A radar monitoring and early warning method and system for open-pit mine slope
By combining ground-based radar and vehicle-mounted radar, a combined early warning model is constructed, which solves the problem of insufficient reliability of single-category radar measurement, and improves the accuracy and reliability of the judgment of slope stability of open-pit mines.
Patent Information
- Application Number
- CN202411266311.2
- Authority / Receiving Office
- CN · China
- Patent Type
- Patents(China)
- Current Assignee / Owner
- Filing Date
- 2024-09-11
- Publication Date
- 2024-12-03
- Estimated Expiration
- 2044-09-11
AI Technical Summary
The reliability of single-class radar measurement in the prior art has insufficient results in low accuracy in judging the stability of open-pit mine slopes.
The combination of ground-based radar and vehicle-mounted radar is used to classify the data through a classifier, build the ground-based radar and vehicle-mounted radar early warning model, and synchronize and correct it to form a combined early warning model to improve the accuracy of stability judgment.
Multi-category radar measurement is realized, which reduces the random error of single-mode data and improves the reliability and accuracy of slope stability monitoring.
Smart Images

Figure CN118795442B_ABST
Abstract
Description
Technical Field
[0001] The invention relates to the technical field of slope monitoring, and in particular to an open-pit mine slope radar monitoring and early warning method and a system thereof. Background Art
[0002] In the process of development and mining of various open-pit mines, they are often affected by some mining disasters. Among them, the safety of slope rock mass is the most important factor restricting the production efficiency and safety of mines. Slope is a slope formed by natural or artificial excavation. It is one of the most basic geological environments in human engineering activities and the most common engineering form in engineering construction. As one of the three major global geological disasters (earthquakes, floods, collapses, landslides and mudslides), slope instability and collapse is one of the main disaster forms often encountered in the mining process, posing a serious threat to national property and the lives of miners. Therefore, the correct understanding of slopes, reasonable design, timely monitoring, and appropriate management to minimize the disasters caused by slope instability are issues that must be considered by engineering design and construction personnel.
[0003] At present, for monitoring the displacement changes of open-pit mine slopes, traditional measurement methods such as levels and total stations can be used, but the real-time performance is not high, the manpower cost is high, and the safety of the surveyors is difficult to guarantee. Alternatively, radar non-contact monitoring method can be used to perform real-time measurement without contacting the slope to be measured, which ensures safety.
[0004] In the prior art, radar measurement of slopes usually only uses single-type radar measurement. The measurement reliability of this single-mode data is insufficient, which will reduce the accuracy of slope stability judgment. Summary of the invention
[0005] The purpose of the present invention is to provide an open-pit mine slope radar monitoring and early warning method and system thereof, so as to solve the technical problem that single-category radar measurement in the prior art has insufficient measurement reliability and reduces the accuracy of slope stability judgment.
[0006] In order to solve the above technical problems, the present invention specifically provides the following technical solutions:
[0007] A radar monitoring and early warning method for an open-pit mine slope, characterized in that it comprises the following steps:
[0008] Use ground-based radar and vehicle-mounted radar to monitor the slope of the open-pit mine, and obtain ground-based radar data and vehicle-mounted radar data for judging the stability of the open-pit mine slope;
[0009] Using a classifier to perform classification training on the ground-based radar data and the vehicle-mounted radar data, and correspondingly obtaining a ground-based radar early warning model and a vehicle-mounted radar early warning model for judging the stability of an open-pit mine slope;
[0010] The ground-based radar early warning model and the vehicle-mounted radar early warning model are synchronously corrected to obtain a combined early warning model for correcting the judgment accuracy of open-pit mine slope stability.
[0011] Furthermore, the method for constructing the ground-based radar early warning model includes:
[0012] Analyzing and labeling the stability status of the open pit mine slope on the ground-based radar data to obtain a stability label for the ground-based radar data;
[0013] Using a classifier to perform classification learning on the ground-based radar data and the stability labels of the ground-based radar data, to obtain a ground-based radar early warning model that outputs the stability labels according to the ground-based radar data;
[0014] The ground-based radar early warning model is:
[0015] Lg=SVM(RadarData_g);
[0016] Where Lg is the stable label prediction value of the ground-based radar warning model, RadarData_g is the ground-based radar data, and SVM is the classifier;
[0017] Among them, assuming that the loss function value of the ground-based radar early warning model is Lossg, then:
[0018] Lossg=MSE(Lg,Lr);
[0019] Where Lg is the stable label prediction value of the ground-based radar warning model, Lr is the stable label true value of the ground-based radar data, MSE is the mean square error function, and MSE(Lg,Lr) is the mean square error between Lg and Lr.
