Identification Method and System for Automatically Detecting Geological Disaster Monitoring Equipment Based on LiDAR

Abstract

The present invention belongs to the technical field of geological disaster monitoring, and discloses an identification method and system for automatically discovering geological disaster monitoring equipment based on lidar. The method determines the variability of the connection between the relative position and geological disaster information; adds error parameters to the data-driven geological disaster monitoring equipment movement trajectory evolution model and trains it to predict the movement change and its variance of geological disaster information in adjacent connection areas, and calculates the variability of the connection of geological disaster information in the prediction result; at the same time, obtains the credibility index of the prediction; performs trajectory correction based on lidar; constructs a lidar digital-analog positioning and identification model based on Kalman filtering, obtains the positioning and identification result of the automatically discovered geological disaster monitoring equipment, and obtains the geological disaster monitoring information at this position according to the positioning and identification result. The present invention solves the problem of still being able to obtain accurate geological disaster evolution information after the position of the geological disaster monitoring equipment changes, which has practical significance for the monitoring project.

Description

Technical Field

[0001] The present invention belongs to the technical field of geological disaster monitoring, and particularly relates to an identification method and system for automatically discovering geological disaster monitoring equipment based on lidar. Background Art

[0002] After a new geological disaster monitoring device is deployed at a corresponding geographical location, it is necessary to manually configure the driver corresponding to the new geological disaster monitoring device on the server so that the server can analyze the geological disaster monitoring data sent by the new geological disaster monitoring device. However, this method requires manual configuration, with low flexibility and low efficiency.

[0003] To solve the above problems, the prior art invention patent (publication number CN114625382A, publication date 20220614) provides a method for automatically discovering geological disaster monitoring equipment. After a new geological disaster monitoring device is deployed at a corresponding location, the device information of the newly deployed geological disaster monitoring device is added to the database in real time. The server obtains the device information of the newly deployed geological disaster monitoring device from the database and determines whether the driver corresponding to the newly deployed geological disaster monitoring device has been installed according to the device information. If it is determined that the driver corresponding to the newly deployed geological disaster monitoring device has not been installed, the driver corresponding to the newly deployed geological disaster monitoring device is matched and obtained from the driver pool and then installed to obtain and analyze the geological disaster monitoring data sent by the newly deployed geological disaster monitoring device. The automatic installation of the driver for the newly deployed geological disaster monitoring device is realized.

[0004] The lidar system uses light pulses to measure the distance to an object based on the time of flight (TOF) of each light pulse. The light pulse emitted from the light source of the lidar system interacts with a far-side object. A part of the light is reflected from the object and returns to the detector of the lidar system. The distance is estimated based on the time elapsed between the emission of the light pulse and the detection of the returned light pulse. In some examples, a laser emitter generates light pulses. The light pulses are focused by a lens or a lens assembly. The time it takes for the laser pulse to return to the detector installed near the emitter is measured. A high-accuracy distance is obtained based on the time measurement result.

[0005] 3D point clouds are required in many operating scenarios. Many solutions have been used to interrogate the surrounding environment in three dimensions. In some examples, a 2D instrument is typically actuated up and down and / or back and forth on a gimbal. This is commonly referred to in the art as "blinking" or "nodding" the sensor. Therefore, a single-beam lidar unit can be used to capture an entire 3D array of distance points, although one point is captured at a time. In related examples, a prism is used to "divide" the laser pulse into multiple layers, each layer having a slightly different vertical angle. This simulates the above-mentioned nodding effect without actuating the sensor itself.

[0006] In many applications, a sequence of emission pulses is transmitted. The direction of each pulse changes rapidly and continuously in sequence. In these examples, the distance measurement associated with each individual pulse can be regarded as a pixel, and the set of pixels (i.e., "point cloud") that are emitted and captured rapidly and continuously can be presented as an image or analyzed for other reasons (e.g., detecting obstacles). In some examples, viewing software is employed to present the resulting point cloud as an image that appears three-dimensional to the user.

[0007] Through the above analysis, the problems and defects existing in the prior art are as follows: Generally, geological disaster monitoring devices include one or more of a rain gauge, a soil moisture meter, a water level gauge, a piezometer, a crack gauge, a surface displacement monitor, and an inclinometer. The base station for analyzing the data information of geological disaster monitoring devices includes multiple relay devices. These data acquisitions, driving, and analyses are usually based on the fact that the monitoring positions of geological disaster monitoring devices remain unchanged during a certain period for corresponding information acquisition and processing. However, in long-term monitoring tasks, due to geological changes, the fixed positions of geological disaster monitoring devices may change. At this time, all information may change, which may be misinterpreted as a change in geological disasters, giving a wrong signal for subsequent preventive measures. Therefore, relying solely on the prior art cannot solve the problem of obtaining accurate geological disaster evolution information after the positions of geological disaster monitoring devices change. For this reason, the present application proposes an identification method for automatically discovering geological disaster monitoring devices based on lidar, which has practical significance for monitoring projects. Summary of the Invention

[0008] To overcome the problems existing in the related art, the disclosed embodiments of the present invention provide an identification method and system for automatically discovering geological disaster monitoring devices based on lidar.

[0009] The technical solutions are as follows: The identification method for automatically discovering geological disaster monitoring devices based on lidar includes:

[0010] S1, constructing and training a data-driven evolution model for the movement trajectory of geological disaster monitoring devices, and using the trained model and based on the data measured by the sensors of geological disaster monitoring devices, determining the relative position and the variability of the connection of geological disaster information;

[0011] S2, adding error parameters to the data-driven evolution model for the movement trajectory of geological disaster monitoring devices for training, predicting the movement changes and their variances of geological disaster information in adjacent connection regions, and calculating the variability of the connection of geological disaster information in the prediction results to obtain a prediction credibility index;

[0012] S3, performing lidar-based trajectory correction;

[0013] S4. Construct a lidar digital-analog positioning and recognition model based on Kalman filtering. Tightly couple the lidar spectrum ranging observables through extended Kalman filtering, and based on the high-frequency displacement rate vector output by the data-driven geological disaster monitoring equipment movement trajectory evolution model, obtain the positioning and recognition results of the automatically discovered geological disaster monitoring equipment, and obtain the geological disaster monitoring information at this location according to the positioning and recognition results.

[0014] In step S1, construct and train a data-driven geological disaster monitoring equipment movement trajectory evolution model, including:

[0015] Construct the data-driven part. With the data of the acceleration monitor and gyroscope built into the geological disaster monitoring equipment, estimate the movement change speed v of the geological disaster monitoring equipment. The expression is:

[0016] v = [v x v y v z T

[0017] In the formula, v x is the lateral change speed, v y is the longitudinal change speed, v z is the spatial direction change speed, and T is the order;

[0018] Construct the model-driven part. Combine the velocity vector with the ranging result based on the lidar wave, and use extended Kalman filtering to estimate the position, displacement direction, and velocity of the geological disaster monitoring equipment.

[0019] Furthermore, before constructing and training the data-driven geological disaster monitoring equipment movement trajectory evolution model, it is necessary to perform normalization preprocessing based on quaternions, specifically including:

[0020] Use an algorithm based on gradient descent to calculate the attitude quaternion, and use the attitude quaternion to transform the acceleration and angular velocity from the sensor coordinate system to the navigation coordinate system. The vector β in the sensor coordinate system IMU is transformed into β in the navigation coordinate system through the following formula INS , and the expression is:

[0021]

[0022] In the formula, α is the conjugate, is the conjugate complex of the quaternion;

[0023] Quaternions are used to rotate vectors and convert them into Euler angles. The definition order of the sensor attitude angles of the geological disaster monitoring equipment is <z, x, y>, where z is the direction angle, x is the pitch angle, and y is the roll angle.

