A rock mass structural surface recognition method, device, equipment and readable storage medium based on digital image recognition technology

Abstract

The present invention provides a rock mass structural surface recognition method, device, equipment and readable storage medium based on digital image recognition technology, which relates to the technical field of electrical digital data processing. The present invention further evaluates the consistency of responses of different recognition areas of a rock mass slope to changes in illumination by analyzing the angle between the normal vector of a representative point and the direction of a light source. This process helps to identify and understand the impact of illumination on the surface characteristics of the rock mass. By assigning a judgment threshold to the spectral characteristics of each recognition area and generating a hue index in combination with an average similarity coefficient, a three-dimensional point cloud model can be more accurately corrected, thereby coping with the challenges brought by different environmental illumination conditions, and further providing a more accurate recognition strategy for rock mass structural surfaces.

Description

Technical Field

[0001] The present invention relates to the technical field of electronic digital data processing, and in particular to a rock mass structural surface recognition method, device, equipment and readable storage medium based on digital image recognition technology. Background Art

[0002] With the frequent occurrence of geological disasters and the increasing demand for mountain development, the importance of rock structural surface identification technology in the fields of slope stability analysis, mine development, and civil engineering has become increasingly prominent; in recent years, research on rock structural surface identification has gradually shifted from traditional manual measurement and empirical judgment to automated methods based on digital technology. Existing technologies, such as the slope rock structural surface identification method with publication number CN117556703B, use a three-dimensional point cloud data generation model and process the slope through steps such as cube segmentation and outlier removal. These methods achieve the identification of rock structural surfaces by constructing deep learning models, and improve the recognition efficiency and accuracy to a certain extent; however, these technologies still rely on basic geometric feature extraction and lack adaptability to changes in ambient lighting, which may lead to reduced recognition accuracy under different lighting conditions;

[0003] The slope rock structure surface recognition method of CN117556703B adopts a three-dimensional point cloud data generation model and processes the slope through steps such as cube segmentation and outlier removal. These methods realize the recognition of rock structure surfaces by constructing a deep learning model, and improve the recognition efficiency and accuracy to a certain extent. However, these technologies still rely on basic geometric feature extraction and are not adaptable enough to changes in ambient lighting, which may lead to a decrease in recognition accuracy under different lighting conditions.

[0004] Existing deficiencies: In the process of rock structure surface recognition, the existing technology usually only focuses on the extraction and grouping of geometric features, but does not fully consider the impact of ambient lighting on point cloud data; changes in lighting conditions will lead to significant changes in the color and brightness of the rock surface, which will directly affect the training effect and recognition results of the model; although the existing methods realize the classification of rock structure surfaces through cube segmentation and data set construction, there is a relative lack of analysis on the consistency of response to lighting changes, which may cause the model to experience performance fluctuations in practical applications; in addition, the existing technology has little research on point cloud model correction, and has failed to effectively integrate ambient lighting parameters with point cloud data, which makes the model lack the necessary adaptability and accuracy; how to effectively combine ambient lighting parameters with rock structure surface analysis is the direction that needs to be improved at present;

[0005] The above information disclosed in the above Background section is only for enhancement of understanding of the background of the present disclosure and therefore it may contain information that does not form the prior art that is already known to one of ordinary skill in the art. Summary of the invention

[0006] The purpose of the present invention is to provide a rock structure surface recognition method, device, equipment and readable storage medium based on digital image recognition technology to solve the problems raised in the above background technology.

[0007] To achieve the above object, the present invention provides the following technical solutions:

[0008] A rock mass structural surface recognition method based on digital image recognition technology, the specific steps include:

[0009] Step S1: obtaining a three-dimensional point cloud model composed of rock slope point cloud data, and performing region division on the three-dimensional point cloud model to form M identification regions;

[0010] Collect the ambient lighting parameters corresponding to the identification area at the rock slope, which include the light source direction and spectral characteristics; and each point in each identification area contains its corresponding three-dimensional coordinates and original color information;

[0011] Step S2: Based on the original color information, the illumination model is used to calculate the ideal color of the points in the point cloud corresponding to each recognition area;

[0012] Step S3: In each recognition area, a number of points are determined from the corresponding point cloud as representative points; by analyzing the angles between the normal vectors of these representative points and the corresponding light source directions, the average similarity coefficients of these representative points are comprehensively calculated;

[0013] Step S4: assigning a judgment threshold to the spectral characteristics of each identification area, and calculating the ratio of the spectral characteristics actually recorded in each identification area to the corresponding judgment threshold, and combining the ratio with the corresponding average similarity coefficient for analysis to generate a hue index for providing a correction strategy for the corresponding lighting conditions of each identification area;

[0014] On the basis of the illumination conditions being adjusted through the correction strategy, the ambient illumination parameter data of each recognition area is re-collected to update the originally constructed three-dimensional point cloud model, thereby obtaining a corrected three-dimensional point cloud model;

[0015] Step S5: extracting the ambient lighting parameters of each identification area from the corrected three-dimensional point cloud model, using the deep learning model to group the point cloud data and the extracted ambient lighting parameters in each identification area by structural surface, and constructing a data set according to the grouping results, the data set containing rock structural surface information and ambient lighting parameters, wherein the point cloud data and ambient lighting parameters are used as inputs of the model, and the rock structural surface information is used as output of the model, and finally, the data set is used for training to obtain a rock structural surface recognition model;

[0016] Step S6: Input the point cloud data of the rock slope to be predicted and its corresponding ambient lighting parameters into the trained deep learning model, and the model will output the rock structure surface information including the location, characteristics and occurrence information.