[0020] Furthermore, the method for constructing the vehicle-mounted radar early warning model includes:
[0021] Analyzing and labeling the stability status of the open pit mine slope on the vehicle-borne radar data to obtain a stability label of the vehicle-borne radar data;
[0022] The vehicle-mounted radar data is used as an input item of the ground-based radar warning model, and the stability label of the vehicle-mounted radar data is used as an output item of the ground-based radar warning model, and the ground-based radar warning model is subjected to transfer learning to obtain a vehicle-mounted radar warning model that outputs the stability label according to the vehicle-mounted radar data;
[0023] The vehicle-mounted radar early warning model is:
[0024] Lc=SVM(RadarData_c);
[0025] Where Lc is the stable label prediction value of the vehicle radar warning model, RadarData_c is the vehicle radar data, and SVM is the classifier;
[0026] Among them, assuming that the loss function value of the vehicle-mounted radar warning model is Lossc, then:
[0027] Lossc=MSE(Lc,Lg)-MSE(RadarData_c,RadarData_g);
[0028] Where Lg is the stable label prediction value of the ground-based radar warning model, Lc is the stable label prediction value of the vehicle-mounted radar warning model, RadarData_g is the ground-based radar data, RadarData_c is the vehicle-mounted radar data, MSE is the mean square error function, MSE(Lg,Lc) is the mean square error between Lg and Lc, and MSE(RadarData_c,RadarData_g) is the mean square error between RadarData_c and RadarData_g.
[0029] Furthermore, the analysis and annotation method of the ground-based radar data includes:
[0030] Using the ground-based radar data, image analysis is performed on the crack properties on the open-pit mine slope to obtain crack characteristics of the open-pit mine slope;
[0031] The stability of the ground-based radar data whose crack characteristics are greater than the crack threshold is marked as a non-steady-state label, and the stability of the ground-based radar data whose crack characteristics are less than or equal to the crack threshold is marked as a stable-state label;
[0032] The analysis and annotation method of the vehicle-mounted radar data includes:
[0033] Using the vehicle-borne radar data to perform image analysis on crack properties on the open-pit mine slope, to obtain crack characteristics of the open-pit mine slope;
[0034] The stability of the vehicle-mounted radar data whose crack characteristics are greater than the crack threshold is marked as a non-steady-state label, and the stability of the vehicle-mounted radar data whose crack characteristics are less than or equal to the crack threshold is marked as a stable-state label.
[0035] Furthermore, the method for obtaining crack characteristics of the open-pit mine slope includes:
[0036] Mapping the ground-based radar data in a three-dimensional coordinate space to construct a three-dimensional image of the open-pit mine slope, and detecting the crack edge in the three-dimensional image of the open-pit mine slope through an edge extraction algorithm, and then calculating the position, width, length and depth characteristics of the crack from the crack edge;
[0037] The vehicle-mounted radar data is mapped in a three-dimensional coordinate space to construct a three-dimensional image of the open-pit mine slope, and the crack edge is detected in the three-dimensional image of the open-pit mine slope through an edge extraction algorithm, and the position, width, length and depth characteristics of the crack are calculated from the crack edge.
[0038] Furthermore, the method for constructing the combined early warning model includes:
[0039] For the loss functions of the ground-based radar warning model and the vehicle-mounted radar warning model, respective correction factors are set;
[0040] The ground-based radar warning model and the vehicle-mounted radar warning model are corrected and trained by the correction factor to obtain the combined warning model;
[0041] The combined early warning model is:
[0042] LRadar=Softmax[SVM(RadarData_g),SVM(RadarData_c)];
[0043] Among them, assuming that the loss function value of the combined early warning model is LossRadar, then:
[0044] LossRadar=Wg*Lossg+Wc*Lossc;
[0045] Wherein, LRadar is the stability label prediction value of the combined warning model, Wg is the correction factor of the loss function of the ground-based radar warning model, Wc is the correction factor of the loss function of the vehicle-mounted radar warning model, Softmax is the softmax function, Lossg is the loss function value of the ground-based radar warning model, Lossc is the loss function value of the vehicle-mounted radar warning model, RadarData_g is the ground-based radar data, RadarData_c is the vehicle-mounted radar data, and SVM is the classifier.
[0046] Furthermore, the method for setting the correction factor includes:
[0047] Track the training completion of ground-based radar warning models and vehicle-mounted radar warning models;
[0048] A correction factor is set based on the training completion degree. for:
[0049] ;in, ; Where: , is the correction factor of the loss function of the ground-based radar early warning model at the tth moment, W c(t) is the correction factor of the loss function of the vehicle-borne radar early warning model at the tth moment; , C g (t-1) is the training completion degree of the ground-based radar early warning model at the t-1th moment, C c (t-1) is the training completion degree of the vehicle-borne radar warning model at the t-1th moment; Loss i (t-1)[Loss g (t-1),Loss c (t-1)],Loss i (t-2)[Loss g (t-2),Loss c (t-2)], Loss g (t-1), Loss g (t-2) are the loss functions of the ground-based radar early warning model at the t-1st and t-2nd moments, respectively. c (t-1), Loss c (t-2) are the loss functions of the vehicle-borne radar warning model at the time t-1 and t-2 respectively;
[0050] T is the total time of training the model, i , j , t are counting variables, g is the identifier of the ground-based radar, and c is the identifier of the vehicle-mounted radar.
