Low-power-consumption sound source localization method based on two-stage data fusion

Abstract

The invention discloses a low-power-consumption sound source localization method based on two-stage data fusion, and relates to the technical field of low-power-consumption sound source localization methods based on two-stage data fusion, and the method comprises the following steps: laying sensor nodes, and constructing a wireless sensor network structure of a cluster head node selection method based on an energy residual value; constructing a sound source signal model based on RSS; constructing a data structure model for collecting physical quantities, and constructing a similar function model for first-stage data fusion; carrying out first-level data fusion, and removing redundancy of data collected by the sensor nodes; constructing a similar function model based on Euclidean distance; second-level data fusion is carried out, and data redundancy among sensor nodes is removed; and constructing a target function, and completing positioning of a sound source position by adopting an EM algorithm. According to the low-power-consumption sound source localization method based on two-stage data fusion, the problems that in a traditional sound source localization method, energy consumption is high, and the service life of the whole wireless sensor network is short are solved.

Description

Technical Field

[0001] The present invention relates to the technical field of low-power sound source localization, and more specifically, to the technical field of a low-power sound source localization method based on two-level data fusion. Background Art

[0002] Sound source localization technology is a technology that uses acoustic electronic devices to receive sound wave signals emitted by sound sources, and calculates and analyzes the detected sound signals to find the location of the sound source. Sound source localization in wireless sensor networks is to randomly spread sensor nodes in the monitoring area to receive the sound field information of the target sound source, and determine the location of the target based on the detected information.

[0003] At present, according to the physical quantities collected by sensor nodes, the sound source localization methods in wireless sensor networks are mainly divided into two categories: sound source localization methods based on TDOA (time difference of arrival) and RSS (received signal energy strength). The TDOA-based method requires precise time synchronization, which places stringent requirements on the software and hardware environment of the wireless sensor network and related protocols. The RSS-based method is simple to implement and directly detects the existence of the sound source through the signal energy strength detected by the sensor node, without requiring additional energy consumption. Therefore, compared with the TDOA-based method, the RSS-based sound source localization method is more suitable.

[0004] The characteristics of sound source localization in wireless sensor networks are low energy consumption, high positioning accuracy, and long network life cycle. Since the environmental conditions of wireless sensor network deployment may interfere with sensor readings and even damage sensor nodes, a considerable number of redundant nodes must be configured in the application of wireless sensor networks to overcome node failure and measurement inaccuracy. In wireless sensor networks, a large number of sensor nodes continuously collect field data according to the set cycle. These data have the following characteristics: the data collected by a single sensor node multiple times in a short period of time have a high similarity, and the data collected by adjacent sensor nodes at similar times have a high similarity. Therefore, the original data collected by the sensor nodes have a high similarity. Transmitting all the data collected by all sensor nodes to the aggregation node is not only meaningless, but will consume too much energy of the sensor nodes and affect the network life.

[0005] Data fusion is a very effective technical means to reduce data communication between sensor nodes, reduce energy consumption, and extend the life of the network in wireless sensor networks. The more data source nodes there are in a wireless sensor network, the more energy the data fusion technology saves. At the same time, the larger the network scale, the more energy the data fusion technology saves. Based on this, the present invention proposes a two-level data fusion method to reduce energy consumption and extend the service life of the entire wireless sensor network. In view of the above problems, the present invention proposes a solution. Summary of the invention

[0006] The purpose of the present invention is to address the deficiencies in the above-mentioned technology. The present invention relates to a low-power sound source localization method based on two-level data fusion in a wireless sensor network. The method overcomes the problems of high energy consumption and short service life of the entire wireless sensor network in traditional sound source localization methods, thereby achieving the purpose of reducing energy consumption, extending network service life, and improving sound source localization accuracy.