[0017] A rock mass structural surface recognition device based on digital image recognition technology, the device is used to execute the rock mass structural surface recognition method based on digital image recognition technology, comprising:

[0018] Region division and parameter acquisition module: used to obtain a three-dimensional point cloud model composed of rock slope point cloud data, and divide the three-dimensional point cloud model into regions to form M identification regions; and collect the ambient light parameters on the M identification regions, including the light source direction and spectral characteristics; each point in each identification region contains its corresponding three-dimensional coordinates and original color information;

[0019] Ideal color determination module: used to calculate the ideal colors of several points in the point cloud corresponding to each recognition area based on the original color information using the illumination model;

[0020] Average similarity coefficient generation module: used to determine several points from the corresponding point cloud as representative points in each identification area; by analyzing the angle between the normal vector of these representative points and the corresponding light source direction, and comprehensively calculating the average similarity coefficient of these representative points, the response consistency of different identification areas on the rock slope to the change of illumination is evaluated;

[0021] Tone index generation module: used to assign a judgment threshold to the spectral characteristics of each recognition area, and to express the spectral characteristics actually recorded in each recognition area and the corresponding judgment threshold in a ratio manner, and to combine and analyze the ratio with the corresponding average similarity coefficient to generate a tone index for providing a correction strategy for the corresponding lighting conditions of each recognition area;

[0022] Re-collecting the ambient lighting parameter data of each recognition area to update the originally constructed three-dimensional point cloud model, thereby obtaining a corrected three-dimensional point cloud model;

[0023] Rock mass structural surface recognition model construction module: extract the ambient lighting parameters of each recognition area from the corrected three-dimensional point cloud model, use the deep learning model to group the point cloud data and the extracted ambient lighting parameters in each recognition area, and build a data set based on the grouping results. The data set contains rock mass structural surface information and ambient lighting parameters, where the point cloud data and ambient lighting parameters are used as the input of the model, and the rock mass structural surface information is used as the output of the model. Finally, the data set is used for training to obtain a rock mass structural surface recognition model;

[0024] Rock mass structural surface information output module: used to input the point cloud data of the rock mass slope to be predicted and its corresponding ambient lighting parameters into the trained deep learning model. The model will output the rock mass structural surface information including location, characteristics and occurrence information.

[0025] A device comprises a memory, a processor and a computer program stored in the memory and executable on the processor. When the processor executes the computer program, the steps of the rock structure surface recognition method based on digital image recognition technology are implemented.

[0026] A readable storage medium stores a computer program, and when the computer program is executed by a processor, the steps of the rock structure surface recognition method based on digital image recognition technology are implemented.

[0027] Compared with the prior art, the beneficial effects of the present invention are: by analyzing the angle between the normal vector of the representative point and the direction of the light source, the consistency of the response of different identification areas of the rock slope to the illumination change is further evaluated. This process helps to identify and understand the influence of illumination on the surface characteristics of the rock mass; by assigning a judgment threshold to the spectral characteristics of each identification area and generating a hue index based on the average similarity coefficient, the three-dimensional point cloud model can be corrected more accurately, thereby coping with the challenges brought by different environmental illumination conditions, and further providing a more accurate identification strategy for the rock structure surface. BRIEF DESCRIPTION OF THE DRAWINGS

[0028] Figure 1 It is a schematic diagram of the overall method flow of the present invention;

[0029] Figure 2 It is a block diagram of the device execution modules of the present invention. DETAILED DESCRIPTION

[0030] In order to make the objectives, technical solutions and advantages of the present invention more clearly understood, the present invention is further described in detail below in conjunction with specific embodiments.

[0031] It should be noted that, unless otherwise defined, the technical terms or scientific terms used in the present invention should be understood by people with ordinary skills in the field to which the present invention belongs. The words "first", "second" and similar words used in the present invention do not indicate any order, quantity or importance, but are only used to distinguish different components. "Include" or "comprise" and similar words mean that the elements or objects appearing before the word include the elements or objects listed after the word and their equivalents, without excluding other elements or objects. "Connect" or "connected" and similar words are not limited to physical or mechanical connections, but may include electrical connections, whether direct or indirect. "Up", "down", "left", "right" and the like are only used to indicate relative positional relationships. When the absolute position of the described object changes, the relative positional relationship may also change accordingly.

[0032] Embodiment 1:

[0033] See also Figure 1 , the present invention provides a technical solution:

[0034] A rock mass structural surface recognition method based on digital image recognition technology, the specific steps include:

[0035] Step S1: obtaining a three-dimensional point cloud model composed of rock slope point cloud data, and performing region division on the three-dimensional point cloud model to form M identification regions;

[0036] Collect the ambient lighting parameters corresponding to the identification area at the rock slope, which include the light source direction and spectral characteristics; and each point in each identification area contains its corresponding three-dimensional coordinates and original color information;

[0037] Step S2: Based on the original color information, the illumination model is used to calculate the ideal color of the points in the point cloud corresponding to each recognition area;

[0038] Step S3: In each recognition area, a number of points are determined from the corresponding point cloud as representative points; by analyzing the angles between the normal vectors of these representative points and the corresponding light source directions, the average similarity coefficients of these representative points are comprehensively calculated;

[0039] Step S4: assigning a judgment threshold to the spectral characteristics of each identification area, and calculating the ratio of the spectral characteristics actually recorded in each identification area to the corresponding judgment threshold, and combining the ratio with the corresponding average similarity coefficient for analysis to generate a hue index for providing a correction strategy for the corresponding lighting conditions of each identification area;

[0040] On the basis of the illumination conditions being adjusted through the correction strategy, the ambient illumination parameter data of each recognition area is re-collected to update the originally constructed three-dimensional point cloud model, thereby obtaining a corrected three-dimensional point cloud model;

[0041] Step S5: extracting the ambient lighting parameters of each identification area from the corrected three-dimensional point cloud model, using the deep learning model to group the point cloud data and the extracted ambient lighting parameters in each identification area by structural surface, and constructing a data set according to the grouping results, the data set containing rock structural surface information and ambient lighting parameters, wherein the point cloud data and ambient lighting parameters are used as inputs of the model, and the rock structural surface information is used as output of the model, and finally, the data set is used for training to obtain a rock structural surface recognition model;

[0042] Step S6: Input the point cloud data of the rock slope to be predicted and its corresponding ambient lighting parameters into the trained deep learning model, and the model will output the rock structure surface information including the location, characteristics and occurrence information.

