A Deep Learning-Based Automatic Identification and Detection System for Geological Structure Fractures

Abstract

The present invention discloses an automatic recognition and detection system for geological structure fissures based on deep learning, which relates to the technical field of automatic recognition and detection of geological structure fissures, and includes a fissure image analysis module, a distortion intelligent evaluation module, a morphology adaptive correction module, a correction efficiency evaluation module, and an identification data management module; the distortion intelligent evaluation module, which obtains the imaging parameter information of each area in the fissure image in real time, judges the morphological distortion degree of each area in the fissure image under the condition of perspective distortion, and divides each area in the fissure image into a low-distortion area, a medium-distortion area, and a high-distortion area according to the judgment result; the morphology adaptive correction module, which constructs a fissure morphology correction mechanism according to the division result of each area in the fissure image. The present invention solves the problem of fissure morphology distortion caused by the change of shooting angle, realizes the accurate recognition and automatic correction of fissures, and improves the detection accuracy and stability.

Description

Technical Field

[0001] The present invention relates to the technical field of automatic identification and detection of geological structure fissures, and particularly relates to an automatic identification and detection system for geological structure fissures based on deep learning. Background Art

[0002] Geological structure fissures refer to cracks or fractures in the earth's crust or rock formations caused by tectonic forces or other natural factors, and are one of the important rock deformation characteristics in geology. Characteristics such as the morphology, distribution, and width of fissures are of great significance for studying aspects such as groundwater flow, seismic activity, and mineral resource distribution. During geological exploration and engineering construction, accurately identifying and detecting these fissures is the basis for ensuring project safety, formulating reasonable mining plans, and conducting disaster prediction. However, traditional fissure identification methods usually rely on manual observation or simple image processing techniques, which are not only inefficient and time-consuming, but also easily affected by human factors, resulting in poor accuracy. Therefore, automatic identification and detection of geological structure fissures based on deep learning has become an effective solution. Deep learning algorithms, especially convolutional neural networks (CNNs), can self-learn through a large amount of geological image data and extract complex fissure features, avoiding the manual intervention and limitations of traditional methods. In addition, deep learning models have significant advantages in large-scale data processing, complex pattern recognition, and adaptive learning, and can be automatically optimized under different geological scenarios, improving the accuracy and real-time performance of fissure detection, further enhancing the intelligence level of the overall detection system, and meeting the requirements of modern geological exploration for efficient and precise detection technologies.

[0003] Existing deep learning-based automatic identification and detection technologies for geological structure fissures usually adopt deep learning algorithms such as convolutional neural networks (CNNs), and automatically extract the characteristic information of fissures by training a large amount of geological image data. Specifically, this technology first obtains a large amount of geological images or three-dimensional data through remote sensing technology, lidar or geological exploration equipment, and these data may include information such as different rock surfaces, fissure distributions, and geological layers. Then, the deep learning model is used to process these raw data. First, the convolutional layer automatically extracts low-level features in the image, such as edges, textures, and shapes, and then gradually abstracts through deeper network structures to extract high-level features related to fissures, such as the morphology, width, length, and distribution pattern of fissures. The deep learning model is trained with a large amount of labeled data to continuously optimize its parameters, enabling the system to accurately identify fissure characteristics under different geological conditions and adaptively process changes in different scenarios. The trained model can automatically process new geological images and output information such as the location, quantity, and type of fissures in real time, thus realizing efficient and accurate automatic identification and detection of fissures. In addition, in order to handle different types of geological data, existing technologies may also combine image enhancement technologies, data preprocessing methods, and multi-modal data fusion technologies to further improve the detection accuracy and system versatility, making the application effect of this technology in complex geological environments more excellent.

[0004] The existing technology has the following deficiencies:

[0005] During the geological exploration process, handheld cameras or drones and other shooting devices are usually used to obtain fissure images on the rock surface from different angles. In the case of large changes in the shooting angle, the morphology of the fissures will be distorted due to the perspective effect. Since the shooting angle of the shooting device is not always perpendicular to the rock surface, the width, length, and orientation of the fissures in the image are stretched or compressed, resulting in the distortion of the true geometric characteristics of the fissures in the image. The existing deep learning-based automatic identification and detection technologies for geological structure fissures mainly rely on two-dimensional image feature extraction and lack the ability to adapt to the morphological distortion caused by the change in the perspective. They cannot accurately restore the true morphology of the fissures, resulting in large errors in the identification of the size and direction of the fissures at different shooting angles, which in turn affects the correct assessment of the fissure expansion trend. This kind of error will lead to misjudgments of the true scale and direction of the fissures during the geological disaster assessment or engineering construction planning process, which may cause an underestimation of the fissure scale, thus ignoring potential geological risks and leading to safety hazards; on the contrary, if the fissures are overestimated, it may lead to unnecessary engineering reinforcement, increase construction costs, and affect the economy and feasibility of the project.

[0006] The above information disclosed in the background art section is only used to enhance the understanding of the background of the present disclosure, so it may include information that does not constitute the prior art known to those of ordinary skill in the art. Summary of the Invention

[0007] The objective of the present invention is to provide an automatic recognition and detection system for geological structure fissures based on deep learning to solve the problems in the above-mentioned background technology.

[0008] To achieve the above objective, the present invention provides the following technical solution: An automatic recognition and detection system for geological structure fissures based on deep learning, comprising a fissure image analysis module, a distortion intelligent evaluation module, a morphology adaptive correction module, a correction efficiency evaluation module, and an identification data management module;

[0009] The fissure image analysis module, during geological exploration, scans the rock surface through a photographing device to obtain the image information of fissures in real time, and detects whether the fissures are distorted due to perspective effect in the obtained fissure image information. When it is detected that the fissure morphology is distorted due to perspective effect, the fissure image is evenly divided into several regions;

[0010] The distortion intelligent evaluation module obtains the imaging parameter information of each region in the fissure image in real time, analyzes it after obtaining, judges the degree of morphological distortion of each region in the fissure image under the condition of perspective distortion, and divides each region in the fissure image into a low-distortion region, a medium-distortion region, and a high-distortion region according to the judgment result;

[0011] The morphology adaptive correction module constructs a fissure morphology correction mechanism according to the division result of each region in the fissure image, and performs corresponding correction measures on the low-distortion region, the medium-distortion region, and the high-distortion region respectively;

[0012] The correction efficiency evaluation module obtains the correction error information of each region in the fissure image in real time during the process of the fissure morphology correction mechanism correcting each region in the fissure image, analyzes it after obtaining, evaluates whether the correction effect of the fissure morphology correction mechanism on each region in the fissure image can reach the expectation, and optimizes the fissure morphology correction mechanism according to the evaluation result;

[0013] The identification data management module completes the automatic recognition and detection of fissures according to the optimized fissure morphology correction mechanism, and records and stores the results of fissure recognition and detection and the application process of the correction mechanism for subsequent monitoring, analysis, and optimization of fissure morphology changes.

[0014] Preferably, in the distortion intelligent evaluation module, after obtaining the imaging parameter information of each region in the fissure image in real time, preprocess it; extract the light intensity feature information and contour structure information in the imaging parameter information of each region in the preprocessed fissure image, and analyze them after extraction to generate the fissure perspective distortion coefficient and fissure contour distortion index of each region respectively; construct a morphological distortion model for the generated fissure perspective distortion coefficient and fissure contour distortion index of each region, generate the morphological distortion coefficient of each region, and analyze it after generation to judge the morphological distortion degree of each region in the fissure image under the condition of perspective distortion, and divide each region in the fissure image into a low distortion region, a medium distortion region and a high distortion region according to the judgment result.

[0015] Preferably, the acquisition logic of the fissure perspective distortion coefficient and fissure contour distortion index of each region is as follows:

[0016] Extract the light intensity feature information in the imaging parameter information of each region in the preprocessed fissure image, specifically including the pixel brightness value, the offset of the centroid coordinate, and the average gradient intensity of the edge pixels of each region in the fissure image at different times within a period of time, and calibrate them respectively as 、 and , represents the pixel brightness value of the th region in the fissure image at the th moment within a period of time, represents the offset of the centroid coordinate of the th region in the fissure image at the th moment within a period of time, represents the average gradient intensity of the edge pixels of the th region in the fissure image at the th moment within a period of time, , , and are all positive integers;

[0017] Calculate the fissure perspective distortion coefficient of each region. The specific calculation formula is as follows:

[0018] ;

[0019] In the formula, is the fissure perspective distortion coefficient of the th region;

[0020] Extract the contour structure information from the imaging parameter information of each region in the pre-processed fracture image, specifically including the total length of the fracture contours at different moments within a period of time for each region in the fracture image, the difference between the maximum and minimum values of the internal pixels, and the amount of change in edge curvature, and label them respectively as , and , represents the total length of the fracture contour at the -th moment within a period of time for the -th region in the fracture image, represents the difference between the maximum and minimum values of the internal pixels at the -th moment within a period of time for the -th region in the fracture image, represents the amount of change in edge curvature at the -th moment within a period of time for the -th region in the fracture image;

[0021] Calculate the fracture contour distortion index for each region. The specific calculation formula is as follows:

[0022] ;

[0023] In the formula, is the fracture contour distortion index for the -th region.