[0007] In order to achieve the above-mentioned purpose, the concept of the present invention is as follows: firstly, a wireless sensor network structure of a cluster head node selection method based on energy residual value is constructed; a sound source signal model based on RSS is constructed; then a data structure model of physical quantity collected by sensor nodes is constructed, and a similarity function model of first-level data fusion is further constructed; on this basis, each sensor node performs first-level data fusion to remove the redundancy of the data collected by the sensor node itself; then a distance function model based on Euclidean distance is constructed, and the cluster head node performs second-level data fusion to remove the data redundancy between sensor nodes; finally, the signal model is remodeled according to the selected data, an objective function is constructed, and the EM algorithm is used to complete the positioning of the sound source position.

[0008] To achieve the above object, the present invention provides the following technical solutions: A low-power sound source localization method based on two-level data fusion comprises the following steps: arranging sensor nodes, acquiring the energy consumed by data transmission between sensor nodes in real time, selecting cluster head nodes based on the energy residual value method, and constructing a wireless sensor network structure according to the cluster head nodes; constructing a sound source signal model based on RSS by detecting the energy value of the signal of the sensor node at time k in the wireless sensor network; acquiring the amount of data collected by the sensor in a period of time based on the sound source signal model to construct a data structure model of collected physical quantities, and constructing a similarity function model of first-level data fusion according to the data structure model; performing first-level data fusion according to the similarity function model of the first-level data fusion, removing the redundancy of the data collected by the sensor node itself, and obtaining measurement data; collecting data transmitted to two different sensor nodes in the same cluster, constructing a similarity function model based on the Euclidean distance according to the data of the two different sensor nodes, removing the redundancy of the received data of different sensor nodes, and performing second-level data fusion on the data; after the second-level data fusion, acquiring all the data received by the sink node, constructing an objective function based on all the data received by the sink node, and completing the location of the sound source by using the EM algorithm.

[0009] In a preferred embodiment, the construction of the RSS-based sound source signal model comprises the following specific steps: randomly deploying A rectangular coordinate system is established with a sensor node and a sound source; the unknown quantity of the nth sensor node position and the sound source at time K is obtained based on the rectangular coordinate system; and a sound source signal model based on RSS is constructed according to the unknown quantity of the nth sensor node position and the sound source at time K.

[0010] In a preferred embodiment, the data structure model of the collected physical quantity is constructed, and the similarity function model of the first-level data fusion is constructed according to the data structure model. The specific steps are as follows: K time periods are evenly divided into a number of equal time periods, a total of m data of all time periods are collected through the nth sensor, the energy value collected is obtained by using the sound source signal model and the data structure model is constructed, and the output vector ; If the sound source does not change or changes very slowly in a short period of time, the data collected by the sensor node There will be a higher redundancy; through the sensor nodes need to search for vector The similarity function of the first-level data fusion is constructed by using similar data in the dataset.

[0011] In a preferred embodiment, during the data collection phase, for the sensor nodes, obtain two new measured data and , use the weight value of the measurement information to represent similar measurement values ​​in the measurement data vector; The similarity of the two newly measured data is determined. If the two newly measured data are similar, they will be determined as similar values. The weight value will be increased by 1, and the newly measured data Will be discarded by the sensor node.

[0012] In a preferred embodiment, the similarity function model based on Euclidean distance is constructed by the following specific steps: obtaining two different sensor nodes in the same cluster and , respectively obtain the data vectors passed to the cluster head node after the first level of data fusion ; Construct a similarity function model based on Euclidean distance based on the data after the first-level data fusion.

[0013] In a preferred embodiment, the second level data fusion is performed on the data, and the specific steps are as follows: the data vector of the nth sensor node after the first level data fusion and the data vector of the nth sensor node after the first level data fusion are combined. The data vector of the sensor node after the first level data fusion is split into two parts, one part is the intersection part contained in both data vectors, and the other part is the different part contained in each data vector; the intersection part is defined to obtain the intersection definition vector, and the different part of the data after the first level data fusion of the nth sensor node is defined to obtain and The different parts of the data after the first level of data fusion of the sensor nodes are defined ; The data vector and data vector Converted into a data set, according to the data set and Get the collection amount If the Euclidean distance is less than the set distance threshold, the amount of collection are similar, there is data redundancy, and the cluster head node selects one of them for data transmission according to the needs and abandons the other one.