[0043] The point cloud data in the 3D point cloud model is preprocessed to remove noise and outliers and generate cleaned point cloud data to improve the accuracy of subsequent processing.

[0044] It is further explained that a high-precision laser scanner, model LeicaRTC360, is used to collect the three-dimensional point cloud data of the rock slope, ensuring that the accuracy of the point cloud acquisition is not less than ±1mm; in this process, it is ensured that the density of the point cloud reaches at least 1000 points per square meter;

[0045] The collected point cloud data is converted into standard point cloud formats LAS or PLY through the data processing software CloudCompare, and the data integrity is confirmed to ensure that the data loss rate does not exceed 5%;

[0046] The voxel grid method is used to divide the three-dimensional point cloud data into M recognition areas, and the M recognition areas are represented as {1, 2, ..., i, ..., M}, where i represents the i-th recognition area, and i∈{1, 2, ..., M}, and the size of each recognition area is set to 0.5mx0.5mx0.5m (V=0.125m³); and the goal of the division is set to keep the number of point clouds in each recognition area to reach or exceed N1 points; in this embodiment, N1 is 200; ensure that the number of point clouds in each recognition area is uniform;

[0047] In the rock mass side wave corresponding to each identification area i, a photometer is used to collect the ambient light parameters, and the ambient light parameters of the i-th identification area, including the light source direction and spectral characteristics, are marked in sequence as ;

[0048] The light source direction The azimuth and elevation angles of the light source are used to numerically represent the direction of the light source. Represented as a vector ;in, represents the azimuth of the i-th recognition area; represents the pitch angle of the i-th recognition area;

[0049] For the azimuth of the i-th recognition area , the azimuth is the horizontal angle of the light source relative to north, The value range is between 0° and 360°, and the specific definitions are as follows:

[0050] =0°, indicating true north;

[0051] =90°, indicating due east;

[0052] =180°, indicating due south;

[0053] =270°, indicating due west;

[0054] Azimuth The calculation formula is as follows:

[0055] ;

[0056] Where X and Y are the horizontal coordinates of the light source relative to the i-th recognition area;

[0057] For the pitch angle of the i-th recognition area , the pitch angle is the vertical angle of the light source relative to the horizontal plane, The value range is [-90°, 90°], and the specific definitions are as follows:

[0058] =0°, indicating a horizontal light source; =90°, indicating a vertical light source; =-90°, indicating a vertical light source;

[0059] The pitch angle is calculated using the following formula:

[0060] ;

[0061] Where Z is the vertical coordinate of the light source;

[0062] Direct the light source Represented as a vector .

[0063] For the light source direction : Record the azimuth and elevation angles of the light source, ensuring that the recorded error does not exceed 5°;

[0064] For spectral characteristics :Use a spectrophotometer to ensure that the recorded wavelength range is between 400nm and 700nm, specifically expressed as 400≤ ≤700;

[0065] Spectral characteristics of the i-th identification area The spectral characteristics are expressed by the wavelength range and light intensity distribution of the spectrum. Specifically expressed as ;in, represent the wavelength range and light intensity distribution of the i-th identification area respectively.

[0066] The wavelength range of the i-th identification area is expressed as ;in Respectively represent the minimum wavelength and maximum wavelength of the i-th identification area;

[0067] For the light intensity distribution of the i-th recognition area , It is expressed as follows:

[0068] ;

[0069] in, Represents the light intensity corresponding to the hth wavelength under the i-th identification area ;

[0070] And h∈{1, 2,…, H}, the larger the value of h is, the larger the corresponding wavelength is.

[0071] Record the lighting characteristics of each recognition area to ensure that the matching degree between the ambient lighting parameters and the recognition area reaches more than 95%.

[0072] To further illustrate, the three-dimensional coordinates of the points in the point cloud corresponding to the i-th recognition area are expressed as ;

[0073] in, is the three-dimensional coordinate representation of the jth point in the point cloud corresponding to the i-th recognition area; for The corresponding three-dimensional coordinates; N1 is the total number of points in the point cloud corresponding to the i-th recognition area;

[0074] The original color information of the points in the point cloud corresponding to each recognition area is represented by the values ​​of the RGB three-color channels. The original color information is represented in the following way:

[0075] The original color information is represented using the RGB color model, with the value of each color channel ranging from 0 to 255. The original color information of the i-th recognition area is represented as ,in They represent the intensity of the red, green and blue channels of the jth point in the point cloud corresponding to the i-th recognition area, respectively. Represents the original color information of the jth point in the i-th recognition area.

[0076] The point cloud data is preprocessed using statistical outlier removal (SOR) and voxel grid downsampling (VGD):

[0077] Noise removal: Set the k value to 20, and the number of identified outliers should not exceed 2% of the total point cloud;

[0078] Downsampling: Set the voxel grid size to 0.1m, ensuring that the number of point clouds in each voxel is not less than 5 points. The number of point clouds is expected to decrease by 50% after downsampling;

[0079] The accuracy and clarity of the cleaned point cloud data should be guaranteed to be above 95% complete to ensure accuracy in the subsequent recognition process.