[0024] Preferably, construct a morphological distortion model for the fracture perspective distortion coefficient and the fracture contour distortion index of each generated region, and generate the morphological distortion coefficient for each region through weighted summation. The specific calculation formula is: , where and are the non-zero weight coefficients of the fracture perspective distortion coefficient and the fracture contour distortion index of each region respectively, and ;

[0025] Determine the preset morphological distortion coefficient threshold interval , and after determination, compare it with the morphological distortion coefficient of each generated region. According to the comparison result, judge the morphological distortion degree of each region in the fracture image under the perspective distortion condition, and divide each region in the fracture image into a low distortion region, a medium distortion region, and a high distortion region according to the judgment result. The specific comparison analysis and division are as follows:

[0026] If , the morphological distortion degree of this area in the fissure image under perspective distortion conditions is at a low level, and this area is divided into a low-distortion area;

[0027] If , the morphological distortion degree of this area in the fissure image under perspective distortion conditions is at a medium level, and this area is divided into a medium-distortion area;

[0028] If , the morphological distortion degree of this area in the fissure image under perspective distortion conditions is at a high level, and this area is divided into a high-distortion area.

[0029] Preferably, in the morphological adaptive correction module, according to the division results of each area in the fissure image, a fissure morphology correction mechanism is constructed. Specifically: according to the division results of the low-distortion area, medium-distortion area, and high-distortion area, different image transformation, feature correction, and edge optimization parameters are set respectively to form a fissure morphology correction mechanism; based on the morphological distortion coefficients of each area and the performance of the image processing algorithm, through preset rules, the image transformation method, feature correction strategy, and adjustment range of edge optimization are automatically determined;

[0030] Corresponding correction measures are taken for the low-distortion area, medium-distortion area, and high-distortion area respectively. Specifically: within the low-distortion area, use the standard image transformation parameters in the fissure morphology correction mechanism to perform lightweight geometric correction to reduce perspective error; within the medium-distortion area, keep the standard feature correction parameters in the fissure morphology correction mechanism, and combine geometric adjustment and texture enhancement strategies to ensure the accuracy of the fissure boundary; within the high-distortion area, use the enhanced edge optimization parameters in the fissure morphology correction mechanism, and combine deep learning reconstruction algorithms and multi-scale image registration methods to enhance the recovery accuracy of the fissure morphology to reduce recognition deviations caused by severe distortion.

[0031] Preferably, in the correction efficiency evaluation module, during the process of the fissure morphology correction mechanism correcting each area in the fissure image, the correction error information of each area in the fissure image is obtained in real time and preprocessed after acquisition; the correction offset information and morphological consistency information in the preprocessed correction error information are extracted and analyzed after extraction, and the fissure correction offset coefficient and fissure correction consistency index of each area are generated respectively; a correction effect evaluation model is constructed for the generated fissure correction offset coefficient and fissure correction consistency index of each area, and the correction evaluation coefficient of each area is generated and analyzed after generation to evaluate whether the correction effect of the fissure morphology correction mechanism on each area in the fissure image can meet the expectations, and the fissure morphology correction mechanism is optimized according to the evaluation results.

[0032] Preferably, the acquisition logic of the crack correction offset coefficient and the crack correction consistency index for each region is as follows:

[0033] Extract the correction offset information from the preprocessed correction error information, specifically including the pixel density change rate, the boundary offset vector length, and the gray-scale gradient variance of each region in the crack images at different times during the correction process of the crack morphology correction mechanism, and label them respectively as , and , represents the pixel density change rate of the th region in the crack image at the th moment during the correction process of the crack morphology correction mechanism, represents the boundary offset vector length of the th region in the crack image at the th moment during the correction process of the crack morphology correction mechanism, represents the gray-scale gradient variance of the th region in the crack image at the th moment during the correction process of the crack morphology correction mechanism, , , and are all positive integers;

[0034] Calculate the crack correction offset coefficient for each region. The specific calculation formula is as follows:

[0035] ;

[0036] In the formula, is the crack correction offset coefficient of the th region;

[0037] Extract the morphological consistency information from the preprocessed correction error information, specifically including the edge pixel coincidence degree, the change rate of the edge gradient, and the change rate of the crack contour length of each region in the crack images at different times during the correction process of the crack morphology correction mechanism, and label them respectively as , and , represents the edge pixel coincidence degree of the th region in the crack image at the th moment during the correction process of the crack morphology correction mechanism, represents the edge pixel coincidence degree of the th region in the crack image at the The change rate of the edge gradient of a region, Indicates that during the calibration process of the fracture morphology calibration mechanism for a period of time At the moment, the change rate of the fracture contour length in the fracture image of the th region;

[0038] Calculate the fracture calibration consistency index of each region. The specific calculation formula is as follows:

[0039] ;

[0040] In the formula, Is the fracture calibration consistency index of the th region.

[0041] Preferably, for the fracture calibration offset coefficient And the fracture calibration consistency index Of each generated region, construct a calibration effect evaluation model, and generate the calibration evaluation coefficient Of each region through weighted summation. The specific calculation formula is: , where And Are the non-zero weight coefficients of the fracture calibration offset coefficient And the fracture calibration consistency index Of each region respectively, and ;

[0042] Determine the calibration evaluation coefficient threshold Of each region set in advance, and compare it with the generated calibration evaluation coefficient Of each region after determination. According to the comparison result, evaluate whether the calibration effect of the fracture morphology calibration mechanism on each region in the fracture image can meet the expectation, and optimize the fracture morphology calibration mechanism according to the evaluation result. The specific comparison and analysis are as follows:

[0043] If , the calibration effect of the fracture morphology calibration mechanism on this region in the fracture image can meet the expectation, and there is no need to optimize the fracture morphology calibration mechanism;

[0044] If , the correction effect of the crack morphology correction mechanism on this area in the crack image cannot meet the expectations, and it is necessary to optimize the crack morphology correction mechanism, specifically including: adjusting the image transformation parameters to optimize the perspective distortion correction in the morphology correction process and reduce the geometric distortion caused by the change of the shooting angle; improving the feature correction algorithm to improve the accuracy of the crack boundary by enhancing the feature point matching and texture analysis accuracy and reducing the boundary misalignment and detail loss; enhancing the edge optimization strategy by adopting multi-scale edge detection and adaptive smoothing techniques, so as to improve the accuracy and consistency of crack recognition and ensure that the correction effect meets the expected standard.

[0045] In the above technical solution, the technical effects and advantages provided by the present invention are as follows:

[0046] 1. By introducing modules such as crack image analysis, distortion evaluation, morphology correction, correction efficiency evaluation, and recognition data management, the present invention realizes the accurate recognition and detection of geological structure cracks under perspective distortion conditions. Compared with the traditional crack recognition method that relies on two-dimensional image features, the present invention can obtain the light intensity characteristics, contour structure, and correction error information of the crack area from multiple dimensions, and dynamically evaluate the degree of crack morphology distortion based on the combination of deep learning and mathematical modeling. By establishing the crack perspective distortion coefficient and the crack contour distortion index, the distortion degree of each area can be quantified to ensure that the true geometric shape of the crack can be accurately restored under different shooting perspectives, thus effectively avoiding the problem of morphological distortion caused by the perspective effect.

[0047] 2. The advantage of the present invention lies in the adoption of an adaptive crack morphology correction mechanism, which sets different image transformation, feature correction, and edge optimization parameters for low, medium, and high distortion areas respectively, automatically adjusts the correction strategy, and improves the accuracy and robustness of crack detection. By real-time obtaining the imaging parameter information of the crack image and using deep learning reconstruction and multi-scale image registration techniques, the blurring and misalignment of the crack boundary are effectively reduced. At the same time, the correction effect is continuously monitored by the intelligent evaluation module to ensure the stability and consistency of crack correction. In addition, based on the evaluation model of the correction offset coefficient and the correction consistency index, the quantitative evaluation of the correction quality is realized, automatically judging whether the correction meets the expectations, and optimizing according to the evaluation results, greatly reducing the need for human intervention and improving the intelligent level of the system.

[0048] 3. The present invention also features efficient data management and traceability. The identification data management module can record and store the crack detection results and the calibration application process, supporting long-term crack morphology monitoring and trend analysis. Through the accumulation of historical data, the system can adaptively adjust calibration parameters and optimize the identification algorithm to adapt to changes in different geological environments and crack morphologies, improving the accuracy of geological disaster prediction. Generally speaking, this technical solution can not only significantly improve the accuracy of crack identification, reduce the identification errors caused by perspective distortion, but also provide scientific and reliable data support for geological exploration and engineering construction, enhancing operation efficiency and safety, and having high engineering application value. BRIEF DESCRIPTION OF THE DRAWINGS

[0049] 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 embodiments. Obviously, the drawings described below are only some embodiments recorded in the present invention, and those of ordinary skill in the art can also obtain other drawings based on these drawings.

[0050] Figure 1 It is a schematic diagram of the modules of an automatic crack identification and detection system for geological structures based on deep learning according to the present invention. DETAILED DESCRIPTION OF THE EMBODIMENTS

[0051] Now, the exemplary embodiments will be described more comprehensively with reference to the drawings. However, the exemplary embodiments can be implemented in various forms and should not be construed as limited to the examples set forth herein; rather, these exemplary embodiments are provided so that the present disclosure will be more complete and comprehensive, and the concept of the exemplary embodiments will be fully conveyed to those skilled in the art.

