CN111970716B - Fluid boundary extraction method based on classified hyperplane search in underwater acoustic sensor network - Google Patents

Fluid boundary extraction method based on classified hyperplane search in underwater acoustic sensor network Download PDF

Info

Publication number
CN111970716B
CN111970716B CN202010818276.6A CN202010818276A CN111970716B CN 111970716 B CN111970716 B CN 111970716B CN 202010818276 A CN202010818276 A CN 202010818276A CN 111970716 B CN111970716 B CN 111970716B
Authority
CN
China
Prior art keywords
event
unit cell
fluid
backbone
nodes
Prior art date
Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
Active
Application number
CN202010818276.6A
Other languages
Chinese (zh)
Other versions
CN111970716A (en
Inventor
刘立
韩光洁
Current Assignee (The listed assignees may be inaccurate. Google has not performed a legal analysis and makes no representation or warranty as to the accuracy of the list.)
Changzhou Campus of Hohai University
Original Assignee
Changzhou Campus of Hohai University
Priority date (The priority date is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the date listed.)
Filing date
Publication date
Application filed by Changzhou Campus of Hohai University filed Critical Changzhou Campus of Hohai University
Priority to CN202010818276.6A priority Critical patent/CN111970716B/en
Publication of CN111970716A publication Critical patent/CN111970716A/en
Application granted granted Critical
Publication of CN111970716B publication Critical patent/CN111970716B/en
Active legal-status Critical Current
Anticipated expiration legal-status Critical

Links

Images

Classifications

    • H—ELECTRICITY
    • H04—ELECTRIC COMMUNICATION TECHNIQUE
    • H04W—WIRELESS COMMUNICATION NETWORKS
    • H04W16/00—Network planning, e.g. coverage or traffic planning tools; Network deployment, e.g. resource partitioning or cells structures
    • H04W16/18—Network planning tools
    • H04W16/20—Network planning tools for indoor coverage or short range network deployment
    • H—ELECTRICITY
    • H04—ELECTRIC COMMUNICATION TECHNIQUE
    • H04W—WIRELESS COMMUNICATION NETWORKS
    • H04W40/00—Communication routing or communication path finding
    • H04W40/02—Communication route or path selection, e.g. power-based or shortest path routing
    • H—ELECTRICITY
    • H04—ELECTRIC COMMUNICATION TECHNIQUE
    • H04W—WIRELESS COMMUNICATION NETWORKS
    • H04W84/00—Network topologies
    • H04W84/18—Self-organising networks, e.g. ad-hoc networks or sensor networks
    • H—ELECTRICITY
    • H04—ELECTRIC COMMUNICATION TECHNIQUE
    • H04B—TRANSMISSION
    • H04B13/00—Transmission systems characterised by the medium used for transmission, not provided for in groups H04B3/00 - H04B11/00
    • H04B13/02—Transmission systems in which the medium consists of the earth or a large mass of water thereon, e.g. earth telegraphy
    • Y—GENERAL TAGGING OF NEW TECHNOLOGICAL DEVELOPMENTS; GENERAL TAGGING OF CROSS-SECTIONAL TECHNOLOGIES SPANNING OVER SEVERAL SECTIONS OF THE IPC; TECHNICAL SUBJECTS COVERED BY FORMER USPC CROSS-REFERENCE ART COLLECTIONS [XRACs] AND DIGESTS
    • Y02—TECHNOLOGIES OR APPLICATIONS FOR MITIGATION OR ADAPTATION AGAINST CLIMATE CHANGE
    • Y02D—CLIMATE CHANGE MITIGATION TECHNOLOGIES IN INFORMATION AND COMMUNICATION TECHNOLOGIES [ICT], I.E. INFORMATION AND COMMUNICATION TECHNOLOGIES AIMING AT THE REDUCTION OF THEIR OWN ENERGY USE
    • Y02D30/00—Reducing energy consumption in communication networks
    • Y02D30/70—Reducing energy consumption in communication networks in wireless communication networks

Abstract

The invention discloses a fluid boundary extraction method based on classified hyperplane search in an underwater acoustic sensor network, which comprises a fluid coverage area search calibration stage, an elliptical envelope determination stage, an annular area division stage, a backbone unit cell selection stage and a fluid boundary extraction stage. The invention can extract the fluid boundary with high precision, and has the advantages of quick deployment, short-distance, real-time and seamless detection and tracking of the streaming media target, and the like.