[0014] In a preferred embodiment, after the second-level data fusion, all the data received by the sink node are obtained, an objective function is constructed based on all the data received by the sink node, and the EM algorithm is used to locate the sound source position, as follows: all the data received by the sink node are obtained, and the target function is constructed based on all the data received by the sink node. , , Three matrices; according to , , The three matrices are used to obtain matrix equations based on the sound source signal model; the Gaussian probability density characteristics and properties of the ambient noise are obtained, and the joint probability density function of the matrix equation is obtained according to the Gaussian probability density characteristics and properties of the ambient noise; the negative log-likelihood function is obtained based on the joint probability density function of the matrix equation as an expression equivalent to a quadratic form; the estimated value of the location information of the sound source is converted into the maximum value of the maximum likelihood function that solves the negative log-likelihood function as an expression equivalent to a quadratic form, and the solution of the maximum likelihood function estimate is further converted into minimizing the following formula to obtain ; Based on the EM algorithm, the energy intensity per unit distance is As a hidden variable, let , get the hidden variable; bring the hidden variable into , obtain the corrected log-likelihood function; minimize the corrected log-likelihood function to be equivalent to the maximum value, and use The variable search algorithm is optimized to obtain the estimated information of the sound source location.

[0015] One or more technical solutions provided in the embodiments of the present application have at least the following technical effects or advantages: Compared with the existing sound source localization method, the low-power sound source localization method based on two-level data fusion in the wireless sensor network of the present invention has obvious advantages: the method adopts the cluster head node selection method based on the energy residual value and the two-level data fusion technology at different levels, reduces the data communication between sensor nodes, avoids the long-distance wireless transmission of a large amount of data, saves energy consumption, and extends the service life of the entire wireless sensor network. On this basis, the signal model is remodeled to construct the objective function, and the EM algorithm is used to complete the positioning of the sound source position. BRIEF DESCRIPTION OF THE DRAWINGS

[0016] Figure 1 A schematic structural diagram of a low-power sound source localization method based on two-level data fusion provided in an embodiment of the present application.

[0017] Figure 2 This is a structural diagram of the wireless sensor network of the present invention.

[0018] Figure 3 The data structure model of the physical quantity collected by the sensor node of the present invention. DETAILED DESCRIPTION

[0019] The following will be combined with the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0020] In order to better understand the technical solution of the present invention, the following is further described in detail: The present invention provides a low-power sound source localization method based on two-level data fusion in a wireless sensor network. Two-level data fusion technology of different levels is performed on cluster member sensor nodes and cluster head sensor nodes respectively, the signal model is remodeled to construct the objective function, and the EM algorithm is used to complete the location of the sound source. The specific implementation steps are as follows: S1, deploy sensor nodes, obtain the energy consumed by data transmission between sensor nodes in real time, select cluster head nodes based on the energy residual value method, and build a wireless sensor network structure based on the cluster head nodes, as follows: Wireless sensor networks are usually composed of sensor nodes, aggregation nodes, Internet, management nodes, etc. A large number of micro sensor nodes are usually scattered in the monitoring area. The nodes form a network through self-organization and transmit the detected data in a multi-hop routing manner. Wireless sensor networks are based on cluster routing protocols and divide the network into several clusters. Each cluster consists of a cluster head and multiple cluster members. The cluster head node is responsible for forwarding data between clusters, while the cluster member nodes are responsible for data collection. Since the cluster head node is responsible for data forwarding, that is, sending a large amount of detection data received in the cluster to the aggregation node or other cluster head nodes, it will consume more energy.

