Waveform-based real-time positioning method for explosion sound signals

Abstract

The present application relates to a kind of waveform-based explosion sound signal real-time positioning method, belong to acoustic signal positioning technical field, solve the technical problem that positioning exists time delay, cannot give the positioning result in real time, it includes according to the sound measuring point in target area and target area range, grid template is divided;Using acoustic wave forward modeling, obtain the theoretical waveform database with each grid point as virtual sound source;Theoretical waveform database is implemented dimension reduction processing, and fast search index library is established;Real-time signal stream matrix is generated, and the steps of defining positioning detection strategy, evaluation.The real-time positioning method is used for the real-time positioning of multiple target explosion sound signals.

Description

Technical Field

[0001] This invention belongs to the field of acoustic signal detection and positioning technology, specifically relating to a real-time positioning method for waveform-based explosion sound signals. Background Technology

[0002] With the development of technology, the demand for real-time acoustic positioning technology in various complex environments and emergencies is increasing. Among these, acoustic signal positioning technology plays a crucial role in many fields. The strong sound waves generated during an explosion can be detected by acoustic instruments, allowing the location of the sound source to be determined. Examples include monitoring illegal car horn use, safety monitoring in chemical plants, earthquake rescue, monitoring illegal mining, and investigation of other violations. However, traditional positioning technologies require manual supervision and suffer from time delays, failing to provide real-time positioning results.

[0003] In their 2013 paper, "Research on the Application of Sound Localization," Zhou Xintong and Li Hongkai presented the principle of explosion sound signal localization using traditional methods. Within the explosion area, the location of the sound source on the ground and the time of the explosion can be obtained by solving a system of nonlinear equations using the spatial relative relationships of three or more pre-deployed sensors and the relative time delay of the recorded explosion sound wave signals. However, this traditional sound signal localization method requires real-time manual monitoring and suffers from time delays.

[0004] Currently, no reliable and stable method has been proposed by researchers for real-time location of explosion sound signals. However, there are still methods and technologies that can be used in the field of real-time earthquake monitoring. Summary of the Invention

[0005] To overcome the shortcomings of existing positioning methods, such as time delay and inability to provide real-time positioning results, this invention proposes a real-time positioning method based on waveform-based explosion sound signals.

[0006] The technical solution adopted by this invention to solve its technical problem is:

[0007] A real-time localization method for explosion sound signals based on waveforms includes the following steps:

[0008] Step 1: Initialization. A grid template needs to be created based on the acoustic measurement points within the study area and the region of interest. Place n sensor receivers (receiving nodes) arbitrarily within the area, recording their positions as set to P1(xx). 1, yy1), P2(xx) 2, yy2), ..., P n (xx n, yy n Meanwhile, the coordinates of two corner points C1(x) of the mesh template are given. 1, y1), C2(x 2,y2), and then divide it into multiple grid points according to the given spacing width.

[0009] Step 2: Obtain a theoretical waveform database with each grid point as a virtual sound source through forward modeling of sound waves.

[0010] Step 3: Establish a fast search index for the theoretical waveform database.

[0011] Step 4: Based on the traditional real-time acoustic signal acquisition system, the real-time signal flow matrix is ​​input into the automatic positioning system for processing. The real-time signal flow matrix undergoes a K-order nearest neighbor algorithm for fast search and evaluation using a custom detection strategy, ultimately providing the real-time positioning result.

[0012] The multiple grid points divided according to the spacing width are uniformly distributed, and their specific coordinates can be represented as (x1, y1), (x1+width, y1), (x1+2*width, y1), ..., (x2, y2).

[0013] The acoustic wave forward modeling simulation, which uses each grid point as a virtual sound source, employs a one-dimensional second-order acoustic wave equation finite difference forward modeling method to generate a theoretical waveform database. The travel time is calculated as t. 1,1 =distance{(x1,y1),(xx1,yy1)} / V s , where t 1,1 Let V represent the travel time between the first grid point and the first receiving node, and let distance{(x1,y1),(xx1,yy1)} represent the distance between the first grid point (x1,y1) and the first receiving node (xx1,yy1). s The speed of sound waves.

