analysis system for profile survey. Background Art

[0002] Geological profile survey refers to the vertical profile observation and survey of geological bodies such as rocks and soils on the surface or underground by geologists or geological engineers. This kind of survey is usually carried out to understand the underground geological structure, lithology, soil type, etc., so as to make reasonable decisions and plans in the fields of engineering construction, resource exploration, environmental protection, etc.

[0003] Traditional systems may rely on manual acquisition of microscopic images of geological profiles, which may lead to inconsistencies and subjectivity in image quality, thus affecting the accuracy of subsequent data processing and analysis. Moreover, traditional systems may lack advanced image processing technologies and are unable to effectively remove noise and interference in images, nor can they effectively segment and enhance images, which may lead to a decline in data quality and unreliability of analysis results. In traditional systems, the identification of soil and rock types usually relies on manual analysis and judgment, which may be affected by individual subjective abilities and experience, resulting in inconsistencies and reliability problems in identification results. In addition, traditional systems may adopt traditional data storage methods such as files or databases, which pose risks to data security and integrity, are vulnerable to tampering and damage, and affect the credibility and security of data. Finally, traditional systems may rely on manual operations and analysis, with low work efficiency, and may have problems such as errors and missed detections, resulting in long survey and analysis processes and high costs. Summary of the Invention

[0004] The object of the present invention is to propose a microscopic image data acquisition and analysis system for profile survey in view of the problems existing in the background art.

[0005] The technical solution of the present invention: A microscopic image data acquisition and analysis system for profile survey, including an image acquisition unit, an image denoising unit, an image segmentation unit, an image display unit, a soil and rock detection unit, and an image enhancement unit.

[0006] The image acquisition unit is used to acquire an image of the target geological profile through a digital camera, so as to obtain a microscopic image of the target geological profile, and transmit the microscopic image of the target geological profile to the image denoising unit.

[0007] The image denoising unit receives the microscopic image of the target geological profile transmitted by the image acquisition unit, and performs image denoising on the microscopic image of the target geological profile through an image denoising method, so as to obtain a standard microscopic image of the target geological profile, and transmit the standard microscopic image of the target geological profile to the image segmentation unit.

[0008] The image segmentation unit receives the standard target geological profile microscopic image transmitted by the image denoising unit, and performs image segmentation on the standard target geological profile microscopic image through a trained image segmentation model, so as to obtain a soil-rock profile image and a stratum profile image, and transmits the soil-rock profile image and the stratum profile image to the image display unit, transmits the soil-rock profile image to the soil-rock detection unit, and transmits the stratum profile image to the image enhancement unit;

[0009] The image display unit receives the soil-rock profile image and the stratum profile image transmitted by the image segmentation unit, and displays the soil-rock profile image and the stratum profile image on a display terminal.

[0010] Preferably, the soil-rock detection unit receives the soil-rock profile image transmitted by the image segmentation unit and the template image group transmitted by the blockchain storage unit, and matches the soil-rock profile image with the template image group. If the soil-rock profile image matches the template image in the template image group, it is determined that the soil-rock type in the soil-rock profile image is the soil-rock type in the template image; otherwise, the soil-rock profile image is transmitted to the expert detection unit.

[0011] Preferably, the image enhancement unit receives the stratum profile image transmitted by the image segmentation unit, and performs image enhancement on the stratum profile image through an image enhancement method, so as to obtain an enhanced stratum profile image, and transmits the enhanced stratum profile image to the feature extraction unit.

[0012] Preferably, the feature extraction unit receives the enhanced stratum profile image transmitted by the image enhancement unit, extracts the median values of the red gray value, the green gray value, and the blue gray value in the RGB color space of the enhanced stratum profile image, converts the median values of the red gray value, the green gray value, and the blue gray value into the median values of H, S, and V in the HSV color space, and extracts the panchromatic gray value of the enhanced stratum profile image, and transmits the median values of H, S, and V and the panchromatic gray value to the stratum identification unit.

[0013] Preferably, the expert detection unit receives the soil-rock profile image transmitted by the soil-rock detection unit, and transmits the soil-rock profile image to the smart terminal of the target expert. The target expert identifies the soil-rock type in the soil-rock profile image in the smart terminal, so as to obtain the soil-rock type in the soil-rock profile image, and transmits the soil-rock profile image and the corresponding soil-rock type to the blockchain storage unit.

[0014] Preferably, the blockchain storage unit receives the soil profile image transmitted by the expert detection unit and the corresponding soil type. The template image group is stored in the blockchain. The template image group includes multiple template images and the corresponding soil types, and stores the soil profile image and the corresponding soil type into the template image group, so as to update the template image group, and transmit the updated template image group to the soil detection unit.

[0015] Preferably, the formation identification unit receives the median values of H, S, and V and the panchromatic gray value transmitted by the feature extraction unit, and inputs the median values of H, S, and V and the panchromatic gray value into the trained soil classification model based on the support vector machine. The soil layer type corresponding to the median values of H, S, and V and the panchromatic gray value is output through the trained soil classification model based on the support vector machine, so as to obtain the soil layer type corresponding to the formation profile.