[0080] It is further explained that the Lambertian Reflection Model is selected as the illumination model for calculating the color of the point cloud. This model can effectively simulate the color changes of the rock surface under different lighting conditions;

[0081] This model assumes that the light intensity at a point is proportional to the angle of the incident light and the surface normal, which is applicable to most natural surfaces. The rock mass structural surface in this embodiment belongs to a natural surface;

[0082] For the i-th recognition area, obtain the three-dimensional coordinate representation of the j-th point in its corresponding point cloud ;

[0083] Use the principal component analysis (PCA) method in the normal estimation algorithm to calculate the normal information of the point cloud in the i-th recognition area; associate the normal information with the three-dimensional coordinates of each point to obtain the normal vector of the j-th point in the i-th recognition area. ;

[0084] According to the ambient lighting parameters collected in step S1, the light source direction of the current recognition area i is extracted and spectral characteristics ;

[0085] Specifically extract the azimuth of the light source direction and pitch angle , and the wavelength range of the spectral characteristics and light intensity distribution ;

[0086] Calculate the ideal color for each point, including:

[0087] For each point, calculate its ideal color using the following formula :

[0088] ;

[0089] in, Represented as three-dimensional coordinates The original color information of the corresponding points; is the diffuse reflectance, ; is the dot product of the incident light source direction and the normal vector;

[0090] Get the spectral reflectance curve of the current identification area i, find the wavelength corresponding to the maximum reflectance from the spectral reflectance curve, and record it as the optimal wavelength ; Further, according to the optimal wavelength in the light intensity distribution set , calculation and optimal wavelength The Euclidean distance is then filtered out to find the light intensity corresponding to the minimum Euclidean distance. , h1∈{1,2,…,H}. The specific steps are as follows:

[0091] 2.1) Determine the optimal wavelength:

[0092] ;

[0093] in is the spectral reflectance curve;

[0094] 2.2) Matching light intensity:

[0095] For each pair in the light intensity distribution set , calculate the optimal wavelength With each and use the absolute difference as the distance metric:

[0096] ;

[0097] 2.3) Find the light intensity corresponding to the minimum distance:

[0098] Filter out light intensity h corresponding to the minimum distance:

[0099] ;

[0100] The calculated ideal color Stored in the corresponding point cloud data structure for subsequent rock mass structural surface identification and analysis.

[0101] Further explanation, the average similarity coefficient and the acquisition of the hue index specifically include:

[0102] 1.1) Select representative points:

[0103] In each recognition area i, N2 points are randomly selected from the point cloud data as representative points, and the normal vector of the j1th representative point selected in the i-th recognition area is marked as , , N2<N1, and N2≥5; the selected points must be evenly distributed in space to avoid local features affecting the overall analysis; the three-dimensional coordinates of the j1th representative point selected in the i-th recognition area are expressed as ;

[0104] 1.2) Calculation of normal vector:

[0105] Use the normal estimation algorithm based on principal component analysis PCA to calculate the normal vector of each representative point ;

[0106] The specific steps are as follows:

[0107] For each representative point , determine the points within the range of d = 0.05m, and use these points as the neighborhood point set, and set the total number of points in the neighborhood point set to , The first points;

[0108] Calculate the covariance matrix of neighborhood points :

[0109] ;

[0110] in The first The coordinates of the points, is the mean of the points in the neighborhood, and T is the transposition operation;

[0111] Find the covariance matrix The eigenvalues ​​and eigenvectors of , select the eigenvector corresponding to the minimum eigenvalue as the normal vector of the j1th representative point selected in the i-th recognition area ;

[0112] is the vector Transpose.

[0113] 1.3) Calculate the angle between the normal vector and the light source direction:

[0114] Direct the light source Convert to unit vector :

[0115] ;

[0116] For each representative point , calculate the angle between the normal vector and the unit light source direction :

[0117] ;

[0118] 1.4) Calculate the average similarity coefficient:

[0119] The average similarity coefficient of the i-th recognition area is defined as , is the average of the cosine values ​​of the angles of all representative points, and the calculation formula is as follows:

[0120] ;

[0121] Average similarity coefficient The value range of is [0,1]:

[0122] like , indicating that the normal vectors of all representative points are completely consistent with the direction of the light source, and the response consistency is strong;

[0123] like , it means that the normal vector is completely opposite to the direction of the light source, and the response consistency is weak;

[0124] like The closer it is to 1, the more consistent the normal vectors of all representative points are with the light source direction, and the stronger the response consistency;

[0125] like The closer it is to 0, the more opposite the normal vector is to the light source, and the weaker the response consistency.

[0126] Further explanation: when identifying rock mass structural surfaces, the wavelength range for each identification area i is and light intensity distribution Set the corresponding judgment thresholds respectively and ;

[0127] Threshold and Set through experiments and historical data; set judgment thresholds The value range of ,and The interval is In the present embodiment for , to correspond to the visible light range;

[0128] Threshold of light intensity distribution : Set the light intensity distribution threshold for , ensure that the recognition area meets the minimum lighting standards;

[0129] Calculate the ratio of the spectral characteristics actually recorded in each identification area to the corresponding judgment threshold:

[0130] The wavelength range and its judgment threshold The ratio of , the calculation formula is as follows:

[0131] ;

[0132] For the ratio of light intensity distribution to its threshold :

[0133] ;

[0134] in, represents the light intensity corresponding to the hth wavelength under the i-th identification area;

[0135] h∈{1, 2, …, H}, the larger the value of h, the larger the corresponding wavelength;

[0136] Combination , Perform analysis to generate the hue index of the i-th identified area , the calculation formula is as follows:

[0137] ;

[0138] in, is the correction factor, , Used to avoid the situation where the numerator is 0; The specific value is determined by the expert group system; and Also used to define the hue index The valid value range of is (0,1];

[0139] Set the Hue Index The valid value range of is (0,1];

[0140] like A value of 1 indicates that the ambient light is sufficient and the spectral characteristics meet expectations, the recognition effect is optimal, and the rock structure surface can be accurately identified;

[0141] like , indicating that the lighting condition and response are consistent, but it is necessary to fine-tune the light source position or increase the point cloud acquisition density; fine-tuning the light source position means adding auxiliary light sources;

[0142] like , indicating that the lighting conditions are poor, which will lead to poor recognition results. At this time, the lighting conditions need to be re-evaluated;

[0143] According to the hue index The following correction strategy is formulated:

[0144] like , indicating that the lighting conditions are good; continue the current analysis process, record the current lighting and spectral data, and ensure that the rock mass structural surface can be effectively identified at a rate of more than 90% without additional correction;

[0145] like , indicating that the lighting conditions are average, and the lighting direction and intensity are optimized.