[0051] Furthermore, the vehicle-mounted radar data and the ground-based radar data are normalized. Furthermore, the present invention provides an open-pit mine slope radar monitoring and early warning system, which is applied to the open-pit mine slope radar monitoring and early warning method, and the system includes:
[0052] A data acquisition unit, used to monitor the open-pit mine slope using ground-based radar and vehicle-mounted radar, and obtain ground-based radar data and vehicle-mounted radar data for judging the stability of the open-pit mine slope;
[0053] The data processing unit uses a combined early warning model to determine the stability of the open pit slope based on ground-based radar data and vehicle-mounted radar data;
[0054] The data storage unit is used to store the combined early warning model.
[0055] Furthermore, the pre-establishment method of the combined early warning model in the data storage unit includes:
[0056] For the loss functions of the ground-based radar warning model and the vehicle-mounted radar warning model, respective correction factors are set;
[0057] The ground-based radar warning model and the vehicle-mounted radar warning model are corrected and trained by the correction factor to obtain the combined warning model;
[0058] The combined early warning model is:
[0059] LRadar=Softmax[SVM(RadarData_g),SVM(RadarData_c)];
[0060] Among them, assuming that the loss function value of the combined early warning model is LossRadar, then:
[0061] LossRadar=Wg*Lossg+Wc*Lossc;
[0062] Wherein, LRadar is the stability label prediction value of the combined warning model, Wg is the correction factor of the loss function of the ground-based radar warning model, Wc is the correction factor of the loss function of the vehicle-mounted radar warning model, Softmax is the softmax function, Lossg is the loss function value of the ground-based radar warning model, Lossc is the loss function value of the vehicle-mounted radar warning model, RadarData_g is the ground-based radar data, RadarData_c is the vehicle-mounted radar data, and SVM is the classifier.
[0063] Compared with the prior art, the present invention has the following beneficial effects:
[0064] The present invention utilizes a classifier to perform classification training on the ground-based radar data and the vehicle-mounted radar data, and correspondingly obtains a ground-based radar early warning model and a vehicle-mounted radar early warning model for judging the stability of an open-pit mine slope. The ground-based radar early warning model and the vehicle-mounted radar early warning model are synchronously corrected to obtain a combined early warning model for correcting the judgment accuracy of the open-pit mine slope stability, thereby realizing multi-category radar measurement and improving the reliability of slope stability monitoring. BRIEF DESCRIPTION OF THE DRAWINGS
[0065] In order to more clearly illustrate the implementation methods of the present invention or the technical solutions in the prior art, the following briefly introduces the drawings required for the implementation methods or the description of the prior art. Obviously, the drawings in the following description are only exemplary, and for ordinary technicians in this field, other implementation drawings can be derived from the provided drawings without creative work.
[0066] Figure 1 A flow chart of an open-pit mine slope monitoring and early warning method provided by an embodiment of the present invention;
[0067] Figure 2 A block diagram of an open-pit mine slope monitoring and early warning system provided in an embodiment of the present invention. DETAILED DESCRIPTION
[0068] The following will be combined with the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.
[0069] like Figure 1 As shown, the present invention provides an open-pit mine slope radar monitoring and early warning method, comprising the following steps:
[0070] Use ground-based radar and vehicle-mounted radar to monitor the slope of the open-pit mine, and obtain ground-based radar data and vehicle-mounted radar data for judging the stability of the open-pit mine slope;
[0071] The ground-based radar data and vehicle-mounted radar data are classified and trained using a classifier, and the ground-based radar early warning model and vehicle-mounted radar early warning model for judging the stability of open-pit mine slopes are obtained accordingly;
[0072] The ground-based radar early warning model and the vehicle-mounted radar early warning model are synchronously corrected to obtain a combined early warning model for correcting the judgment accuracy of open-pit mine slope stability.
[0073] In order to improve the accuracy of stability monitoring of open-pit mine slopes, the present invention combines ground-based radar and vehicle-mounted radar to obtain monitoring data for judging stability. Multiple categories of radar data are mutually constrained to avoid data randomness caused by single-mode monitoring. For example, the data monitored by the ground-based radar is abnormal (data items are missing or data collection errors, etc.). If the ground-based radar data is used alone to judge stability, it will cause misjudgment of stability. In this case, the present invention uses vehicle-mounted radar to synchronously monitor stability and constrain the stability monitoring of the ground-based radar, thereby reducing the random error of the single-mode data and improving the accuracy of stability judgment.
[0074] Specifically, the present invention uses a classifier to classify and learn ground-based radar data to construct a ground-based radar early warning model, which can determine the stability of open-pit mine slopes based on ground-based radar data. The machine learning model is used to realize automatic monitoring of the stability of open-pit mine slopes, improve monitoring efficiency, and meet the timeliness requirements of slope monitoring.
[0075] Similarly, the present invention performs transfer learning on the ground-based radar warning model, which can determine the stability of the open-pit mine slope based on the vehicle-mounted radar data. Transfer learning can enable the vehicle-mounted radar warning model to inherit the advantageous performance of the ground-based radar warning model, ensuring the identity, high accuracy and efficiency of the vehicle-mounted radar warning model and the ground-based radar warning model in judging stability.