[0052] The present invention provides an automatic crack identification and detection system for geological structures based on deep learning as Figure 1 shown, which includes a crack image analysis module, a distortion intelligent evaluation module, a morphology adaptive calibration module, a calibration efficiency evaluation module, and an identification data management module;

[0053] The crack image analysis module, during the geological exploration process, scans the rock surface through a photographing device to obtain real-time image information of the cracks, and detects whether the cracks are distorted due to the perspective effect in the obtained crack image information. When it is detected that the crack morphology is distorted due to the perspective effect, the crack image is evenly divided into several regions;

[0054] During the geological exploration process, the rock surface is scanned by a photographing device to obtain the image information of fractures in real time. Devices such as high-resolution RGB cameras, multispectral cameras, or lidar (LiDAR) can be used to achieve comprehensive coverage of the rock surface. These sensors are carried by drones or handheld devices and scanned row by row over the exploration area under a predefined trajectory or manual control to ensure the acquisition of fracture images from different angles. To achieve real-time data acquisition, the system software should integrate an image acquisition module, use the device SDK (such as the camera control API or the LiDAR data stream interface) to obtain images, and combine with an inertial measurement unit (IMU) and a GPS module to obtain the shooting position, attitude, and altitude information to ensure the geographical accuracy of the data. In addition, to reduce the acquisition deviation caused by device jitter or complex terrain, the system can apply real-time image stitching technology (such as image stitching algorithms based on SIFT or ORB feature points) during the image acquisition stage to generate a complete fracture surface image, and at the same time extract the features of the fracture area through Canny edge detection or deep learning object detection algorithms (such as YOLO or Faster R-CNN) to achieve high-precision fracture information acquisition.

[0055] After obtaining the fracture image information, it is necessary to detect whether the fracture is distorted due to the perspective effect, and in the case of detected distortion, divide the fracture image evenly into several regions. The distortion degree of the image can be evaluated by using the homography matrix calculation algorithm through obtaining the imaging parameters of the image (including shooting angle, focal length, flight altitude, etc.) and the geometric features of the fracture edge. The system first performs image registration, matches the feature points of the image through the RANSAC algorithm, obtains the deformation parameters of the fracture area, and combines with the perspective projection model to calculate the morphological distortion degree of the fracture. If the distortion degree exceeds the set threshold, a grid division or adaptive segmentation algorithm (such as K-means clustering or region growing algorithm) is used to divide the fracture image into multiple regions to ensure the accuracy of subsequent processing. The perspective distortion characteristics of each region are jointly determined by indicators such as edge curvature, width change rate, and surface texture consistency. The system generates a distortion evaluation coefficient based on these indicators and classifies the image regions accordingly to ensure the adoption of adaptive correction strategies under different distortion degrees.

[0056] The reason for this is that in the actual exploration environment, due to the complexity of the terrain, the perspective changes of the imaging equipment, and the irregularity of the rock surface, the fracture images often have significant perspective distortion, resulting in the stretching or compression of the geometric shape of the fractures in the images. If these distortions are not detected and corrected in a timely manner, the width, orientation, and distribution characteristics of the fractures cannot be accurately extracted, which will in turn affect the risk assessment of geological disasters and the rationality of engineering design. By performing real-time distortion detection and regional division in the early stage of image acquisition, potential problems can be identified in advance, and an accurate basis can be provided for subsequent fracture identification and correction. The purpose of regional division is to implement different correction measures under different degrees of distortion, so as to restore the true shape of the fractures to the greatest extent while ensuring the computational efficiency, ensure the accuracy and reliability of the final identification results, improve the adaptability and robustness of the detection system, and reduce misjudgments and omissions caused by distortion.

[0057] The distortion intelligent evaluation module obtains the imaging parameter information of each region in the fracture image in real time, analyzes it after acquisition, judges the degree of morphological distortion of each region in the fracture image under the condition of perspective distortion, and divides each region in the fracture image into a low-distortion region, a medium-distortion region, and a high-distortion region according to the judgment result;

[0058] In this embodiment, in the distortion intelligent evaluation module, after obtaining the imaging parameter information of each region in the fracture image in real time, preprocess it; extract the light intensity feature information and contour structure information in the imaging parameter information of each region in the preprocessed fracture image, analyze it after extraction, and generate the fracture perspective distortion coefficient and fracture contour distortion index of each region respectively; construct a morphological distortion model for the generated fracture perspective distortion coefficient and fracture contour distortion index of each region, generate the morphological distortion coefficient of each region, analyze it after generation, judge the degree of morphological distortion of each region in the fracture image under the condition of perspective distortion, and divide each region in the fracture image into a low-distortion region, a medium-distortion region, and a high-distortion region according to the judgment result.

[0059] Obtaining the imaging parameter information of each region in the fracture image in real time can be achieved by combining image segmentation and feature extraction. First, use common image segmentation techniques (such as K-means clustering, watershed algorithm, or semantic segmentation networks based on deep learning like U-Net) to divide the fracture image into multiple regions. Then, on the divided regions, use image processing tools (such as OpenCV, MATLAB, etc.) to extract the imaging parameter information of each region, such as brightness histogram, edge intensity, pixel gradient direction, etc. The specific implementation methods of these parameter information can extract edge features by calling the Sobel operator or Laplacian operator, obtain the pixel brightness distribution using histogram statistics, and use the optical flow method (such as the Lucas-Kanade optical flow algorithm) to track the movement trajectory of pixels to record the dynamic changes of each region in real time. In addition, to ensure the real-time nature of the acquired data, a multi-threaded processing method can be adopted and combined with GPU accelerated computing to update the imaging parameters in the region in real time while acquiring the fracture image.

[0060] The purpose of preprocessing is to improve the accuracy and stability of subsequent fracture recognition and analysis, and reduce the influence of factors such as noise, uneven illumination, and insufficient contrast on the recognition accuracy. When performing preprocessing, operations such as noise removal, image enhancement, contrast adjustment, and edge strengthening are usually required. Noise removal can use spatial domain methods such as Gaussian filtering and median filtering to remove the random noise generated by environmental interference in the fracture image; image enhancement can enhance the local contrast of the fracture region through adaptive histogram equalization (CLAHE) to make small fractures clearer; contrast adjustment can use Gamma transformation to optimize the image brightness to make the acquisition of imaging parameters more accurate; edge strengthening can use the Laplacian operator or non-maximum suppression (NMS) to highlight the fracture edge and improve the robustness of feature extraction. These preprocessing steps can be sequentially executed in the software through a pipeline processing method to ensure that all fracture image regions reach the best image quality standard before entering the next step of analysis.

[0061] The extraction of the light intensity feature information and the contour structure information in the imaging parameter information of each region in the preprocessed fissure image can be achieved by combining image feature analysis and computer vision algorithms. First, when extracting the light intensity feature information, pixel statistics and transformation methods can be used to extract indicators such as the brightness distribution, average brightness, and standard deviation of the region from the grayscale histogram of the image to reflect the illumination intensity and uniformity of each region. At the same time, the local binary pattern (LBP) method is used to extract the texture features of the region to capture the light intensity change pattern of the fissure region. In addition, Fourier transform can be used to analyze the frequency components of the image to evaluate the illumination change trend of the region and avoid high-frequency noise interference. Second, when extracting the contour structure information, the fissure boundary can be identified through edge detection operators (such as Canny, Sobel, or Prewitt operators), and morphological operations (such as erosion, dilation, opening, and closing operations) are used to optimize the edge integrity, and then indicators such as edge length, radius of curvature, and contour complexity are calculated. In addition, the Hough transform can be applied to detect the linear or curved features of the fissure, obtain the directionality and geometric morphology of the fissure, and at the same time use the minimum circumscribed rectangle method to extract parameters such as the aspect ratio and area of the region to reflect the morphological characteristics of the fissure. These extraction operations can be implemented through software tools such as OpenCV and MATLAB, and the parallel processing of each region in the image is carried out in a batch processing manner to ensure the real-time performance and accuracy of the extraction results.

[0062] In this embodiment, the acquisition logic of the fissure perspective distortion coefficient and the fissure contour distortion index of each region is as follows:

[0063] Extract the light intensity feature information in the imaging parameter information of each region in the preprocessed fissure image, specifically including the pixel brightness values, the offset of the centroid coordinates, and the average gradient intensity of the edge pixels of each region in the fissure image at different times within a period of time, and are respectively calibrated as 、 and , represents the pixel brightness value of the th region in the fissure image at the th moment within a period of time, represents the offset of the centroid coordinates of the th region in the fissure image at the th moment within a period of time, represents the average gradient intensity of the edge pixels of the th region in the fissure image at the th moment within a period of time, , , and are all positive integers;

[0064] The pixel brightness values, centroid coordinate offsets, and average gradient intensities of edge pixels in each region of the crack image at different moments within a period of time can be obtained in real time by combining image processing and computer vision techniques, and these operations can be automatically performed in a software environment (such as OpenCV, MATLAB, etc.). First, the pixel brightness values can be obtained by converting the crack image into a grayscale image or extracting single-channel information (such as the Y channel or V channel) in an RGB image, then using region division methods (such as contour detection based on connected regions or image segmentation based on the watershed algorithm) to determine the boundaries of each region, and using pixel statistical methods to calculate the average brightness, maximum brightness, and minimum brightness of each region to ensure that the changes in light intensity within the region can be recorded in real time at different time points. Secondly, the centroid coordinate offset can be obtained through morphological processing and region detection techniques, that is, using binaryzation, morphological opening operation, and contour detection in the image to extract the boundary of the crack region, and calculating the centroid coordinates of this region (such as calculating the geometric center position of the region through image moments), then tracking the centroid coordinates in consecutive time frames (such as using optical flow method or Kalman filter), and calculating the centroid offset at different moments to evaluate the position change of the crack under the change of viewing angle. Finally, the average gradient intensity of edge pixels can be obtained by edge detection algorithms (such as Sobel operator, Canny operator) to extract edge pixel points within each region, and calculating the gradient amplitude of these pixels, then statistically averaging the gradient values of each frame to evaluate the sharpness change of the crack boundary under distortion conditions. To ensure real-time performance, these acquisition processes can be accelerated by combining multi-threaded parallel processing with GPU, updating the feature data of each region in real time while collecting images, and using time series analysis methods (such as sliding window technology) to ensure the stability and continuity of the data. In this way, the imaging parameter information of each region in the crack image can be accurately obtained at different moments, providing reliable data support for subsequent distortion evaluation and correction.