[0024] ​In step S1, based on the data measured by the sensors of the geological disaster monitoring equipment, the relative position and the connection variability of geological disaster information are determined, including:

[0025] Using the sliding window method, with a fixed-size sliding window, the data of the geological disaster monitoring equipment sensors are divided into independent sequences; the window size n of the sequence is 100 frames, and the shortest distance of the sliding window is 5 frames; the velocity vector is predicted by the deep neural network of each sequence, the position is generated through the merging link module, and the current position is updated to 100 frames; the prediction is updated using overlapping windows, the output frequency is increased to 40Hz, and the position obtained by processing the velocity vector through a low-pass filter is used to smoothly reconstruct the predicted trajectory.

[0026] Using the publicly available standard dataset and the self-collected dataset as the training set and the test set, during the training phase, the features and labels of each window are randomly rotated to enhance the displacement direction independence of the data; the training objective is to minimize the mean square error between the estimated value and the true value provided by the dataset, and the best parameters are obtained through training with the ADAM optimizer. After training on the GPU, the model is directly applied to the local geological disaster monitoring equipment sensors to calculate the displacement trajectory of the geological disaster monitoring equipment.

[0027] In step S2, the connection variability of the geological disaster information of the prediction result is calculated to obtain the prediction credibility index, including:

[0028] The total connection variability of the geological disaster information predicted by the network is defined as σ total , and the expression is:

[0029] σ total = var p(y|x) (y)

[0030] In the formula, var p(y|x) (y) is the predicted value of the probability distribution in the total connection variability of the geological disaster information, and p(y|x) is the conversion function between the roll angle y and the pitch angle x;

[0031] According to the source of the connection variability of the geological disaster information, it is divided into the data connection variability of the geological disaster information σ total and the model connection variability of the geological disaster information σ model . The data connection variability of the geological disaster information indicates the anti-noise performance of the network for the input data;

[0032] The connection variability of the geological disaster information is based on the Bayesian belief network, with the help of the expectation propagation EP framework, using the density filtering ADF method, and the general network layer of ResNet18 is overwritten by mathematical reasoning. The connection variability network layer of the geological disaster information is defined as follows:

[0033] For any network layer c (i) = h(i) (c (i-1) , θ (i) ), the geological disaster information connection variability propagation layer is implemented based on the following formula:

[0034]

[0035] In the formula, h (i) is the i-th propagation layer, c (i-1) is the (i - 1)-th network layer, θ (i) is the variability value of the i-th network layer, is the sampling prediction value of the geological disaster information connection variability propagation layer, is the variance in the sampling prediction of the geological disaster information connection variability propagation layer, is the sampling prediction method in the (i - 1)-th network layer, is the method of the sampling prediction variance in the (i - 1)-th network layer;

[0036] For the model geological disaster information connection variability, the expectation and variance are output through Monte Carlo sampling. The dropout layer after the fixed activation layer ReLU is in the training mode, and the parameters will be randomly discarded; the noise variance is initialized and propagated together with the test data to obtain several groups of sampling prediction values b and the corresponding variance ζ; the credibility of the network for the output is evaluated to obtain the model geological disaster information connection variability.

[0037] Furthermore, evaluating the credibility of the network for the output to obtain the model geological disaster information connection variability includes: the network undergoes several propagations and Monte Carlo samplings to obtain several groups of prediction values b i and variance ζ i , and calculate the model uncertainty γ model , and the expression is:

[0038] γ model = var(b i )

[0039] In the formula, var() is the variance calculation;

[0040] After Monte Carlo sampling, the geological disaster information connection variability of the data is expressed as the mean of several sampling prediction values, and further the total uncertainty γ total of the network, and the expression is:

[0041] γ total = γ data + γ model

[0042] In the formula, γ data is the uncertainty of the current model.

[0043] In step S3, lidar-based trajectory correction is performed, including:

[0044] When the lidar magnetic field in the environment is stable, there is only one stable lidar magnetic field resultant vector, and the three-dimensional magnetic vector of the magnetometer changes with the attitude of the IMU; when the attitude change of the IMU and the change of the magnetic vector do not match, under the influence of the strong magnet, the lidar magnetic field environment is unstable; use the gyroscope to calculate the attitude change of the IMU and represent it in the form of a quaternion a i ; use the quaternion to transform the magnetic vector m i into calculate the modulus of the vector difference between m i+1 and as the matching degree su; the larger the matching degree, the greater the attitude difference between the IMU and the magnetic vector, and the more unstable the lidar magnetic field; m i+1 is the (i + 1)-th magnetic vector, is the transformed (i + 1)-th magnetic vector;

[0045] The lidar magnetic field strength reflects the stability of the lidar environment. Take the sliding standard deviation std of the lidar magnetic field strength as the condition for determining the stability of the lidar magnetic field; set the size of the sliding window to 100 frames, and define the stability of the lidar magnetic field as:

[0046]

[0047] In the formula, is the Reynolds derivative, is the stability value of the lidar magnetic field within the window size d, is the sliding average value, thrshd1 is the first window of the sliding window size d, and thrshd2 is the second window of the sliding window size d;

[0048] When the average sliding standard deviation and the average matching degree are less than the threshold, the lidar magnetic field is completely stable, then set the stability of the lidar magnetic field to 1, and the range of the stability of the lidar magnetic field is from 0 to 1, indicating from unstable to stable;

[0049] Select the stable lidar magnetic field environment and low geological disaster information, and connect the associated quantities on the straight-line sequence with variability as the anchor point g k ; if the posture of the sensor of the geological disaster monitoring device is not changed, each associated quantity of the straight-line sequence will fluctuate within a small range; when the fluctuation is greater than the threshold, it is determined that the relative horizontal attitude relationship between the geological disaster monitoring device sensor and the person has changed, and at this time, g needs to be updated k ; use g k to calculate the correction angle of the straight-line sequence The expression is:

[0050]

[0051] In the formula, ft i is the average stability of the lidar magnetic field, and g i is the i-th anchor point;

[0052] By using the neural network shortest distance calculation method, the correction angle of the straight-line sequence is predicted to achieve adaptive correction of the trajectory.

[0053] In step S4, a lidar digital-analog positioning and recognition model based on Kalman filtering is constructed, including:

[0054] S4.1, constructing a system model and an observation model, adopting the data-driven shortest distance calculation method in the local coordinate system A (x A , y A , α A ), and selecting the northeast celestial coordinate system as the absolute coordinate system B (e B , n B , α B ), where α is the displacement direction, and the system state vector to be estimated X p is described as:

[0055] X p = [e p n p θ p T

[0056] In the formula, (e p , n p ) are the east and north coordinates of the target in the B coordinate system at time P, and θ p is the displacement direction angle of the A coordinate system relative to the B coordinate system in the right-hand system, including the initial deviation angle between the two coordinate systems at the moment when the positioning service is started and the cumulative error deviation angle of the sensor; driven by the data of the shortest distance calculation method, the state transition of the system is as follows:

[0057]

[0058] In the formula, h() is the function to be estimated for the state vector, O p is the initial value of the state at time p, e p-1 is the east direction coordinate at time p-1, n p-1 is the north direction coordinate at time p-1, θ p-1 is the displacement direction angle of the A coordinate system relative to the B coordinate system in the right-hand system at time p-1, X p-1 is the system state vector to be estimated at time p-1, Δx p , Δyp They are the displacement increments in the horizontal and vertical directions of the shortest distance calculation method from time p - 1 to time p in the local coordinate system, respectively.

[0059] The lidar spectrum positioning estimation adopts the time - division multiple access and frequency - division multiple access strategies to form a lidar spectrum signal network; the observation data containing at least two lidar spectra are solved, and the observation equation is obtained through geometric relationships. The expression is:

[0060]

[0061] In the formula, C p is the observation data at time p, q() is the time - division multiple access estimation function, E p is the frequency - division multiple access estimated value, F1 is the first time - division address, F2 is the second time - division address, F i-1 is the (i - 1)th time - division address, F i is the ith time - division address, F m-1 is the time - division address of the lidar spectrum estimated in the (m - 1)th epoch, F m is the time - division address of the lidar spectrum estimated in the mth epoch, i is the number, g is the sound propagation speed at temperature T, and m is the number of lidar spectra estimated in the epoch.

[0062] The model is initialized and the model parameters are dynamically adjusted.