[0146] Further explanation: the optimization of lighting direction includes:

[0147] Determine the representative points of each recognition area as , set and calculate the normal vector The average direction of ; and determine the light source direction of each recognition area i ;

[0148] calculate With light source direction Angle ;

[0149] like , indicating that the lighting direction is good and no adjustment is needed;

[0150] like , adjust the direction of the light source so that The reduction rate reached 10%-15%;

[0151] like , then you need to reposition the light source so that the angle is less than 30°, that is, adjust the angle by at least 30°;

[0152] For optimization of light intensity:

[0153] Measure the light intensity of each identification area at the current hth wavelength ;

[0154] If the ideal color of N2 representative points in each recognition area exceeds GL170% of the representative points When the displayed color is dark, the RGB value is lower than 100, and the auxiliary light source is used to increase the current light intensity. 20%;

[0155] If the ideal color of the N2 representative points in each recognition area exceeds GL160% of the representative points When the display color is normal, the RGB value is between 100-200, and the current light intensity is maintained;

[0156] If the ideal color of the N2 representative points in each recognition area exceeds GL150% of the representative points When the displayed color is too bright, and the RGB value is higher than 200, the current light intensity is reduced. 10%;

[0157] Among them, GL1>GL2>GL3, and GL1≥70, GL2≥59, GL3≥48; this embodiment is based on the experimental demonstration of the expert group system, and sets the values ​​of GL1, GL2, and GL3 to 70, 60, and 50 respectively;

[0158] Ensure that after optimization, each point has the ideal color The average RGB value is in the range of 120-200;

[0159] Increase the density of point cloud collection by at least 10% to ensure that the amount of rock structure information that can be captured under different lighting conditions is increased by more than 30%.

[0160] Further explanation: if , indicating that the lighting conditions are poor and need to be re-evaluated. Specific measures include:

[0161] Increase the brightness of the light source to ensure that the light intensity reaches or exceeds ;

[0162] Use auxiliary light sources to improve the current lighting uniformity by 80%. Auxiliary light sources include reflectors or diffusers.

[0163] The environmental lighting parameter data is re-collected and lighting correction is performed according to the ideal color model to ensure that the rock structure surface can be captured with an accuracy of more than 95% under ideal lighting conditions.

[0164] Further explanation: performing illumination correction according to the ideal color model specifically includes:

[0165] Get the ideal color of N2 representative points in each recognition area , and these ideal colors Take the average to get the average ideal color ;

[0166] Under actual lighting conditions, the color values ​​of N2 representative points in each recognition area are measured, and the color values ​​of the representative points in each recognition area are averaged to obtain the current color , calculate the average ideal color With current color The color difference between them is calculated as follows:

[0167] ;

[0168] like , then the color difference is acceptable and the current lighting state is maintained;

[0169] like , then perform illumination correction, with the help of auxiliary light source, to adjust the current light intensity Make a first-level adjustment; the first-level adjustment is to increase the current light intensity 15%;

[0170] like , then for the current light intensity Make secondary adjustments, which is to increase the current light intensity More than 25% of the total to ensure that the lighting uniformity reaches 80%;

[0171] After the illumination correction is performed, the goal is to ensure that the average RGB value of each recognition area after correction is consistent with the average ideal color. The difference between them is less than Mc, and in this embodiment, Mc=15% is set;

[0172] After the correction is completed, the captured rock structure information is re-evaluated to ensure that the effective recognition rate reaches more than 95%.

[0173] Further explanation: the acquisition and use of the rock mass structural surface identification model specifically includes:

[0174] The method for obtaining the rock mass structural surface recognition model is the same as the deep learning model in the authorized patent application with publication number CN117556703B, and will not be described in detail;

[0175] Arrange the ambient lighting parameters of each recognition area into a feature vector to form a feature set that can be used as input for the deep learning model;

[0176] Choose TensorFlow or PyTorch as the deep learning framework;

[0177] Build a convolutional neural network (CNN) model with multiple convolutional and pooling layers to extract features;

[0178] The extracted ambient lighting parameters are combined with the corresponding point cloud data to form an input data set, and numpy is used for data preprocessing;

[0179] Pair the annotated rock mass structural surface information with the extracted features to form a training set and a validation set, with a ratio of 80% training and 20% validation;

[0180] The Adam optimizer was used for model training, the learning rate was set to 0.001, 50 epochs were trained, and the batch size was set to 32; thus, a trained rock mass structural surface recognition model was formed;

[0181] Monitor the loss function and validation set accuracy in real time to prevent overfitting;

[0182] After training is completed, use the model saving function provided by the framework to save the model parameters to a file;

[0183] The model performance is evaluated on the validation set to generate indicators such as accuracy, recall, and F1 value to ensure the effectiveness of the model.