[0065] Calculate the perspective distortion coefficients of the cracks in each region. The specific calculation formula is as follows:

[0066] ;

[0067] In the formula, is the perspective distortion coefficient of the crack in the th region;

[0068] The reason for designing this calculation formula is to fully consider the comprehensive influence of factors such as light intensity change, position offset, and edge gradient in each region of the crack image under perspective distortion conditions on the distortion degree. First, the exponential operation is used in the formula to process the average gradient intensity , because the edge gradient intensity can reflect the clarity and sharpness of the crack boundary, and the exponential operation can amplify the change of the edge intensity to highlight the greater impact that the high-gradient region may be subjected to under perspective distortion. Secondly, the logarithmic operation aims to balance the relationship between the pixel brightness value and the centroid coordinate offset, avoiding the impact of overly large or small values on the overall calculation. Among them, the logarithmic function has the characteristic of smooth growth, which can suppress the outliers brought by extreme illumination or offset. At the same time, adding "1" avoids the problem of the denominator being zero. The in the denominator represents the influence of the centroid offset. The larger the offset, the more significant the perspective distortion of the region, thereby reducing the influence weight of the light intensity on the distortion coefficient. Finally, by taking the average over the time dimension , the calculation result becomes more stable in time, which can eliminate the random fluctuations within a short period and ensure that the crack perspective distortion coefficient can more accurately reflect the distortion degree of the region over the entire time period. To sum up, the calculation method of this formula can comprehensively evaluate the perspective distortion degree of each region by exponentially amplifying the edge distortion characteristics, logarithmically balancing the weights of illumination and position offset, and combining the method of time averaging.

[0069] The size of the crack perspective distortion coefficient of the th region directly reflects the degree of morphological distortion of the region under perspective distortion conditions. The larger the value of the coefficient, the more serious the deformation impact on the region under the perspective effect. The width, length and geometric consistency of the edge features of the crack have been distorted to a large extent, resulting in an increase in the recognition error; on the contrary, if the value is small, it indicates that the perspective distortion degree of the region is low, and the crack morphology is closer to its true geometric features and can be recognized and evaluated more accurately. Specifically, the crack perspective distortion coefficient combines the light intensity feature information and the spatial displacement information. When the edge gradient intensity is high and the light intensity fluctuates greatly, and the centroid offset also increases significantly, the coefficient value will increase, indicating that significant morphological distortion has occurred in the region and further correction is required; while when the gradient, illumination and position offset remain relatively stable, the coefficient value is low, indicating that the region morphology is relatively stable and belongs to the low-distortion region.

[0070] Extract the contour structure information in the imaging parameter information of each region in the preprocessed crack image, specifically including the total length of the crack contour at different times within a period of time in each region of the crack image, the difference between the maximum and minimum values of the internal pixels, and the amount of change in the edge curvature, and calibrate them as , and , represents the The total length of the crack profile at a certain moment within a period of time for a region The difference between the maximum and minimum values of the internal pixels at a certain moment within a period of time for a region Denote the th region in the crack image within a period of time The difference between the maximum and minimum values of the internal pixels at a certain moment within a period of time Denote the th region in the crack image within a period of time The change amount of the edge curvature at a certain moment

[0071] The total length of the crack profile, the difference between the maximum and minimum values of the internal pixels, and the change amount of the edge curvature at different moments within a period of time for each region in the crack image can be obtained in real time through image processing and computer vision techniques and implemented in a software environment (such as OpenCV, MATLAB). First, the total length of the crack profile can be obtained by extracting the crack boundary through edge detection (such as the Canny operator), identifying the region contour using a contour detection algorithm (such as findContours), and combining it with a perimeter calculation function (such as arcLength) to obtain the real-time contour length. Second, the difference between the maximum and minimum values of the internal pixels can be obtained through grayscale histogram analysis. First, convert the crack region into a grayscale image, and then use pixel statistical methods to calculate the difference between the maximum and minimum pixel values to reflect the change in regional contrast. Finally, the change amount of the edge curvature can be obtained by combining edge detection with a curvature calculation method (such as second-order difference or polynomial fitting). Calculate the local curvature of the boundary pixels and analyze its change in consecutive frames. Use the optical flow method to track key points to monitor the dynamic change of the edge shape. These methods can be executed in real time through multi-threaded processing and GPU acceleration technology, providing accurate data for subsequent crack distortion analysis.

[0072] Calculate the crack profile distortion index for each region. The specific calculation formula is as follows:

[0073] ;

[0074] In the formula, is the crack profile distortion index for the th region.

[0075] This calculation formula is in this form to comprehensively evaluate the degree of change in the contour shape of each region in the crack image under perspective distortion conditions and ensure the scientificity and accuracy of the evaluation process. First, the purpose of the exponential operation is to amplify the influence of the total length of the crack profile on the distortion, and at the same time consider the difference between the maximum and minimum values of the internal pixels As a normalization factor, it prevents extreme values from affecting the calculation results. Adding "1" to the denominator can avoid division-by-zero errors and ensure the stability of the calculation. This operation can reflect the change in the length of the crack profile due to perspective distortion. At the same time, through the amplification effect of the exponential operation, it can more sensitively capture the subtle changes in the contour deformation. Secondly, the logarithmic operation is used to balance the change in edge curvature 's impact on the overall distortion degree. Since there may be large fluctuations in the local curvature of the crack edge, the logarithmic operation can smooth its impact, ensure the stable growth of the value, and avoid the distortion effect being ignored due to too small a value. In addition, adding "1" can prevent the problem that the logarithm cannot be calculated when it is zero, ensuring the robustness of the mathematical calculation. Finally, the way of taking the time average is to balance the distortion fluctuations at different time points, reduce the interference of abnormal data at a single moment, make the calculation results more globally representative, and ensure the crack profile distortion index can accurately reflect the distortion situation of the area over the entire time range. Therefore, this calculation method can comprehensively and accurately evaluate the morphological distortion degree of the crack area while ensuring the stability of data calculation, providing a reliable data basis for subsequent distortion correction.

[0076] The size of the crack profile distortion index of the th area directly reflects the morphological distortion degree of this area under perspective distortion conditions. The larger the value, the more significant the change in the total length of the crack profile at different time points. At the same time, the contrast difference of internal pixels is large, and the edge curvature changes violently, indicating that the contour shape of the crack has undergone large deformation and distortion due to perspective distortion, resulting in an increase in the irregularity of the crack edge and an increase in the recognition difficulty. Therefore, a high value of indicates that there is a relatively serious morphological distortion in this area, and corresponding correction measures need to be taken, such as morphological restoration or geometric correction; on the contrary, if the value is small, it means that the crack profile length is stable, the pixel contrast change is small, and the edge shape is relatively smooth, indicating that this area is less affected by perspective distortion and the morphology is basically stable, and the true geometric features of the crack can be recognized more accurately.

[0077] In this embodiment, for the crack perspective distortion coefficients and the crack profile distortion indices of each generated area, a morphological distortion model is constructed, and the morphological distortion coefficient of each area is generated through weighted summation. The specific calculation formula is: , where and are respectively the crack perspective distortion coefficients and the crack profile distortion indices of each area non-zero weight coefficients, and ;

[0078] The purpose of constructing the morphological distortion model by weighted summation is to comprehensively evaluate the overall impact of each region of the fracture image on perspective distortion and contour distortion, so as to ensure that the evaluation of morphological distortion is both comprehensive and accurate. The fracture perspective distortion coefficient mainly measures the geometric distortion caused by the change of shooting angle, such as morphological stretching or compression, while the fracture contour distortion index is used to evaluate morphological characteristics such as the curvature change and structural integrity of the fracture boundary. Therefore, it is necessary to reasonably allocate weights to the two to balance the influence of different distortion factors. By introducing non-zero weight coefficients and , and satisfying , the system can flexibly adjust the contribution ratio of the two according to different geological conditions. For example, when the perspective distortion is significant, increase the weight of , and when the morphological change of the fracture edge is large, increase the weight of . In the specific implementation process, first, through image processing techniques (such as edge detection, feature extraction, etc.), the and of each region are obtained in real time, and then a linear weighted operation is performed at the software level to finally generate the morphological distortion coefficient , and compare it with the preset threshold interval, and divide the fracture region into low-distortion, medium-distortion, and high-distortion regions according to the comparison result, providing a scientific basis for subsequent automatic correction.

[0079] Determine the preset threshold interval of the morphological distortion coefficient , and after determination, compare it with the morphological distortion coefficients of each generated region, judge the morphological distortion degree of each region in the fracture image under perspective distortion conditions according to the comparison result, and divide each region in the fracture image into low-distortion regions, medium-distortion regions, and high-distortion regions according to the judgment result. The specific comparison analysis and division are as follows:

[0080] If , the morphological distortion degree of this region in the fracture image under perspective distortion conditions is low, and this region is divided into a low-distortion region;

[0081] This situation means that the perspective distortion in this area has little impact on the fracture morphology. The width, length, and edge features of the fractures basically remain in their original state, without obvious deformation or distortion. At this time, the extraction and recognition accuracy of fracture features are relatively high, which can accurately reflect the actual geometric features of the fractures, thus providing reliable data support for subsequent geological analysis, disaster assessment, and engineering design, reducing the possibility of misjudgment, and not requiring additional correction processing, saving computational resources and time.

[0082] If , the degree of morphological distortion of this area in the fracture image under perspective distortion conditions is at a medium level, and this area is classified as a medium distortion area;

[0083] This situation means that the degree of morphological distortion of this area under perspective distortion conditions is at a medium level, and certain degrees of stretching, compression, or edge curvature changes begin to appear in the geometric features of the fracture morphology, which may have a certain impact on the fracture recognition accuracy. At this time, the system needs to perform moderate morphological correction on this area to reduce the influence of distortion on the fracture geometric information, thereby improving the reliability of recognition. If the correction is not carried out in a timely manner, it may lead to evaluation deviations in the subsequent fracture propagation trend, affecting the accuracy of geological disaster risk prediction.