[0063] Furthermore, the model initialization includes:

[0064] Combining the observation information with the lidar simple particle filter to calculate the initial plane coordinates of the geological disaster monitoring equipment sensor; the coordinate data contained in the observation information gives the regional range where the geological disaster monitoring equipment sensor is located. A certain number of particles are evenly distributed within this range. Through resampling operations, given the particle attribute [e B n B T , the particle set is initialized as:

[0065]

[0066] In the formula, e min , e max , n min , n max are the regional boundaries delimited by the initial observation information, N is the number of particles, e l is the lth particle attribute in the east direction, n l is the lth particle attribute in the north direction, and L() is the particle attribute function under the constraint conditions of the regional boundaries delimited by the initial observation information.

[0067] ​The overall spectral estimation error of the lidar follows a normal distribution. According to the prediction residual δ, the weight of each particle is:

[0068]

[0069] σ = C0 - C l

[0070] In the formula, w l is the weight of the l-th particle, R l is the radius of the l-th particle distribution, exp{} is the residual prediction function, σ l is the prediction residual of the l-th particle, C0, C l are the initial observation and the predicted observation respectively;

[0071] After normalizing the weights of the particle set and using the random resampling method to obtain a new particle set, an average processing is performed to obtain the initial value estimation:

[0072]

[0073] In the formula, e0 is the original particle attribute in the east direction, n0 is the original particle attribute in the north direction, e l is the l-th particle attribute in the east direction, n l is the l-th particle attribute in the north direction;

[0074] Perform dynamic adjustment of the model parameters, including:

[0075] On the premise of the given initial error covariance matrix of the system W0 = I 3×3 , make the displacement direction angle converge rapidly, enhance the basic weight of the observation information in the early stage of filtering, and amplify the process noise, and adjust the observation noise corresponding to each lidar observation data. The expression is:

[0076]

[0077] In the formula, is the i-th diagonal element of the observation noise covariance matrix, is the initial diagonal element of the observation noise, is the prediction residual of the (p + 1)-th lidar spectrum, is the prediction residual of the p-th lidar spectrum, P0 is the variance inflation threshold;

[0078] The geological disaster information connection variability factor of the data-driven shortest distance calculation method model gives guidance on adjusting the state and observation weight in a timely manner when detecting the attitude anomaly update caused by the lidar mutation.

[0079] Another object of the present invention is to provide an identification system for automatically discovering geological disaster monitoring equipment based on lidar. The system implements the identification method for automatically discovering geological disaster monitoring equipment based on lidar. The system includes:

[0080] A module for determining the connection variability of position and geological disaster information, which is used to construct and train an evolution model of the movement trajectory of geological disaster monitoring equipment based on data driving. Using the trained model and based on the data measured by the sensors of the geological disaster monitoring equipment, the connection variability of relative position and geological disaster information is determined;

[0081] A credibility index prediction module, which is used to add error parameters to the evolution model of the movement trajectory of geological disaster monitoring equipment based on data driving for training, predict the movement change and its variance of geological disaster information in adjacent connection areas, calculate the connection variability of geological disaster information in the prediction result, and obtain the predicted credibility index;

[0082] A trajectory correction module, which is used to perform lidar-based trajectory correction;

[0083] A geological disaster monitoring information identification module, which is used to construct a lidar digital-analog positioning identification model based on Kalman filtering, closely couple the lidar spectrum ranging observables through extended Kalman filtering, and based on the high-frequency displacement rate vector output by the evolution model of the movement trajectory of geological disaster monitoring equipment based on data driving, obtain the positioning identification result of automatically discovering geological disaster monitoring equipment, and obtain the geological disaster monitoring information at this position according to the positioning identification result.

[0084] Combining all the above technical solutions, the beneficial effects of the present invention are as follows: The present invention can monitor the monitoring position information of geological disaster monitoring equipment in a certain period in real time, correctly identify the changing positions, and can seamlessly connect with the information at the original position, further obtain accurate information on the evolution of geological disasters, and will not be misinterpreted as a large abnormal change in geological disasters, giving an accurate signal for later prevention measures, solving the problem of still being able to obtain accurate information on the evolution of geological disasters after the position of the geological disaster monitoring equipment changes, which has practical significance for the monitoring project. BRIEF DESCRIPTION OF THE DRAWINGS

[0085] The accompanying drawings here are incorporated into the specification and form a part of this specification, showing embodiments consistent with the present disclosure, and are used together with the specification to explain the principles of the present disclosure;

[0086] Figure 1 is a flowchart of the identification method for automatically discovering geological disaster monitoring equipment based on lidar provided by an embodiment of the present invention;

[0087] Figure 2It is a schematic diagram of an identification system for automatically discovering geological disaster monitoring equipment based on lidar provided by an embodiment of the present invention;

[0088] In the figure: 1. Module for determining the variability of the connection between position and geological disaster information; 2. Credibility index prediction module; 3. Trajectory correction module; 4. Geological disaster monitoring information identification module. Specific implementation manners

[0089] To make the above objects, features, and advantages of the present invention more obvious and understandable, the specific implementation manners of the present invention will be described in detail below with reference to the accompanying drawings. Many specific details are set forth in the following description in order to fully understand the present invention. However, the present invention can be implemented in many other ways different from those described herein, and those skilled in the art can make similar improvements without departing from the connotation of the present invention. Therefore, the present invention is not limited by the specific implementations disclosed below.

[0090] Example 1, as Figure 1 shown, the identification method for automatically discovering geological disaster monitoring equipment based on lidar provided by an embodiment of the present invention includes:

[0091] S1. Construct and train a data-driven evolution model of the movement trajectory of geological disaster monitoring equipment. Using the trained model and based on the data measured by the sensors of the geological disaster monitoring equipment, determine the variability of the connection between the relative position and geological disaster information;

[0092] S2. Add error parameters to the data-driven evolution model of the movement trajectory of geological disaster monitoring equipment for training, predict the movement change and its variance of geological disaster information in adjacent connection areas, calculate the variability of the connection of geological disaster information in the prediction result, and obtain the predicted credibility index;

[0093] S3. Perform lidar-based trajectory correction;

[0094] S4. Construct a lidar digital-analog positioning and identification model based on the Kalman filter. Tightly couple the lidar spectrum ranging observations through the extended Kalman filter, and based on the high-frequency displacement rate vector output by the data-driven evolution model of the movement trajectory of geological disaster monitoring equipment, obtain the positioning and identification result of automatically discovering geological disaster monitoring equipment, and obtain the geological disaster monitoring information at this position according to the positioning and identification result.

[0095] Example 2, the identification method for automatically discovering geological disaster monitoring equipment based on lidar provided by an embodiment of the present invention includes:

[0096] Step 1. A multi-source fusion positioning architecture based on the combination of digital and analog;

[0097] The estimation of the location of the geological disaster monitoring device is achieved by fusing lidar spectral ranging and the sensor data of the geological disaster monitoring device. The data processing architecture for estimating the location of the geological disaster monitoring device mainly consists of two parts;

[0098] (1) Construct a data-driven part. By using the data from the acceleration monitor and gyroscope built into the geological disaster monitoring device, estimate the moving change speed v of the geological disaster monitoring device. The expression is:

[0099] v = [v x v y v z T

[0100] In the formula, v x is the lateral change speed, v y is the longitudinal change speed, v z is the spatial direction change speed, and T is the order;

[0101] (2) Construct a model-driven part. Combine the velocity vector with the ranging result based on the lidar wave, and use the Extended Kalman Filter to estimate the location, displacement direction, and velocity of the geological disaster monitoring device.

[0102] Due to the characteristics of low power consumption and high data update frequency of the geological disaster monitoring device sensor, it is widely used in positioning. However, the measurement results of low-cost geological disaster monitoring device sensors (such as acceleration monitors, gyroscopes, and magnetometers) are easily affected by drift errors. For example, when the acceleration is directly double-integrated to calculate the displacement, the error almost grows geometrically. Using the measurement values of the geological disaster monitoring device sensor as the input, the present invention adopts a deep learning-based method to estimate the movement trajectory of the geological disaster monitoring device. This is a data-driven shortest distance calculation method, which is completely different from the model-driven method using image detection.