[0014] The method described above for building a fast search index library for theoretical waveform databases uses PCA and multiple random K-dimensional tree (MRKD) methods to reduce the dimensionality and generate the dimensionality-reduced fast search index library.

[0015] The custom detection strategy involves evaluating the system's performance when searching real-time data using the K-order nearest neighbor algorithm. The system will then provide the top few closest matching results. These results need to be judged and detected. If an event is identified, its location and occurrence time are automatically provided. Unlike traditional methods, this invention does not use correlation coefficients to determine the likelihood of an event. Instead, it uses the differences between the peak values ​​of the actual data and the search results to assess the discrepancies. This strategy enables automatic identification of multiple overlapping events.

[0016] The beneficial effects of this invention are:

[0017] A waveform-based real-time localization method for explosion sound signals is proposed, which introduces acoustic forward modeling, multiple random K-dimensional tree method, and K-order nearest neighbor algorithm to achieve real-time detection and localization of multi-target explosion sound signals. Simultaneously, a customized detection strategy is used to automatically identify multi-event aliasing, optimizing the entire algorithm. Attached Figure Description

[0018] Figure 1 This is a schematic diagram of a receiving node according to an embodiment of the present invention;

[0019] Figure 2 This is a flowchart of the real-time detection and positioning method according to an embodiment of the present invention;

[0020] Figure 3 This is a schematic diagram of the real-time data stream processing process according to an embodiment of the present invention;

[0021] Figure 4 This is a distribution diagram of the target area and receiving nodes according to an embodiment of the present invention;

[0022] Figure 5 This is a diagram showing the identification result of the first explosion signal in an embodiment of the present invention;

[0023] Figure 6 This is a diagram showing the identification result of the second explosion signal in an embodiment of the present invention. Detailed Implementation

[0024] The present invention will now be described in further detail with reference to the accompanying drawings and specific embodiments.

[0025] Example 1

[0026] A waveform-based real-time explosion sound signal localization method can achieve real-time localization and simultaneously identify and locate multiple aliased explosion sound signals. Invalid interference signals can also be automatically identified and skipped. The method includes the following steps:

[0027] Step 1: Initialization, which requires dividing the grid template based on the acoustic measurement points within the study area and the area of ​​interest. In this embodiment, the target area is as follows: Figure 4 As shown, six sensor receivers, or receiving nodes, will be randomly placed within the area. Figure 4 (As shown by the blue triangle in the middle), the recorded positions are set as LZC1(xx1,yy1), LZC6(xx2,yy2), ..., LSB(xx6,yy6). Simultaneously, the coordinates of two corner points of the mesh template (x1,y1) and (x2,y2) are given, and then the mesh is divided into multiple grid points according to the given spacing width, as shown... Figure 4The black cross indicates that the grid points are evenly distributed according to the spacing width, and their specific coordinates can be represented as (x1, y1), (x1+width, y1), (x1+2*width, y1), ..., (x2, y2).

[0028] Step 2: Obtain a theoretical waveform database with each grid point as a virtual sound source through acoustic wave forward modeling. For acoustic wave forward modeling, the first-order finite difference method is used to perform forward modeling of the acoustic wave equation. The first-order wave equation is expressed as follows:

[0029]

[0030] Where y is a function of space and time, i.e., the wave field function. The space x and t are divided into equally spaced grids x1, x2, ..., xn and t1, t2, ..., tn using the finite difference method, with step sizes Δx and Δt respectively. With the grid, the above second derivative can be expressed in finite difference terms as follows:

[0031]

[0032] Summarized as follows:

[0033] y i,n+1 =2y i,n -y i,n-1 +C 2 (y i-1,n -2y i,n +y i+1,n )

[0034] in It is a dimensionless constant, and Vs is the velocity of sound. Therefore, when t is known... n and t n-1 The sound wave function at time t can be used to obtain t. n+1 The sound wave function at time t.

[0035] The forward modeling method described above is used to simulate the relationship between each grid point and the measurement node. The multiple grid points, divided according to the spacing width, are uniformly distributed, and their coordinates can be represented as (x1, y1), (x1 + width, y1), (x1 + 2 * width, y1), ..., (x2, y2). The travel time is calculated as t. 1,1 =distance{(x1,y1),(xx1,yy1)} / V s , where t 1,1 Let V represent the travel time between the first grid point and the first receiving node, and let distance{(x1,y1),(xx1,yy1)} represent the distance between the first grid point (x1,y1) and the first receiving node (xx1,yy1). sThe velocity of sound waves was used to obtain a theoretical waveform database.