[0016] Preferably, the image denoising method includes the following steps:

[0017] A1. Respectively extract the luminance components of the L channel of the target geological profile microscopic image and the reference image, and calculate the histogram of the luminance components of the L channel. The calculation formula of the histogram of the luminance components of the L channel is as follows:

[0018]

[0019] where H LK (μ) represents the percentage of the value of the luminance histogram at the luminance value of μ in the total number of pixels, w and h represent the width and height of the image, and L R (j, i) represents the luminance value at the position (j, i) in the reference image;

[0020] A2. Based on the histogram of the luminance components of the L channel, calculate the cumulative histogram corresponding to the histogram of the luminance components of the L channel. The calculation formula of the cumulative histogram is as follows:

[0021]

[0022] where represents the value of the luminance histogram at the luminance value of μ, and H LK (i) represents the percentage of the value of the luminance histogram at the luminance value of i in the total number of pixels;

[0023] A3. Compare the brightness values of each pixel in the reference image and the microscopic image of the target geological section, so as to obtain the corresponding brightness transformation relationship between the two. Establish a mapping function between the brightness value and the cumulative histogram through the corresponding brightness transformation relationship. Adjust the microscopic image of the target geological section at each same brightness value based on the mapping function between the brightness value and the cumulative histogram, so that the microscopic image of the target geological section has the same cumulative histogram as the reference image.

[0024] Preferably, the image segmentation model includes a backbone module, an auxiliary module, and a position attention pooling module. The backbone module uses U-Net, which is used for feature extraction and image segmentation. The auxiliary module is used to take the output feature map of each layer of the backbone encoder of the U-Net as input and fuse the shallow local detail information and the deep global information. The position attention pooling module is used to take the restored feature map of the backbone decoder of the U-Net and the enhanced semantic feature map generated by the auxiliary module as input, and assist the backbone module to capture spatial position dependencies and establish channel mapping associations.

[0025] Preferably, the auxiliary module performs 3×3 convolution on the first feature map F output by each layer of the backbone encoder of the U-Net y to obtain a second feature map Conv(F y ), where y represents the layer in the backbone encoder of the U-Net, and Conv() represents the convolution operation. And through bilinear interpolation, the upsampling of the second feature map Conv(F y ) is unified to the same resolution, so as to obtain the first feature map F y ' corresponding to the second feature map Conv(F y ), and F y ' = Bilinear(Conv(F y ))), where Bilinear() represents the bilinear interpolation function. Concatenate and fuse all the first feature maps F y ' into a tensor, and obtain a multi-scale feature map B with enhanced semantic information through convolution operation, and B = Conv([F1', F2', F3', F4']).

[0026] Preferably, the position attention pooling module performs spatial pyramid pooling on the restored feature map A output by the backbone decoder of the U-Net and the multi-scale feature map B output by the auxiliary module, so as to obtain a first feature matrix K1 of the restored feature map A and a second feature matrix of the multi-scale feature map B, expand the first feature matrix into a first feature vector K1, expand the second feature matrix into a second feature vector K2, convert the first feature vector K1 and the second feature vector K2 into a second feature map D and a third feature map G respectively through a fully connected layer, perform 1×1 convolution on the restored feature map A, so as to obtain a fourth feature map E, transpose the fourth feature map E and perform matrix correlation operation with the third feature map G, so as to obtain a position attention feature map F, and where F ij represents the influence of the i-th position in the third feature map G on the i-th position in the third feature map G, G i represents the i-th position in the third feature map G, E j represents the j-th position in the fourth feature map E, K represents the total number of feature vectors in the third feature map G and the fourth feature map E, perform matrix correlation operation on the second feature map D and the position attention feature map F, so as to obtain a third feature map, and fuse the restored feature map A and the third feature map, so as to obtain a fused feature map X, and where X j represents the j-th position in the fused feature map X, A j represents the j-th position in the restored feature map A, D j represents the j-th position in the second feature map D, T represents the total number of feature vectors in the second feature map D.

[0027] Preferably, matching the soil-rock profile image with the template image group includes the following steps:

[0028] B1. Extract the first SURF feature points of the soil-rock profile image through the SURF algorithm, and extract the second SURF feature points of the template image;

[0029] B2. Positively match the first SURF feature points with the second SURF feature points through the FLANN algorithm, so as to obtain a positive matching set, and inversely match the second SURF feature points with the first SURF feature points through the FLANN algorithm, so as to obtain an inverse matching set;

[0030] B3. Extract the symmetric intersection points of the forward matching set and the reverse matching set to obtain the symmetric intersection, which includes multiple pairs of matching points. Count the total number of pairs of matching points in the symmetric intersection, and compare the total number of pairs of matching points in the symmetric intersection with a preset threshold. If it is greater than the preset threshold, execute step B4; otherwise, it indicates that the soil-rock profile image does not match the template image group.

[0031] B4. Filter the symmetric intersection through the RANSAC algorithm to obtain a pure symmetric intersection. Calculate the similarity of multiple pairs of matching points in the pure symmetric intersection based on the Euclidean distance, and calculate the average similarity of the pure symmetric intersection based on the similarities of multiple pairs of matching points.