[0184] Furthermore, new rock slope point cloud data are collected to ensure that the data structure is the same as that during training;

[0185] Collect the corresponding ambient lighting parameters and format them;

[0186] Perform the same preprocessing steps on the data to be predicted, including denoising and formatting, to ensure data consistency;

[0187] According to the feature extraction method in step S5, the ambient light parameters of the area to be predicted are calculated to form a feature vector;

[0188] Input the point cloud data of the area to be predicted and the calculated ambient lighting parameters into the trained deep learning model;

[0189] Deep learning model outputs include the following:

[0190] Position: predict the coordinates of the structural surface in three-dimensional space;

[0191] Features: Identify the morphological features of the structural surface, including crack width, direction, and inclination;

[0192] Occurrence information: output the relative position and interaction relationship of the structural surface in the geological layer;

[0193] Perform geometric correction on the model output to ensure that the output structural surface position is consistent with the actual situation;

[0194] Compare with actual observation data, or ask geological experts to cross-verify to ensure the accuracy and rationality of the output information.

[0195] Embodiment 2:

[0196] See also Figure 2 , a rock mass structural surface recognition device based on digital image recognition technology, the device is used to execute the rock mass structural surface recognition method based on digital image recognition technology, comprising:

[0197] Region division and parameter acquisition module: used to obtain a three-dimensional point cloud model composed of rock slope point cloud data, and divide the three-dimensional point cloud model into regions to form M identification regions; and collect the ambient light parameters on the M identification regions, including the light source direction and spectral characteristics; each point in each identification region contains its corresponding three-dimensional coordinates and original color information;

[0198] Ideal color determination module: used to calculate the ideal colors of several points in the point cloud corresponding to each recognition area based on the original color information using the illumination model;

[0199] Average similarity coefficient generation module: used to determine several points from the corresponding point cloud as representative points in each identification area; by analyzing the angle between the normal vector of these representative points and the corresponding light source direction, and comprehensively calculating the average similarity coefficient of these representative points, the response consistency of different identification areas on the rock slope to the change of illumination is evaluated;

[0200] Tone index generation module: used to assign a judgment threshold to the spectral characteristics of each recognition area, and to express the spectral characteristics actually recorded in each recognition area and the corresponding judgment threshold in a ratio manner, and to combine and analyze the ratio with the corresponding average similarity coefficient to generate a tone index for providing a correction strategy for the corresponding lighting conditions of each recognition area;

[0201] Re-collecting the ambient lighting parameter data of each recognition area to update the originally constructed three-dimensional point cloud model, thereby obtaining a corrected three-dimensional point cloud model;

[0202] Rock mass structural surface recognition model construction module: extract the ambient lighting parameters of each recognition area from the corrected three-dimensional point cloud model, use the deep learning model to group the point cloud data and the extracted ambient lighting parameters in each recognition area, and build a data set based on the grouping results. The data set contains rock mass structural surface information and ambient lighting parameters, where the point cloud data and ambient lighting parameters are used as the input of the model, and the rock mass structural surface information is used as the output of the model. Finally, the data set is used for training to obtain a rock mass structural surface recognition model;

[0203] Rock mass structural surface information output module: used to input the point cloud data of the rock slope to be predicted and its corresponding ambient lighting parameters into the trained deep learning model. The model will output the rock mass structural surface information including location, characteristics and occurrence information.

[0204] Embodiment three:

[0205] A device comprises a memory, a processor and a computer program stored in the memory and executable on the processor. When the processor executes the computer program, the steps of the rock structure surface recognition method based on digital image recognition technology are implemented.

[0206] Embodiment 4:

[0207] A readable storage medium stores a computer program, and when the computer program is executed by a processor, the steps of the rock structure surface recognition method based on digital image recognition technology are implemented.

[0208] The above formulas are all dimensionless and numerical calculations. The formula is a formula for the most recent real situation obtained by collecting a large amount of data and performing software simulation. The preset parameters in the formula are set by technicians in this field according to actual conditions.

[0209] The above embodiments may be implemented in whole or in part by software, hardware, firmware or any other combination thereof. When implemented by software, the above embodiments may be implemented in whole or in part in the form of a computer program product. Those skilled in the art may appreciate that the units and algorithm steps of each example described in conjunction with the embodiments disclosed herein may be implemented by electronic hardware, or a combination of computer software and electronic hardware. Whether these functions are performed by hardware or software methods depends on the specific application and design constraints of the technical solution.

[0210] 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, and may be located in one place or distributed on multiple network units. Some or all of the units may be selected according to actual needs to achieve the purpose of the solution of this embodiment.

[0211] The above description is only a specific implementation manner of the present application, but the protection scope of the present application is not limited thereto. Any technician familiar with the technical field can easily think of changes or substitutions within the technical scope disclosed in the present application, which should be included in the protection scope of the present application.

Claims

1. A rock mass structural surface recognition method based on digital image recognition technology, characterized in that: The specific steps include: Step S1: obtaining a three-dimensional point cloud model composed of rock slope point cloud data, and performing region division on the three-dimensional point cloud model to form M identification regions; Collect the ambient lighting parameters corresponding to the identification area at the rock slope, which include the light source direction and spectral characteristics; and each point in each identification area contains its corresponding three-dimensional coordinates and original color information; Step S2: Based on the original color information, the illumination model is used to calculate the ideal color of the points in the point cloud corresponding to each recognition area; Step S3: In each recognition area, a number of points are determined from the corresponding point cloud as representative points; by analyzing the angles between the normal vectors of these representative points and the corresponding light source directions, the average similarity coefficients of these representative points are comprehensively calculated; Step S4: assigning a judgment threshold to the spectral characteristics of each identification area, and calculating the ratio of the spectral characteristics actually recorded in each identification area to the corresponding judgment threshold, and combining the ratio with the corresponding average similarity coefficient for analysis to generate a hue index for providing a correction strategy for the corresponding lighting conditions of each identification area; On the basis of the illumination conditions being adjusted through the correction strategy, the ambient illumination parameter data of each recognition area is re-collected to update the originally constructed three-dimensional point cloud model, thereby obtaining a corrected three-dimensional point cloud model; Step S5: extracting the ambient lighting parameters of each identification area from the corrected three-dimensional point cloud model, using the deep learning model to group the point cloud data and the extracted ambient lighting parameters in each identification area by structural surface, and constructing a data set according to the grouping results, the data set containing rock structural surface information and ambient lighting parameters, wherein the point cloud data and ambient lighting parameters are used as inputs of the model, and the rock structural surface information is used as output of the model, and finally, the data set is used for training to obtain a rock structural surface recognition model; Step S6: Input the point cloud data of the rock slope to be predicted and its corresponding ambient lighting parameters into the trained deep learning model, and the model will output the rock structure surface information including the location, characteristics and occurrence information.