[0036] Step 3: Establish a fast search index for the theoretical waveform database. After obtaining the theoretical waveform database through forward modeling, it needs to be dimensionality reduced. Here, PCA is used to reduce the dimensionality of the original theoretical waveform database, followed by a fast search using the Multiple Random K-tree (MRKD) method. Assume the database to be dimensionality reduced is X. n×m Where n is the number of grid points and m is the number of sampling points. First, the covariance matrix convar(X) of the data matrix needs to be calculated. m×m :

[0037]

[0038] covar(X)P=PA,

[0039] A=diag{λ1,λ2,…,λ m},

[0040] p = {p1, p2, ..., p} m},

[0041] Where p is the covariance matrix containing m eigenvalue vectors, and A is the diagonal eigenvalue matrix. This simplifies the characteristics of the original data, allowing the principal eigenvectors to replace the original data, thus achieving dimensionality reduction.

[0042] A K-dimensional tree is a high-dimensional index tree data structure that stores K-dimensional data. Constructing a K-dimensional tree on a K-dimensional dataset represents a partition of the K-dimensional space formed by that dataset, primarily used for processing and searching data points in multidimensional space. This data structure is a binary tree where each node is a K-dimensional numerical point. Each point represents a hyperplane perpendicular to the coordinate axis of the current partition dimension, dividing the space into two parts: one in its left subtree and the other in its right subtree. It has the form of a binary search tree, where each node corresponds to a point in the dimensional space. The fast index library obtained in step 2 is processed using MRKD to obtain a dimensionality-reduced fast search index library. The K-order nearest neighbor algorithm can then be used to quickly search and judge real-time data streams. The K-order nearest neighbor algorithm, also known as KNN, is a basic classification and regression method, its core function being to solve supervised classification problems. The algorithm works on the premise that there exists a sample dataset, and each sample is manually labeled. When new, unlabeled data is input, the algorithm compares the features corresponding to the new data and then extracts the classification label from the data whose features are most similar to the input data. Here, K refers to the number of samples; that is, in real-time detection, the algorithm selects the K most similar data points from the sample set.

[0043] Step 4: Based on the traditional real-time acoustic signal acquisition system, the real-time signal flow matrix is ​​input into the automatic positioning system for processing. The real-time signal flow matrix undergoes a K-order nearest neighbor algorithm for fast search and evaluation using a custom detection strategy, ultimately providing the real-time positioning result.

[0044] The specific operations for real-time data streams are as follows:

[0045] First, the real-time data stream is transformed into feature data based on the eigenvectors of PCA. Then, the distance between the points in the fast search index and the real-time feature data points is calculated. Next, the points are sorted in ascending order of distance. Then, the K points with the smallest distance to the current point are selected. Finally, the original template data corresponding to these K points is determined.

[0046] After the system provides the K most similar search results, it needs to evaluate them. If the given recognition criteria are met, it is determined to be an explosion event; otherwise, the system will automatically classify it as background noise or other signals. Of course, the given recognition criteria can adapt to the following situations:

[0047] Scenario 1: Identification of a single explosion signal. For a single signal, there is a corresponding explosion peak signal on each sensor signal, with corresponding times T1, T2, ..., Tn. The system automatically calculates the matching degree between the true peak signal Tr1, Tr2, ..., Trn of each receiving node and its corresponding K most similar results, and integrates the accumulated error. If the value is below a certain threshold, it is judged as a suspected event, and the accurate explosion time and corresponding explosion location are given based on calculations.

[0048] Scenario 2: Identification of two superimposed explosion signals, such as Figure 5 , Figure 6 As shown. When two explosion sound signals exist within the same window, there are two corresponding explosion peak signals on each sensor signal, corresponding to times T11, T12, T21, T22, ..., Tn1, Tn2. The system automatically calculates the matching degree between the true peak signals Tr11, Tr12, Tr21, Tr22, ..., Trn1, Trn2 of each receiving node and the corresponding K most similar results. For each result in K, the minimum error between it and the true explosion signal is calculated and accumulated as the final judgment criterion, which can be specifically expressed as:

[0049]

[0050]

[0051] If the cumulative error is below a given threshold, the event is classified as a suspected incident, and the system calculates the exact explosion time and location. Otherwise, the system automatically skips the determination and proceeds to identify the next real-time data stream.

