A safety monitoring method and system for rock slope

Abstract

The present application relates to the technical field of rock monitoring, in particular to a safety monitoring method and system for rock slope, which generates an abnormal focusing area through gradient change and density mutation, builds a basic entry for early risk identification, strengthens the positioning accuracy of crack sensitive section, enhances the identification accuracy and dynamic intervention ability of potential risk area of rock slope, extracts the linear features of boundary response through convolutional neural network, realizes the identification and reconstruction of main extension path of crack, and combines direction sorting and structure continuity analysis to complete the closed fitting and stable connection node labeling of crack path, provides traceable structure skeleton for subsequent trend analysis, realizes dynamic modeling of crack deflection evolution through comparison and identification of displacement direction and angle mutation point, enables to build closed path grid in crack direction reversal area, and realizes dynamic response layout of crack control structure.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of rock monitoring, in particular to a safety monitoring method and system for rock slope. BACKGROUND

[0002] The technical field of rock monitoring refers to a method of detecting, analyzing and evaluating the state of rock and constituent bodies through various physical, chemical and mechanical means. It is widely used in geological exploration, mineral exploitation, civil engineering and slope stability analysis fields. The goal is to obtain real-time information about the changes inside rock bodies, identify potential risks and hidden dangers, improve engineering safety and avoid catastrophic accidents. Common monitoring methods include sensor technology, seismic wave testing, temperature monitoring and stress-strain detection. Dynamic data about rock structure can be obtained through monitoring methods for analysis and early warning.

[0003] A safety monitoring method for rock slope aims to monitor the stability, stress-strain condition and displacement change parameters of rock slope in real time. The purpose is to evaluate the safety state of rock slope through regular and continuous monitoring data, so as to discover potential disaster risks and structural changes in time, and take effective prevention and repair measures. The method can effectively avoid landslides, collapses and engineering accidents caused by rock slope instability, protect life and property safety, and provide scientific basis for related engineering construction and design.

[0004] The existing technology mainly uses sensor data collection, stress-strain curve analysis and structural crack detection physical monitoring as the main way, which has limitations in spatial coverage and structural depth interpretation. Image data is not fully utilized and is only used as an auxiliary identification means for surface crack deformation. There is a lack of system linkage with the evolution process of structural behavior. Sensor collection is limited by installation location and physical interference, which may result in data loss and abnormal fluctuations in complex terrain and superficial weathering layer conditions. It is difficult to construct continuous and reliable trend trajectories. The multi-point monitoring results lack a global integration mechanism, making it difficult to reflect the structural response characteristics under the joint action of multiple variables. In the identification of crack development direction and evaluation of regional instability state of slope, the data is isolated and the judgment is delayed. SUMMARY

[0005] The purpose of the present application is to solve the problems existing in the prior art and to provide a safety monitoring method and system for rock slope.

[0006] In order to achieve the above purpose, the present application adopts the following technical scheme: a safety monitoring method for rock slope, comprising the following steps:

[0007] S1: By the linearity structure continuity, the regional contour edge curvature variation degree and the boundary direction extension characteristic shown in the rock slope surface image, using difference ratio and gradient detection, the boundary closedness change section is screened out, and the density mutation area is marked, and the structure abnormal focusing partition is obtained;

[0008] S2: Based on the structure abnormal focusing partition, after using the convolutional neural network to extract the boundary response structure linearity feature, the contour line extension direction is sorted and the cross section is excluded, the crack main extension path is located, the continuity is verified and the closed structure is fitted, the stable connection section is marked, and the crack expansion main path layer is obtained;

[0009] S3: Based on the crack expansion main path layer, the displacement direction and the extension angle change value of the multi-time sequence path are compared, the offset stable section is identified and the path expansion trend is judged, the continuous mutation node is extracted, and the crack expansion trend trajectory set is generated;

[0010] S4: Based on the crack expansion trend trajectory set, the multi-variable direction feature of the monitoring point is extracted by combining the artificial neural network, the rock mass monitoring point data is matched, the relationship between the trajectory path and the inclination and stress change direction is judged, the variable direction synchronization section and the path offset aggregation area are located, and the instability induced area set is constructed;

[0011] S5: Based on the instability induced area set, the path splitting node angle mutation value is counted and the clustering area is divided, the closed control network is constructed through the continuous splitting path, the direction reversal closed structure is formed, and the dynamic crack control grid structure is obtained.

[0012] As a further scheme of the application, the structure abnormal focusing partition includes boundary closedness change area, pixel density mutation area and contour shape change area, the crack expansion main path layer includes main extension path section, closed structure section and continuous connection node, the crack expansion trend trajectory set includes path displacement direction sequence, extension angle change sequence and mutation node sequence, the instability induced area set includes trajectory variable direction synchronization section, path offset aggregation area and direction relationship coupling area, and the dynamic crack control grid structure includes direction reversal path section, closed connection node and control grid boundary unit.

[0013] As a further scheme of the application, the specific steps for obtaining the structure abnormal focusing partition are:

[0014] By the linearity structure continuity, the regional contour edge curvature variation degree and the boundary direction extension characteristic shown in the rock slope surface image, the edge region horizontal and vertical gray gradient value is extracted and the gray jump intensity of each boundary section is calculated, the gradient difference value of the continuous section is located, and the boundary structure mutation segment set is generated;

[0015] Based on the boundary structure mutation fragment set, the transverse and longitudinal pixel distribution density in the image region and the average deviation value of the density aggregation region are calculated, the range of the high-density block segment boundary is labeled, and the structure abnormal focusing partition is obtained.