[0103] Step 2: Construct a data-driven evolution model for the movement trajectory of the geological disaster monitoring device;

[0104] ​The shortest distance calculation method relies on the sensors of geological disaster monitoring equipment to deduce and track the position trajectory of the geological disaster monitoring equipment. The traditional shortest distance calculation method mainly uses the shortest distance model formula to estimate the shortest distance, and gyroscopes and magnetometers to estimate the direction, and then deduce the position of each straight line segment. Since the shortest distance model needs to adjust parameters according to the characteristics of the geological disaster monitoring equipment, and the direction estimation needs to limit the monitoring posture of the sensors of the geological disaster monitoring equipment, this results in the inability of the traditional shortest distance calculation method to be used in general scenarios. The data-driven shortest distance calculation method rewrites the probabilistic neural network as a geological disaster information connection variability neural network based on mathematical inference, and uses a large amount of preprocessed sensor data of geological disaster monitoring equipment and the real trajectory to train a neural network model that adapts to different postures and different geological disaster monitoring equipment. This model is driven by the sensor data of the geological disaster monitoring equipment and outputs the relative position and the geological disaster information connection variability. Due to the cumulative error in the displacement direction of the shortest distance calculation method, by using the lidar magnetic field stability detection algorithm and combining the geological disaster information connection variability output by the neural network, automatic correction is performed at the appropriate time.

[0105] 2.1 Normalization preprocessing based on quaternions;

[0106] Before inputting the sensor data of the geological disaster monitoring equipment into the neural network, the data needs to be preprocessed to reduce the complexity of the data so that the network can better fit the displacement model of the geological disaster monitoring equipment. After calculating the attitude quaternion using the algorithm based on gradient descent through the 9-axis sensor data, the acceleration and angular velocity are converted from the sensor coordinate system to the navigation coordinate system using the attitude quaternion. The vector β in the sensor coordinate system IMU can be converted to β in the navigation coordinate system through the following formula INS , and the expression is:

[0107]

[0108] In the formula, α is the conjugate, is the conjugate complex of the quaternion;

[0109] Quaternions are used to rotate vectors and convert them into Euler angles. The definition order of the attitude angles of the sensors of the geological disaster monitoring equipment is <z, x, y>, where z is the direction angle, x is the pitch angle, and y is the roll angle.

[0110] Since the initial coordinate systems of each motion trajectory are different, and in most actual positioning application scenarios, absolute global coordinates are required. Therefore, it is necessary to find the rotation relationship between the navigation coordinate system and the global coordinate system of each sequence. Through quaternions and attitude angles, the present invention can calculate the rotation relationship from the navigation coordinate system to the global coordinate system.

[0111] Quaternions can only be used to rotate vectors and cannot calculate the angle of their own rotation. It is necessary to convert them into Euler angles. The order of the definition of the attitude angles of the sensors of the geological disaster monitoring equipment is <z, x, y>, that is, the displacement direction angle, the pitch angle, and the roll angle. Therefore, the quaternion is converted into Euler angles in this order. The Z axes of the navigation coordinate system and the global coordinate system are the same, and there is only a difference in the Y axis on the horizontal plane. Therefore, the difference between the Euler angle and the displacement direction angle of the attitude angle is taken to convert the motion trajectory from the navigation coordinate system to the global coordinate system.

[0112] 2.2 Displacement trajectory calculation of geological disaster monitoring equipment based on deep neural network;

[0113] The traditional shortest distance calculation method algorithm obtains the displacement by performing double integration based on the data of the gyroscope and the acceleration monitor, and there is a relatively serious problem of error accumulation. The present invention uses a lightweight deep neural network to solve the problem of infinite error accumulation in traditional inertial navigation, and uses the ResNet18 residual neural network framework to learn the inertial tracking of the sensor data of the geological disaster monitoring equipment.

[0114] The physical model of the traditional inertial navigation system is based on Newton's mechanism. The sensors of low-cost geological disaster monitoring equipment have high noise, and the displacement trajectory derivation has extremely strong continuity. Therefore, errors will accumulate rapidly during the displacement trajectory calculation process. The models established on the estimated shortest distance and the shortest distance direction (such as the shortest distance calculation method) contain implicit motion models and will fail due to the evolution law of geological objects or the change of the use environment. In short, inertial navigation based on Zero Velocity Update (ZUPT) and navigation methods such as the shortest distance calculation method based on steps will be restricted by motion dynamics and sensor connection assumptions, while deep learning methods can extract high-level feature representations (velocity vectors) according to the periodic nature of the position change of the geological disaster monitoring equipment and do not depend on geometric theories. At the same time, the deep learning framework based on the ResNet18 architecture reduces the computational complexity to a certain extent, providing the possibility for deployment to the sensors of geological disaster monitoring equipment.

[0115] The present invention adopts a sliding window method to improve the output rate of neural network prediction. By using a sliding window with a fixed size, the sensor data of the geological disaster monitoring equipment will be divided into independent sequences. The window size n of the sequence is 100 frames (1 second), and the shortest distance of the sliding window is 5 frames. The velocity vector is predicted by the deep neural network of each sequence, and the position is generated through the merging link module. The current position is updated to the previous 100 frames. The prediction is updated using overlapping windows, and the output frequency will be increased to 40 Hz. Then, the position obtained from the velocity vector is further processed by a low-pass filter to smoothly reconstruct the predicted trajectory.

[0116] The present invention uses public standard data sets and self-collected data sets as training sets and test sets. During the training phase, the features and labels of each window are randomly rotated to enhance the independence of the displacement direction of the data. The training goal is to minimize the mean square error between the estimated value and the true value provided by the data set. The ADAM optimizer is used to train to obtain the optimal parameters. The learning rate can be set to 1e -5 , the batch size is set to 256. After training on the GPU, the model can be directly applied to the local geological disaster monitoring equipment sensor to infer the displacement trajectory of the geological disaster monitoring equipment.

[0117] 2.3 Geohazard information connection and variability network based on EP framework;

[0118] Deep neural network navigation is generally evaluated by the mean square error between the estimated value and the true value. However, the true value trajectory cannot be obtained during actual deployment, so the accuracy of the network prediction cannot be evaluated (even if the prediction deviation is large). Therefore, the present invention studies the variability of geological hazard information connection of deep learning habitual navigation, which represents the credibility of the output of the deep neural network model. The estimation of geological hazard information connection variability can quantify the degree of trust model prediction. By adding error parameters to the neural network framework and training, the movement changes and variances of geological hazard information in adjacent connection areas can be predicted, thereby inferring the variability of geological hazard information connection of the prediction results, and then providing credibility indicators for different prediction segments for subsequent fusion algorithms. The residual network displacement trajectory calculation framework is extended to the Bayesian model, and the probability network is overwritten as a geological hazard information connection variability network based on mathematical inference, so as to quickly estimate the variability of geological hazard information connection of inertial navigation in an unsupervised manner.

[0119] The total geological hazard information connection variability of network prediction is defined as σ total , the expression is:

[0120] σ total =var p(y|x) (y)

[0121] In the formula, var p(y|x) (y) is the predicted value of probability distribution in the connection change of total geological hazard information, and p(y|x) is the conversion function between roll angle y and pitch angle x;

[0122] According to the source of geological hazard information connection variability, it is divided into data geological hazard information connection variability σ data and the model geological hazard information connection variability σ model The variability of data geological disaster information connection indicates the network's anti-noise performance for input data, and the model uncertainty indicates the network's confidence level in the prediction.