[0021] The energy consumed in transmitting data between the sensor nodes is specifically expressed as follows:

[0022] In the formula, Indicates the energy consumed by a radio device to start the transmission circuit. Indicates the energy consumed by the transmission amplifier startup circuit. is the transmission distance, is the number of bits of transmitted data, is the threshold value, Indicates the distance Transmission between two sensor nodes The energy consumed by a bit of data.

[0023] It should be noted that, since cluster head nodes forward a lot of data and consume energy quickly, the present invention adopts a method based on the energy residual value to select cluster head nodes, and updates the cluster head nodes at regular intervals. The sensor node with the largest energy residual value in each cluster becomes the new cluster head, extending the life cycle of the cluster head node and the service life of the entire wireless sensor network. Since the cluster head nodes are selected and continuously updated based on the cluster routing protocol and the energy residual value of the sensor nodes, this network structure has a longer life cycle. The schematic diagram of the wireless sensor network structure of the cluster head node selection method based on the energy residual value is shown in FIG. Figure 2 shown.

[0024] S2, constructing a sound source signal model based on RSS by detecting the energy value of the signal of the sensor node in the wireless sensor network at time k. The specific steps are as follows; Randomly deploy in a monitoring area of ​​a wireless sensor network sensor nodes and 1 sound source to establish a rectangular coordinate system; Obtain the unknown quantities of the nth sensor node position and the sound source at time K based on the rectangular coordinate system; A sound source signal model based on RSS is constructed according to the position of the nth sensor node and the unknown quantity of the sound source at time K.

[0025] The specific calculation formula of the sound source signal model is as follows:

[0026]

[0027] In the formula, For wireless sensor networks The sensor nodes at time The energy value of the detected signal, For the The known quantities of nodes, represents the gain of the sensor, is the position of the nth sensor node, For the sound source The unknown quantity of time, Represents the energy intensity of the sound source per unit distance, is Gaussian noise.

[0028] It should be noted that the task of sound source localization is to estimate these unknown quantities. For the convenience of the following expression, the time variables are omitted in the subsequent formulas. .

[0029] S3, based on the sound source signal model, the amount of data collected by the sensor within a period of time is acquired to build a data structure model of the collected physical quantity, and a similarity function model of the first level data fusion is constructed according to the data structure model.

[0030] In wireless sensor networks, a large number of sensor nodes continuously collect detection data according to the set cycle. If the sound source does not change or changes very slowly in a short period of time, the collected detection data has the following characteristics: the data collected by a single sensor node multiple times in a short period of time has a high degree of similarity, and the data collected by adjacent sensor nodes at similar times also has a high degree of similarity. Therefore, the data collected by the sensor nodes has a high degree of redundancy. In order to reduce energy consumption and extend the service life of the wireless sensor network, it is necessary to perform data fusion at different levels to reduce the amount of data transmission.

[0031] If for a period of time Divide the time into several equal time periods , , , each time period is divided into At different moments, each sensor node collects data in an evenly divided time period. times, such as When the value is 6, , , Expressed as , , ,like Figure 2 As shown, if the time period is divided , , If it is shorter, the data collected at the 6 times in each time period have a high degree of similarity.

[0032] The data structure model of the collected physical quantities is constructed, and the similarity function model of the first-level data fusion is constructed according to the data structure model. The specific steps are as follows: Divide the K time period into several equal time periods, collect m data in all time periods through the nth sensor, use the sound source signal model to obtain the collected energy value and build a data structure model, and output the vector , which can be expressed as , where the definition .

[0033] If the sound source does not change or changes very slowly in a short period of time, the data collected by the sensor node There will be a higher degree of redundancy; Through the sensor nodes need to search for vector The similarity function for the first level data fusion is defined with similar data in .

[0034] The specific formula of the similarity function of the first-level data fusion is as follows:

[0035] In the formula, For the Sensor nodes collect data The similarity function of , is the threshold value.