[0016] As a further scheme of the present application, the specific steps for obtaining the crack propagation main path layer are:

[0017] Based on the structure abnormal focusing partition, a linear contour response intensity value atlas in the boundary image region is extracted using a convolutional neural network, the direction difference value of the coordinates of the two ends of the boundary line segment is extracted and the direction angle amplitude is sorted, the crossing line segments with an angle greater than a set limit value are screened out and the directionally continuous line segments are retained, and a unidirectional continuous contour set is generated.

[0018] Based on the unidirectional continuous contour set, the distance difference between the end points of adjacent contour paths is calculated and the synchronicity of the direction angle is judged, similar direction path segments are connected and the interrupted positions are fitted and completed, and a crack propagation path chain group is generated.

[0019] Based on the crack propagation path chain group, the closed edge jump difference between the path connection nodes is evaluated and the curvature continuity is judged, the stable path unit of the continuous segment from the starting point to the ending point is marked, and the crack propagation main path layer is obtained.

[0020] As a further scheme of the present application, the specific execution process of the convolutional neural network is that the structure abnormal image region in the focusing partition is focused layer by layer using a multi-layer convolution structure, the local region gray value is extracted using a fixed size convolution kernel sliding on the image matrix, the feature atlas group is generated and then connected to the activation operation, the linear edge response is retained, the multi-scale convolution kernel is superimposed to cover different scale contour structures, the convolution output is stacked and normalized, the continuous boundary direction change intensity value matrix in the image is extracted, and the region contour peak response path is extracted according to the linear response amplitude sorting in the extraction matrix.

[0021] As a further scheme of the present application, the specific steps for generating the crack propagation trend trajectory set are:

[0022] Based on the crack propagation main path layer, the horizontal and vertical pixel coordinate difference of the same path node in the continuous image is calculated, a direction vector sequence is constructed, the angle change amplitude between adjacent frames is extracted, and a path propagation direction change group is generated.

[0023] Based on the path propagation direction change group, the continuous segment length of the mutation fragment in the direction sequence is counted and the peak value offset fragment position is extracted, the path direction mutation point set is marked, and the crack propagation trend trajectory set is obtained.

[0024] As a further scheme of the present application, the specific process for constructing the instability induction region set is:

[0025] Based on the crack propagation trend trajectory set, the three-dimensional coordinate difference between each trajectory node and the rock mass monitoring point is calculated, a space distance threshold is set, the monitoring points falling within the range are screened, and a time value sequence is extracted, and a path monitoring matching data set is generated;

[0026] Based on the path monitoring matching data set, a multivariate directional feature mode is identified by using an artificial neural network, a numerical difference between a path direction and an inclination change direction is compared, a consistent segment is marked, a path angle turning trend is matched, and a direction consistent section is screened out, and a trajectory synchronous offset section group is obtained.

[0027] Based on the trajectory synchronous offset section group, the horizontal and vertical coordinate density of the position in each group node is counted, a local aggregation grid is constructed, a region with an overlap rate greater than a set threshold is extracted, and a boundary is closed, and a set of instability inducing regions is obtained.

[0028] As a further scheme of the present application, the specific execution process of the artificial neural network is as follows: the path direction, the inclination change direction, the stress change amplitude, and the time sequence interval in the path monitoring matching data set are taken as input variables, a multilayer neural structure is constructed, the weights of each node are initialized and the activation threshold is allocated, the input layer is weighted and transferred to the hidden layer and the activated nodes are activated, the direction change response mode under each variable combination is extracted, and then the output is transferred to the output layer for node state normalization, and a directional feature label matrix corresponding to the multivariate combination is generated.

[0029] As a further scheme of the present application, the specific steps for constructing the set of instability inducing regions are as follows:

[0030] Based on the set of instability inducing regions, the path splitting node angle change value sequence is extracted, the continuous segment angle difference increasing trend is identified, the position coordinate density is calculated, the aggregation boundary section is screened, and a splitting angle offset clustering segment group is generated.

[0031] Based on the splitting angle offset clustering segment group, the reverse segment angle sequence of the path splitting starting point and the ending point in each group is determined, and a set of reverse closed path line segments is extracted, a boundary line network is generated, and a closed grid is generated to obtain a dynamic crack control grid structure.

[0032] A safety monitoring system for rock slope, which is used to execute the safety monitoring method for rock slope described above, the system comprises:

[0033] An image feature extraction module: through the linearity structure continuity, the regional contour edge curvature change degree and the boundary direction extension characteristic shown in the rock slope surface image, the boundary closed segment gray level jump is extracted, the density aggregation area difference is analyzed, and a structure abnormal focusing partition is obtained.