[0123] The geological hazard information connection variability scheme is based on the Bayesian belief network. With the help of the expectation propagation (EP) framework, the assumed density filtering (ADF) method is used to overwrite the general network layer of ResNet18 through mathematical reasoning. The geological hazard information connection variability network layer is defined as follows:

[0124] Definition 1: For any network layer c (i) =h (i) (c (i-1) ,θ (i) ), the geological disaster information connection and changeability propagation layer is approximately realized based on the following formula:

[0125]

[0126] In the formula, h (i) is the i-th propagation layer, c (i-1) is the i-1th network layer, θ (i) is the change value of the i-th network layer, To connect the geological disaster information with the variable transmission layer sampling prediction value, The variance in the sampling prediction of the variable propagation layer for the geological hazard information connection is is the sampling prediction method in the i-1th network layer, is the method of sampling prediction variance in the i-1th network layer;

[0127] The geological hazard information connection variability network enables the noise variance initialized to the order of 1e-3 to propagate forward in the network, probabilizing the network layer of the deep model, and finally the network will obtain the mean and variance. For a single propagation, the mean μ represents the network prediction value, and the variance v represents the final output of the initial noise forward propagation, which is the data geological hazard information connection variability.

[0128] For the variability of model geological disaster information connection, the network outputs the expectation and variance through Monte Carlo sampling. The dropout layer after the fixed activation layer ReLU is in training mode, and the parameters are randomly discarded. The noise variance is initialized and propagated together with the test data to obtain several groups of sampled prediction values ​​b and corresponding variance ζ. The credibility of the network output (i.e., the variability of model geological disaster information connection) is evaluated through Definition 2.

[0129] Definition 2: After several propagations and Monte Carlo samplings as in Definition 1, the network obtains several sets of prediction values ​​b i and variance ζ i , calculate the model uncertainty γ model , the expression is:

[0130] γ model = var(b i )

[0131] where var() is the variance calculation;

[0132] After Monte Carlo sampling, the variability of the connection of data geological disaster information is expressed as the mean of several sampling prediction values, and then the total uncertainty γ of the network total , and the expression is:

[0133] γ total = γ data + γ model

[0134] γ data = mean(ζ i )

[0135] where γ data is the uncertainty of the current model, and mean() is the mean calculation.

[0136] Intuitively, the variability of the connection of data geological disaster information is the result of the propagation of noise in the network and interacts with the input data in the network layer. The model uncertainty can be regarded as a test of the prediction stability of the network in the case of discarding parameters. In the calculation of the total variability of the connection of geological disaster information in the network, the network framework of the variability of the connection of geological disaster information is adopted, which avoids the computational complexity of the Bayesian network and can give the credibility index of the network prediction without affecting the network structure and output and without supervision, thus providing a reference for the subsequent fusion algorithm.

[0137] 2.4 Trajectory correction based on lidar;

[0138] Since there is an accumulated error in the displacement direction in the trajectory predicted by the neural network and this error cannot be eliminated by the model, it is necessary to combine lidar information to correct the trajectory. The displacement direction angle in the lidar magnetic field information has a certain correlation with the body direction when going straight. The displacement direction angle is independent of time and is not affected by cumulative drift. It is an absolute quantity and can be used to correct the trajectory of the shortest distance calculation method. In order to obtain stable displacement direction angle information, it is necessary to design a lidar magnetic field stability detection algorithm to screen out stable and available displacement direction angles. And the variability of the connection of geological disaster information output by the neural network is used to screen out the available body directions of geological disaster monitoring equipment.

[0139] When the lidar magnetic field in the environment is stable, there is only one stable lidar magnetic field resultant vector, and the three-dimensional magnetic vector of the magnetometer changes with the attitude of the IMU; when the attitude change of the IMU does not match the change of the magnetic vector, under the influence of the strong magnet, the lidar magnetic field environment is unstable; use the gyroscope to calculate the attitude change of the IMU and represent it in the form of a quaternion a i ; use the quaternion to transform the magnetic vector m i into calculate m i+1 and of the modulus of the vector difference as the matching degree su; the larger the matching degree, the greater the attitude difference between the IMU and the magnetic vector, and the lidar magnetic field is unstable; m i+1 is the (i + 1)-th magnetic vector, is the transformed (i + 1)-th magnetic vector;

[0140] At the same time, the lidar magnetic field strength can also reflect the stability of the lidar environment. Take the sliding standard deviation std of the lidar magnetic field strength as one of the conditions for determining the lidar magnetic field stability. The size of the sliding window is set to 100 frames (0.5 s). Finally, the stability of the lidar magnetic field is defined by the following formula:

[0141]

[0142] In the formula, is the Reynolds derivative, is the stability value of the lidar magnetic field within the window size d, is the sliding average value, thrshd1 is the first window of the sliding window size d, and thrshd2 is the second window of the sliding window size d;

[0143] When the average sliding standard deviation and the average matching degree are less than the threshold, the lidar magnetic field is completely stable, then the stability of the lidar magnetic field is set to 1, and the stability of the lidar magnetic field ranges from 0 to 1, indicating from unstable to stable;

[0144] Through the straight-line sequence, the present invention can obtain the horizontal rotation relationship between the body direction of the geological disaster monitoring device and the global coordinate system. And the displacement direction angle represents the horizontal rotation relationship between the IMU coordinate system and the global coordinate system. Since the human body is in a periodic motion state during displacement, the straight-line displacement direction angle almost fluctuates around a value. Directly take the average value of the straight-line displacement direction angle and find that it can well reflect the horizontal orientation of the IMU during straight-line movement. Use the difference between the average displacement direction angle of each straight-line sequence and the direction of the average shortest distance calculation method of the neural network to represent the correlation quantity c i of the lidar magnetic field information and the body direction.

[0145] Select the associated quantity on the straight-line sequence that connects the stable lidar magnetic field environment and the low geological disaster information connection variability as the anchor point g k If the posture of the sensor of the geological disaster monitoring device is not changed, the associated quantity of each straight-line sequence should only fluctuate within a small range. When the fluctuation is greater than a certain threshold, it can be determined that the relative horizontal posture relationship between the sensor of the geological disaster monitoring device and the person has changed significantly. At this time, g needs to be updated k After that, use g k to calculate the correction angle of the straight-line sequence

[0146]

[0147] In the formula, ft i is the average stability of the lidar magnetic field, and g i is the i-th anchor point;

[0148] By using the neural network shortest distance calculation method, predict the correction angle of the straight-line sequence Realize the adaptive correction of the trajectory

[0149] Step 3, construct a lidar digital-analog positioning recognition model based on Kalman filtering;

[0150] Compared with the single data source positioning method with many limitations, tightly coupling and complementing data of different dimensions and attributes can usually achieve better positioning accuracy and stability. The data-driven shortest distance calculation method algorithm has given fine-grained relative dynamic information in a short time. Therefore, only by supplementing with global observation information, a model can be constructed with mathematical expressions to provide position services in the absolute coordinate system. Since lidar spectrum positioning estimation is a high-precision, high-privacy, and high-security positioning technology, and it has the characteristics of being compatible with mass intelligent geological disaster monitoring device sensors and supporting an unlimited number of concurrent users, etc., the present invention takes the shortest distance calculation method and lidar digital-analog positioning data as examples, and gives a general fusion positioning model based on the Kalman filtering framework

[0151] Construct a system model and an observation model. Adopt the local coordinate system A system (x A , y A , α A ) for the data-driven shortest distance calculation method. Select the northeast celestial coordinate system as the absolute coordinate system B system (e B , n B , α B ). Among them, α is the displacement direction, and the system state vector to be estimated X p is described as:

[0152] X p = [ep n p θ p T

[0153] Where, (e p ,n p ) are the east and north coordinates of the target in the B coordinate system at time P, and θ p is the displacement direction angle of the A coordinate system relative to the B coordinate system in the right - hand system, including the initial deviation angle between the two coordinate systems at the moment when the positioning service is started and the cumulative error deviation angle of the sensor; Driven by the data of the shortest - distance calculation method, the state transition of the system is as follows:

[0154]

[0155] Where, h() is the function to be estimated of the state vector, O p is the initial value of the state at time p, e p-1 is the east - direction coordinate at time p - 1, n p-1 is the north - direction coordinate at time p - 1, θ p-1 is the displacement direction angle of the A coordinate system relative to the B coordinate system in the right - hand system at time p - 1, X p-1 is the state vector to be estimated of the system at time p - 1, Δx p , Δy p are the horizontal and vertical direction output displacement increments of the shortest - distance calculation method from time p - 1 to time p in the local coordinate system respectively;