[0084] If , the degree of morphological distortion of this area in the fracture image under perspective distortion conditions is at a high level, and this area is classified as a high distortion area.

[0085] This situation means that this area is severely affected by perspective distortion, and the morphological features of the fractures are severely distorted, which may lead to problems such as fracture contour breakage, splicing errors, and morphological misjudgment. Such areas require more complex geometric correction methods, such as multi-view image fusion, deep learning reconstruction, or model matching methods based on known fracture morphologies, to restore the true geometric features of the fractures. If this area is not effectively corrected, it may lead to serious engineering safety hazards, such as misestimation of fracture scale, affecting decision-making in key projects such as mining and tunnel construction.

[0086] The pre-set threshold interval of the morphological distortion coefficient can be determined through statistical analysis, machine learning modeling, and experimental calibration, etc., to ensure that the threshold can accurately reflect the degree of morphological distortion of the fracture image under different perspective distortion conditions. First, statistical analysis methods can be used to process a large amount of fracture image data, extract the fracture perspective distortion coefficients and fracture contour distortion indices of each area, and calculate their statistical indicators such as mean and variance, and divide the critical values of low, medium, and high distortion areas by setting quantiles or standard deviation ranges. Secondly, machine learning modeling can be adopted, using a large amount of labeled data to train classification models (such as support vector machine SVM, random forest, or neural network), and inputting the and Data, learning form distortion coefficient The distribution patterns under different distortion degrees and automatically generate the optimal threshold interval. In specific implementation, the classification boundary of the model can be optimized through cross-validation to ensure the stability and accuracy of the threshold interval. Finally, with the help of the experimental calibration method, on the reference image with known true fracture morphology, by manually selecting low, medium, and high distortion regions and calculating the corresponding values, an empirical threshold library is formed, and a reasonable threshold interval is determined by the method of similarity matching in the new image. These methods can be implemented in software environments (such as Python, MATLAB, etc.) through automated data analysis, model training, and verification processes, so as to ensure the scientificity, objectivity, and adaptability of the form distortion coefficient threshold.

[0087] The form adaptive correction module constructs a fracture form correction mechanism according to the division results of each region in the fracture image, and performs corresponding correction measures on the low distortion region, medium distortion region, and high distortion region respectively;

[0088] In this embodiment, in the form adaptive correction module, according to the division results of each region in the fracture image, a fracture form correction mechanism is constructed, specifically: according to the division results of the low distortion region, medium distortion region, and high distortion region, different image transformation, feature correction, and edge optimization parameters are set respectively to form a fracture form correction mechanism; this fracture form correction mechanism automatically determines the image transformation method, feature correction strategy, and the adjustment range of edge optimization based on the form distortion coefficient of each region and the performance of the image processing algorithm through preset rules;

[0089] In the morphological adaptive correction module, according to the division results of each region in the fissure image, a fissure morphology correction mechanism is constructed, which can be achieved by combining adaptive adjustment based on image processing algorithms, optimization of machine learning models, and parameter adjustment driven by rules, so as to ensure the correction accuracy and adaptability of the fissure morphology under different distortion degrees. In the specific implementation process, first, lightweight correction is performed on the low-distortion region through image processing techniques (such as affine transformation, perspective transformation, and geometric transformation), and the preset morphological transformation parameters are used to adjust the scale, rotation, and perspective of the fissure image to correct slight morphological deviations. Secondly, in the medium-distortion region, combined with feature correction techniques, the geometric features of the fissure are extracted and compared through edge detection (such as Canny operator) and feature point matching (such as SIFT, ORB, etc.), and a deformation model (such as Thin Plate Spline, TPS) is used for morphological correction to ensure the accuracy and consistency of the fissure boundary features. In addition, in the high-distortion region, deep learning algorithms (such as convolutional neural network CNN or autoencoder network Autoencoder) are used to learn and reconstruct the fissure morphology, combined with multi-scale image registration techniques (such as pyramid optical flow method), to perform refined correction on the severely distorted region and enhance the integrity and recognizability of the fissure morphology. The implementation of this method can minimize the impact of perspective distortion on fissure recognition, ensure that the system can obtain accurate fissure geometric information under different geological conditions and shooting angles, improve the correction efficiency at the same time, and provide high-quality data support for subsequent fissure analysis and prediction.

[0090] Corresponding correction measures are taken for the low-distortion region, medium-distortion region, and high-distortion region respectively. Specifically: in the low-distortion region, the standard image transformation parameters in the fissure morphology correction mechanism are used to perform lightweight geometric correction to reduce perspective error; in the medium-distortion region, the standard feature correction parameters in the fissure morphology correction mechanism are maintained, combined with geometric adjustment and texture enhancement strategies to ensure the accuracy of the fissure boundary; in the high-distortion region, the enhanced edge optimization parameters in the fissure morphology correction mechanism are used, combined with deep learning reconstruction algorithms and multi-scale image registration methods to enhance the recovery accuracy of the fissure morphology and reduce the recognition deviation caused by severe distortion.

[0091] Corresponding correction measures are taken for the low-distortion region, medium-distortion region, and high-distortion region respectively, which can be achieved by combining adaptive adjustment based on image processing, feature extraction and matching techniques, and deep learning models to ensure accurate correction of the fissure morphology under different distortion degrees, reduce recognition errors, and improve detection accuracy. Specifically, in the low-distortion region, lightweight geometric correction methods such as affine transformation and perspective transformation are used for correction based on standard image transformation parameters to fine-tune the scale, rotation, and perspective relationship of the image, reduce geometric distortion caused by perspective deviation, ensure the basic consistency of the fissure morphology, and at the same time use smoothing techniques such as bilinear interpolation to prevent information loss during the correction process; in the medium-distortion region, optimization is carried out through a method combining feature correction, geometric adjustment, and texture enhancement. Techniques such as edge detection (e.g., Canny operator), Harris corner detection, or SIFT feature extraction are used to identify the boundary features of the fissure and perform feature point matching. Subsequently, a deformation correction model (e.g., Thin Plate Spline, TPS) is used for non-rigid adjustment, and at the same time, histogram equalization, gamma correction, etc. are used to enhance the image texture information to improve the clarity of the fissure boundary and ensure the integrity of its geometric morphology; in the high-distortion region, it is necessary to combine deep learning reconstruction algorithms and multi-scale image registration methods. The fissure morphology features are deeply learned through a pre-trained deep neural network (e.g., U-Net, CNN) to reconstruct the original fissure morphology. At the same time, the optical flow method or pyramid Kanade matching algorithm is used for multi-scale image registration to extract the global morphology of the fissure from images of different scales, further reducing the structural errors caused by severe distortion. The purpose of adopting these methods is to provide hierarchical and precise correction strategies for different degrees of morphological distortion, ensure the authenticity and reliability of the fissure morphology, and then improve the accuracy of subsequent fissure recognition and analysis, providing more accurate data support for geological disaster prediction and engineering decision-making.

[0092] The correction efficiency evaluation module, during the process of the fissure morphology correction mechanism correcting each region in the fissure image, obtains the correction error information of each region in the fissure image in real time, analyzes it after obtaining, evaluates whether the correction effect of the fissure morphology correction mechanism on each region in the fissure image can meet the expectations, and optimizes the fissure morphology correction mechanism according to the evaluation results;

[0093] In this embodiment, in the calibration efficiency evaluation module, during the process of the crack morphology calibration mechanism calibrating each region in the crack image, the calibration error information of each region in the crack image is obtained in real time and preprocessed after being obtained; the calibration offset information and the morphological consistency information in the preprocessed calibration error information are extracted and analyzed after being extracted, and the crack calibration offset coefficient and the crack calibration consistency index of each region are generated respectively; a calibration effect evaluation model is constructed for the generated crack calibration offset coefficient and crack calibration consistency index of each region, the calibration evaluation coefficient of each region is generated, and the generated calibration evaluation coefficient is analyzed to evaluate whether the calibration effect of the crack morphology calibration mechanism on each region in the crack image can meet the expectation, and the crack morphology calibration mechanism is optimized according to the evaluation result.

[0094] In the calibration efficiency evaluation module, the real-time acquisition of the calibration error information of each region in the crack image can be achieved through image registration, feature matching, and statistical analysis techniques. The specific methods include using the phase correlation method or the optical flow method to align the crack images before and after calibration to ensure the consistency of the positions of each region, and detecting the changes in the crack contour, texture, and edge features through feature matching algorithms such as SIFT and ORB, and calculating the calibration error, such as centroid offset, contour overlap, and gradient change. The necessity of preprocessing is that the calibration error information may contain noise and outliers, and direct analysis may affect the accuracy. Therefore, normalization, denoising, and smoothing processing are required to improve the data quality. The specific preprocessing includes: (1) using Gaussian filtering to reduce random noise and smooth the crack boundary; (2) adopting min-max normalization to unify the data range and enhance comparability; (3) using the box plot method to remove outliers to prevent interference with the calibration effect evaluation. Through these preprocessing measures, the stability and accuracy of the data are ensured, providing a reliable basis for the subsequent optimization of the calibration mechanism.