[0034] Path recognition construction module: based on the structure abnormal focus partition, the linear profile response atlas in the boundary area is extracted by using a convolutional neural network, the profile line segment direction is sorted and the connection relationship is recognized, the direction deviated segment is screened out and the continuous segment is connected, and a crack propagation main path layer is obtained;

[0035] Trend analysis operation module: based on the crack propagation main path layer, the direction angle displacement sequence of the path node in the multi-time sequence image is constructed, the path direction change amplitude and the fluctuation region length between continuous frames are calculated, the angle mutation node position is extracted, and a crack propagation trend trajectory set is generated;

[0036] State association judgment module: based on the crack propagation trend trajectory set, the multi-variable direction mode of the monitoring point is recognized in combination with an artificial neural network, the corresponding relationship of the inclination direction, the stress direction and the path trend is compared, the direction synchronous node segment and the displacement aggregation segment are marked, and a loss of stability induced region set is constructed;

[0037] Grid structure formation module: based on the loss of stability induced region set, the direction mutation value of the path splitting point is counted and a clustering distribution area is generated, the continuous direction mutation path is reconstructed and the closed boundary line and the grid block are divided, and a dynamic crack control grid structure is obtained.

[0038] Compared with the prior art, the advantages and positive effects of the present application are that:

[0039] 1. In the present application, the abnormal focus area is generated by gradient change and density mutation, the basis entry for early risk identification is constructed, the positioning accuracy of the crack sensitive section is strengthened, and the identification accuracy and dynamic intervention ability of the potential risk area of the rock slope are enhanced;

[0040] 2. In the present application, the linear features of the boundary response are extracted by a convolutional neural network, the identification and reconstruction of the main crack propagation path are realized, and the closed fitting and stable connection node labeling of the crack path are completed by combining direction sorting and structure continuity analysis, which provides a traceable structure skeleton for subsequent trend analysis;

[0041] 3. In the present application, the displacement direction and the angle mutation point are compared and recognized, the dynamic modeling of the crack displacement evolution is realized, the closed path grid can be constructed in the crack direction reversal area, and the dynamic response layout of the crack control structure is realized. BRIEF DESCRIPTION OF DRAWINGS

[0042] Figure 1 It is a work flow diagram of the present application;

[0043] Figure 2 It is a system flow chart of the present application. DETAILED DESCRIPTION

[0044] In order to make the objects, technical solutions and advantages of the present application clearer, the present application will be further described in detail below with reference to the drawings and embodiments. It should be understood that the specific embodiments described herein are only used to explain the present application and not to limit the present application.

[0045] Embodiment one

[0046] Please refer to Figure 1 The present application provides a technical solution: a safety monitoring method for rock slope, comprising the following steps:

[0047] S1: Through the linearity structure continuity, regional contour edge curvature change degree and boundary direction extension characteristics shown in the rock slope surface image, using difference comparison and gradient detection, the boundary closedness change section is screened out, and the density mutation area is marked, and the structure abnormal focusing partition is obtained;

[0048] S2: Based on the structure abnormal focusing partition, after using convolutional neural network to extract the boundary response structure linearity characteristics, the contour line extension direction is sorted and the cross section is excluded, the crack main extension path is located, the continuity is verified and the closed structure is fitted, the stable connection section is marked, and the crack expansion main path layer is obtained;

[0049] S3: Based on the crack expansion main path layer, comparing the displacement direction and extension angle change value of the multi-time sequence path, identifying the offset stable section and judging the path expansion trend, extracting the continuous mutation node, and generating the crack expansion trend trajectory set;

[0050] S4: Based on the crack expansion trend trajectory set, combining artificial neural network to extract the multivariate direction characteristics of the monitoring point, matching the rock mass monitoring point data, judging the relationship between the trajectory path and the inclination, stress change direction, locating the variable direction synchronization section and the path offset aggregation area, and constructing the instability induced area set;

[0051] S5: Based on the instability induced area set, the angle mutation value of the path splitting node is counted and the clustering area is divided, the closed control network is constructed through the continuous splitting path, the direction reversal closed structure is formed, and the dynamic crack control grid structure is obtained.

[0052] The structure abnormal focusing partition includes the boundary closedness change area, the pixel density mutation area and the contour shape change area, the crack expansion main path layer includes the main extension path section, the closed structure section and the continuous connection node, the crack expansion trend trajectory set includes the path displacement direction sequence, the extension angle change sequence and the mutation node sequence, the instability induced area set includes the trajectory variable direction synchronization section, the path offset aggregation area and the direction relationship coupling area, and the dynamic crack control grid structure includes the direction reversal path section, the closed connection node and the control grid boundary unit.

[0053] The specific steps for obtaining the structure abnormal focusing partition are:

[0054] By the linearity structure continuity, the regional contour edge curvature variation degree and the boundary direction extension characteristic shown in the rock slope surface image, the edge region horizontal and vertical gray scale gradient values are extracted and the gray scale jump intensity of each boundary segment is calculated, the gradient difference value of the continuous segment is located, and the boundary structure mutation segment set is generated;

[0055] Based on the boundary structure mutation segment set, the horizontal and vertical pixel distribution density in the image region and the average deviation value of the density aggregation region are calculated, the range of the high density block segment boundary is marked, and the structure abnormal focusing partition is obtained;

[0056] Based on the linearity structure continuity, the regional contour edge curvature variation degree and the boundary direction extension characteristic shown in the rock slope surface image, the original image is converted to gray scale, the sliding window size is set to the standardized size, the boundary gradient of the horizontal and vertical channels is scanned respectively, the image pixel points are traversed by fixed step, the boundary jump intensity distribution map is constructed by using the neighborhood gray scale intensity variation, the continuous boundary segment of the high intensity variation region is extracted in the region connection mode, the background noise area of the jump dense segment is excluded, and the closed jump structure segment is screened according to the linear distribution direction, and the boundary structure mutation segment set is generated;

[0057] Based on the boundary structure mutation segment set, the equidistant grid division mechanism is established in the horizontal and vertical directions of the image, the window grid size range is set to be proportional to the image size, the pixel number in each grid region is counted to construct a two-dimensional density layer, the average density of each grid block is compared and analyzed, the concentrated region with density exceeding the preset threshold is identified, the boundary envelope of the continuous high density region is delimited based on the number and spatial distribution structure of the concentrated region, the spatial position of the density mutation block is recorded, the abnormal marker index layer is established, and is superimposed on the original image for verification, and the structure abnormal focusing partition is obtained.