[0076] Furthermore, in order to synchronize the progress of slope stability monitoring by the on-board radar warning model and the ground-based radar warning model, that is, to achieve synchronization of slope stability prediction accuracy between the on-board radar warning model and the ground-based radar warning model, the present invention balances the model accuracy performance of the on-board radar warning model and the ground-based radar warning model during the slope monitoring process through synchronization correction to avoid a large prediction gap between the on-board radar warning model and the ground-based radar warning model. The greater the prediction gap between the two, the greater the difference in slope stability judgment results output by the two, which further leads to the need for human participation in the analysis and judgment of the judgment results of the two to determine the final result of the slope stability. Therefore, the present invention synchronizes and corrects the two, unifies the prediction results of the on-board radar warning model and the ground-based radar warning model, and the output results corroborate / constrain each other to accurately obtain the slope stability results, that is, to improve the accuracy of slope stability monitoring.
[0077] Specifically, by tracking the loss functions of the vehicle-mounted radar warning model and the ground-based radar warning model, the training progress, or the accuracy performance, of the vehicle-mounted radar warning model and the ground-based radar warning model can be monitored. The descent rate of the loss function is calculated through loss function tracking. The faster the descent rate of the loss function is, the faster the training progress of the model is and the higher the accuracy performance is.
[0078] Therefore, the present invention sets a correction factor, which is bound to the loss function, to reduce the proportion of the loss function of the model with faster training progress / higher precision performance in the total loss, thereby lowering the model training progress or precision performance, so that the model with slow training progress and poor precision performance has a high proportion in the total loss, thereby improving its training progress or precision performance, thereby realizing the correction of the vehicle-mounted radar warning model and the ground-based radar warning model to synchronized operation, that is, the vehicle-mounted radar warning model and the ground-based radar warning model perform stability monitoring with similar precision performance, and the vehicle-mounted radar warning model and the ground-based radar warning model obtain unified and accurate judgment.
[0079] In order to improve the accuracy of stability monitoring of open-pit mine slopes, the present invention combines ground-based radar and vehicle-mounted radar to obtain monitoring data for judging stability. Multiple categories of radar data are mutually constrained to avoid the randomness of data generated by single monitoring. The vehicle-mounted radar is used to synchronously monitor stability, constrain the stability monitoring of the ground-based radar, thereby reducing the random error of single-mode data and improving the accuracy of stability judgment. The details are as follows:
[0080] The construction method of the ground-based radar early warning model includes:
[0081] Analyze and annotate the stability of the open-pit mine slope using ground-based radar data to obtain stability labels for the ground-based radar data;
[0082] The ground-based radar data and the stability labels of the ground-based radar data are classified and learned by using a classifier, so as to obtain a ground-based radar early warning model that outputs the stability labels according to the ground-based radar data;
[0083] The ground-based radar early warning model is:
[0084] Lg=SVM(RadarData_g);
[0085] Where Lg is the stable label prediction value of the ground-based radar warning model, RadarData_g is the ground-based radar data, and SVM is the classifier;
[0086] Among them, the loss function of the ground-based radar warning model is: Lossg=MSE(Lg,Lr), where Lossg is the loss function value of the ground-based radar warning model, Lg is the stable label prediction value of the ground-based radar warning model, Lr is the stable label true value of the ground-based radar data, MSE is the mean square error function, and MSE(Lg,Lr) is the mean square error between Lg and Lr. The construction method of the vehicle-mounted radar warning model includes:
[0087] Analyze and annotate the stability of the open-pit mine slope using the vehicle-mounted radar data to obtain the stability label of the vehicle-mounted radar data;
[0088] The vehicle-mounted radar data is used as the input of the ground-based radar warning model, and the stability label of the vehicle-mounted radar data is used as the output of the ground-based radar warning model. Transfer learning is performed on the ground-based radar warning model to obtain a vehicle-mounted radar warning model that outputs stability labels based on the vehicle-mounted radar data.
[0089] The vehicle-mounted radar warning model is:
[0090] Lc=SVM(RadarData_c);
[0091] Where Lc is the stable label prediction value of the vehicle radar warning model, RadarData_c is the vehicle radar data, and SVM is the classifier;
[0092] Among them, the loss function of the vehicle radar warning model is: Lossc=MSE(Lc,Lg)-MSE(RadarData_c,RadarData_g);
[0093] Wherein, Lossc is the loss function value of the vehicle-mounted radar warning model, Lg is the stable label prediction value of the ground-based radar warning model, Lc is the stable label prediction value of the vehicle-mounted radar warning model, RadarData_g is the ground-based radar data, RadarData_c is the vehicle-mounted radar data, MSE is the mean square error function, MSE(Lg,Lc) is the mean square error between Lg and Lc, and MSE(RadarData_c,RadarData_g) is the mean square error between RadarData_c and RadarData_g. The present invention uses a classifier to classify and learn the ground-based radar data to construct a ground-based radar warning model, which can determine the stability of the open-pit mine slope according to the ground-based radar data, and uses a machine learning model to realize the automatic monitoring of the stability of the open-pit mine slope, improve the monitoring efficiency, and meet the timeliness requirements of slope monitoring. The present invention performs transfer learning on the ground-based radar early warning model, and can judge the stability of the open-pit mine slope according to the vehicle-mounted radar data. Transfer learning can make the vehicle-mounted radar early warning model inherit the advantageous performance of the ground-based radar early warning model, and ensure the identity, high accuracy and efficiency of the vehicle-mounted radar early warning model and the ground-based radar early warning model in judging stability.