[0095] Extracting correction offset and morphological consistency information from the preprocessed correction error information can be achieved through techniques such as image feature analysis, morphological processing, and statistical modeling. Specifically, correction offset information can be extracted by calculating the displacement of key points in the crack image before and after correction using optical flow or feature point matching methods (such as SIFT and ORB). Combined with centroid calculation techniques, the overall offset of pixel positions within the region can be obtained, and the overall displacement trend of the crack edge can be measured through vector analysis. Conversely, morphological consistency information can be extracted by extracting the crack boundary based on contour detection (such as Canny edge detection) and quantifying the overlap of the contours before and after correction using the Hausdorff distance or Dice similarity coefficient to assess the consistency of the correction. Furthermore, texture analysis techniques (such as the gray-level co-occurrence matrix (GLCM)) can be used to calculate features such as contrast and energy within the crack region to further assess the internal consistency of the corrected crack region. During the extraction process, a sliding window method is used to analyze the crack image region by region to extract local feature changes. Regions with significant offset and consistency deviations are then screened by setting a threshold to ensure accurate assessment of the correction effect. These methods can be automatically executed in image processing software (such as OpenCV and MATLAB) for efficient and accurate correction error extraction.

[0096] In this embodiment, the logic for obtaining the crack correction offset coefficient and crack correction consistency index of each region is as follows:

[0097] The correction offset information in the preprocessed correction error information is extracted, specifically including the pixel density change rate, boundary offset vector length and grayscale gradient variance of each area in the crack image at different times during a period of time during the correction process of the crack morphology correction mechanism, and calibrated as 、 and , Indicates the period of time during which the crack morphology correction mechanism is correcting The first moment in the crack image The pixel density change rate of the region, Indicates the period of time during which the crack morphology correction mechanism is correcting The first moment in the crack image The length of the boundary offset vector of the region, Indicates the period of time during which the crack morphology correction mechanism is correcting The first moment in the crack image The gray gradient variance of the region, , , and are all positive integers;

[0098] During the calibration process of the crack morphology calibration mechanism, the pixel density change rate, the length of the boundary offset vector, and the gray gradient variance of each region in the crack images at different moments within a period of time can be achieved through image processing and feature analysis techniques. The pixel density change rate can extract the crack region through image binarization (such as Otsu threshold method) and count the change of the number of pixels per unit area over time to evaluate the expansion or contraction trend of the crack morphology. The length of the boundary offset vector can extract the crack boundary using edge detection (such as Canny operator) and track the boundary changes at different moments through feature point matching (such as SIFT, ORB algorithms), calculate the offset distance of the centroid, so as to quantify the adjustment amplitude of the crack boundary. The acquisition of the gray gradient variance depends on image gradient operation (such as Sobel operator), calculate the gray gradient of the crack region at different time points and count its variance to measure the texture consistency inside the calibrated region. To improve the accuracy of the data, the sliding window block processing can be used to process the image, combined with the time series smoothing algorithm to remove noise and ensure the stability and reliability of the acquired data. These methods can be automatically implemented through software tools (such as OpenCV, MATLAB), providing accurate data support for crack calibration evaluation.

[0099] Calculate the crack calibration offset coefficient of each region. The specific calculation formula is as follows:

[0100] ;

[0101] In the formula, is the crack calibration offset coefficient of the th region;

[0102] The reason for adopting this calculation formula is mainly to comprehensively and accurately evaluate the offset degree of each region during the crack morphology calibration process, so as to quantify the accuracy of the calibration effect. First, in the numerator part of the formula, the product of the pixel density change rate and the length of the boundary offset vector is adopted. Such a processing method can comprehensively consider the expansion or contraction of the crack region morphology before and after calibration, and at the same time combine the overall change of the boundary position, making the calculation of the offset degree more comprehensive and avoiding local deviation that may be caused by a single index. In addition, the square operation of the product term is introduced to amplify larger offset errors, making the regions with significant deviations more prominent in the calculation process, which helps to quickly identify high-error regions and ensure the sensitivity of the evaluation. Second, in the denominator part, The variance of the gray - level gradient is used as the normalized factor for correction. Adding "1" prevents the problem of zero denominator and ensures the computational stability is maintained even when the gradient change is small. The variance of the gray - level gradient measures the texture consistency of the crack area before and after correction. The larger the gradient change, the greater the morphological change in the corrected crack area. Therefore, the weight of its influence on the overall correction offset should be reduced to avoid misjudgment caused by over - correction. Finally, the overall calculation result is smoothed by taking the square - root operation to reduce the influence of extreme values on the calculation result, while improving the robustness and stability of the calculation. Ultimately, by taking the average value in the time dimension, it is ensured that the evaluation result has temporal consistency and representativeness, avoiding the deviation at a single time point from affecting the overall judgment and ensuring the scientificity and reliability of the correction effect.

[0103] The crack correction offset coefficient of the th region directly reflects the quality of the correction effect of the crack morphology correction mechanism on this region. The smaller the value, the higher the consistency of the pixel density distribution, boundary position, and gray - level gradient of this region before and after correction, indicating that the distortion of the crack morphology has been effectively corrected, the correction effect is good, and it approaches the expected goal. On the contrary, if

[0104] the value is large, it indicates that there is still a significant offset in the crack morphology of this region after correction, manifested as excessive pixel density change, significant deviation of the boundary position, or inconsistent texture features, indicating that the correction mechanism fails to fully eliminate perspective distortion or morphological distortion, the correction effect is not ideal, and it may be necessary to further optimize the correction parameters or adopt a more refined correction strategy.

[0104] Extract the morphological consistency information in the pre - processed correction error information, specifically including the edge pixel coincidence degree, the change rate of the edge gradient, and the change rate of the crack contour length in each region of the crack images at different moments during a period when the crack morphology correction mechanism is performing correction, and they are respectively calibrated as 、 and , represents the edge pixel coincidence degree of the th region in the crack image at the th moment during a period when the crack morphology correction mechanism is performing correction, represents the change rate of the edge gradient of the th region in the crack image at the th moment during a period when the crack morphology correction mechanism is performing correction, represents the change rate of the crack contour length of the th region in the crack image at the th moment during a period when the crack morphology correction mechanism is performing correction;

[0105] During the calibration process of the crack morphology calibration mechanism, the edge pixel coincidence degree, the change rate of the edge gradient, and the change rate of the crack contour length in different regions of the crack images at different times within a period can be achieved through image processing techniques. The edge pixel coincidence degree can extract the crack boundary through edge detection algorithms (such as Canny or Sobel operators), and use feature matching methods (such as ORB, SIFT) to register the images before and after calibration, calculate the proportion of matching pixels, and evaluate the boundary consistency. The change rate of the edge gradient can calculate the gray gradient of the crack boundary at different time points through gradient operations (such as Sobel operators), and measure the sharpness change through normalized difference to identify whether the calibration causes edge blurring or oversharpening. The change rate of the crack contour length can calculate the perimeter change of the crack contour before and after calibration through contour detection (such as contour extraction based on morphological closing operation) to evaluate the degree of morphological stretching or compression. These methods can be automatically executed with the help of image processing libraries such as OpenCV to ensure real-time and accurate acquisition of calibration evaluation data.

[0106] Calculate the crack calibration consistency index for each region. The specific calculation formula is as follows:

[0107] ;

[0108] In the formula, is the crack calibration consistency index of the th region.

[0109] The reason for adopting this calculation formula is to comprehensively and accurately evaluate the impact of the crack morphology calibration mechanism on the morphological consistency of each region and ensure that different degrees of distortion are reasonably measured. First, the edge pixel coincidence degree in the numerator part measures the matching degree of the crack edge pixels before and after calibration, directly reflecting the consistency of the crack morphology. The higher the value, the better the boundary calibration effect. However, the change in boundary sharpness also affects the consistency. Therefore, the edge gradient change rate is introduced in the denominator for normalization to prevent the excessive influence on the matching degree when the gradient change amplitude is large and avoid misjudgment caused by too high or too low edge sharpness. Second, taking the logarithmic operation on the crack contour length change rate is aimed at smoothing the influence of the contour length change before and after calibration, avoiding sudden deviations in the overall consistency evaluation caused by excessive morphological stretching or compression, and ensuring that small changes can also be effectively identified. Finally, the calculation results at all times are smoothed through time averaging operation to reduce the accidental error of single measurement and improve the stability and reliability of the evaluation results. In summary, this formula combines the boundary coincidence degree, the gradient change rate, and the contour change rate to ensure a comprehensive evaluation of the calibration consistency from three aspects of the edge, texture, and overall morphology, and guarantee the comparability and accuracy of the crack morphology at different calibration stages.

[0110] The fracture correction consistency index of the region directly reflects the morphological consistency restoration effect of the fracture morphology correction mechanism on this region. The smaller the index, the higher the consistency of the fracture morphology before and after correction, the higher the matching degree of edge pixels, the smaller the change in edge gradient, and the reasonable adjustment range of contour length, indicating a better correction effect, accurate retention of fracture characteristics, and meeting the expected goal; conversely, if the value is large, it means that there are still significant inconsistencies after fracture correction, such as low edge matching degree, drastic gradient change, and large change in contour length, indicating that there may be a large deviation in the corrected fracture morphology, the true characteristics of the fracture are not accurately restored, and the correction effect is not ideal.

[0111] In this embodiment, the fracture correction offset coefficient and the fracture correction consistency index of each generated region are used to construct a correction effect evaluation model, and the correction evaluation coefficient of each region is generated by weighted summation. The specific calculation formula is: , where and are the non-zero weight coefficients of the fracture correction offset coefficient and the fracture correction consistency index of each region respectively, and ;

[0112] In the correction effect evaluation model, the weight coefficients and are respectively used to measure the importance of the fracture correction offset coefficient and the fracture correction consistency index in the overall correction evaluation. reflects the influence of the offset degree of the fracture region on the correction effect. If the correction process causes a large boundary displacement, a higher weight needs to be assigned to highlight its influence; then reflects the contribution of morphological consistency to the evaluation. If the fracture morphology needs to be highly maintained after correction, the weight should be increased. The sum of the two is fixed at 1 to ensure the balance of the evaluation coefficient. The specific values of the weights can be adjusted according to experimental data or experience, usually optimized through historical evaluation results to achieve adaptability to different fracture conditions.