[0058] The specific steps for obtaining the crack propagation main path layer are as follows:

[0059] Based on the structure abnormal focusing partition, the convolutional neural network is used to extract the linear contour response intensity value atlas in the boundary image region, the direction angle amplitude of the coordinate difference direction of the two ends of the boundary line segment is sorted, the intersecting line segments with the direction angle greater than the set limit value are screened out, and the unidirectional continuous contour set is generated;

[0060] Based on the unidirectional continuous contour set, the distance difference between the end points of adjacent contour paths is calculated and the direction angle synchronism is judged, the similar direction path segment is extended and connected, and the interrupted position is fitted and completed, and the crack propagation path chain group is generated.

[0061] Based on the crack propagation path chain group, the closing edge jump difference between the path connection nodes is evaluated, the curvature continuity is judged, the stable path unit of the starting and ending point continuous paragraph is marked, and the crack propagation main path layer is obtained;

[0062] Based on the structural anomaly focusing partition, a convolutional neural network is used to extract the linear contour response intensity value atlas in the boundary image region, the image region is standardized and then input into the network model, the model is composed of three convolutional layers and two pooling layers, the first convolutional layer has a convolution kernel size of three by three, a stride of one, a padding method of the same, and uses ReLU as the activation function, the pooling layer uses maximum pooling with a window size of two by two, then the second convolutional layer is connected and configured repeatedly, and after the third convolutional layer outputs, the feature response atlas is output through the fully connected layer, the image input size is preset to 256 by 256 pixels, the batch size is set to 16, the number of training rounds is set to 50, the learning rate is set to 0.001, the edge response intensity of each pixel in the image is scored during processing, and the linear boundary direction significant region is extracted according to the convolution result, the direction angle amplitude of the coordinate difference of the two ends of the boundary line segment is sorted, the cross line segment with an angle greater than the set limit value is filtered out, and the one-way continuous contour set is generated;

[0063] Based on the one-way continuous contour set, the distance difference between the end points of adjacent contour paths is calculated and the synchronism of the direction angle is judged, first, the end point coordinates of each path in the path set are extracted, and a coordinate comparison matrix is constructed according to the path number, the Euclidean distance between the start and end points of each pair of adjacent path segments is calculated, and the end point difference threshold is set to 20 pixels, the direction angle of the path pair below the threshold is calculated, the direction angle is calculated by connecting the start and end points of the line segment to form a vector, the direction angle synchronism judgment range is set to less than 20 degrees, the paths that meet the range are connected, the connection order is sorted according to the path number, the order is extended, the positions with path breaks are completed, the fitting angle is generated by fitting the average value of the direction angles of the front and rear path segments, and the fitting path segment is generated, the fitting segment length is set to the average length of the front and rear paths, the fitting segment is connected to the front and rear path nodes and then renumbered and added to the path set, and the crack propagation path chain group is generated;

[0064] Based on the crack propagation path chain group, the closed edge jump difference between the path connection nodes is evaluated and the curvature continuity is judged. The edge jump value of each node in the path node sequence is calculated by the gray intensity difference of adjacent nodes. The local jump identification window is set to five continuous nodes. The standard deviation of the jump value in each window is calculated and the maximum value is recorded. The path segment higher than the set jump threshold is marked. Five points at both ends of the path segment are extracted for curvature judgment. The curvature continuity judgment is based on the angle change rate being less than 30% of the average angle change rate of each segment. The path segment that meets the condition is marked as a curvature stable segment. The start and end point nodes of the path segment that meets the conditions of low amplitude edge jump and curvature stability are confirmed. The continuous segment number in the path connection is marked. The crack propagation main path layer is obtained.

[0065] The specific execution process of the convolutional neural network is that a multi-layer convolutional structure is used to focus on the structure abnormal image area in the partition layer by layer, a fixed size convolution kernel is used to slide on the image matrix to extract the local area gray value, the feature map group is generated, the activation operation is connected, the linear edge response is retained, the multi-scale convolution kernel is superimposed to cover different scale contour structures, the convolution output is stacked and normalized, the continuous boundary direction change intensity value matrix in the image is extracted, and the region contour peak response path is extracted according to the linear response amplitude in the extraction matrix.