[0094] Among them, the loss function of the vehicle-mounted radar warning model in the present invention is composed of two parts: MSE (Lc, Lg) and MSE (RadarData_c, RadarData_g). MSE (Lc, Lg) measures the degree of proximity between the output of the vehicle-mounted radar warning model and the ground-based radar warning model. Therefore, using MSE (Lc, Lg) as the loss function makes the output of the vehicle-mounted radar warning model and the ground-based radar warning model the closest. That is, the vehicle-mounted radar warning model inherits the superior performance of the ground-based radar warning model. MSE (RadarData_c, RadarData_g) measures the degree of proximity between the vehicle-mounted radar warning model and the ground-based radar warning model. The input proximity of the model is determined by MSE (RadarData_c, RadarData_g). Therefore, MSE (RadarData_c, RadarData_g) is used as the loss function to minimize the input proximity between the vehicle-mounted radar warning model and the ground-based radar warning model. MSE (Lc, Lg) is combined with MSE (RadarData_c, RadarData_g) to train the vehicle-mounted radar warning model. The vehicle-mounted radar warning model can use vehicle-mounted radar data that is different from the ground-based radar data, and can obtain the prediction results obtained by the ground-based radar warning model using the ground-based radar data, which ensures the identity of the vehicle-mounted radar warning model and the ground-based radar warning model in judging stability, high accuracy and efficiency.
[0095] Transfer learning refers to any learning strategy that uses the knowledge gained in solving one problem (problem A) to subsequently solve another separate problem (problem B). By definition, transfer learning involves the concepts of domain and task. The purpose of transfer learning is to learn model a and then use the knowledge gained in learning to learn model b. For the radar monitoring and early warning of the present invention, the most commonly used method of transfer learning is to pre-train the model on the ground-based radar (ground-based radar early warning model), and then transfer the model training on the ground-based radar to the vehicle-mounted radar to obtain the vehicle-mounted radar early warning model.
[0096] The analysis and annotation methods of ground-based radar data include:
[0097] The crack properties on the open-pit mine slope were analyzed by using ground-based radar data to obtain the crack characteristics of the open-pit mine slope;
[0098] The stability of the ground-based radar data with crack characteristics greater than the crack threshold is labeled as a non-steady-state label, and the stability of the ground-based radar data with crack characteristics less than or equal to the crack threshold is labeled as a steady-state label;
[0099] The analysis and annotation methods of vehicle-mounted radar data include:
[0100] The crack properties on the open-pit mine slope were analyzed by using vehicle-mounted radar data to obtain the crack characteristics of the open-pit mine slope;
[0101] The stability of the vehicle-borne radar data with crack characteristics greater than the crack threshold is marked as a non-steady-state label, and the stability of the vehicle-borne radar data with crack characteristics less than or equal to the crack threshold is marked as a steady-state label.
[0102] The methods for obtaining crack characteristics of open-pit mine slopes include:
[0103] The ground-based radar data is mapped in a three-dimensional coordinate space to construct a three-dimensional image of the open-pit mine slope. The crack edges are detected in the three-dimensional image of the open-pit mine slope through an edge extraction algorithm (Canny algorithm). The position, width, length and depth characteristics of the cracks are then calculated from the crack edges.
[0104] The vehicle-mounted radar data is mapped in the three-dimensional coordinate space to construct a three-dimensional image of the open-pit mine slope. The crack edges are detected in the three-dimensional image of the open-pit mine slope through the edge extraction algorithm, and the position, width, length and depth characteristics of the cracks are calculated from the crack edges.
[0105] In the present invention, a three-dimensional spatial coordinate system is established, and x, y, and z axis data are converted respectively according to radar scanning data to generate a three-dimensional image of the slope; according to the positional relationship between the radar and the slope, a spatial coordinate system is established with the direction of the laser radar scanning as the x axis, the direction of the distance sinusoidal component as the y axis, and the plane perpendicular to x and y as the z axis; the starting position of the laser radar is set as the reference point M (0, 0, h0), h0 represents the vertical height between the laser radar and the open-pit mine slope, and the three-dimensional coordinates (Xk, Yk, Zk) of the scanning point K are calculated using the trigonometric function method according to the distance and angle information output by the radar scanning data point K; in the three-dimensional coordinate system of the present invention, Xk is equal to the product of the scanning period T and the vehicle speed V, and corresponds to all the values of Yk and Zk in the entire period. The value of Yk is the product of the distance from the laser radar to the scanning point and the sine value of the scanning angle, and Zk is the distance from the laser radar to the scanning point and the cosine value of the scanning angle. The three-dimensional coordinates of the real-time scanning point coordinate K on the slope surface are obtained (Xk, Yk, Zk), where Xk = T*V, Yk = dk*sinθk, and Zk = dk*cosθk. According to the three-dimensional coordinates, the three-dimensional image of the slope is reconstructed using programming software (Python, Matlab, etc.). The image is denoised by Gaussian function; the Sobel operator is used to calculate the change amplitude and direction of the gradient value to calibrate the crack edge; non-maximum suppression is performed, and the gradient values of all points in the gradient direction are compared, and the gradient value with the largest value is selected as the image edge. The image edge is found using the double threshold method. For images between the double thresholds, the eight-connected domain method is used to complete the edge; the slope cracks are identified and displayed based on the Canny algorithm.