[0032] B5. Compare the average similarity with a preset similarity threshold. If it is greater than the preset similarity threshold, it indicates that the soil-rock profile image matches the template image group; otherwise, it indicates that the soil-rock profile image does not match the template image group.

[0033] Preferably, the expression for the median value of H in the HSV color space is as follows:

[0034]

[0035] where R, G, and B represent the median values of the red gray value, green gray value, and blue gray value in the RGB color space, and H represents the median value of H in the HSV color space.

[0036] The expression for the median value of S in the HSV color space is as follows:

[0037]

[0038] where S represents the median value of S in the HSV color space.

[0039] The expression for the median value of V in the HSV color space is as follows:

[0040]

[0041] where V represents the median value of V in the HSV color space.

[0042] The expression for the panchromatic gray value of the enhanced formation profile image is as follows:

[0043] DN = ω1R + ω2G + ω3B;

[0044] where DN represents the panchromatic gray value of the enhanced formation profile image, and ω1, ω2, and ω3 respectively represent preset weight parameters.

[0045] Preferably, the image enhancement method includes the following steps:

[0046] C1. Select a sliding window, traverse the formation profile image through the sliding window, calculate the fluctuation condition within the sliding window, and characterize the clarity of all pixels within the sliding window through the fluctuation condition. The expression of the fluctuation condition is as follows:

[0047]

[0048] where V ar (x,y) represents the degree of fluctuation of the pixel at coordinates (x,y) in the formation profile image, I(x,y) represents the gray value of the pixel at coordinates (x,y) in the formation profile image, u(x,y) represents the mathematical expectation of the pixel point, and

[0049] C2. Record the fluctuation conditions of all sliding windows within the formation profile image to obtain a fluctuation matrix, traverse each of the fluctuation conditions in the fluctuation matrix, and thus obtain the maximum fluctuation condition in the fluctuation matrix;

[0050] C3. Assign the pixel values of the sliding window corresponding to the maximum fluctuation condition to the pixel values of the entire formation profile image.

[0051] Compared with the prior art, the above technical solution of the present invention has the following beneficial technical effects:

[0052] 1. In the present invention, the image acquisition unit uses a digital camera to collect microscopic images of geological profiles, providing initial data for subsequent data processing and analysis, ensuring the acquisition of high-quality images. The image denoising unit is responsible for removing noise and interference in the images, ensuring that the image quality is clearer and more stable, thereby improving the accuracy of subsequent image processing and analysis and ensuring the reliability of data results. The image segmentation unit, based on a trained model, segments the denoised microscopic images to extract relevant information about soil and rock and the formation. Image segmentation can effectively convert complex image data into analyzable elements, providing a better analysis basis for the system. The image display unit displays the processed image information on a display terminal, providing a visual analysis interface for operators and experts, thereby making the interaction more intuitive and helping users understand the data and make decisions. The soil and rock detection unit is used to identify the segmented soil and rock profile images and match them with stored template images to determine the types of soil and rock, thereby automatically detecting the types of soil and rock, reducing manual participation, and improving the recognition efficiency.

[0053] 2. The image enhancement unit of the present invention is responsible for enhancing the formation profile image to ensure that the image is clearer, thereby improving the recognition accuracy. Image enhancement helps to better extract features and improve the quality of subsequent analysis. When the expert detection unit cannot automatically identify the soil and rock types, it transmits the image to an expert for manual identification. The intervention of the expert can ensure the accuracy and credibility of the system results. The feature extraction unit extracts specific features from the enhanced formation profile image, which can be used for further formation identification. The formation identification unit uses the trained support vector machine model to automatically identify the formation type, thereby improving the efficiency and accuracy of formation identification. The blockchain storage unit provides a secure and tamper-proof storage method for saving the soil and rock profile images and the identification results, and can update the template image group, thereby ensuring the credibility and security of the data. BRIEF DESCRIPTION OF THE DRAWINGS

[0054] Figure 1 It is a schematic flow chart of the overall system in an embodiment proposed by the present invention;

[0055] Figure 2 It is a schematic flow chart of the image denoising method in an embodiment proposed by the present invention;

[0056] Figure 3 It is a schematic flow chart of the matching between the soil and rock profile image and the template image group in an embodiment proposed by the present invention;

[0057] Figure 4 It is a schematic flow chart of the image enhancement method in an embodiment proposed by the present invention.