Description

Fluid boundary extraction method based on classified hyperplane search in underwater acoustic sensor network
Technical Field
The invention relates to a fluid boundary extraction method based on classified hyperplane search in an underwater acoustic sensor network, and belongs to the field of underwater acoustic sensors.
Background
With the increasing demand of human life style for energy, the frequency and scale of offshore industrial activities in the global range are increasing, the human economic activities in coastal zone areas are highly concentrated, and unexpected leakage of petroleum, nuclear waste water and the like and large-area outbreaks of harmful algae such as red tide, water bloom and the like are easy to occur.
At present, the monitoring networking mode of the fluid target can be divided into four dimensions of land, sea, air and sky according to space regions. However, there are many disadvantages of monitoring the fluid target, such as remote sensing observation is constrained by revisit time and spatial resolution, and it is unable to capture the fluid target formed in burst in time and provide motion monitoring in full life cycle; for example, autonomous underwater vehicles have high equipment cost, are only suitable for small-scale fluid tracking on the coasts of ports, wharfs and the like, and need to keep synchronous movement with the fluid to avoid target loss.
Disclosure of Invention
In order to overcome the defects and shortcomings of the prior art, the invention discloses a fluid boundary extraction method based on classified hyperplane search in an underwater acoustic sensor network.
The technical scheme adopted in the invention is as follows:
a fluid boundary extraction method based on classified hyperplane search in an underwater acoustic sensor network comprises the following steps:
s1: fluid coverage area search calibration phase
S1-1: the sensor nodes are clustered according to geographic positions, each cluster at least comprises 4 nodes, and different monitoring areas are formed according to the node clustering; the sensor node covered by the fluid target can sense the existence of the fluid, and is called an event node, and a cluster where the event node is located is called an event cluster;
s1-2: all event nodes send notification messages to the cluster heads of the event cluster i, the cluster heads determine the maximum and minimum values of the transverse coordinates and the longitudinal coordinates in all event nodes in the cluster according to the collected event node IDs and the coordinates, the maximum and minimum values are respectively marked as Xmin_ia, xmax_ib, ymin_ic and Ymax_id, and a local quadruple quadricycle_i= [ L_ia, R_ib, D_ic and U_id ] is generated according to the node coordinates corresponding to the maximum and minimum values of the transverse coordinates and the longitudinal coordinates, and the quadruple is used for representing the coverage range of fluid in the up, down, left and right directions in the cluster;
s1-3: each event cluster in the range of the underwater acoustic sensor network exchanges respective local quaternions, and each event cluster updates own quaternion according to the coordinate value in the received quaternion until the local quaternion of all event clusters in the network is not changed any more; at this time, the four-tuple maintained by the event cluster in the network represents the coverage range of the fluid in four directions of the whole network, namely the up direction, the down direction, the left direction and the right direction; a circumscribed rectangle that can contain the fluid coverage area is determined using the quadruple.
S2: elliptical envelope determination stage
S2-1: determining an elliptical envelope according to the length-width ratio of the circumscribed rectangle; if the aspect ratio of the circumscribed rectangle is more than or equal to 2, determining the circumscribed ellipse of the rectangle as an envelope; if the aspect ratio of the circumscribed rectangle is smaller than 2, determining that the circumscribed circle of the rectangle is taken as an envelope;
s3: annular region dividing stage
S3-1: setting the number of layers of a desired annular region as H, and uniformly dividing H concentric ellipses by taking an elliptical envelope as an outermost layer to obtain H annular regions;
s3-2: and dividing the annular area of each layer according to the organization mode of the full binary tree. The ith annular area is uniformly divided into 2 i-1 A sub-region. Eventually, 2 will be contained within the elliptical envelope H -1 sub-partition of different sizes. All the sub-partitions are called as unit cells, and are positioned at the outermost layer 2 H-1 The individual unit cells are called leaf unit cells.
S4 backbone cell selection
S4-1: selecting a backbone unit cell from the unit cells formed in step S3-2, wherein the conditions for satisfying the backbone unit cell are: (1) leaf cells containing event nodes; (2) the device contains event nodes, and at least one of the left child unit cells and the right child unit cells has no event node; both of the above conditions are met and one is selected as the backbone unit cell.
S5: fluid boundary extraction stage
S5-1: all sensor nodes in each backbone unit cell are distributed and organized according to geographic positions to form a learning community. Each learning community contains the same number of nodes, so that the same scale but different numbers of learning communities are arranged in different backbone units. Each learning community still structurally maintains the organization form of a full binary tree;