[0106] In order to synchronize the progress of slope stability monitoring by the vehicle-mounted radar warning model and the ground-based radar warning model, that is, to achieve synchronization of slope stability prediction accuracy between the vehicle-mounted radar warning model and the ground-based radar warning model, the present invention balances the model accuracy performance of the vehicle-mounted radar warning model and the ground-based radar warning model during the slope monitoring process through synchronization correction, so as to avoid a large prediction gap between the vehicle-mounted radar warning model and the ground-based radar warning model. The larger the prediction gap between the two, the greater the difference in slope stability judgment results output by the two, which further leads to the need for human participation in the analysis and judgment of the judgment results of the two to determine the final result of slope stability. Therefore, the present invention synchronizes and corrects the two, unifies the prediction results of the vehicle-mounted radar warning model and the ground-based radar warning model, and outputs results that corroborate / constrain each other, so as to accurately obtain the slope stability result, that is, to improve the accuracy of slope stability monitoring, as follows:
[0107] The construction method of the combined early warning model includes:
[0108] For the loss functions of the ground-based radar warning model and the vehicle-mounted radar warning model, respective correction factors are set;
[0109] The ground-based radar warning model and the vehicle-mounted radar warning model are corrected and trained by the correction factor to obtain a combined warning model.
[0110] The combined early warning model is:
[0111] LRadar=Softmax[SVM(RadarData_g),SVM(RadarData_c)];
[0112] Among them, the loss function of the combined early warning model is:
[0113] LossRadar=Wg*Lossg+Wc*Lossc;
[0114] Where LRadar is the stability label prediction value of the combined warning model, Wg is the correction factor of the loss function of the ground-based radar warning model, Wc is the correction factor of the loss function of the vehicle-mounted radar warning model, Softmax is the softmax function, Lossg is the loss function value of the ground-based radar warning model, Lossc is the loss function value of the vehicle-mounted radar warning model, RadarData_g is the ground-based radar data, RadarData_c is the vehicle-mounted radar data, and SVM is the classifier. The setting methods of the correction factors include:
[0115] Track the training completion of ground-based radar warning models and vehicle-mounted radar warning models;
[0116] Negatively correlated to the training completion, a correction factor is set: ;in, ;
[0117] In the formula, , is the correction factor of the loss function of the ground-based radar early warning model at the tth moment, W c (t) is the correction factor of the loss function of the vehicle-borne radar warning model at the tth moment, C i (t-1)[C g (t-1),C c (t-1)],C g (t-1) is the training completion degree of the ground-based radar early warning model at the t-1th moment, C c (t-1) is the training completion degree of the vehicle-borne radar warning model at the t-1th moment, Loss i (t-1)[Loss g (t-1),Loss c (t-1)],Loss i (t-2)[Lossg (t-2),Loss c (t-2)], Loss g (t-1), Loss g (t-2) are the loss functions of the ground-based radar early warning model at the t-1st and t-2nd moments, respectively. c (t-1), Loss c (t-2) are the loss functions of the vehicle-borne radar warning model at the t-1th and t-2th moments, respectively. T is the total time length of the training model. i , j , t are all count variables, g is the identifier of the ground-based radar, and c is the identifier of the vehicle-mounted radar. i =g, W i (t) = W g (t), C i (t-1)=C g (t-1), Loss i (t-1)=Loss g (t-1), Loss i (t-2) = Loss g (t-2); ;in, ;when i = c, W i (t) = W c (t), C i (t-1)=C c (t-1), Loss i (t-1)=Loss c (t-1), Loss i (t-2) = Loss c (t-2); ;in, .
[0118] The present invention sets a correction factor, which is bound to the loss function, so as to reduce the proportion of the loss function of the model with faster training progress / higher precision performance in the total loss, thereby lowering the model training progress or precision performance, so that the model with slow training progress and poor precision performance has a high proportion in the total loss, and improves its training progress or precision performance, thereby realizing the correction of the vehicle-mounted radar warning model and the ground-based radar warning model to synchronized operation, that is, the vehicle-mounted radar warning model and the ground-based radar warning model perform stability monitoring with similar precision performance, and the vehicle-mounted radar warning model and the ground-based radar warning model obtain unified and accurate judgment.