[0058] Reference numerals: 1, image acquisition unit; 2, image denoising unit; 3, image segmentation unit; 4, image display unit; 5, soil and rock detection unit; 6, image enhancement unit; 7, expert detection unit; 8, feature extraction unit; 9, formation identification unit; 10, blockchain storage unit. DETAILED DESCRIPTION OF THE EMBODIMENTS

[0059] Embodiment 1, as Figure 1 shown, a microscopic image data acquisition and analysis system for profile investigation proposed by the present invention includes a soil and rock detection unit 5, an image enhancement unit 6, an expert detection unit 7, a feature extraction unit 8, a formation identification unit 9, and a blockchain storage unit 10, and further includes:

[0060] An image acquisition unit 1, which is used to acquire an image of the target geological profile through a digital camera to obtain a microscopic image of the target geological profile, and transmit the microscopic image of the target geological profile to the image denoising unit 2;

[0061] The image denoising unit 2 receives the microscopic image of the target geological section transmitted by the image acquisition unit 1, and performs image denoising on the microscopic image of the target geological section through an image denoising method, so as to obtain a standard microscopic image of the target geological section, and transmits the standard microscopic image of the target geological section to the image segmentation unit 3;

[0062] The image segmentation unit 3 receives the standard microscopic image of the target geological section transmitted by the image denoising unit 2, and performs image segmentation on the standard microscopic image of the target geological section through a trained image segmentation model, so as to obtain a soil-rock section image and a stratum section image, and transmits the soil-rock section image and the stratum section image to the image display unit 4, transmits the soil-rock section image to the soil-rock detection unit 5, and transmits the stratum section image to the image enhancement unit 6;

[0063] The image display unit 4 receives the soil-rock section image and the stratum section image transmitted by the image segmentation unit 3, and displays the soil-rock section image and the stratum section image on the intelligent terminal.

[0064] In the present invention, the intelligent terminal generally refers to a mobile device with certain intelligent computing and communication capabilities, such as a smart phone, a tablet computer, a wearable device, etc. In the microscopic image data acquisition and analysis system for profile investigation, the role of the intelligent terminal is to serve as an interface for users to interact with the system. Users can control the system, view data, perform analysis and other operations through the intelligent terminal.

[0065] In an optional embodiment, the soil-rock detection unit 5 receives the soil-rock section image transmitted by the image segmentation unit 3 and the template image group transmitted by the blockchain storage unit 10, and matches the soil-rock section image with the template image group. If the soil-rock section image matches the template image in the template image group, it is determined that the soil-rock type in the soil-rock section image is the soil-rock type in the template image; otherwise, the soil-rock section image is transmitted to the expert detection unit 7.

[0066] It should be noted that the template image group is a set of pre-prepared images used to represent different types of soil-rock section images. These template images can include various different soil-rock types. Each template image represents a specific soil-rock type and has its characteristics and features.

[0067] In an optional embodiment, the image enhancement unit 6 receives the stratum section image transmitted by the image segmentation unit 3, and performs image enhancement on the stratum section image through an image enhancement method, so as to obtain an enhanced stratum section image, and transmits the enhanced stratum section image to the feature extraction unit 8.

[0068] In an alternative embodiment, the feature extraction unit 8 receives the enhanced formation profile image transmitted by the image enhancement unit 6, extracts the median values of the red grayscale value, green grayscale value, and blue grayscale value in the RGB color space of the enhanced formation profile image, converts the median values of the red grayscale value, green grayscale value, and blue grayscale value into the median values of H, S, and V in the HSV color space, extracts the panchromatic grayscale value of the enhanced formation profile image, and transmits the median values of H, S, and V and the panchromatic grayscale value to the formation identification unit 9.

[0069] In an alternative embodiment, the expert detection unit 7 receives the soil and rock profile image transmitted by the soil and rock detection unit 5, transmits the soil and rock profile image to the intelligent terminal of the target expert, and the target expert identifies the soil and rock type in the soil and rock profile image in the intelligent terminal, thereby obtaining the soil and rock type in the soil and rock profile image, and transmitting the soil and rock profile image and the corresponding soil and rock type to the blockchain storage unit 10.

[0070] It should be noted that the target expert uses the intelligent terminal to identify the soil and rock type in the soil and rock profile image and determine the soil and rock type in the soil and rock profile image. This process usually requires the expert to make a judgment based on his or her professional knowledge and experience.

[0071] In an alternative embodiment, the blockchain storage unit 10 receives the soil and rock profile image transmitted by the expert detection unit 7 and the corresponding soil and rock type. The blockchain stores a template image group, which includes multiple template images and the corresponding soil and rock types, stores the soil and rock profile image and the corresponding soil and rock type in the template image group, thereby updating the template image group, and transmitting the updated template image group to the soil and rock detection unit 5.

[0072] It should be noted that blockchain storage is a way of using blockchain technology for data storage and management. Blockchain is a decentralized and distributed ledger technology that allows data to be stored on multiple nodes, and these nodes jointly participate in the recording, storage, and verification of data.

[0073] In an alternative embodiment, the formation identification unit 9 receives the median values of H, S, and V and the panchromatic grayscale value transmitted by the feature extraction unit 8, inputs the median values of H, S, and V and the panchromatic grayscale value into the trained support vector machine-based soil classification model, and outputs the soil layer type corresponding to the median values of H, S, and V and the panchromatic grayscale value through the trained support vector machine-based soil classification model, thereby obtaining the soil layer type corresponding to the formation profile.

[0074] It should be noted that the soil classification model based on support vector machine is a machine learning algorithm used for classifying soil types. Support vector machine is a supervised learning algorithm, and its basic principle is to find an optimal hyperplane in the feature space to separate data of different categories, maximizing the interval between data points of different categories and the hyperplane.