s5-2: and designing a longitudinal cascade type learning strategy and a transverse cascade type learning strategy by using father-son type and brother type adjacency relations maintained by using a full binary tree by each learning community, enabling nodes in different learning communities to take own position information and whether fluid is monitored as samples, and performing cooperative training in a support vector machine to obtain a classification hyperplane for maximizing the interval between event nodes and non-event nodes, and taking the classification hyperplane as an extraction result of the fluid boundary.
Preferably, the manner of updating the quadruple of each event cluster in the step S1-3 is as follows: and comparing the coordinates of the local quaternion with the coordinates of the received quaternion in the upper, lower, left and right directions, and updating the quaternion of the local quaternion by using the node coordinates corresponding to the maximum and minimum values of the horizontal coordinate and the vertical coordinate obtained after comparison.
Preferably, in the above-described elliptical envelope determination stage, since the ellipse becomes a circle when the eccentricity is 0, a circular envelope may be regarded as a special case of an elliptical envelope.
Preferably, in the above step S3-2, the specific division manner of the annular area of each layer according to the organization manner of the full binary tree is that the annular area of the innermost layer, i.e. the annular area of the first layer, is called the root area. The second layer annular region is uniformly divided into 2 regions of equal area as left and right sub-partitions of the root region. The annular region of the third layer is uniformly divided into 4 area phasesEtc. as left and right sub-partitions of the two partitions in the second layer, respectively. The above process is equivalent to the process of generating left and right child nodes by father nodes in the full binary tree, and is continuously executed until the H layer where the elliptical envelope is located, and the H layer is uniformly divided into 2 H-1 Sub-partitions. Eventually, 2 will be contained within the elliptical envelope H -1 sub-partition of different sizes.
Preferably, all the sensor nodes of the backbone cells described in step S5-1 are uniformly distributed according to the density of geographical distribution, in the following manner:
the unit cell areas of the brother relations are the same, the node distribution number is the same, and the learning community number is the same. For cells having a parent-child relationship, the area of the parent cell is 2 times that of the child cell, and therefore, the number of sensor nodes in the parent cell is 2 times that of the child cell, and similarly, the number of learning communities of the parent cell is 2 times that of the child cell.
Preferably, the nodes in the different learning communities in the step S5-2 take their own position information and whether the fluid is detected as samples, and take { (x) 1 ,y 1 )，(x 2 ,y 2 ),…,(x m ,y m ) Is input { z } 1 ,z 2 ,…,z m And the output is a sample set formed by the output. Wherein (x) i ，y i ) Representing the Cartesian coordinates of the sensor node i; z i Tag, z, representing sensor node i i E { +1, -1}, when z i When the value is +1, the node i is an event node, and when z i And when the value is-1, the node i is a non-event node.
Preferably, in the step S5-2, the training of the classification hyperplane by using the support vector machine is to input the training samples into a classification decision function, where the classification decision function is:
Figure GDA0004045363670000051
wherein N is sv Representing the number of support vectors, alpha i The Lagrangian multiplier corresponding to the support vector, K (,) is a Gaussian kernel function.Let sigmoid [ f (x)]The obtained classification decision function represents a classification hyperplane, i.e., an estimation result of the position of the fluid boundary.
Preferably, the vertical cascade learning strategy in step S5-2 described above is directed to a set of backbone cells where a parent-child relationship exists. Starting from the top of the vertical cascade structure, the parent backbone unit sends its own training generated local support vectors within each learning community to the child backbone unit. The child backbone unit combines the training samples of all school communities with the received local support vectors, trains the local support vectors of the child backbone unit, and then continuously transmits the local vectors downwards. Repeating the above process until the bottom end of the longitudinal cascade structure. And feeding back the local support vector at the bottom end of the cascade structure to the top end of the cascade structure. And iterating the process until the support vector at the top end of the longitudinal cascade structure converges.
Preferably, the lateral cascade learning strategy in step S4-3 described above is directed to a set of backbone unit cells that have sibling relationships. Starting from the beginning of the horizontal cascade structure, the backbone cells send their respective intra-learning community training generated local support vectors to their sibling backbone cells. The brother backbone unit cell combines the training samples of all school communities with the received local support vectors, trains the local support vectors of the brother backbone unit cell, and then continuously transmits the local vectors backwards. The above process is repeated until the end of the lateral cascade structure. And feeding back the local support vector at the tail end of the cascade structure to the beginning of the cascade structure. And iterating the process until the support vector at the top end of the transverse cascade structure is converged.