[0066] The convolutional neural network is according to the formula:

[0067]

[0068] Among them: represents the multi-factor weighted edge response intensity value of the image at position , represents the gray value of the image at position , represents the weight coefficient of the convolution kernel at position , represents the gradient amplitude value of the image at position , represents the direction consistency measure value of the image at position , represents the density aggregation degree value of the image at position , represents the weighted coefficient of the gray channel response, represents the weighted coefficient of the gradient channel response, represents the weighted coefficient of the direction consistency channel, represents the weighted coefficient of the density aggregation degree channel, represents the vertical index position of the image, represents the horizontal index position of the image, represents the vertical direction offset step index of the convolution window, This represents the horizontal stride index of the convolution window. This indicates the half-side size of the convolution kernel;

[0069] Execution process: First, a convolutional neural network is used to slide through the image region pixel by pixel to calculate the position of each pixel. Edge response strength value The original grayscale value is obtained by weighted fusion of four feature channels. Gradient magnitude Orientation consistency metric With density and degree of polymerization index The system traverses the local neighborhood of each image pixel. Use weighting coefficients at each offset position. Controlling the contribution ratio of different channels to the edge response, convolution kernel weights By balancing the spatial structural distribution characteristics, a weighted convolution model can accurately extract the structural backbone of crack paths in complex slope images.

[0070] The specific steps for generating the crack propagation trend trajectory set are as follows:

[0071] Based on the main path layer of the crack extension, the difference between the horizontal and vertical pixel coordinates of the same path nodes in continuous images is calculated, and a direction vector sequence is constructed. The angle change amplitude between adjacent frames is extracted to generate a path extension direction change group.

[0072] Based on the path extension direction change group, the continuous segment length of the abrupt segment in the direction sequence is statistically analyzed and the peak offset segment position is extracted. The set of directional abrupt point points in the path is marked to obtain the crack propagation trend trajectory set.

[0073] Based on the crack extension main path layer, the difference between the horizontal and vertical pixel coordinates of the same path nodes in continuous images is calculated, and a direction vector sequence is constructed. The node coordinate set of each crack path in two frames of images is read, and the horizontal and vertical position indices of the same numbered path nodes in each group are extracted in the image matrix. A coordinate difference dictionary is established according to the frame sequence number, and a two-dimensional coordinate vector of the node difference is constructed. The direction angle value is obtained by calculating the arctangent of the x and y components of the node difference, with the unit set to degrees and the precision retained to one decimal place. The direction angle vector sequence is sorted by node number and stored in the vector matrix. Then, the direction difference between continuous vector groups is calculated, and the difference between the vector angle of the current frame and the vector angle of the previous frame is calculated to obtain the direction change amplitude. The absolute value is taken and stored as an angle change sequence. An inter-frame angle change matrix is ​​established for each path to generate a path extension direction change group.

[0074] Based on the path extension direction change group, the length of the continuous segment of the mutation segment in the direction sequence is counted and the peak value offset segment position is extracted, the obtained angle change sequence is segmented, the mutation threshold is set to twenty degrees, the window length is set to five frames, the mean value of the angle change value of each frame in the sliding window is calculated, the window starting frame number whose average change amplitude exceeds the threshold is recorded, and the frame corresponding to the maximum change value in the window is marked as the peak value frame, an index table is established for the mutation window number and the peak value frame index, and is marked as a mutation segment group, the path node number in the segment group is associated, a path internal mutation node list is constructed, the mutation segment sequence in each path is combined with the corresponding node coordinates and recorded as a position annotation dictionary, and a crack propagation trend trajectory set is generated.

[0075] The specific process of constructing the instability inducing region set is:

[0076] Based on the crack propagation trend trajectory set, the three-dimensional coordinate difference between each trajectory node and the rock mass monitoring point is calculated and the spatial distance threshold is set, the monitoring points falling within the range are screened and the time value sequence is extracted, and a path monitoring matching data set is generated;

[0077] Based on the path monitoring matching data set, the multivariate directional feature mode is recognized by using an artificial neural network, the numerical difference between the path direction and the inclination change direction is compared, and the segments with the same symbol are marked, the path angle turning trend is matched, the direction consistent section is screened out, and a trajectory synchronous offset section group is obtained;

[0078] Based on the trajectory synchronous offset section group, the horizontal and vertical coordinate density of the position in each node set is counted and a local aggregation grid is constructed, a region with an overlap rate greater than a set threshold is extracted and the boundary is closed, and a instability inducing region set is obtained;

[0079] Based on the crack propagation trend trajectory set, the three-dimensional coordinate difference between each trajectory node and the rock mass monitoring point is calculated and the spatial distance threshold is set, the three-dimensional coordinate array of the trajectory node and the three-dimensional coordinate array of the monitoring point are imported, the x, y and z coordinates of each trajectory node and the corresponding coordinates of each monitoring point are subtracted in turn to obtain the coordinate difference in three axes, the difference value is squared and added, the square root is calculated to obtain the three-dimensional Euclidean distance value, the spatial threshold is set to 5.0, the monitoring point number less than or equal to the threshold is screened and recorded by traversing the distance array, the corresponding time sequence is extracted after matching the record and sorted according to the time stamp to establish a data table, each group of trajectory nodes and the time sequence of the hit monitoring points are combined into independent subsets, and a path monitoring matching data set is generated;

[0080] Based on the path monitoring matching data set, a multi-variable directional feature mode is recognized by using an artificial neural network, a numerical difference value of the path direction and the inclination angle change direction is compared, a consistent segment is marked, a path angle turning trend is matched, and a consistent section is screened out, a multi-layer feedforward neural network structure is constructed, the number of input layer nodes is set to 4, corresponding to the path direction angle, the inclination angle change amplitude, the stress direction amplitude and the time sequence number, the hidden layer is set to two layers, the first hidden layer contains 16 neurons, the second hidden layer contains 8 neurons, the activation function uses the ReLU function, the output layer contains 2 neurons, and the label value belonging to the consistent section is output respectively, the optimizer uses SGD, the learning rate is set to 0.01, the batch size is set to 32, after the feature standardization normalization processing of each data subset is performed, the network is input for forward propagation and back propagation training, each trajectory path is executed by a sliding window type for classification prediction, a segment with a continuous label of positive is marked, and a trajectory synchronous offset section group is generated;