[0036] It should be noted that the threshold value can be determined according to the actual environmental conditions. If the value is 1, the amount of collection In order to maintain the integrity of the measurement information, the weight value of the measurement information is used to represent the similar measurement values ​​in the measurement data vector, which is expressed as .

[0037] S4, performing first-level data fusion according to the similarity function model of the first-level data fusion, removing the redundancy of the data collected by the sensor node itself, and obtaining measurement data, the measurement data is ; In wireless sensor networks, a large number of sensor nodes continuously collect detection data according to a set period. The data collected by a single sensor node multiple times in a short period of time has a high degree of similarity. In order to reduce the amount of data transmission, the first level of data fusion is required. The specific steps are as follows: During the data collection phase, sensor nodes, make similarity judgment on the newly measured data; If the newly measured data is judged to be similar values, that is, and Similar, then The weight value of will be increased by 1, that is +1, new measured data Will be discarded by the sensor node.

[0038] After the first level of data fusion, the measurement data vector Will be converted into a non-redundant data vector with weights ,in , For example, for sensor nodes, the collected data vector is , if the threshold Taking 0.1, the measured data vector after the first level of data fusion is ,in .

[0039] For sensor nodes, after the first level of data fusion, the measurement data vector is becomes , data redundancy is removed, the amount of data transmitted will be greatly reduced, the amount of data transmission is reduced, energy consumption is saved, and the service life of the wireless sensor network is extended.

[0040] S5, collects data transmitted to two different sensor nodes in the same cluster, constructs a similarity function model based on Euclidean distance according to the data of the two different sensor nodes, and performs second-level data fusion on the data.

[0041] For a single sensor node, after the first level of data fusion to remove data redundancy, the measured data is sent to the cluster head node. However, for the cluster head node, the cluster head will also receive data sent from other sensor nodes in the cluster. There is still data redundancy between different sensor nodes. In order to further save the energy consumption of the cluster head node and extend the life cycle of the cluster head node, the cluster head node still needs to remove the redundancy of the data received from different sensor nodes.

[0042] The similarity function model based on Euclidean distance is constructed, and the specific steps are as follows: Get two different sensor nodes in a cluster and , respectively obtain and pass to the cluster head node and The data vector after the first data fusion ; A similarity function model based on Euclidean distance is constructed according to the data after the first data fusion, wherein: , ,in , .

[0043] The specific formula of the similarity function model based on Euclidean distance is as follows:

[0044] In the formula, For two sensor nodes belonging to the same cluster, the amount of data collected The similarity function of is the threshold value, is the data vector of the nth sensor node after the first level of data fusion, For the The data vector of the sensor nodes after the first level of data fusion.

[0045] It should be noted that the threshold value can be determined according to the actual environmental conditions. If the value of the function similarity function is 1, the collection amounts of the two sensors are similar.

[0046] The specific steps of performing the second-level data fusion on the data are as follows: The data vector of the nth sensor node after the first level of data fusion and the The data vectors of the sensor nodes after the first level of data fusion are split into two parts, one part is the intersection part contained in both data vectors, and the other part is the different parts contained in each data vector; The intersection part is defined to obtain the intersection definition vector, and the different part of the data after the first level data fusion of the nth sensor node is defined to obtain and The different parts of the data after the first level of data fusion of the sensor nodes are defined ; The data vector and data vector Converted into a data set, according to the data set and Get the collection amount The Euclidean distance of If the Euclidean distance is less than the set distance threshold, the amount of collection are similar, there is data redundancy, and the cluster head node selects one of them for data transmission according to the needs and abandons the other one.