[0052] Scenario 3: Identification of Multiple Superimposed Explosion Signals. When multiple (m) explosion sound signals exist within the same window, each sensor signal corresponds to m explosion peak signals, with corresponding times T11, T12, ..., T1m, ..., Tn1, Tn2, ..., Tnm. The system automatically calculates the matching degree between the actual peak signals Tr11, Tr12, ..., Tr1m, ..., Trn1, Trn2, ..., Trnm of each receiving node and its corresponding K most similar results. For each result in K, the minimum error between it and the actual explosion signal is calculated and accumulated as the final judgment criterion, which can be expressed as follows:

[0053]

[0054]

[0055]

[0056] If the cumulative error is below a given threshold, the event is classified as a suspected incident, and the system calculates the exact explosion time and location. Otherwise, the system automatically skips the determination and proceeds to identify the next real-time data stream.

[0057] Furthermore, embodiments of the present invention can also be extended to subsonic and supersonic flight tests. This invention can automatically identify the shock wave signals of supersonic flying objects using the aforementioned customized judgment strategy, identifying, judging, and locating only events closest to the template. Embodiments of the present invention can also be extended to traffic systems for real-time assessment and recording of illegal horn use on fixed road surfaces, and to specific practical applications such as real-time monitoring of illegal mining in fixed mountainous areas.

Claims

1. A real-time localization method for explosion sound signals based on waveforms, characterized in that, Includes the following steps: Step 1: Divide the area into grid templates based on the acoustic measurement points and the target area's extent. Within the target area, n sensors, i.e., n receiving nodes, are placed; the positions of the n sensors are set as P1(x1, y1), P2(x2, y2), ..., P... n (XX n yy n ); Given the coordinates of two corner points of the grid template, C1(x1, y1) and C2(x2, y2), the grid is divided into several grid points according to the given spacing width. The grid points are evenly distributed and their specific coordinates are (x1, y1), (x1+width, y1), (x1+2*width, y1), ..., (x2, y2). Step 2: Using forward modeling of acoustic waves, a theoretical waveform database is obtained with each grid point as a virtual sound source. Acoustic forward modeling is used to simulate the relationship between each grid point and the receiving node, based on the grid point coordinates. The running time is calculated using the following formula: t 1，1 =distance{(x1,y1),(xx1,yy1)} / V s In the above formula, t 1，1 Let V be the travel time between the first grid point and the first receiving node, and let distance{(x1, y1), (xx1, yy1)} be the distance between the first grid point (x1, y1) and the first receiving node (xx1, yy1). s The speed of sound waves; Thus, a theoretical waveform database with each grid point as a virtual sound source is obtained; Step 3: Perform dimensionality reduction processing on the theoretical waveform database and establish a fast search index library: PCA is used to reduce the dimensionality of the theoretical waveform database obtained from forward modeling of acoustic waves, resulting in a fast index library. A multi-random K-dimensional tree method is used to perform a fast search on the fast index library obtained by dimensionality reduction, resulting in a fast search index library. Step 4: Generate a real-time signal flow matrix, define the positioning detection strategy, and make evaluations. Based on the real-time signals acquired by the acoustic signal real-time acquisition system, a real-time signal flow matrix is ​​generated; a positioning detection strategy is defined, and based on a fast search index library, an automatic positioning system is used to evaluate the real-time signal flow matrix to obtain the real-time positioning result.

2. The real-time location method for explosion sound signals according to claim 1, characterized in that, Step 2 further includes: The first-order finite difference method is used to perform forward modeling of the sound wave equation, which is as follows: In the above formula, y(x, t) is a function of coordinates and time, i.e., the wave field function; Using the finite difference method, space x and time t are divided into equally spaced grids x1, x2, ..., xt. N and t1, t2, ..., t N The step sizes are Δx and Δt, respectively; the mesh is obtained; Using finite difference to express the first-order wave equation of sound, we obtain: y i，n+1 =2y i，n -y i，n-1 +C 2 (y i-1，n -2y i，n +y i+1，n ) In the above formula, It is a dimensionless constant, where c is the velocity of sound; y i.，n+1 Let i be the wave field of the i-th grid at time n+1, where i is the grid number; Based on t n and t n-1 The sound wave function at time t can be used to calculate t. n+1 The sound wave function at time t.