[0113] Determine the correction evaluation coefficient threshold set in advance for each region, and after determination, compare it with the correction evaluation coefficient A comparison is performed to evaluate whether the correction effect of the crack morphology correction mechanism on each area in the crack image can meet expectations based on the comparison results. The crack morphology correction mechanism is optimized based on the evaluation results. The specific comparison analysis is as follows:

[0114] like , the crack morphology correction mechanism can achieve the expected correction effect on this area in the crack image, and there is no need to optimize the crack morphology correction mechanism;

[0115] This indicates that the fracture morphology correction mechanism has achieved the desired correction effect in this area. The fracture morphology, boundaries, and texture characteristics are highly consistent with the actual situation, and the correction error is within an acceptable range, eliminating the need for further adjustment of the correction strategy. This ensures accurate identification and assessment of fracture structures in geological exploration or engineering applications, reducing additional computational overhead and unnecessary adjustments, thereby improving the system's operational efficiency and reliability.

[0116] like The crack morphology correction mechanism cannot achieve the expected correction effect on this area in the crack image, and the crack morphology correction mechanism needs to be optimized, including: adjusting the image transformation parameters to optimize the perspective distortion correction in the morphology correction process and reduce the geometric distortion caused by changes in shooting angles; improving the feature correction algorithm, by enhancing the accuracy of feature point matching and texture analysis, improving the accuracy of the crack boundary, reducing boundary dislocation and detail loss; enhancing the edge optimization strategy, adopting multi-scale edge detection and adaptive smoothing technology, thereby improving the accuracy and consistency of crack identification, and ensuring that the correction effect meets the expected standards.

[0117] This situation means that the fracture morphology correction mechanism has failed to achieve the expected correction effect in this area. Problems such as blurred fracture boundaries, distorted fracture shapes, or dimensional distortion may exist, resulting in reduced recognition accuracy. The impact of this is that if the correction mechanism is not optimized in a timely manner, it may lead to incorrect identification of fracture characteristics, which in turn affects the accuracy of geological hazard assessments or engineering construction planning, increasing potential safety hazards or construction costs. Therefore, the correction mechanism needs to be adjusted to ensure that the corrected fracture morphology is consistent with the actual situation.

[0118] The optimization of the crack morphology correction mechanism can be achieved by combining image processing, deep learning, and optimization algorithms, so as to reduce geometric distortion caused by changes in shooting angles and improve the accuracy and consistency of crack recognition. First, in terms of adjusting image transformation parameters, a method based on homography matrix estimation (such as the RANSAC algorithm) can be adopted. By detecting feature point pairs in the images before and after correction, the optimal perspective transformation matrix is calculated to perform geometric correction on the crack images, thereby compensating for the morphological distortion caused by perspective deviation. Second, in terms of improving the feature correction algorithm, deep learning techniques (such as CNN convolutional neural networks) can be introduced to perform high-precision learning and matching on the texture features of the crack area. Feature point detection algorithms such as SIFT and ORB are used to enhance the accuracy of key point positioning. Combining with the optical flow method to evaluate the boundary changes after correction, ensuring the accuracy of the crack boundary, and reducing boundary misalignment and detail loss. Finally, in terms of enhancing the edge optimization strategy, multi-scale edge detection techniques (such as the combination of Canny operator and LoG operator) can be adopted to detect the crack edges at different scales, ensuring the integrity of boundary details. Combining with adaptive smoothing algorithms (such as bilateral filtering or guided filtering) to balance the sharpness and smoothness of the edges, reducing noise interference, and improving the consistency of the crack morphology. Through the above methods, the correction parameters can be automatically adjusted at the software level, making the correction of the crack morphology more in line with the actual situation, avoiding overcorrection or undercorrection, and ultimately ensuring the accuracy and reliability of the recognition results.

[0119] The presetting of the correction evaluation coefficient thresholds for each region can be achieved by combining statistical analysis, machine learning, and optimization algorithms. First, based on a large amount of historical crack image data, mean-variance analysis or confidence interval estimation is used to calculate the reasonable ranges of the crack correction offset coefficient and the consistency index, and empirical thresholds are initially set. Second, through K-means clustering or Gaussian mixture model (GMM), the corrected crack regions are classified to automatically identify the threshold ranges of different distortion degrees. Finally, genetic algorithms or particle swarm optimization algorithms are applied, combined with the feedback of the correction effect to dynamically adjust the thresholds to ensure accuracy and robustness. These methods can be automatically executed through software tools (such as OpenCV, Scikit-learn) to achieve intelligent and adaptive adjustment of threshold setting.

[0120] The recognition data management module completes the automatic recognition and detection of cracks according to the optimized crack morphology correction mechanism, and records and stores the results of crack recognition and detection and the application process of the correction mechanism for subsequent monitoring, analysis, and optimization of crack morphology changes.

[0121] The implementation of the recognition data management module can be achieved by combining database management, automated data processing, and deep learning technologies to ensure the accurate storage and efficient management of fracture recognition and detection results. First, a database management system (such as MySQL, MongoDB) is used to establish a structured storage scheme to classify and store fracture detection results, calibration parameters, and evaluation data, and timestamp indexing is used for subsequent querying and retrospective analysis. Second, data processing tools (such as Pandas, NumPy) are used to preprocess the stored data, including data cleaning, normalization, and trend analysis, to support long-term monitoring and change analysis of fracture morphology. Finally, in combination with deep learning frameworks (such as TensorFlow, PyTorch), the stored fracture images and detection results are trained and optimized to continuously improve the recognition accuracy, and predictive analysis is performed based on historical data to identify potential fracture propagation trends. This approach can not only achieve the efficient management of fracture recognition results but also optimize the subsequent recognition process in a data-driven manner, ensuring the self-adaptability and long-term monitoring ability of the system and improving the reliability of geological safety assessment.

[0122] The above formulas are all dimensionless and take their numerical values for calculation. The formulas are obtained by collecting a large amount of data for software simulation to get a formula closest to the actual situation. The preset parameters in the formulas are set by those skilled in the art according to the actual situation.

[0123] The above embodiments can be implemented in whole or in part by software, hardware, firmware, or any other combination. When implemented using software, the above embodiments can be implemented in whole or in part in the form of a computer program product. The computer program product includes one or more computer instructions or computer programs. When the computer instructions or computer programs are loaded or executed on a computer, the processes or functions described in the embodiments of the present application are generated in whole or in part. The computer can be a general-purpose computer, a special-purpose computer, a computer network, or other programmable devices. The computer instructions can be stored in a computer-readable storage medium or transmitted from one computer-readable storage medium to another. For example, the computer instructions can be transmitted from one website, computer, server, or data center to another website, computer, server, or data center by wired or wireless (such as infrared, wireless, microwave, etc.) means. The computer-readable storage medium can be any available medium that the computer can access or a data storage device such as a server or data center that contains one or more collections of available media. The available media can be magnetic media (such as floppy disks, hard disks, magnetic tapes), optical media (such as DVDs), or semiconductor media. The semiconductor media can be solid-state drives.

[0124] It should be understood that in various embodiments of the present application, the sequence numbers of the above processes do not indicate the order of execution, and the execution order of each process should be determined according to its function and internal logic, and should not constitute any limitation to the implementation process of the embodiments of the present application.

[0125] Those of ordinary skill in the art can realize that the units and algorithm steps of each example described in combination with the embodiments disclosed herein can be implemented by electronic hardware, or a combination of computer software and electronic hardware. Whether these functions are executed in a hardware or software manner depends on the specific application and design constraints of the technical solution. Professional technicians can use different methods to implement the described functions for each specific application, but such implementation should not be considered to exceed the scope of the present application.

[0126] In several embodiments provided by the present application, it should be understood that the disclosed systems and methods can be implemented in other ways. For example, the above-described embodiments are merely illustrative. For example, the division of the units is only a logical function division, and there can be other division methods in actual implementation. For example, multiple units or components can be combined or integrated into another system, or some features can be ignored or not executed. Another point is that the displayed or discussed couplings or direct couplings or communication connections to each other can be indirect couplings or communication connections through some interfaces, devices or units, and can be in electrical, mechanical or other forms.

[0127] The units described as separate components may or may not be physically separated, and the components shown as units may or may not be physical units, that is, they can be located in one place, or distributed to multiple network units. Some or all of the units can be selected according to actual needs to achieve the purpose of the solution of this embodiment.

[0128] In addition, the functional units in each embodiment of the present application can be integrated into one processing unit, or each unit can exist physically alone, or two or more units can be integrated into one unit.

[0129] As described above, it is only the specific implementation manner of the present application, but the protection scope of the present application is not limited thereto. Any person skilled in the art within the technical scope disclosed by the present application can easily think of changes or substitutions, which should all be covered by the protection scope of the present application. Therefore, the protection scope of the present application should be subject to the protection scope of the claims.