[0075] Example 2, as Figures 2 - 4 shown, a microscopic image data acquisition and analysis system for profile survey proposed by the present invention, compared with Example 1, this embodiment further includes:

[0076] An image denoising method, including the following steps:

[0077] A1. Respectively extract the luminance components of the L channel of the target geological profile microscopic image and the reference image, and calculate the histogram of the luminance components of the L channel. The calculation formula of the histogram of the luminance components of the L channel is as follows:

[0078]

[0079] where H LK (μ) represents the percentage of the value of the luminance histogram at the luminance value of μ in the total number of pixels, w and h represent the width and height of the image, and L R (j, i) represents the luminance value at the position (j, i) in the reference image;

[0080] A2. Based on the histogram of the luminance components of the L channel, calculate the cumulative histogram corresponding to the histogram of the luminance components of the L channel. The calculation formula of the cumulative histogram is as follows:

[0081]

[0082] where represents the value of the luminance histogram at the luminance value of μ, and H LK (i) represents the percentage of the value of the luminance histogram at the luminance value of i in the total number of pixels;

[0083] A3. Compare the luminance values of each pixel in the reference image and the target geological profile microscopic image to obtain the corresponding luminance transformation relationship between the two. Establish a mapping function between the luminance value and the cumulative histogram through the corresponding luminance transformation relationship. Adjust the target geological profile microscopic image at each same luminance value based on the mapping function between the luminance value and the cumulative histogram, so that the target geological profile microscopic image has the same cumulative histogram as the reference image.

[0084] In an optional embodiment, the image segmentation model includes a backbone module, an auxiliary module, and a position attention pooling module. The backbone module adopts U-Net, which is used for feature extraction and image segmentation. The auxiliary module is used to take the output feature map of each layer of the backbone encoder of U-Net as input, and fuse the shallow local detail information and the deep global information. The position attention pooling module is used to take the restored feature map of the backbone decoder of U-Net and the enhanced semantic feature map generated by the auxiliary module as input, and assist the backbone module to capture spatial position dependencies and establish channel mapping associations.

[0085] It should be noted that U-Net is a classic neural network architecture commonly used in image segmentation tasks. It consists of a series of convolutional layers and upsampling layers, and its overall structure is U-shaped. The structure of U-Net includes an encoder and a decoder. The encoder is used to extract features from the input image and gradually reduce the spatial resolution. It usually consists of convolutional layers and pooling layers, and gradually extracts the high-level semantic features of the image by stacking these layers. The decoder is used to upsample the feature map extracted by the encoder and gradually restore the spatial resolution. The decoder gradually enlarges the feature map through transposed convolution and performs feature fusion with the corresponding layer of the encoder, thereby gradually restoring the detail information. The skip connection is an important design in U-Net. It connects the feature map of each layer in the encoder with the feature map of the corresponding layer in the decoder. This can help the decoder better utilize the low-level and high-level features in the encoder for accurate pixel-level prediction.

[0086] In an optional embodiment, the auxiliary module performs 3×3 convolution on the first feature map F output by each layer of the backbone encoder of U-Net y to obtain a second feature map Conv(F y ), where y represents the level in the backbone encoder of U-Net, Conv() represents the convolution operation, and the upsampling of the second feature map Conv(F y ) is unified to the same resolution through bilinear interpolation, so as to obtain the first feature map F y ' corresponding to the second feature map Conv(F y ), and F y ' = Bilinear(Conv(F y ))), where Bilinear() represents the bilinear interpolation function. All the first feature maps F y ' are concatenated and fused into a tensor, and a multi-scale feature map B with enhanced semantic information is obtained through convolution operation, and B = Conv([F1', F2', F3', F4']).

[0087] In an alternative embodiment, the position attention pooling module performs spatial pyramid pooling on the restored feature map A output by the backbone decoder of the U-Net and the multi-scale feature map B output by the auxiliary module, thereby obtaining a first feature matrix K1 of the restored feature map A and a second feature matrix of the multi-scale feature map B, expanding the first feature matrix into a first feature vector K1, expanding the second feature matrix into a second feature vector K2, converting the first feature vector K1 and the second feature vector K2 into a second feature map D and a third feature map G respectively through a fully connected layer, performing 1×1 convolution on the restored feature map A, thereby obtaining a fourth feature map E, transposing the fourth feature map E and performing matrix correlation operation with the third feature map G, thereby obtaining a position attention feature map F, and where F ij represents the influence of the i-th position in the third feature map G on the i-th position in the third feature map G, G i represents the i-th position in the third feature map G, E j represents the j-th position in the fourth feature map E, K represents the total number of feature vectors in the third feature map G and the fourth feature map E, performing matrix correlation operation on the second feature map D and the position attention feature map F, thereby obtaining a third feature map, and fusing the restored feature map A with the third feature map, thereby obtaining a fused feature map X, and where X j represents the j-th position in the fused feature map X, A j represents the j-th position in the restored feature map A, D j represents the j-th position in the second feature map D, T represents the total number of feature vectors in the second feature map D.