2. The rock mass structural surface recognition method based on digital image recognition technology according to claim 1 is characterized in that: The voxel grid method is used to divide the 3D point cloud data into M recognition areas, and the M recognition areas are represented as {1, 2, ..., i, ..., M}, where i represents the i-th recognition area and i∈{1, 2, ..., M}, and the goal of the division is to retain more than N1 points of the point cloud in each recognition area; In the rock mass side wave corresponding to each identification area i, a photometer is used to collect the ambient light parameters, and the ambient light parameters of the i-th identification area, including the light source direction and spectral characteristics, are marked in sequence as ; The light source direction The azimuth and elevation angles of the light source are used to numerically represent the direction of the light source. Represented as a vector ;in, represents the azimuth of the i-th recognition area; represents the pitch angle of the i-th recognition area; Spectral characteristics of the i-th identification area The spectral characteristics are expressed by the wavelength range and light intensity distribution of the spectrum. Specifically expressed as ;in, represent the wavelength range and light intensity distribution of the i-th identification area respectively; The wavelength range of the i-th identification area is expressed as ;in Respectively represent the minimum wavelength and maximum wavelength of the i-th identification area; The light intensity distribution of the i-th recognition area is expressed as follows: ； in, Represents the light intensity corresponding to the hth wavelength under the i-th identification area ; And h∈{1, 2, …, H}, the larger the value of h is, the larger the corresponding wavelength is.

3. The rock mass structural surface recognition method based on digital image recognition technology according to claim 2 is characterized in that: The three-dimensional coordinates of the points in the point cloud corresponding to the i-th recognition area are expressed as ; in, is the three-dimensional coordinate representation of the jth point in the point cloud corresponding to the i-th recognition area; for The corresponding three-dimensional coordinates; N1 is the total number of points in the point cloud corresponding to the i-th recognition area; The original color information of the points in the point cloud corresponding to each recognition area is represented by the values ​​of the RGB three-color channels. The original color information is represented in the following way: Use RGB color mode to represent the original color information; the original color information of the i-th recognition area is represented as ,in They represent the intensity of the red, green and blue channels of the jth point in the point cloud corresponding to the i-th recognition area, respectively. Represents the original color information of the jth point in the i-th recognition area.

4. The rock mass structural surface recognition method based on digital image recognition technology according to claim 3 is characterized in that: Select the Lambert illumination model as the illumination model for calculating the color of the point cloud. For the i-th recognition area, obtain the three-dimensional coordinate representation of the j-th point in its corresponding point cloud. ; Use the principal component analysis (PCA) method in the normal estimation algorithm to calculate the normal information of the point cloud in the i-th recognition area; associate the normal information with the three-dimensional coordinates of each point to obtain the normal vector of the j-th point in the i-th recognition area. ; According to the collected ambient lighting parameters, extract the light source direction of the current recognition area i and spectral characteristics ; Specifically extract the azimuth of the light source direction and pitch angle , and the wavelength range of the spectral characteristics and light intensity distribution ; Get the spectral reflectance curve of the current identification area i, find the wavelength corresponding to the maximum reflectance from the spectral reflectance curve, and record it as the optimal wavelength ; In the light intensity distribution collection , calculation and optimal wavelength The Euclidean distance is then filtered out to find the light intensity corresponding to the minimum Euclidean distance. , h1∈{1,2,…,H}; Calculate the ideal color for each point, including: For each point, calculate its ideal color using the following formula : ； in, Represented as three-dimensional coordinates The original color information of the corresponding points; is the diffuse reflectance, ; is the dot product of the incident light source direction and the normal vector.

5. The rock mass structural surface recognition method based on digital image recognition technology according to claim 4 is characterized in that: The average similarity coefficient and the acquisition of the hue index include: 1.1) Select representative points: In each recognition area i, N2 points are randomly selected from the point cloud data as representative points, and the normal vector of the j1th representative point selected in the i-th recognition area is marked as , , N2<N1, and N2≥5; The three-dimensional coordinates of the j1th representative point selected in the i-th recognition area are expressed as , the representative points are referred to as ; 1.2) Calculation of normal vector: Use the normal estimation algorithm based on principal component analysis PCA to calculate the normal vector of the j1th representative point in the i-th recognition area ; 1.3) Calculate the angle between the normal vector and the light source direction: Direct the light source Convert to unit vector : ； For each representative point , calculate the angle between the normal vector and the unit light source direction : ； 1.4) Calculate the average similarity coefficient: The average similarity coefficient of the i-th recognition area is defined as , is the average of the cosine values ​​of the angles of all representative points, and the calculation formula is as follows: ； Average similarity coefficient The value range of is [0,1]: like , indicating that the normal vectors of all representative points are completely consistent with the direction of the light source, and the response consistency is strong; like , it means that the normal vector is completely opposite to the direction of the light source, and the response consistency is weak; like The closer it is to 1, the more consistent the normal vectors of all representative points are with the light source direction, and the stronger the response consistency; like The closer it is to 0, the more opposite the normal vector is to the light source, and the weaker the response consistency.