Claims

1. An automatic recognition and detection system for geological structure fractures based on deep learning, characterized in that, It includes a fracture image analysis module, a distortion intelligent evaluation module, a morphology adaptive correction module, a correction efficiency evaluation module, and an identification data management module; The fracture image analysis module scans the rock surface through a photographing device during geological exploration to obtain real-time image information of fractures, and detects whether the fractures are distorted due to perspective effect in the obtained fracture image information. When it is detected that the fracture morphology is distorted due to perspective effect, the fracture image is evenly divided into several regions; The distortion intelligent evaluation module obtains real-time imaging parameter information of each region in the fracture image, analyzes it after obtaining, judges the degree of morphological distortion of each region in the fracture image under perspective distortion conditions, and divides each region in the fracture image into a low-distortion region, a medium-distortion region, and a high-distortion region according to the judgment result; In the distortion intelligent evaluation module, after obtaining the imaging parameter information of each region in the fracture image in real time, it performs preprocessing on it; extracts the light intensity feature information and contour structure information in the imaging parameter information of each region in the preprocessed fracture image, analyzes them after extraction, and respectively generates the fracture perspective distortion coefficient and the fracture contour distortion index of each region; constructs a morphological distortion model for the generated fracture perspective distortion coefficient and fracture contour distortion index of each region, generates the morphological distortion coefficient of each region, analyzes it after generation, judges the degree of morphological distortion of each region in the fracture image under perspective distortion conditions, and divides each region in the fracture image into a low-distortion region, a medium-distortion region, and a high-distortion region according to the judgment result; The acquisition logic of the fracture perspective distortion coefficient and the fracture contour distortion index of each region is as follows: Extract the light intensity feature information in the imaging parameter information of each region in the pre - processed fissure image, specifically including the pixel brightness values, the offset of the centroid coordinates, and the average gradient intensity of the edge pixels at different moments within a period of time for each region in the fissure image, and label them respectively as , and , represents the pixel brightness value of the th region in the fissure image at the moment within a period of time, represents the offset of the centroid coordinates of the th region in the fissure image at the moment within a period of time, represents the average gradient intensity of the edge pixels of the th region in the fissure image at the moment within a period of time, , , and are all positive integers; Calculate the fracture perspective distortion coefficient of each region. The specific calculation formula is as follows: ; wherein, is the fracture perspective distortion coefficient of the th region; The contour structure information in the imaging parameter information of each area in the preprocessed crack image is extracted, including the total length of the crack contour of each area in the crack image at different times within a period of time, the difference between the maximum and minimum values of the internal pixels, and the change in edge curvature, and they are calibrated as 、 and , Indicates the crack image area within a period of time The total length of the crack profile at time , Indicates the crack image area within a period of time The difference between the maximum and minimum values of the internal pixels at the moment, Indicates the crack image area within a period of time The change of edge curvature at the moment; Calculate the fracture contour distortion index of each region. The specific calculation formula is as follows: ; wherein, is the crack profile distortion index of the th region; The morphology adaptive correction module constructs a fracture morphology correction mechanism according to the division result of each region in the fracture image, and performs corresponding correction measures on the low-distortion region, the medium-distortion region, and the high-distortion region respectively; The correction efficiency evaluation module obtains real-time correction error information of each region in the fracture image during the process of the fracture morphology correction mechanism correcting each region in the fracture image, analyzes it after obtaining, evaluates whether the correction effect of the fracture morphology correction mechanism on each region in the fracture image can reach the expectation, and optimizes the fracture morphology correction mechanism according to the evaluation result; The identification data management module completes the automatic identification and detection of fractures according to the optimized fracture morphology correction mechanism, and records and stores the results of fracture identification and detection and the application process of the correction mechanism for subsequent monitoring, analysis, and optimization of fracture morphology changes.

2. The deep learning-based automatic identification and detection system for geological structure cracks according to claim 1 is characterized in that: The perspective distortion coefficient of the fractures in each generated region and the fracture contour distortion index Construct a shape distortion model, and generate the shape distortion coefficient of each region through weighted summation , and the specific calculation formula is: , where and are the perspective distortion coefficients of the fractures in each region and the fracture contour distortion index are non-zero weight coefficients, and ; Determine the pre-set threshold interval of the morphological distortion coefficient , and after determination, compare it with the morphological distortion coefficients of the generated regions . According to the comparison results, judge the morphological distortion degree of each region in the fissure image under the condition of perspective distortion, and divide each region in the fissure image into a low-distortion region, a medium-distortion region, and a high-distortion region according to the judgment results. The specific comparison analysis and division are as follows: like , the morphological distortion degree of this area in the crack image under the perspective distortion condition is low, and this area is divided into a low distortion area; If , the morphological distortion degree of this area in the fissure image under the perspective distortion condition is medium, and this area is divided into a medium distortion area; If , the morphological distortion degree of this area in the fissure image under perspective distortion conditions is high, and this area is divided into a high distortion area.

3. The automatic identification and detection system for geological structure fissures based on deep learning according to claim 2, characterized in that, In the morphological adaptive correction module, a crack morphology correction mechanism is constructed based on the results of the division of each area in the crack image. Specifically, different image transformation, feature correction and edge optimization parameters are set according to the division results of low-distortion areas, medium-distortion areas and high-distortion areas to form a crack morphology correction mechanism. Based on the morphological distortion coefficients of each area and the performance of the image processing algorithm, the crack morphology correction mechanism automatically determines the image transformation method, feature correction strategy and edge optimization adjustment range through pre-set rules. Corresponding correction measures are taken for low-distortion areas, medium-distortion areas, and high-distortion areas respectively. Specifically, in low-distortion areas, the standard image transformation parameters in the crack morphology correction mechanism are used to perform lightweight geometric correction to reduce perspective error; in medium-distortion areas, the standard feature correction parameters in the crack morphology correction mechanism are maintained, combined with geometric adjustment and texture enhancement strategies to ensure the accuracy of crack boundaries; in high-distortion areas, the enhanced edge optimization parameters in the crack morphology correction mechanism are used, combined with deep learning reconstruction algorithms and multi-scale image registration methods to enhance the restoration accuracy of crack morphology and reduce recognition bias caused by severe distortion.

4. The automatic identification and detection system for geological structure fissures based on deep learning according to claim 3, characterized in that, In the correction effectiveness evaluation module, during the process of the crack morphology correction mechanism correcting each area in the crack image, the correction error information of each area in the crack image is obtained in real time and preprocessed after acquisition; the correction offset information and morphological consistency information in the preprocessed correction error information are extracted, and analyzed after extraction to generate the crack correction offset coefficient and crack correction consistency index of each area respectively; a correction effect evaluation model is constructed for the generated crack correction offset coefficient and crack correction consistency index of each area, and the correction evaluation coefficient of each area is generated, and analyzed after generation to evaluate whether the correction effect of the crack morphology correction mechanism on each area in the crack image can meet expectations, and the crack morphology correction mechanism is optimized according to the evaluation results.

5. The automatic recognition and detection system for geological structure fissures based on deep learning according to claim 4, characterized in that The logic for obtaining the crack correction offset coefficient and crack correction consistency index of each region is as follows: Extract the calibration offset information from the preprocessed calibration error information, specifically including the pixel density change rate, boundary offset vector length, and gray gradient variance of each region in the fracture image at different moments within a period of time during the calibration process of the fracture morphology calibration mechanism, and label them respectively as 、 and , represents the pixel density change rate of the th region in the fracture image at the th moment within a period of time during the calibration process of the fracture morphology calibration mechanism, represents the boundary offset vector length of the th region in the fracture image at the th moment within a period of time during the calibration process of the fracture morphology calibration mechanism, represents the gray gradient variance of the th region in the fracture image at the th moment within a period of time during the calibration process of the fracture morphology calibration mechanism, , , and are all positive integers; Calculate the crack correction offset coefficient for each area. The specific calculation formula is as follows: Where, For the The crack correction offset coefficient for each region; Extract the morphological consistency information from the preprocessed calibration error information, specifically including the edge pixel coincidence degree, the change rate of the edge gradient, and the change rate of the crack contour length in each region of the crack image at different times during the calibration process of the crack morphology calibration mechanism, and label them respectively as , and , indicating the edge pixel coincidence degree of the -th region in the crack image at the -th moment during the calibration process of the crack morphology calibration mechanism, indicating the change rate of the edge gradient of the -th region in the crack image at the -th moment during the calibration process of the crack morphology calibration mechanism, indicating the change rate of the crack contour length of the -th region in the crack image at the -th moment during the calibration process of the crack morphology calibration mechanism; Calculate the crack correction consistency index of each area. The specific calculation formula is as follows: ; wherein, is the fracture correction consistency index of the th region.

6. The automatic recognition and detection system for geological structure fissures based on deep learning according to claim 5, characterized in that, Fissure correction offset coefficients for each generated region and fissure correction consistency index Construct a correction effect evaluation model, and generate a correction evaluation coefficient for each region through weighted summation , and the specific calculation formula is: , where and are respectively the fissure correction offset coefficients of each region and the fissure correction consistency index are non-zero weight coefficients, and ; Determine the calibration evaluation coefficient thresholds for each preset region , and after determination, compare with the calibration evaluation coefficients of each generated region , evaluate whether the calibration effect of the crack morphology calibration mechanism on each region in the crack image can meet the expectation according to the comparison result, and optimize the crack morphology calibration mechanism according to the evaluation result. The specific comparison and analysis are as follows: like , the crack morphology correction mechanism can achieve the expected correction effect on this area in the crack image, and there is no need to optimize the crack morphology correction mechanism; If , the correction effect of the crack morphology correction mechanism on this area in the crack image cannot meet the expectations, and it is necessary to optimize the crack morphology correction mechanism, specifically including: adjusting the image transformation parameters to optimize the perspective distortion correction in the morphology correction process and reduce the geometric distortion caused by the change of the shooting angle; improving the feature correction algorithm to improve the accuracy of the crack boundary by enhancing the feature point matching and texture analysis accuracy and reducing the boundary misalignment and detail loss; enhancing the edge optimization strategy by adopting multi-scale edge detection and adaptive smoothing techniques, so as to improve the accuracy and consistency of crack recognition and ensure that the correction effect meets the expected standards.

Images