[0088] In an alternative embodiment, matching the soil-rock profile image with a template image group includes the following steps:

[0089] B1. Extracting first SURF feature points of the soil-rock profile image through the SURF algorithm, and extracting second SURF feature points of the template image;

[0090] B2. Forwardly matching the first SURF feature points with the second SURF feature points through the FLANN algorithm, thereby obtaining a forward matching set, and reversely matching the second SURF feature points with the first SURF feature points through the FLANN algorithm, thereby obtaining a reverse matching set;

[0091] B3. Extract the symmetric intersection points of the forward matching set and the reverse matching set to obtain the symmetric intersection, which includes multiple pairs of matching points. Count the total number of pairs of matching points in the symmetric intersection and compare it with a preset threshold. If it is greater than the preset threshold, execute step B4; otherwise, it indicates that the soil-rock profile image does not match the template image group.

[0092] B4. Filter the symmetric intersection through the RANSAC algorithm to obtain a pure symmetric intersection. Calculate the similarity of multiple pairs of matching points in the pure symmetric intersection based on the Euclidean distance, and calculate the average similarity of the pure symmetric intersection based on the similarities of multiple pairs of matching points.

[0093] B5. Compare the average similarity with a preset similarity threshold. If it is greater than the preset similarity threshold, it indicates that the soil-rock profile image matches the template image group; otherwise, it indicates that the soil-rock profile image does not match the template image group.

[0094] It should be noted that the SURF algorithm is a computer vision algorithm for image feature detection and description; the RANSAC algorithm is an iterative method for estimating the parameters of a mathematical model, especially suitable for data sets with a large number of outliers; FLANN can quickly find the point in the data set that is closest to the query point given a data set and a query point; the symmetric intersection refers to finding those pairs of feature points that match each other in both forward and reverse matches in the forward matching and reverse matching. This means that if the feature point A of the soil-rock profile image corresponds to the feature point B of the template image in the forward match, and at the same time the feature point B of the template image also corresponds to the feature point A of the soil-rock profile image in the reverse match, then this pair of matching points is part of the symmetric intersection.

[0095] In an optional embodiment, the expression for the median value of H in the HSV color space is as follows:

[0096]

[0097] where R, G, and B represent the median values of the red grayscale value, green grayscale value, and blue grayscale value in the RGB color space, and H represents the median value of H in the HSV color space.

[0098] The expression for the median value of S in the HSV color space is as follows:

[0099]

[0100] where S represents the median value of S in the HSV color space.

[0101] The expression for the median value of V in the HSV color space is as follows:

[0102]

[0103] Among them, V represents the V median value in the HSV color space;

[0104] The expression of the panchromatic gray value for enhancing the formation profile image is as follows:

[0105] DN = ω1R + ω2G + ω3B;

[0106] Among them, DN represents the panchromatic gray value of the enhanced formation profile image, and ω1, ω2, and ω3 respectively represent preset weight parameters.

[0107] In an alternative embodiment, an image enhancement method includes the following steps:

[0108] C1. Select a sliding window, traverse the formation profile image through the sliding window, and calculate the fluctuation situation within the sliding window, and characterize the clarity of all pixels within the sliding window through the fluctuation situation. The expression of the fluctuation situation is as follows:

[0109]

[0110] Among them, V ar (x, y) represents the degree of fluctuation of the pixel with coordinates (x, y) in the formation profile image, I(x, y) represents the gray value of the pixel with coordinates (x, y) in the formation profile image, u(x, y) represents the mathematical expectation of the pixel point, and

[0111]

[0112] C2. Record the fluctuation situations of all sliding windows within the formation profile image to obtain a fluctuation matrix, and traverse each fluctuation situation in the fluctuation matrix to obtain the maximum fluctuation situation in the fluctuation matrix;

[0113] C3. Assign the pixel values of the sliding window corresponding to the maximum fluctuation situation to the pixel values of the entire formation profile image.

[0114] The embodiments of the present invention have been described in detail above in conjunction with the accompanying drawings. However, the present invention is not limited thereto. Various changes can be made without departing from the spirit of the present invention within the scope of knowledge possessed by those skilled in the art to which the present invention pertains.

Claims

1. A microscopic image data acquisition and analysis system for profile survey, comprising an image acquisition unit (1), an image denoising unit (2), an image segmentation unit (3), an image display unit (4), a soil and rock detection unit (5) and an image enhancement unit (6), characterized in that, The image acquisition unit (1) is used to acquire an image of a target geological profile through a digital camera, so as to obtain a microscopic image of the target geological profile, and transmit the microscopic image of the target geological profile to the image denoising unit (2); The image denoising unit (2) receives the microscopic image of the target geological profile transmitted by the image acquisition unit (1), and performs image denoising on the microscopic image of the target geological profile by an image denoising method, so as to obtain a standard microscopic image of the target geological profile, and transmit the standard microscopic image of the target geological profile to the image segmentation unit (3); The image segmentation unit (3) receives the standard microscopic image of the target geological profile transmitted by the image denoising unit (2), and performs image segmentation on the standard microscopic image of the target geological profile by a trained image segmentation model, so as to obtain a soil and rock profile image and a stratigraphic profile image, and transmit the soil and rock profile image and the stratigraphic profile image to the image display unit (4), transmit the soil and rock profile image to the soil and rock detection unit (5), and transmit the stratigraphic profile image to the image enhancement unit (6); The image display unit (4) receives the soil and rock profile image and the stratigraphic profile image transmitted by the image segmentation unit (3), and displays the soil and rock profile image and the stratigraphic profile image on a display terminal.