[0156] The lidar spectrum positioning estimation adopts the time - division multiple access and frequency - division multiple access strategies to form a lidar spectrum signal network; The observation data including at least 2 lidar spectra are calculated, and the observation equation is obtained through geometric relations. The expression is:

[0157]

[0158] Where, C p is the observation data at time p, q() is the time - division multiple access estimation function, E p is the frequency - division multiple access estimated value, F1 is the first time - division address, F2 is the second time - division address, F i-1 is the (i - 1) - th time - division address, F i is the i - th time - division address, F m-1 is the time - division address of the lidar spectrum estimated in the (m - 1) - th current epoch, F m is the time - division address of the lidar spectrum estimated in the m - th current epoch, i is the number, g is the sound propagation speed at temperature T, and m is the number of lidar spectra estimated in the current epoch;

[0159] The model is initialized and the model parameters are dynamically adjusted. ​

[0160] Model initialization, including:

[0161] Combining the observation information with the lidar simple particle filter to calculate the initial plane coordinates of the sensors of the geological disaster monitoring equipment; the coordinate data contained in the observation information gives the area range where the sensors of the geological disaster monitoring equipment are located. A certain number of particles are evenly distributed within this range. Through resampling operations, given the particle attribute [e B n B T , then the particle set is initialized as:

[0162]

[0163] In the formula, e min , e max , n min , n max are the regional boundaries delimited by the initial observation information, N is the number of particles, e l is the l-th particle attribute in the east direction, n l is the l-th particle attribute in the north direction, and L() is the particle attribute function under the constraint of the regional boundaries delimited by the initial observation information;

[0164] The overall lidar spectrum estimation error follows a normal distribution. According to the prediction residual δ, the weight of each particle is:

[0165]

[0166] σ = C0 - C l

[0167] In the formula, w l is the weight of the l-th particle, R l is the radius of the l-particle distribution, exp{} is the residual prediction function, σ l is the prediction residual of the l-th particle, and C0, C l are the initial observation and the predicted observation respectively;

[0168] After normalizing the weights of the particle set and using the random resampling method to obtain a new particle set, an average processing is performed to obtain the initial value estimation:

[0169]

[0170] In the formula, w l is the weight of the l-th particle, R l is the radius of the l-particle distribution, exp{} is the residual prediction function, σ l is the prediction residual of the l-th particle, and C0, C l ​are the initial observation and the predicted observation respectively; where e0 is the original particle attribute in the east direction, n0 is the original particle attribute in the north direction, e l is the l-th particle attribute in the east direction, n l is the l-th particle attribute in the north direction;

[0171] Perform dynamic adjustment of model parameters, including:

[0172] On the premise that the given initial error covariance matrix of the system W0 = I 3×3 make the displacement direction angle converge rapidly, enhance the basic weight of the observation information in the early stage of filtering, and amplify the process noise, and adjust the observation noise corresponding to each lidar observation data. The expression is:

[0173]

[0174] In the formula, is the i-th diagonal element of the observation noise covariance matrix, is the initial diagonal element of the observation noise, is the predicted residual of the (p + 1)-th lidar spectrum, is the predicted residual of the p-th lidar spectrum, and P0 is the variance inflation threshold;

[0175] The geological disaster information connection variability factor of the data-driven shortest distance calculation method model gives timely guidance on the adjustment state and observation weight when detecting the attitude anomaly update caused by lidar mutation.

[0176] As can be seen from the above embodiments, in terms of data-driven, the present invention constructs a neural network of the shortest distance calculation method based on deep learning, trains the observation characteristics of the acceleration monitor and gyroscope, learns the position change speed vector of the geological disaster monitoring equipment, and accurately calculates the displacement trajectory of the geological disaster monitoring equipment. The traditional shortest distance calculation method scheme statistically calculates the shortest distance in units of meters. The data-driven shortest distance calculation method can learn the entire process of movement, output at a 40Hz speed vector, and realize the position update of the high-frequency shortest distance calculation method, giving full play to the advantage of the high data update rate of the geological disaster monitoring equipment sensor. In the model part, through the extended Kalman filter, the high-precision lidar spectrum ranging observation quantity is tightly coupled with the high-frequency displacement rate vector output by the shortest distance calculation method network, and the positioning accuracy of 1 cm is achieved with a position update rate of 40Hz. Compared with the traditional pure model-driven fusion positioning method, the present invention has obvious advantages in terms of positioning update rate, positioning accuracy, and positioning stability.

[0177] The data-driven shortest distance calculation method of the present invention solves the double integration problem, avoids the problem of rapid error accumulation caused by low-cost IMUs, and at the same time releases various constraints during displacement.

[0178] The technical solution of the present invention solves the problem of realizing high-precision and high-update-rate location services in underground spaces, enabling underground positioning to be further realized in practical applications.

[0179] The technical solution of the present invention analyzes the gait characteristics of multi-modal inertial sensing data, constructs a motion prediction network based on hybrid supervised representation learning and an assumed density filtering algorithm based on variational inference, trains a self-supervised high-precision displacement trajectory estimation model with the ability to evaluate and output the variability of geological disaster information connection, fuses the model output with lidar information for heuristic recalibration, effectively suppresses the accumulation and propagation of inertial navigation system errors, and breaks through the accurate displacement trajectory estimation of data-driven geological disaster monitoring equipment in complex environments; a positioning method based on extended Kalman filter fusion is proposed, which organically couples model-driven accurate ranging and data-driven accurate displacement trajectories, and establishes a new paradigm of coupled positioning driven by both data and models.

[0180] Example 3, as Figure 2 shown, the identification system for automatically discovering geological disaster monitoring equipment based on lidar provided by the embodiment of the present invention includes:

[0181] A determination module 1 for the connection variability of position and geological disaster information, which is used to construct and train a data-driven evolution model of the movement trajectory of geological disaster monitoring equipment, and use the trained model and based on the data measured by the sensors of the geological disaster monitoring equipment to determine the relative position and the connection variability of geological disaster information;

[0182] A credibility index prediction module 2, which is used to add error parameters to the data-driven evolution model of the movement trajectory of geological disaster monitoring equipment for training, predict the movement changes and their variances of geological disaster information in adjacent connection areas, calculate the connection variability of geological disaster information in the prediction results, and obtain the predicted credibility index;

[0183] A trajectory correction module 3, which is used to perform lidar-based trajectory correction;

[0184] A geological disaster monitoring information identification module 4, which is used to construct a lidar digital-analog positioning identification model based on Kalman filter, tightly couple the lidar spectrum ranging observations through extended Kalman filter, and based on the high-frequency displacement rate vector output by the data-driven evolution model of the movement trajectory of geological disaster monitoring equipment, obtain the positioning identification result of automatically discovering geological disaster monitoring equipment, and obtain the geological disaster monitoring information at this position according to the positioning identification result.

[0185] In the above embodiments, the descriptions of each embodiment have their own emphases. For parts not detailed or recorded in a certain embodiment, reference may be made to the relevant descriptions of other embodiments.

[0186] For the content such as information interaction and execution process between the above-mentioned device / unit, since it is based on the same concept as the method embodiment of the present invention, for its specific functions and the technical effects brought, reference can be specifically made to the method embodiment part, and details will not be repeated here.

[0187] Those skilled in the art can clearly understand that, for the convenience and conciseness of description, only the above-mentioned division of each functional unit and module is used as an example. In actual applications, the above functions can be allocated to different functional units and modules according to needs, that is, the internal structure of the device is divided into different functional units or modules to complete all or part of the functions described above. Each functional unit and module in the embodiment can be integrated in a processing unit, or each unit can exist physically alone, or two or more units can be integrated in one unit. The above integrated unit can be implemented in the form of hardware or in the form of a software functional unit. In addition, the specific names of each functional unit and module are only for the convenience of mutual distinction and do not limit the protection scope of the present invention. The specific working process of the units and modules in the above system can refer to the corresponding process in the foregoing method embodiment.

[0188] The embodiment of the present invention also provides a computer device, which includes: at least one processor, a memory, and a computer program stored in the memory and executable on the at least one processor. When the processor executes the computer program, the steps in any of the above method embodiments are implemented.