[0047] The intersection part is defined to obtain the intersection definition vector, and the specific calculation formula is as follows:

[0048] The data The different parts are defined to obtain the data vector , the specific calculation formula is as follows:

[0049] The data The different parts are defined to obtain the data vector , the specific calculation formula is as follows:

[0050] The collection amount The specific calculation formula of the Euclidean distance is as follows:

[0051] In the formula, For two sensor nodes belonging to the same cluster, the amount of data collected The Euclidean distance of The data vector of the nth sensor node after the first level of data fusion, No. The data vector of the sensor nodes after the first level of data fusion, For data The data vector obtained by defining the different parts, For data The different parts are defined to obtain the data vector, Define the intersection part to get the intersection definition vector, The amount of data collected m times for the nth sensor node, The amount of data collected m times for the nth sensor node.

[0052] It should be noted that Represents the operation of finding the difference between two data sets, and obtains any two data measurement values ​​through the sound source signal model And the similarity function of the first-level data fusion, if , , and the similarity function of the first-level data fusion ,but ; S6, after the second-level data fusion, all the data received by the sink node are obtained, an objective function is constructed based on all the data received by the sink node, and the EM algorithm is used to complete the location of the sound source, as follows: Get all the data received by the sink node and define , , Three matrices; according to , , The three matrices obtain matrix equations based on the sound source signal model; Obtaining Gaussian probability density characteristics and properties of environmental noise, and obtaining a joint probability density function of a matrix equation according to the Gaussian probability density characteristics and properties of the environmental noise; The joint probability density function based on the matrix equation is equivalent to a quadratic expression of the negative log-likelihood function; The estimated value of the location information of the sound source is converted into the maximum value of the maximum likelihood function of the expression equivalent to a quadratic form of the negative logarithmic likelihood function, and the solution of the maximum likelihood function estimate is further converted into minimizing the following formula to obtain ; Based on the EM algorithm, the energy intensity per unit distance is As a hidden variable, let , get the hidden variables; Substitute the hidden variables into In the equation, we get the corrected log-likelihood function; The corrected log-likelihood function is minimized to a maximum value, and the The variable search algorithm is optimized to obtain the estimated information of the sound source location.

[0053] Said , , The specific expressions of the three matrices are as follows:

[0054]

[0055]

[0056] The matrix equation, the specific calculation formula is as follows:

[0057] The specific calculation formula of the joint probability density function of the matrix equation is as follows:

[0058] The negative log-likelihood function is equivalent to an expression in quadratic form, and the specific formula is as follows:

[0059] The minimization of the following equation yields , the specific formula is as follows:

[0060] The specific calculation formula of the hidden variable is as follows:

[0061] The specific calculation formula of the corrected log-likelihood function is as follows:

[0062] The corrected log-likelihood function is minimized to obtain a maximum value, and the specific calculation formula is as follows:

[0063] In the formula, , , Three matrices, are the Gaussian noise mean and variance, is the energy intensity per unit distance, represents the transpose of a matrix or vector, N is the number of sensors, is the energy collected by the Nth sensor, where =1,2,3,...,R, R is an integer, The maximum likelihood function estimate is transformed into minimization, is the corrected log-likelihood function, is a single matrix, is a projection matrix, is the gain of the Nth sensor, is the transmission distance of the Nth sensor.

[0064] It should be noted that

[0065] The above formulas are all dimensionless and numerical calculations. The formula is a formula for the most recent real situation obtained by collecting a large amount of data and performing software simulation. The preset parameters in the formula are set by technicians in this field according to actual conditions.

[0066] The above embodiments may be implemented in whole or in part by software, hardware, firmware or any other combination. When implemented by software, the above embodiments may be implemented in whole or in part in the form of a computer program product.

[0067] Those of ordinary skill in the art will appreciate that the modules and algorithm steps of each example described in conjunction with the embodiments disclosed herein can be implemented in electronic hardware, or a combination of computer software and electronic hardware. Whether these functions are performed in hardware or software depends on the specific application and design constraints of the technical solution. Professional and technical personnel can use different methods to implement the described functions for each specific application, but such implementation should not be considered to be beyond the scope of this application.