3. The real-time location method for explosion sound signals according to claim 1, characterized in that, Step 3 further includes: The theoretical waveform database is X n×m Where m is the number of sampling points; Calculate the covariance matrix covar(X) of the data matrix. m×m The calculation formula is as follows: covar(X)P=PA, A=diag{λ1,λ2,...,λ m }, p={p1,p2,...,p m }, Where p is the covariance matrix containing m eigenvalue vectors, A is the diagonal eigenvalue matrix, and λ i Let be the i-th eigenvalue.

4. The real-time location method for explosion sound signals according to claim 1, characterized in that, Step 4 further includes: Real-time signal streams are also known as real-time data streams. The processing procedure for real-time data streams is as follows: First, based on the eigenvectors of PCA, the real-time data stream is transformed into feature data, and the distance between the points in the fast search index and the real-time feature data points is calculated. Secondly, sort them in ascending order of distance; Next, select the K points with the smallest distance from the current point; Finally, the original template data corresponding to the K points is determined.

5. The real-time location method for explosion sound signals according to claim 1, characterized in that, In step 4, when the real-time signal is a single explosion signal, the identification process is as follows: For a single explosion signal, one sensor signal corresponds to one explosion peak signal, and its corresponding time is T1. T2, ..., Tn; The automatic positioning system calculates the degree of matching between the true peak signals Tr1, Tr2, ..., Trn of each station and their corresponding K most similar results, and calculates the comprehensive cumulative error. The calculation formula is as follows: In the above formula, n is the total number of receiving nodes, and i is the ordinal number of the receiving node; If the cumulative error is lower than a given threshold, it is considered a suspected event. The assessment provides the accurate time and location of the explosion.

6. The real-time location method for explosion sound signals according to claim 1, characterized in that, In step 4, when the real-time signal is two superimposed explosion signals, the identification process is as follows: For two superimposed explosion signals, i.e., when two explosion sound signals exist within the same window, each explosion sound signal corresponds to two explosion peak signals on each sensor signal, corresponding to times T11, T12, T21, T22, ..., Tn1, Tn2; the automatic positioning system calculates the actual peak signals Tr11, Tr12, Tr21, Tr22, ..., Trn1 for each sensor. The degree of matching between Trn2 and its corresponding K most similar results is calculated; for each result in K, the minimum error between it and the actual explosion signal is calculated and accumulated, and the accumulated minimum error is used as the criterion. The accumulated minimum error is as follows: If the cumulative minimum error is lower than the given threshold, it is judged as a suspected event; the accurate explosion time and corresponding explosion location are given based on the calculation. If the cumulative minimum error reaches or exceeds the given threshold, the automatic positioning system skips the judgment and proceeds to the identification of the next real-time data stream.

7. The real-time location method for explosion sound signals according to claim 1, characterized in that, In step 4, when the real-time signal is a combination of multiple superimposed explosion signals, the identification process is as follows: For multiple superimposed explosion signals, i.e., when there are m explosion sound signals within the same window (m is an integer greater than 2), each sensor signal corresponds to m explosion peak signals, with corresponding times T11, T12, ..., T1m, ..., Tn1, Tn2, ..., Tnm. The automatic positioning system calculates the matching degree between the actual peak signals Tr11, Tr12, ..., Tr1m, ..., Trn1, Trn2, ..., Trnm of each sensor and the corresponding K most similar results. For each result in K, the minimum error between it and the actual explosion signal is calculated and accumulated. The accumulated minimum error is used as the judgment criterion. The accumulated minimum error is as follows: ...... If the cumulative minimum error is lower than the given threshold, it is judged as a suspected event, and the accurate explosion time and corresponding explosion location are given based on the calculation. If the cumulative minimum error reaches or exceeds the given threshold, the automatic positioning system skips the judgment and continues to identify the next real-time data stream.

Images