6. The rock mass structural surface recognition method based on digital image recognition technology according to claim 5 is characterized in that: When identifying rock mass structural surfaces, the wavelength range for each identification area i is and light intensity distribution Set the corresponding judgment thresholds respectively and ; Calculate the ratio of the spectral characteristics actually recorded in each identification area to the corresponding judgment threshold: Setting judgment threshold The value range of ,and The interval is within the scope; The wavelength range and its judgment threshold The ratio of , the calculation formula is as follows: ； For the ratio of light intensity distribution to its threshold : ； in, represents the light intensity corresponding to the hth wavelength under the i-th identification area; Combination , Perform analysis to generate the hue index of the i-th identified area , the calculation formula is as follows: ； in, is the correction factor, , Used to avoid the situation where the numerator is 0; Set the Hue Index The valid value range of is (0,1]; like A value of 1 indicates that the ambient light is sufficient and the spectral characteristics meet expectations, the recognition effect is optimal, and the rock structure surface can be accurately identified; According to the hue index The following correction strategy is formulated: like , indicating good lighting conditions; like , indicating that the lighting conditions are average, optimizing the lighting direction and light intensity; like , indicating that the lighting conditions are poor and the lighting conditions need to be re-evaluated.

7. The rock mass structural surface recognition method based on digital image recognition technology according to claim 6 is characterized in that: Optimization of lighting direction, including: Determine the representative points of each recognition area as , set and calculate the normal vector The average direction of ; and determine the light source direction of each recognition area i ; calculate With light source direction Angle ; like , indicating that the lighting direction is good and no adjustment is needed; like , adjust the direction of the light source; like , then the light source needs to be repositioned; Optimization of light intensity includes: Measure the light intensity of each identification area at the current hth wavelength ; If the ideal color of the N2 representative points in each recognition area exceeds GL1%, When the displayed color is dark, the current light intensity is increased through auxiliary light sources; If the ideal color of the N2 representative points in each recognition area exceeds GL2%, When the display color is normal, the current light intensity is maintained; If the ideal color of the N2 representative points in each recognition area exceeds GL3%, When the displayed color is too bright, reduce the current light intensity; Among them, GL1>GL2>GL3; and GL1≥70, GL2≥59, GL3≥48; For re-evaluating lighting conditions, specific measures include: Use auxiliary light sources to improve the uniformity of current lighting. Auxiliary light sources include reflectors or diffusers. Re-collect the ambient lighting parameter data and perform lighting correction according to the ideal color model; Perform lighting correction based on the ideal color model, including: Get the ideal color of N2 representative points in each recognition area , and these ideal colors Take the average to get the average ideal color ; Under actual lighting conditions, the color values ​​of N2 representative points in each recognition area are measured, and the color values ​​of the representative points in each recognition area are averaged to obtain the current color , calculate the average ideal color With current color The color difference between them is calculated as follows: ； like , then the color difference is acceptable and the current lighting state is maintained; like , then perform illumination correction, with the help of auxiliary light source, to adjust the current light intensity Make a first-level adjustment; like , then for the current light intensity Perform a secondary adjustment; the light intensity increased by the primary adjustment is lower than that of the secondary adjustment; After the illumination correction is performed, the goal is to ensure that the average RGB value of each recognition area after correction is consistent with the average ideal color. The difference between them is less than Mc; After the correction is completed, the captured rock structure information is re-evaluated to ensure that the effective recognition rate reaches more than 95%.

8. A rock mass structural surface recognition device based on digital image recognition technology, characterized in that: The device is used to execute the rock structure surface recognition method based on digital image recognition technology according to any one of claims 1 to 7, comprising: Region division and parameter acquisition module: used to obtain a three-dimensional point cloud model composed of rock slope point cloud data, and divide the three-dimensional point cloud model into regions to form M identification regions; and collect the ambient light parameters on the M identification regions, including the light source direction and spectral characteristics; each point in each identification region contains its corresponding three-dimensional coordinates and original color information; Ideal color determination module: used to calculate the ideal colors of several points in the point cloud corresponding to each recognition area based on the original color information using the illumination model; Average similarity coefficient generation module: used to determine several points from the corresponding point cloud as representative points in each identification area; by analyzing the angle between the normal vector of these representative points and the corresponding light source direction, and comprehensively calculating the average similarity coefficient of these representative points, the response consistency of different identification areas on the rock slope to the change of illumination is evaluated; Tone index generation module: used to assign a judgment threshold to the spectral characteristics of each recognition area, and to express the spectral characteristics actually recorded in each recognition area and the corresponding judgment threshold in a ratio manner, and to combine and analyze the ratio with the corresponding average similarity coefficient to generate a tone index for providing a correction strategy for the corresponding lighting conditions of each recognition area; Re-collecting the ambient lighting parameter data of each recognition area to update the originally constructed three-dimensional point cloud model, thereby obtaining a corrected three-dimensional point cloud model; Rock mass structural surface recognition model construction module: extract the ambient lighting parameters of each recognition area from the corrected three-dimensional point cloud model, use the deep learning model to group the point cloud data and the extracted ambient lighting parameters in each recognition area, and build a data set based on the grouping results. The data set contains rock mass structural surface information and ambient lighting parameters, where the point cloud data and ambient lighting parameters are used as the input of the model, and the rock mass structural surface information is used as the output of the model. Finally, the data set is used for training to obtain a rock mass structural surface recognition model; Rock mass structural surface information output module: used to input the point cloud data of the rock slope to be predicted and its corresponding ambient lighting parameters into the trained deep learning model. The model will output the rock mass structural surface information including location, characteristics and occurrence information.

9. A device comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, characterized in that: When the processor executes the computer program, the steps of the rock structure surface identification method based on digital image recognition technology as described in any one of claims 1 to 7 are implemented.

10. A readable storage medium storing a computer program, characterized in that: When the computer program is executed by a processor, the steps of the rock structure surface recognition method based on digital image recognition technology as claimed in any one of claims 1 to 7 are implemented.

Images