[0068] In addition, each functional module in each embodiment of the present application may be integrated into one processing module, or each module may exist physically separately, or two or more modules may be integrated into one module.

[0069] The above is only a specific implementation of the present application, but the protection scope of the present application is not limited thereto. Any person skilled in the art who is familiar with the present technical field can easily think of changes or substitutions within the technical scope disclosed in the present application, which should be included in the protection scope of the present application. Therefore, the protection scope of the present application should be based on the protection scope of the claims.

[0070] Finally: The above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions, improvements, etc. made within the spirit and principles of the present invention should be included in the protection scope of the present invention.

Claims

1. A low-power sound source localization method based on two-level data fusion, characterized in that: The following steps are involved: Deploy sensor nodes, obtain the energy consumed by transmitting data between sensor nodes in real time, select cluster head nodes based on the energy residual value method, and build a wireless sensor network structure based on the cluster head nodes; The sound source signal model based on RSS is constructed by detecting the energy value of the signal of the sensor node at time k in the wireless sensor network; Acquiring the amount of data collected by the sensor within a period of time based on the sound source signal model to build a data structure model of the collected physical quantity, and constructing a similarity function model of the first-level data fusion according to the data structure model; Performing first-level data fusion according to the similarity function model of the first-level data fusion, removing the redundancy of the data collected by the sensor node itself, and obtaining measurement data; Collect data transmitted to two different sensor nodes in the same cluster, construct a similarity function model based on Euclidean distance according to the data of the two different sensor nodes, remove redundancy from the data received from different sensor nodes, and perform second-level data fusion on the data; After the second-level data fusion, all the data received by the sink node are obtained, an objective function is constructed based on all the data received by the sink node, and the EM algorithm is used to complete the positioning of the sound source.

2. A low-power sound source localization method based on two-level data fusion according to claim 1, characterized in that: The specific steps of constructing the sound source signal model based on RSS are as follows: Randomly deploy in a monitoring area of ​​a wireless sensor network sensor nodes and 1 sound source to establish a rectangular coordinate system; Obtain the unknown quantities of the nth sensor node position and the sound source at time K based on the rectangular coordinate system; A sound source signal model based on RSS is constructed according to the position of the nth sensor node and the unknown quantity of the sound source at time K.

3. The low-power sound source localization method based on two-level data fusion according to claim 1, characterized in that: The data structure model of the collected physical quantities is constructed, and the similarity function model of the first-level data fusion is constructed according to the data structure model. The specific steps are as follows: Divide the K time period into several equal time periods, collect m data in all time periods through the nth sensor, use the sound source signal model to obtain the collected energy value and build a data structure model, and output the vector ; If the sound source does not change position in a short period of time and changes very slowly, the data collected by the sensor node There is a high degree of redundancy; Through the sensor nodes need to search for vector The similarity function of the first-level data fusion is constructed by using similar data in the dataset.

4. The low-power sound source localization method based on two-level data fusion according to claim 1, characterized in that: The specific steps of performing the first level data fusion are as follows: During the data collection phase, sensor nodes, obtain two new measured data and , use the weight value of the measurement information to represent similar measurement values ​​in the measurement data vector; The similarity of the two newly measured data is determined. If the two newly measured data are similar, they will be determined as similar values. The weight value will be increased by 1, and the newly measured data Will be discarded by the sensor node.

5. The low-power sound source localization method based on two-level data fusion according to claim 1, characterized in that: The similarity function model based on Euclidean distance is constructed, and the specific steps are as follows: Get two different sensor nodes in the same cluster and , respectively obtain the data vectors passed to the cluster head node after the first level of data fusion ; A similarity function model based on Euclidean distance is constructed according to the data after the first-level data fusion.