2. The microscopic image data acquisition and analysis system for profile investigation according to claim 1, characterized in that The soil and rock detection unit (5) receives the soil and rock profile image transmitted by the image segmentation unit (3) and the template image group transmitted by the blockchain storage unit (10), and matches the soil and rock profile image with the template image group. If the soil and rock profile image matches the template image in the template image group, it is determined that the soil and rock type in the soil and rock profile image is the soil and rock type in the template image; otherwise, the soil and rock profile image is transmitted to the expert detection unit (7); The image enhancement unit (6) receives the stratigraphic profile image transmitted by the image segmentation unit (3), and performs image enhancement on the stratigraphic profile image by an image enhancement method, so as to obtain an enhanced stratigraphic profile image, and transmit the enhanced stratigraphic profile image to the feature extraction unit (8).

3. The microscopic image data acquisition and analysis system for profile survey according to claim 2, characterized in that, The feature extraction unit (8) receives the enhanced stratigraphic profile image transmitted by the image enhancement unit (6), extracts the median values of the red grayscale value, the green grayscale value and the blue grayscale value in the RGB color space of the enhanced stratigraphic profile image, converts the median values of the red grayscale value, the green grayscale value and the blue grayscale value into the median values of H, S and V in the HSV color space, and extracts the panchromatic grayscale value of the enhanced stratigraphic profile image, and transmits the median values of H, S and V and the panchromatic grayscale value to the stratigraphic identification unit (9); The expert detection unit (7) receives the soil-rock profile image transmitted by the soil-rock detection unit (5), and transmits the soil-rock profile image to the intelligent terminal of the target expert. The target expert identifies the soil-rock type in the soil-rock profile image in the intelligent terminal, so as to obtain the soil-rock type in the soil-rock profile image, and transmits the soil-rock profile image and the corresponding soil-rock type to the blockchain storage unit (10).

4. A microscopic image data acquisition and analysis system for profile survey according to claim 3, characterized in that, The blockchain storage unit (10) receives the soil-rock profile image transmitted by the expert detection unit (7) and the corresponding soil-rock type. The template image group is stored in the blockchain. The template image group includes a plurality of template images and the corresponding soil-rock types, and stores the soil-rock profile image and the corresponding soil-rock type into the template image group, so as to update the template image group, and transmit the updated template image group to the soil-rock detection unit (5); The formation identification unit (9) receives the median values of H, S, and V and the panchromatic gray value transmitted by the feature extraction unit (8), and inputs the median values of H, S, and V and the panchromatic gray value into the trained soil classification model based on the support vector machine. The trained soil classification model based on the support vector machine outputs the soil layer type corresponding to the median values of H, S, and V and the panchromatic gray value, so as to obtain the soil layer type corresponding to the formation profile.

5. The microscopic image data acquisition and analysis system for profile survey according to claim 1, characterized in that, The image denoising method includes the following steps: A1. Respectively extract the luminance components of the L channel of the target geological profile microscopic image and the reference image, and calculate the histogram of the luminance components of the L channel. The calculation formula of the histogram of the luminance components of the L channel is as follows: Among them, H LK (μ) represents the percentage of the value of the luminance histogram at the luminance value μ in the total number of pixels, w and h represent the width and height of the image, and L R (j, i) represents the luminance value at the position (j, i) in the reference image; A2. Based on the histogram of the luminance components of the L channel, calculate the cumulative histogram corresponding to the histogram of the luminance components of the L channel. The calculation formula of the cumulative histogram is as follows: Among them, represents the value of the luminance histogram at the luminance value of μ, H LK (i) represents the percentage of the total number of pixels occupied by the value of the luminance histogram at the luminance value of i; A3. Compare the luminance values of each pixel in the reference image and the target geological profile microscopic image, so as to obtain the corresponding luminance transformation relationship between the two. Establish a mapping function between the luminance value and the cumulative histogram through the corresponding luminance transformation relationship. Adjust the target geological profile microscopic image at each same luminance value based on the mapping function between the luminance value and the cumulative histogram, so that the target geological profile microscopic image has the same cumulative histogram as the reference image.

6. The microscopic image data acquisition and analysis system for profile survey according to claim 1, wherein, The image segmentation model includes a backbone module, an auxiliary module, and a position attention pooling module. The backbone module adopts U-Net, and U-Net is used for feature extraction and image segmentation. The auxiliary module is used to take the output feature map of each layer of the backbone encoder of U-Net as input, and fuse the shallow local detail information and the deep global information. The position attention pooling module is used to take the restored feature map of the backbone decoder of U-Net and the enhanced semantic feature map generated by the auxiliary module as input, and assist the backbone module to capture the spatial position dependence relationship and establish the channel mapping association; The auxiliary module performs a 3×3 convolution on the first feature map F output by each layer of the backbone encoder of the U-Net y to obtain a second feature map Conv(F y ), where y represents the level in the backbone encoder of the U-Net, Conv() represents the convolution operation, and the upsampling of the second feature map Conv(F y ) is unified to the same resolution through bilinear interpolation, so as to obtain the first feature map F y ' corresponding to the second feature map Conv(F y '), and F y ' = Bilinear(Conv(F y )), where Bilinear() represents the bilinear interpolation function. All the first feature maps F y ' are concatenated and fused into a tensor, and a multi-scale feature map B with enhanced semantic information is obtained through a convolution operation, and B = Conv([F1', F2', F3', F4']).