[0189] The embodiment of the present invention also provides a computer-readable storage medium, which stores a computer program. When the computer program is executed by a processor, the steps in each of the above method embodiments can be implemented.

[0190] The embodiment of the present invention also provides an information data processing terminal, which is used to provide a user input interface to implement the steps in each of the above method embodiments when executed on an electronic device. The information data processing terminal is not limited to geological disaster monitoring device sensors, computers, and switches.

[0191] The embodiment of the present invention also provides a server, which is used to provide a user input interface to implement the steps in each of the above method embodiments when executed on an electronic device.

[0192] The embodiment of the present invention provides a computer program product. When the computer program product runs on an electronic device, the electronic device can be made to execute the steps in each of the above method embodiments when executed.

[0193] When the integrated unit is implemented in the form of a software functional unit and sold or used as an independent product, it can be stored in a computer-readable storage medium. Based on this understanding, to implement all or part of the processes in the above-described embodiment methods of this application, a computer program can be used to instruct the relevant hardware to complete. The computer program can be stored in a computer-readable storage medium. When the computer program is executed by a processor, the steps of the above-described various method embodiments can be implemented. Among them, the computer program includes computer program code, and the computer program code can be in the form of source code, object code, executable file, or some intermediate form, etc. The computer-readable medium can at least include: any entity or device that can carry the computer program code to the photographing device / terminal device, recording medium, computer memory, read-only memory (ROM), random access memory (RAM), electrical carrier signal, telecommunication signal, and software distribution medium. For example, a USB flash drive, a mobile hard disk, a magnetic disk, or an optical disc, etc.

[0194] As described above, the above is only a relatively preferred specific implementation manner of the present invention, but the protection scope of the present invention is not limited thereto. Any person skilled in the art within the technical scope disclosed by the present invention, any modifications, equivalent replacements, and improvements made within the spirit and principle of the present invention should all be covered within the protection scope of the present invention.

Claims

1. A laser radar-based automatic identification method for geological disaster monitoring equipment, characterized in that: The method includes: S1, construct and train a data-driven geological hazard monitoring equipment movement trajectory evolution model, and use the trained model and the data measured by the geological hazard monitoring equipment sensors to determine the relative position and geological hazard information connection variability; S2, adding the error parameters to the data-driven geological hazard monitoring equipment movement trajectory evolution model for training, predicting the movement change and variance of geological hazard information in adjacent connection areas, calculating the connection variability of geological hazard information in the prediction results, and obtaining the prediction credibility index; S3, performing trajectory correction based on laser radar; S4, constructing a laser radar digital-analog positioning and identification model based on Kalman filtering, by extending the Kalman filtering to tightly couple the laser radar spectrum ranging observation, and based on the high-frequency displacement rate vector output by the data-driven geological disaster monitoring equipment movement trajectory evolution model, obtaining the automatic geological disaster monitoring equipment positioning and identification results, and obtaining the geological disaster monitoring information at the location according to the positioning and identification results; In step S1, a data-driven geological disaster monitoring equipment movement trajectory evolution model is constructed and trained, including: The location estimation of geological disaster monitoring equipment is achieved by fusing the laser radar spectrum ranging and the geological disaster monitoring equipment sensor data; the data processing architecture for the location estimation of geological disaster monitoring equipment consists of two parts; (1) Construct the data-driven part, and use the data of the built-in acceleration monitor and gyroscope of the geological disaster monitoring equipment to estimate the moving speed v of the geological disaster monitoring equipment. The expression is: v=[v x v y v z ] T In the formula, v x is the lateral velocity, v y is the longitudinal velocity, v z is the speed of change in spatial direction, T is the vector transpose; (2) Construct the model driving part, combine the velocity vector with the ranging results based on the lidar wave, and use the extended Kalman filter to estimate the position, displacement direction and velocity of the geological disaster monitoring equipment; In step S2, the variability of the geological disaster information connection of the prediction result is calculated to obtain the prediction credibility index, including: The total geological hazard information connection variability of network prediction is defined as σ total , the expression is: σ total =var p(y|x) (and) In the formula, var p(y|x) (y) is the predicted value of probability distribution in the connection change of total geological hazard information, and p(y|x) is the conversion function between roll angle y and pitch angle x; According to the source of geological hazard information connection variability, it is divided into data geological hazard information connection variability σ total and the model geological hazard information connection variability σ model , the data geological disaster information connection variability indicates the anti-noise performance of the network for the input data; The variability of geological hazard information connection is based on the Bayesian belief network. With the help of the expected propagation EP framework and the density filtering ADF method, the general network layer of ResNet18 is overwritten by mathematical reasoning. The network layer of geological hazard information connection variability is defined as follows: For any network layer c (i) =h (i) (c (i-1) ,θ (i) ), the geological disaster information connection variable propagation layer is realized based on the following formula: In the formula, g (i) is the i-th propagation layer, c (i-1) is the i-1th network layer, θ (i) is the change value of the i-th network layer, To connect the geological disaster information with the variable transmission layer sampling prediction value, The variance in the sampling prediction of the variable propagation layer for the geological hazard information connection is is the sampling prediction method in the i-1th network layer, is the method of sampling prediction variance in the i-1th network layer; For the variability of the connection of model geological disaster information, the expectation and variance are output through Monte Carlo sampling. The dropout layer after the fixed activation layer ReLU is in training mode, and the parameters will be randomly discarded; the noise variance is initialized and propagated together with the test data to obtain several groups of sampled prediction values ​​b and corresponding variance ζ; the credibility of the network for the output is evaluated to obtain the variability of the connection of model geological disaster information.

2. The method for automatically discovering geological disaster monitoring equipment based on laser radar according to claim 1 is characterized in that: Before building and training the data-driven geological disaster monitoring equipment movement trajectory evolution model, the quaternion-based normalization preprocessing is performed, including: The attitude quaternion is calculated using a gradient descent algorithm, and the acceleration and angular velocity are converted from the sensor coordinate system to the navigation coordinate system using the attitude quaternion. The vector β in the sensor coordinate system IMU The following formula is used to convert β into the navigation coordinate system INS , the expression is: In the formula, α is conjugated, is the complex conjugate of the quaternion; Quaternions are used to rotate vectors and convert them into Euler angles. The order of defining the attitude angles of sensors in geological disaster monitoring equipment is:<z,x,y> , where z is the azimuth angle, x is the pitch angle, and y is the roll angle.

3. The method for automatically discovering geological disaster monitoring equipment based on laser radar according to claim 2 is characterized in that: In step S1, based on the data measured by the sensors of the geological disaster monitoring equipment, the relative position and the variability of the connection of geological disaster information are determined, including: The sliding window method is adopted. Using a sliding window of fixed size, the sensor data of geological disaster monitoring equipment is divided into independent sequences; the window size n of the sequence is 100 frames, and the shortest distance of the sliding window is 5 frames; the velocity vector is predicted by the deep neural network of each sequence, and the position is generated by merging the link module, and the current position is updated to 100 frames; the prediction update is performed using overlapping windows, and the output frequency is increased to 40Hz. The position obtained by processing the velocity vector through a low-pass filter is used to smoothly reconstruct the predicted trajectory; Public standard data sets and self-collected data sets are used as training sets and test sets. During the training phase, the features and labels of each window are randomly rotated to enhance the independence of the displacement direction of the data. The training goal is to minimize the mean square error between the estimated value and the true value provided by the data set. The optimal parameters are obtained through ADAM optimizer training. After training on the GPU, the model is directly applied to the local geological disaster monitoring equipment sensor to infer the displacement trajectory of the geological disaster monitoring equipment.

4. The method for automatically discovering geological disaster monitoring equipment based on laser radar according to claim 1 is characterized in that: Evaluate the credibility of the network output and obtain the variability of the model geological disaster information connection, including: the network undergoes several propagations and Monte Carlo sampling to obtain several groups of prediction values ​​b i and variance ζ i , calculate the model uncertainty γ model , the expression is: y model =var(b i ) In the formula, var() is the variance calculation; After Monte Carlo sampling, the variability of the data geological disaster information connection is expressed as the mean of several sampling prediction values, and then the total uncertainty of the network γ total , the expression is: c total =c data +g model In the formula, γ data is the uncertainty of the current model.