[0081] Based on the trajectory synchronous offset section group, the horizontal and vertical coordinate densities of the position of each group node are counted and a local aggregation grid is constructed, a region with an overlap rate greater than a set threshold is extracted and a closed boundary is formed, the trajectory node positions are divided into regions according to the two-dimensional coordinate grid, the length of each grid is set to 10 units, the two-dimensional grid matrix is initialized, the node position coordinates are divided by the grid length and mapped to the grid index after being rounded, the number of nodes in each grid is counted and a density distribution map is established, the density of each grid is normalized, the overlap rate threshold is set to 0.65, the grid number set with a density value greater than the threshold is screened out, the hit grid boundary vertex coordinates are extracted and connected to form a closed figure, the closed figure is stored with the path number as an index, and a set of instability inducing regions is generated.

[0082] The specific execution process of the artificial neural network is as follows: the path direction, the inclination angle change direction, the stress change amplitude and the time sequence interval in the path monitoring matching data set are taken as input variables, a multi-layer neural structure is constructed, the weights of each node are initialized and the activation threshold is allocated, the input layer is weighted and transmitted to the hidden layer and the activated nodes are performed, the direction change response mode under each variable combination is extracted, and then the output is transmitted to the output layer for node state normalization, and a directional feature label matrix corresponding to the multi-variable combination is generated.

[0083] The artificial neural network is according to the formula:

[0084]

[0085] Wherein: represents the output response vector of the i-th layer, represents the feature mapping matrix of the i-th layer, which is used for weighted connection between input and nodes, represents the output response vector of the i-th layer, represents the feature mapping matrix of the i-th layer, which is used for weighted connection between input and nodes, represents the output response vector of the i-th layer, input feature vector of the layer, representing the first bias vector of the layer, representing the first path-monitoring point space synchronous disturbance term introduced in the layer, main direction projection term representing the stress change trend at the path node, contribution control factor of the synchronous disturbance term, influence degree control factor of the stress response term, nonlinear activation function, which is set to in the embodiment, representing the current network layer index number;

[0086] Execution process: first, an initial input feature vector is constructed, which is composed of path direction angle, inclination rate of change, stress main direction amplitude, and time series step number. After normalization transformation, each input is sequentially mapped to the input layer node position, and is transmitted to the first hidden layer. Each neuron performs linear weighted operation of the feature mapping matrix according to the input vector, and introduces the bias term to increase two enhanced structural feature terms, the synchronous disturbance term is fitted by the fluctuation value of the node pairing distance with the monitoring point space in the multi-time sequence trajectory, and the stress trend term represents the projection intensity of the stress change vector of the path node in the same time window on the main direction axis. The above two terms are controlled by factors and to control the weighting degree in the response total, and the values of the two coefficients are iteratively optimized in the training data with the goal of maximum classification accuracy through 5-fold cross-validation, and are set to float within the interval . The synthesis result is used as the input value of the activation function , the nonlinear mapping result is output by using function, and the response vector is obtained. Then it is transmitted to the second hidden layer to continue the same structure transformation, and finally the prediction label belonging to the path direction consistency region is generated in the output layer.

[0087] The specific steps of constructing the instability inducing region set are as follows:

[0088] Based on the instability inducing region set, the path splitting node angle change value sequence is extracted, and the continuous segment angle difference increasing trend is identified. Combined with the position coordinate density calculation, the aggregated boundary segment is screened, and the splitting angle offset clustering segment group is generated.

[0089] Based on the split angle offset clustering fragment group, the angle sequence of the reverse segment of the path split starting point and the ending point in each group is determined, the reverse closed path segment set is extracted, the boundary line network is generated, and the closed grid generation is performed to obtain the dynamic crack control grid structure;

[0090] Based on the instability inducing region set, the angle change value of the continuous path segment in each group node sequence is extracted, the angle sequence list is constructed according to the order of each group angle change value, the sliding window width is set to 3 and the step is set to 1, the angle difference value of each group sliding window data and the angle difference ratio of the previous window are calculated in turn, the amplitude threshold is set to 1.25, the fragment index satisfying the continuous increasing relationship is marked, the two-dimensional histogram matrix is constructed by combining the frequency of each group fragment in the x, y coordinate system, the horizontal and vertical axis pixel scale is set to 5 units, the density of each region is calculated, the coordinate boundary corresponding to the density peak value region is obtained, the coordinate hit boundary segment path index is extracted, and the angle difference increasing clustering set is aggregated to generate the split angle offset clustering fragment group.