[0119] The vehicle-mounted radar data and ground-based radar data are normalized.
[0120] like Figure 2 As shown, the present invention provides an open-pit mine slope radar monitoring and early warning system, which is applied to an open-pit mine slope radar monitoring and early warning method, and the system includes:
[0121] A data acquisition unit, used to monitor the open-pit mine slope using ground-based radar and vehicle-mounted radar, and obtain ground-based radar data and vehicle-mounted radar data for judging the stability of the open-pit mine slope;
[0122] The data processing unit uses a combined early warning model to determine the stability of the open pit slope based on ground-based radar data and vehicle-mounted radar data;
[0123] The data storage unit is used to store the combined early warning model.
[0124] The pre-establishment method of the combined early warning model in the data storage unit includes:
[0125] For the loss functions of the ground-based radar warning model and the vehicle-mounted radar warning model, respective correction factors are set;
[0126] The ground-based radar warning model and the vehicle-mounted radar warning model are corrected and trained by the correction factor to obtain a combined warning model.
[0127] The combined early warning model is:
[0128] LRadar=Softmax[SVM(RadarData_g),SVM(RadarData_c)];
[0129] Among them, the loss function of the combined early warning model is:
[0130] LossRadar=Wg*Lossg+Wc*Lossc;
[0131] Wherein, LRadar is the stability label prediction value of the combined warning model, Wg is the correction factor of the loss function of the ground-based radar warning model, Wc is the correction factor of the loss function of the vehicle-mounted radar warning model, Softmax is the softmax function, Lossg is the loss function value of the ground-based radar warning model, Lossc is the loss function value of the vehicle-mounted radar warning model, RadarData_g is the ground-based radar data, RadarData_c is the vehicle-mounted radar data, and SVM is the classifier.
[0132] The present invention utilizes a classifier to perform classification training on ground-based radar data and vehicle-mounted radar data, and obtains a ground-based radar early warning model and a vehicle-mounted radar early warning model for judging the stability of an open-pit mine slope. The ground-based radar early warning model and the vehicle-mounted radar early warning model are synchronously corrected to obtain a combined early warning model for correcting the judgment accuracy of the open-pit mine slope stability, thereby realizing multi-category radar measurement and improving the reliability of slope stability monitoring.
[0133] The above embodiments are only exemplary embodiments of the present application and are not intended to limit the present application. The protection scope of the present application is defined by the claims. Those skilled in the art may make various modifications or equivalent substitutions to the present application within the essence and protection scope of the present application, and such modifications or equivalent substitutions shall also be deemed to fall within the protection scope of the present application.
Claims
1. A radar monitoring and early warning method for open-pit mine slopes, characterized in that: The following steps are involved: Use ground-based radar and vehicle-mounted radar to monitor the slope of the open-pit mine, and obtain ground-based radar data and vehicle-mounted radar data for judging the stability of the open-pit mine slope; Using a classifier to perform classification training on the ground-based radar data and the vehicle-mounted radar data, and correspondingly obtaining a ground-based radar early warning model and a vehicle-mounted radar early warning model for judging the stability of an open-pit mine slope; The ground-based radar early warning model and the vehicle-mounted radar early warning model are synchronously corrected to obtain a combined early warning model for correcting the judgment accuracy of open-pit mine slope stability; The construction method of the ground-based radar early warning model includes: Analyzing and labeling the stability status of the open pit mine slope on the ground-based radar data to obtain a stability label for the ground-based radar data; Using a classifier to perform classification learning on the ground-based radar data and the stability labels of the ground-based radar data, to obtain a ground-based radar early warning model that outputs the stability labels according to the ground-based radar data; The ground-based radar early warning model is: Lg=SVM(RadarData_g); Where Lg is the stable label prediction value of the ground-based radar warning model, RadarData_g is the ground-based radar data, and SVM is the classifier; Among them, assuming that the loss function value of the ground-based radar early warning model is Lossg, then: Lossg=MSE(Lg,Lr); Where Lg is the stable label prediction value of the ground-based radar warning model, Lr is the stable label true value of the ground-based radar data, MSE is the mean square error function, and MSE(Lg,Lr) is the mean square error between Lg and Lr; The construction method of the vehicle-mounted radar warning model includes: Analyzing and labeling the stability status of the open pit mine slope on the vehicle-borne radar data to obtain a stability label of the vehicle-borne radar data; The vehicle-mounted radar data is used as an input item of the ground-based radar warning model, and the stability label of the vehicle-mounted radar data is used as an output item of the ground-based radar warning model, and the ground-based radar warning model is subjected to transfer learning to obtain a vehicle-mounted radar warning model that outputs the stability label according to the vehicle-mounted radar data; The vehicle-mounted radar early warning model is: Lc=SVM(RadarData_c); Where Lc is the stable label prediction value of the vehicle radar warning model, RadarData_c is the vehicle radar data, and SVM is the classifier; Among them, assuming that the loss function value of the vehicle-mounted radar warning model is Lossc, then: Lossc=MSE(Lc,Lg)-MSE(RadarData_c,RadarData_g); Where Lg is the stable label prediction value of the ground-based radar warning model, Lc is the stable label prediction value of the vehicle-mounted radar warning model, RadarData_g is the ground-based radar data, RadarData_c is the vehicle-mounted radar data, MSE is the mean square error function, MSE(Lg,Lc) is the mean square error between Lg and Lc, and MSE(RadarData_c,RadarData_g) is the mean square error between RadarData_c and RadarData_g; The construction method of the combined early warning model includes: For the loss functions of the ground-based radar warning model and the vehicle-mounted radar warning model, respective correction factors are set; The ground-based radar warning model and the vehicle-mounted radar warning model are corrected and trained by the correction factor to obtain a combined warning model; The combined early warning model is: LRadar=Softmax[SVM(RadarData_g),SVM(RadarData_c)]; Among them, assuming that the loss function value of the combined early warning model is LossRadar, then: LossRadar=Wg*Lossg+Wc*Lossc; Wherein, LRadar is the stability label prediction value of the combined warning model, Wg is the correction factor of the loss function of the ground-based radar warning model, Wc is the correction factor of the loss function of the vehicle-mounted radar warning model, Softmax is the softmax function, Lossg is the loss function value of the ground-based radar warning model, Lossc is the loss function value of the vehicle-mounted radar warning model, RadarData_g is the ground-based radar data, RadarData_c is the vehicle-mounted radar data, and SVM is the classifier.