6. The low-power sound source localization method based on two-level data fusion according to claim 1, characterized in that: The specific steps of performing the second-level data fusion on the data are as follows: The data vector of the nth sensor node after the first level of data fusion and the The data vectors of the sensor nodes after the first level of data fusion are split into two parts, one part is the intersection part contained in both data vectors, and the other part is the different parts contained in each data vector; The intersection part is defined to obtain the intersection definition vector, and the different part of the data after the first level data fusion of the nth sensor node is defined to obtain and The different parts of the data after the first level of data fusion of the sensor nodes are defined ; The data vector and data vector Converted into a data set, according to the data set and Get the collection amount The Euclidean distance of If the Euclidean distance is less than the set distance threshold, the amount of collection are similar, there is data redundancy, and the cluster head node selects one of them for data transmission according to the needs and abandons the other one.

7. The low-power sound source localization method based on two-level data fusion according to claim 1, characterized in that: After the second-level data fusion, all the data received by the sink node are obtained, the objective function is constructed based on all the data received by the sink node, and the EM algorithm is used to complete the location of the sound source, as follows: Get all the data received by the sink node and define matrix, matrix, Matrix three matrices; according to matrix, matrix, The three matrices are based on the source signal model to obtain the matrix equations; Obtaining Gaussian probability density characteristics and properties of environmental noise, and obtaining a joint probability density function of a matrix equation according to the Gaussian probability density characteristics and properties of the environmental noise; The joint probability density function based on the matrix equation is equivalent to a quadratic expression of the negative log-likelihood function; The estimated value of the location information of the sound source is converted into the maximum value of the maximum likelihood function of the expression equivalent to a quadratic form of the negative logarithmic likelihood function, and the solution of the maximum likelihood function estimate is further converted into minimizing the following formula to obtain ; Based on the EM algorithm, the energy intensity per unit distance is As a hidden variable, let , get the hidden variables; Substitute the hidden variables into In the equation, we get the corrected log-likelihood function; The corrected log-likelihood function is minimized to a maximum value, and the The variable search algorithm is optimized to obtain the estimated information of the sound source location.

8. The low-power sound source localization method based on two-level data fusion according to claim 5, characterized in that: The specific formula of the similarity function model based on Euclidean distance is as follows: In the formula, For two sensor nodes belonging to the same cluster, the amount of data collected The similarity function of is the threshold value, is the data vector of the nth sensor node after the first level of data fusion, For the The data vector of the sensor nodes after the first level of data fusion.

9. The low-power sound source localization method based on two-level data fusion according to claim 1, characterized in that: The energy consumed in transmitting data between the sensor nodes is specifically expressed as follows: In the formula, Indicates the energy consumed by a radio device to start the transmission circuit. Indicates the energy consumed by the transmission amplifier startup circuit. is the transmission distance, is the number of bits of transmitted data, is the threshold value, Indicates the distance Transmission between two sensor nodes The energy consumed by a bit of data.

10. The low-power sound source localization method based on two-level data fusion according to claim 7, characterized in that: The specific expressions of the three matrices are as follows: The matrix equation, the specific calculation formula is as follows: The specific calculation formula of the joint probability density function of the matrix equation is as follows: The negative log-likelihood function is equivalent to an expression in quadratic form, and the specific formula is as follows: The minimization formula is as follows: The specific calculation formula of the hidden variable is as follows: The specific calculation formula of the corrected log-likelihood function is as follows: The corrected log-likelihood function is minimized to obtain a maximum value, and the specific calculation formula is as follows: In the formula, matrix, matrix, The matrix is ​​three matrices, are the Gaussian noise mean and variance, is the energy intensity per unit distance, represents the transpose of matrices and vectors, N is the number of sensors, is the energy collected by the Nth sensor, where =1,2,3,...,R, R is an integer, The maximum likelihood function estimate is transformed into minimization, is the corrected log-likelihood function, is the unit matrix, is a projection matrix, is the gain of the Nth sensor, is the transmission distance of the Nth sensor.

Images