7. A microscopic image data acquisition and analysis system for profile survey according to claim 6, characterized in that, The position attention pooling module performs spatial pyramid pooling on the restored feature map A output by the backbone decoder of the U-Net and the multi-scale feature map B output by the auxiliary module, so as to obtain the first feature matrix K1 of the restored feature map A and the second feature matrix of the multi-scale feature map B, expand the first feature matrix into the first feature vector K1, expand the second feature matrix into the second feature vector K2, convert the first feature vector K1 and the second feature vector K2 into the second feature map D and the third feature map G respectively through a fully connected layer, perform 1×1 convolution on the restored feature map A, so as to obtain the fourth feature map E, transpose the fourth feature map E and perform matrix correlation operation with the third feature map G, so as to obtain the position attention feature map F, and where F ij represents the influence of the i-th position in the third feature map G on the i-th position in the third feature map G, G i represents the i-th position in the third feature map G, E j represents the j-th position in the fourth feature map E, K represents the total number of feature vectors in the third feature map G and the fourth feature map E, perform matrix correlation operation on the second feature map D and the position attention feature map F, so as to obtain the third feature map, and fuse the restored feature map A and the third feature map, so as to obtain the fused feature map X, and where X j represents the j-th position in the fused feature map X, A j represents the j-th position in the restored feature map A, D j represents the j-th position in the second feature map D, and T represents the total number of feature vectors in the second feature map D.

8. The microscopic image data acquisition and analysis system for profile survey according to claim 2, characterized in that, Matching the soil-rock profile image with the template image group includes the following steps: B1. Extract the first SURF feature points of the soil-rock profile image through the SURF algorithm, and extract the second SURF feature points of the template image; B2. Forward-match the first SURF feature points with the second SURF feature points through the FLANN algorithm to obtain a forward matching set, and backward-match the second SURF feature points with the first SURF feature points through the FLANN algorithm to obtain a backward matching set; B3. Extract the symmetric intersection points of the forward matching set and the backward matching set to obtain a symmetric intersection. The symmetric intersection includes multiple matching point pairs. Count the total number of the matching point pairs in the symmetric intersection, and compare the total number of the matching point pairs in the symmetric intersection with a preset threshold. If it is greater than the preset threshold, execute step B4; otherwise, it indicates that the soil-rock profile image does not match the template image group; B4. Filter the symmetric intersection through the RANSAC algorithm to obtain a pure symmetric intersection. Calculate the similarity of multiple matching point pairs in the pure symmetric intersection based on the Euclidean distance, and calculate the average similarity of the pure symmetric intersection based on the similarity of multiple matching point pairs; B5. Compare the average similarity with a preset similarity threshold. If it is greater than the preset similarity threshold, it indicates that the soil-rock profile image matches the template image group; otherwise, it indicates that the soil-rock profile image does not match the template image group.

9. A microscopic image data acquisition and analysis system for profile survey according to claim 3, characterized in that, The expression of the H median value in the HSV color space is as follows: wherein, R, G and B represent the median values of the red grayscale value, the green grayscale value and the blue grayscale value in the RGB color space, and H represents the H median value in the HSV color space; The expression of the S median value in the HSV color space is as follows: wherein, S represents the S median value in the HSV color space; The expression of the V median value in the HSV color space is as follows: wherein, V represents the V median value in the HSV color space; The expression of the panchromatic grayscale value of the enhanced formation profile image is as follows: DN = ω1R + ω2G + ω3B; wherein, DN represents the panchromatic grayscale value of the enhanced formation profile image, and ω1, ω2 and ω3 respectively represent preset weight parameters.

10. A microscopic image data acquisition and analysis system for profile survey according to claim 2, characterized in that, The image enhancement method includes the following steps: C1. Select a sliding window, traverse the formation profile image through the sliding window, and calculate the fluctuation situation within the sliding window. Characterize the clarity of all pixels within the sliding window through the fluctuation situation. The expression of the fluctuation situation is as follows: Among them, V ar (x, y) represents the fluctuation degree of the pixel with the coordinate (x, y) in the formation profile image, I(x, y) represents the gray value of the pixel with the coordinate (x, y) in the formation profile image, u(x, y) represents the mathematical expectation of the pixel point, and C2. Record the fluctuation situations of all sliding windows within the formation profile image to obtain a fluctuation matrix. Traverse each of the fluctuation situations in the fluctuation matrix to obtain the maximum fluctuation situation in the fluctuation matrix; C3. Assign the pixel values of the sliding window corresponding to the maximum fluctuation situation to the pixel values of the entire formation profile image.

Images