2. The open-pit mine slope radar monitoring and early warning method according to claim 1 is characterized in that: The analysis and annotation methods of ground-based radar data include: Using the ground-based radar data, image analysis is performed on crack properties on the open-pit mine slope to obtain crack characteristics of the open-pit mine slope; The stability of the ground-based radar data whose crack characteristics are greater than the crack threshold is marked as a non-steady-state label, and the stability of the ground-based radar data whose crack characteristics are less than or equal to the crack threshold is marked as a stable-state label; The analysis and annotation methods of vehicle-mounted radar data include: Using the vehicle-borne radar data, image analysis is performed on crack properties on the open-pit mine slope to obtain crack characteristics of the open-pit mine slope; The stability of the vehicle-mounted radar data whose crack characteristics are greater than the crack threshold is marked as a non-steady-state label, and the stability of the vehicle-mounted radar data whose crack characteristics are less than or equal to the crack threshold is marked as a stable-state label.
3. The open-pit mine slope radar monitoring and early warning method according to claim 2 is characterized in that: The methods for obtaining crack characteristics of open-pit mine slopes include: The ground-based radar data is mapped in a three-dimensional coordinate space to construct a three-dimensional image of the open-pit mine slope. The crack edges are detected in the three-dimensional image of the open-pit mine slope through an edge extraction algorithm, and the position, width, length and depth characteristics of the cracks are calculated from the crack edges. The vehicle-mounted radar data is mapped in the three-dimensional coordinate space to construct a three-dimensional image of the open-pit mine slope. The crack edges are detected in the three-dimensional image of the open-pit mine slope through the edge extraction algorithm, and the position, width, length and depth characteristics of the cracks are calculated from the crack edges.
4. The open-pit mine slope radar monitoring and early warning method according to claim 3 is characterized in that: The correction factor setting methods include: Tracking the training completion of ground-based radar warning models and vehicle-mounted radar warning models; A correction factor is set based on the training completion degree. for: ;in, ; Where: is the correction factor of the loss function of the ground-based radar early warning model at the tth moment, W c (t) is the correction factor of the loss function of the vehicle-borne radar early warning model at the tth moment; C i (t-1)[C g (t-1),C c (t-1)],C g (t-1) is the training completion degree of the ground-based radar early warning model at the t-1th moment, C c (t-1) is the training completion degree of the vehicle-borne radar warning model at the t-1th moment; Loss i (t-1)[Loss g (t-1),Loss c (t-1)],Loss i (t-2)[Loss g (t-2),Loss c (t-2)], Loss g (t-1), Loss g (t-2) are the loss functions of the ground-based radar early warning model at the t-1st and t-2nd moments, respectively. c (t-1), Loss c (t-2) are the loss functions of the vehicle-borne radar warning model at the t-1th and t-2th moments respectively; T is the total time length of the training model, i , j , t are counting variables, g is the identifier of the ground-based radar, and c is the identifier of the vehicle-mounted radar.
5. The open-pit mine slope radar monitoring and early warning method according to claim 1 is characterized in that: The vehicle-mounted radar data and ground-based radar data are normalized.
6. An open-pit mine slope radar monitoring and early warning system, characterized in that: An open-pit mine slope radar monitoring and early warning method applied to any one of claims 1 to 5, the system comprising: A data acquisition unit, used to monitor the open-pit mine slope using ground-based radar and vehicle-mounted radar, and obtain ground-based radar data and vehicle-mounted radar data for judging the stability of the open-pit mine slope; The data processing unit uses a combined early warning model to determine the stability of the open pit slope based on ground-based radar data and vehicle-mounted radar data; The data storage unit is used to store the combined early warning model.