[0091] Based on the split angle offset clustering fragment group, the angle sequence of the reverse segment of the path split starting point and the ending point in each group is determined, the continuous path segment from the starting point to the ending point in each group is selected, the direction vector included angle of the first and last paragraphs is calculated, the included angle is converted into an angle value, the reverse angle threshold is set to 150 degrees, the path group with an included angle greater than or equal to the threshold is extracted, the node coordinate sequence of each path segment is extracted and a line segment set is constructed, the boundary of the above line segment set is extracted, the coordinates of each path endpoint are converted into two-dimensional pixel point coordinates, the boundary box diagram is generated by connecting the paths in order, and the path set is input into the graph structure model for topological connection to construct the adjacency table structure between each line segment. The radial ring property of the closed boundary path is detected, the boundary fragment path with the starting and ending overlapping characteristics is closed, and the dynamic crack control grid structure is generated.

[0092] Please refer to Figure 2 A safety monitoring system for a rock slope, the system comprising:

[0093] An image feature extraction module: through the linearity structure continuity, the regional contour edge curvature change degree and the boundary direction extension characteristic shown in the rock slope surface image, the boundary closed segment gray jump is extracted and the density aggregation area difference is analyzed to obtain the structure abnormal focusing partition;

[0094] A path recognition construction module: based on the structure abnormal focusing partition, the linear contour response spectrum in the boundary region is extracted by using a convolutional neural network, the contour segment direction is sorted and the connection relationship is recognized, the direction deviation segment is excluded and the continuous paragraphs are connected to obtain a crack expansion main path layer;

[0095] Trend analysis operation module: based on the main path layer of crack propagation, the direction angle offset sequence of the path node in the multi-time sequence image is constructed, the path direction change amplitude and the fluctuation area length between the continuous frames are calculated, the angle mutation node position is extracted, and the crack propagation trend trajectory set is generated;

[0096] State association judgment module: based on the crack propagation trend trajectory set, the multi-variable direction mode of the monitoring point is recognized by combining the artificial neural network, the corresponding relationship of the inclination direction, the stress direction and the path trend is compared, the direction synchronization node segment and the offset aggregation segment are marked, and the instability inducing area set is constructed;

[0097] Grid structure forming module: based on the instability inducing area set, the direction mutation value of the path splitting point is counted and the clustering distribution area is generated, the continuous direction mutation path is reconstructed and the closed boundary line and the grid block are divided, and the dynamic crack control grid structure is obtained.

[0098] The above is only a preferred embodiment of the present application, and does not limit the present application in other forms. Any person skilled in the art can use the disclosed technical content to make changes or modifications to equivalent embodiments applied to other fields, but any simple modification, equivalent change and modification made to the above embodiments without departing from the technical solution content of the present application, according to the technical essence of the present application, still belongs to the protection scope of the technical solution of the present application.

Claims

1. A method for safety monitoring of rock slopes, characterized in that, The method comprises the following steps: S1: filtering out the boundary closedness change section and marking the pixel density mutation area by using difference comparison and gradient detection based on the linearity structure continuity, regional contour edge curvature change degree and boundary direction extension characteristics shown in the rock slope surface image, and obtaining the structure abnormal focus partition; S2: based on the structure abnormal focus partition, sorting the contour line extension direction and excluding the cross section after extracting the boundary response structure linearity feature by using the convolutional neural network, positioning the crack main extension path, verifying the continuity and fitting the closed structure, marking the stable connection section, and obtaining the crack expansion main path layer; S3: based on the crack expansion main path layer, comparing the path displacement direction and the extension angle change value in multiple time sequences, identifying the offset stable section and judging the path expansion trend, extracting the continuous mutation node, and generating the crack expansion trend trajectory set; S4: based on the crack expansion trend trajectory set, combining the artificial neural network to extract the multivariate direction feature of the monitoring point, matching the rock mass monitoring point data, judging the relationship between the trajectory path and the inclination and stress change direction, positioning the synchronous section and the path offset aggregation area, and constructing the instability induced area set; S5: based on the instability induced area set, counting the angle mutation value of the path splitting node and dividing the clustering area, constructing the closed control network through the continuous splitting path, forming the direction reversal closed structure, and obtaining the dynamic crack control grid structure; The specific process of constructing the instability induced area set is: Based on the crack expansion trend trajectory set, calculating the three-dimensional coordinate difference value between each trajectory node and the rock mass monitoring point and setting a space distance threshold, screening the monitoring points falling within the range and extracting the time value sequence, and generating the path monitoring matching data set; Based on the path monitoring matching data set, identifying the multivariate direction feature mode by using the artificial neural network, comparing the numerical difference value of the path direction and the inclination change direction and marking the same symbol section, matching the path angle turning trend and screening the direction consistent section, and obtaining the trajectory synchronous offset section group; Based on the trajectory synchronous offset section group, counting the horizontal and vertical coordinate density of the position in each node set and constructing a local aggregation grid, extracting the area with an overlap rate greater than a set threshold and closing the boundary, and obtaining the instability induced area set; The specific steps of obtaining the dynamic crack control grid structure are: Based on the instability induced area set, extracting the path splitting node angle change value sequence and identifying the continuous section angle difference increasing trend, combining the position coordinate density to screen and aggregate the boundary section, generating the splitting angle offset clustering section group; Based on the splitting angle offset clustering section group, determining the reverse section angle sequence of the path splitting starting point and the ending point in each group and extracting the reverse closed path line segment set, generating the boundary line network and then performing closed grid generation, and obtaining the dynamic crack control grid structure.