5. The method for automatically discovering geological disaster monitoring equipment based on laser radar according to claim 1 is characterized in that: In step S3, trajectory correction based on laser radar is performed, including: When the LiDAR magnetic field in the environment is stable, there is only one stable LiDAR magnetic field vector, and the three-dimensional magnetic vector of the magnetometer changes with the attitude of the IMU; when the attitude change of the IMU does not match the change of the magnetic vector, the LiDAR magnetic field environment is unstable under the influence of strong magnets; the attitude change of the IMU is calculated using a gyroscope and expressed in the form of a quaternion. i ; Use quaternion to convert magnetic vector m i Convert to Calculate m i+1 and The modulus of the vector difference is taken as the matching degree su; the larger the matching degree, the greater the attitude difference between the IMU and the magnetic vector, and the unstable magnetic field of the lidar; m i+1 is the i+1th magnetic vector, is the i+1th magnetic vector after transformation; The laser radar magnetic field strength reflects the stability of the laser radar environment. The sliding standard deviation std of the laser radar magnetic field strength is used as the condition for determining the stability of the laser radar magnetic field. The size of the sliding window is set to 100 frames, and the stability of the laser radar magnetic field Defined as: In the formula, Seek guidance for Renault, is the stability value of the lidar magnetic field in the window size d, is the sliding average, thrshd1 is the first window of sliding window size d, and thrshd2 is the second window of sliding window size d; When the average sliding standard deviation and the average matching degree are less than the threshold, the lidar magnetic field is completely stable, and the stability of the lidar magnetic field Set to 1, the stability of the lidar magnetic field The range is from 0 to 1, indicating from unstable to stable; Select the stable LiDAR magnetic field environment and low geological disaster information, and connect the correlation quantity on the variable straight sequence as the anchor point g k If the posture of the geological disaster monitoring equipment sensor is not changed, the correlation value of each straight sequence will fluctuate in a small range; when the fluctuation is greater than the threshold, it is determined that the relative horizontal posture relationship between the geological disaster monitoring equipment sensor and the person has changed, and it is necessary to update g k ; Using g k Calculate the correction angle for a straight sequence The expression is: In the formula, ft i is the average stability of the lidar magnetic field, g i is the i-th anchor point; By using the shortest distance calculation method of the neural network, the correction angle of the straight sequence is predicted Realize adaptive correction of trajectory.

6. The method for automatically discovering geological disaster monitoring equipment based on laser radar according to claim 1 is characterized in that: In step S4, a laser radar digital-analog positioning recognition model based on Kalman filtering is constructed, including: S4.1, build the system model and observation model, and use the data-driven shortest distance calculation method to adopt the local coordinate system A system (x A ,y A , α A ), select the Northeast Sky coordinate system as the absolute coordinate system B (e B , n B , α B ), where α is the displacement direction and the system state vector to be estimated is X p Described as: X p =[e p n p i p ] T In the formula, (e p , n p ) are the east and north coordinates of the target in the B system at time P, θ p It is the right-handed displacement angle of system A relative to system B, including the initial deflection angle of the two coordinate systems at the moment of positioning service startup and the accumulated error deflection angle of the sensor. Driven by the data of the shortest distance calculation method, the state transition of the system is as follows: Where h() is the state vector function to be estimated, O p is the initial state value at time p, e p-1 is the east direction coordinate at time p-1, n p-1 is the north direction coordinate at time p-1, θ p-1 is the right-handed displacement angle of system A relative to system B at time p-1, X p-1 is the system state vector to be estimated at time p-1, Δx p , Δy p They are the displacement increments output in the horizontal and vertical directions of the shortest distance calculation method in the local coordinate system from time p-1 to time p; The laser radar spectrum positioning estimation adopts the time division multiple access and frequency division multiple access strategies to form a laser radar spectrum signal network; the observation data including at least two laser radar spectra are solved, and the observation equation is obtained through geometric relations, which is expressed as follows: In the formula, C p is the observed data at time p, q() is the time division multiple access estimation function, F p is the frequency division multiple access estimated value, F1 is the first time division address, F2 is the second time division address, F i-1 is the i-1th time division address, F i is the i-th time division address, F m-1 is the time division address of the lidar spectrum estimated at the m-1th epoch, F m is the time division address of the lidar spectrum estimated in the mth epoch, i is the number, g is the speed of sound propagation at temperature T, and m is the number of lidar spectra estimated in this epoch; The model is initialized and model parameters are dynamically adjusted.

7. The method for automatically discovering geological disaster monitoring equipment based on laser radar according to claim 6 is characterized in that: Model initialization, including: The observation information is combined with the simple particle filter of the laser radar to calculate the initial value of the plane coordinates of the geological disaster monitoring equipment sensor; the coordinate data contained in the observation information gives the regional range of the geological disaster monitoring equipment sensor, and a certain number of particles are evenly distributed within the range. Through the resampling operation, the particle attributes [e B n B ] T , then the particle set is initialized as: In the formula, e min , e max , n min , n max is the region boundary defined by the initial observation information, N is the number of particles, e l is the property of the lth particle pointing eastward, n l is the property of the lth particle in the north, L() is the particle property function of the boundary constraint condition of the region defined by the initial observation information; The overall error of the LiDAR spectrum estimation follows a normal distribution. According to the prediction residual δ, the weight of each particle is: σ=C0-C l In the formula, w l is the weight of the lth particle, R l is the radius of particle distribution, exp{} is the residual prediction function, σ l is the prediction residual of the lth particle, C0, C l are the initial observation and the predicted observation respectively; After normalizing the particle set weights and using the random resampling method to assign them to a new particle set, the initial value estimate is obtained by averaging: Where, e0 is the original particle property of the east-downward direction, n0 is the original particle property of the north-downward direction, and e l is the property of the lth particle pointing eastward, n l It is the property of the lth particle in the north downward direction; Dynamically adjust model parameters, including: Given the initial system error covariance matrix W0 = I 3×3 Under the premise of , the displacement direction angle converges rapidly, the basic weight of the observation information is improved in the early stage of filtering, and the process noise is amplified, and the observation noise corresponding to each lidar observation data is adjusted. The expression is: In the formula, is the i-th diagonal element of the observation noise covariance matrix, is the initial diagonal element of the observation noise, is the prediction residual of the p+1th lidar spectrum, is the prediction residual of the pth lidar spectrum, P0 is the variance expansion threshold; The geological disaster information connection variability factor of the data-driven shortest distance calculation method model can provide timely guidance on adjusting the status and observation weight when abnormal posture updates caused by sudden changes in the lidar are detected.

8. A laser radar-based automatic detection and identification system for geological disaster monitoring equipment, characterized in that: The method for automatically discovering geological disaster monitoring equipment based on laser radar according to any one of claims 1 to 7 is implemented, and the system comprises: A module (1) for determining the variability of the connection between the position and geological hazard information is used to construct and train a data-driven geological hazard monitoring equipment movement trajectory evolution model, and to determine the relative position and the variability of the connection between the geological hazard information based on the trained model and the data measured by the geological hazard monitoring equipment sensor; The credibility index prediction module (2) is used to add the error parameter to the data-driven geological disaster monitoring equipment movement trajectory evolution model for training, predict the movement change of geological disaster information in adjacent connection areas and its variance, calculate the connection variability of the geological disaster information of the prediction result, and obtain the prediction credibility index; A trajectory correction module (3), used for performing trajectory correction based on laser radar; The geological disaster monitoring information identification module (4) is used to construct a laser radar digital-analog positioning and identification model based on Kalman filtering, and obtain the automatic geological disaster monitoring equipment positioning and identification results based on the high-frequency displacement rate vector output by the extended Kalman filtering and the geological disaster monitoring equipment moving trajectory evolution model driven by the data, and obtain the geological disaster monitoring information of the location according to the positioning and identification results.

Images