2. The method for safety monitoring of rock slope according to claim 1, wherein, The structure abnormality focusing partition includes a boundary closedness change area, a pixel density mutation area and a contour shape change area, the crack propagation main path layer includes a main extension path segment, a closed structure segment and a continuous connection node, the crack propagation trend trajectory set includes a path displacement direction sequence, an extension angle change sequence and a mutation node sequence, the instability induction area set includes a trajectory turning direction synchronization segment, a path deviation aggregation area and a direction relationship coupling area, and the dynamic crack control grid structure includes a direction reversal path segment, a closed connection node and a control grid boundary unit.

3. The method for safety monitoring of rock slope according to claim 1, wherein, The specific steps for obtaining the structure abnormality focusing partition are as follows: By the linearity structure continuity, the area contour edge curvature change degree and the boundary direction extension characteristic shown in the rock slope surface image, the edge region horizontal and vertical gray gradient values are extracted and the gray jump intensity of each boundary segment is calculated, the gradient difference value of the continuous segment is located, and the boundary structure mutation segment set is generated; Based on the boundary structure mutation segment set, the horizontal and vertical pixel distribution density in the image area and the average deviation value of the density aggregation area are calculated, the range of the high-density block segment boundary is marked, and the structure abnormality focusing partition is obtained.

4. The method for safety monitoring of rock slope according to claim 1, wherein, The specific steps for obtaining the crack propagation main path layer are as follows: Based on the structure abnormality focusing partition, the linearity contour response intensity value atlas in the boundary image area is extracted by using a convolutional neural network, the direction of the coordinate difference value of the boundary line segment is extracted and the direction angle amplitude is sorted, the cross line segments with the direction angle greater than a set limit value are screened out and the direction continuous line segments are reserved, and a unidirectional continuous contour set is generated; Based on the unidirectional continuous contour set, the end point distance difference value between adjacent contour paths is calculated and the direction angle synchronization is judged, the similar direction path segments are extended and connected, and the interrupted positions are fitted and completed, and a crack extension path chain group is generated; Based on the crack extension path chain group, the closed edge jump difference value between the path connection nodes is evaluated and the curvature continuity is judged, the stable path unit of the start and end point continuous segment is marked, and the crack propagation main path layer is obtained.

5. The method for safety monitoring of rock slope according to claim 4, characterized in that, The specific execution process of the convolutional neural network is as follows: the structure abnormality image area in the focusing partition is focused layer by layer by using a multi-layer convolution structure, the local area gray change value is extracted by sliding a fixed size convolution kernel on the image matrix, the feature map group is generated and then connected to an activation operation, the linearity edge response is reserved, the multi-scale convolution kernel is superimposed to cover different scale contour structures, the convolution output is stacked and normalized in channels, the continuous boundary direction change intensity value matrix in the image is extracted, and the region contour peak response path is extracted according to the linear response amplitude in the extraction matrix.

6. The method for safety monitoring of rock slope according to claim 1, wherein, The specific steps for generating the crack propagation trend trajectory set are as follows: Based on the crack propagation main path layer, the horizontal and vertical pixel coordinate difference values of the same path node in the continuous image are calculated, and a direction vector sequence is constructed, the angle change amplitude between adjacent frames is extracted, and a path extension direction change group is generated; Based on the path extension direction change group, the continuous segment length of the mutation segment in the direction sequence is counted and the peak value deviation segment position is extracted, the path direction mutation point set is marked, and the crack propagation trend trajectory set is obtained.

7. The method for safety monitoring of rock slope according to claim 1, wherein, The specific implementation process of the artificial neural network is as follows: taking the path direction, the inclination change direction, the stress change amplitude, and the time sequence interval in the path monitoring matching data set as input variables, constructing a multi-layer neural structure, initializing the node weights and assigning the activation threshold, weighting the input layer to the hidden layer and performing the activation node, extracting the direction change response mode under each variable combination, and then outputting to the output layer for node state normalization to generate a directionality feature label matrix corresponding to the multi-variable combination.

8. A safety monitoring system for a rock slope, characterized in that The safety monitoring method for rock slopes according to any one of claims 1-7, the system comprises: An image feature extraction module: through the linearity structure continuity, the regional contour edge curvature change degree and the boundary direction extension characteristic in the rock slope surface image, the boundary closed section gray jump is extracted and the density aggregation area difference is analyzed, and the structure abnormal focusing partition is obtained; A path recognition construction module: based on the structure abnormal focusing partition, the linear contour response spectrum in the boundary area is extracted by using a convolutional neural network, the contour line direction is sorted and the connection relationship is recognized, the direction deviation section is screened out and the continuous section is connected, and a crack propagation main path layer is obtained; A trend analysis operation module: based on the crack propagation main path layer, the direction angle offset sequence of the path nodes in the multi-time sequence image is constructed, the path direction change amplitude and the fluctuation area length between continuous frames are calculated, the angle mutation node position is extracted, and a crack propagation trend trajectory set is generated; A state correlation judgment module: based on the crack propagation trend trajectory set, the multi-variable direction mode of the monitoring point is recognized by combining an artificial neural network, the corresponding relationship of the inclination direction, the stress direction and the path trend is compared, the direction synchronous node section and the offset aggregation segment are marked, and a instability inducing area set is constructed; A grid structure forming module: based on the instability inducing area set, the direction mutation value of the path splitting point is counted and a clustering distribution area is generated, the continuous direction mutation path is reconstructed and the closed boundary line and the grid block are divided, and a dynamic crack control grid structure is obtained.

Images