The beneficial effects are that: the invention provides a fluid boundary extraction method based on classified hyperplane search, which has the following advantages:
the ultrasonic sensor network is utilized to extract the hyperplane, and the deployment area and the fluid movement area are in the same dimension, so that the rapid deployment of the monitoring task is facilitated. And the underwater acoustic sensor network can compensate the technical short plates of satellite remote sensing, underwater vehicle cruising and other observation means to provide long-time, short-distance, seamless and real-time fluid target monitoring and tracking service in the face of fluid targets which are formed suddenly and are distributed in space along with the height change of the motion process.
The fluid boundary extraction method based on the classified hyperplane search can extract the fluid edge with high precision. Meanwhile, training classification hyperplane is carried out by taking learning communities as units, so that the limitation of resources of sensor nodes is fully considered, and the life cycle of the underwater acoustic sensor network is prolonged.
Drawings
FIG. 1 is a schematic illustration of the fluid coverage area search calibration and elliptical envelope determination of the present invention;
FIG. 2 is a schematic diagram of a backbone cell selection according to the present invention;
FIG. 3 is a schematic diagram of a longitudinal cascade learning strategy of the present invention;
fig. 4 is a schematic diagram of a lateral cascading learning strategy of the present invention.
Detailed Description
In order to better understand the technical solutions in the present application, the following description will clearly and completely describe the technical solutions in the embodiments of the present application, and it is obvious that the described embodiments are only some embodiments of the present application, not all embodiments. All other embodiments, which can be made by one of ordinary skill in the art based on the embodiments herein without making any inventive effort, shall fall within the scope of the present application.
A fluid boundary extraction method based on classified hyperplane search in a hydroacoustic sensor network comprises a fluid coverage area search calibration stage, an elliptical envelope determination stage, an annular area division stage, a backbone unit cell selection stage and a fluid boundary extraction stage.
As shown in fig. 1, the fluid coverage area search calibration phase includes:
s1: fluid coverage area search calibration phase
S1-1: the sensor nodes are clustered according to geographic positions, each cluster at least comprises 4 nodes, and different monitoring areas are formed according to the node clustering; the sensor node covered by the fluid target can sense the existence of the fluid, and is called an event node, and a cluster where the event node is located is called an event cluster; as shown in fig. 1, there are three event clusters, cluster head nodes A0, B0, C0, respectively.
S1-2: all event nodes send notification messages to the cluster heads of the event cluster i, the cluster heads determine the maximum and minimum values of the transverse coordinates and the longitudinal coordinates in all event nodes in the cluster according to the collected event node IDs and the coordinates, the maximum and minimum values are respectively marked as Xmin_ia, xmax_ib, ymin_ic and Ymax_id, and a local quadruple quadricycle_i= [ L_ia, R_ib, D_ic and U_id ] is generated according to the node coordinates corresponding to the maximum and minimum values of the transverse coordinates and the longitudinal coordinates, and the quadruple is used for representing the coverage range of fluid in the up, down, left and right directions in the cluster; as shown in fig. 1, taking an event cluster in which a cluster head node A0 is located as an example, event nodes A0 to A4 exist in the cluster, the maximum and minimum values of the abscissa and the ordinate of all event nodes in the cluster are respectively xmin_a2, xmax_a4, ymin_a3 and ymax_a1, and four-element groups quadri_a= [ l_a2, r_a4, d_a3 and u_a1] used for describing distribution in four directions of fluid up, down, left and right in the cluster
S1-3: each event cluster in the whole network range exchanges respective local quaternion, and each event cluster updates the local quaternion according to the coordinate value in the received quaternion until the local quaternion of all event clusters in the network is not changed any more. The local quadruple maintained by the event cluster in the network at this time represents the coverage of the fluid in four directions, up, down, left and right of the whole network. As shown in fig. 1, after the four clusters of events in the figure exchange the four clusters with each other, the four clusters that ultimately describe the distribution range of the fluid in four directions of up, down, left, and right over the whole network are [ l_c2, r_b5, d_c3, u_a1]. A circumscribed rectangle that can contain the fluid coverage area is determined using the quadruple.
As shown in fig. 1, the elliptical envelope determining phase includes:
s2-1: a circumscribed rectangle is determined that can contain the fluid coverage area.
S2-2: an elliptical envelope is determined from the aspect ratio of the circumscribed rectangle. If the aspect ratio of the circumscribed rectangle is more than or equal to 2, determining the circumscribed ellipse of the rectangle as an envelope; and if the aspect ratio of the circumscribed rectangle is smaller than 2, determining that the circumscribed circle of the rectangle is taken as an envelope.
The annular region dividing stage comprises the following steps:
s3-1: setting the number of layers of a desired annular region as H, and uniformly dividing H concentric ellipses by taking an elliptical envelope as an outermost layer to obtain H annular regions;
s3-2: and dividing the annular area of each layer according to the organization mode of the full binary tree. The ith annular area is uniformly divided into 2 i-1 A sub-region. Eventually, 2 will be contained within the elliptical envelope H -1 sub-partition of different sizes. All the sub-partitions are called as unit cells, and are positioned at the outermost layer 2 H-1 The individual unit cells are called leaf unit cells. As shown in fig. 2, the elliptical envelope is divided into 6 layers, and after each layer is further divided, 63 sub-regions are obtained, with numbers 1 to 63, wherein the total number of leaf cells is 32, and the numbers are 32 to 63.
The selection stage of the backbone unit cell comprises the following steps:
s4-1: selecting a backbone unit cell from the unit cells formed in step S3-2, wherein the conditions for satisfying the backbone unit cell are: (1) leaf cells containing event nodes; (2) the device contains event nodes, and at least one of the left child unit cells and the right child unit cells has no event node; both of the above conditions are met and one is selected as the backbone unit cell. As shown in FIG. 2, cells 9, 12-14, 16-18, 20-24, 29-31, 35-36, 47-48 are selected as backbone cells.
S5: fluid boundary extraction stage
S5-1: all sensor nodes in each backbone unit cell are distributed and organized according to geographic positions to form a learning community. Each learning community contains the same number of nodes, so that the same scale but different numbers of learning communities are arranged in different backbone units. Each learning community still structurally maintains the organization form of a full binary tree;
s5-2: and designing a longitudinal cascade type learning strategy and a transverse cascade type learning strategy by using father-son type and brother type adjacency relations maintained by using a full binary tree by each learning community, enabling nodes in different learning communities to take own position information and whether fluid is monitored as samples, and performing cooperative training in a support vector machine to obtain a classification hyperplane for maximizing the interval between event nodes and non-event nodes, and taking the classification hyperplane as an extraction result of the fluid boundary.
The method for updating the quadruple of each event cluster in the step S1-3 is to compare the coordinates of the local quadruple with the received quadruple in the four directions of up, down, left and right, and update the quadruple of the local quadruple by the node coordinates corresponding to the maximum and minimum values of the horizontal and vertical coordinates obtained after comparison.
In the elliptical envelope determination stage, since the ellipse becomes a circle when the eccentricity is 0, a circular envelope can be regarded as a special case of an elliptical envelope.
In step S3-2, the specific dividing mode for dividing the annular area of each layer according to the organization mode of the full binary tree is that the annular area of the innermost layer, namely the annular area of the first layer, is called a root area. The second layer annular region is uniformly divided into 2 regions of equal area as left and right sub-partitions of the root region. The annular region of the third layer is uniformly divided into 4 sub-partitions with equal areas, and the sub-partitions are respectively used as left and right sub-partitions of two partitions in the second layer. The above process is equivalent to the process of generating left and right child nodes by father nodes in the full binary tree, and is continuously executed until the H layer where the elliptical envelope is located, and the H layer is uniformly divided into 2 H-1 Sub-partitions. Eventually, 2 will be contained within the elliptical envelope H -1 sub-partition of different sizes.
The density of all the sensor nodes of the backbone cells in the step S5-1 is uniformly distributed according to the geographical position distribution, and the specific distribution mode is as follows:
the unit cell areas of the brother relations are the same, the node distribution number is the same, and the learning community number is the same. The area of the father unit cell is 2 times that of the child unit cell, the node distribution number of the father unit cell is 2 times that of the child unit cell, and the learning community number of the father unit cell is 2 times that of the child unit cell.
In step S5-2, the nodes in different learning communities take their own position information and whether fluid is monitored as samples, and { (x) 1 ，y 1 )，(x 2 ，y 2 )，……，(x m ，y m ) Sum { z } 1 ，z 2 ，……，z m Together, constitute a sample set. Wherein (x) i ，y i ) Representing the Cartesian coordinates of the sensor node i; z i Tag, z, representing sensor node i i E { +1, -1}, when z i When the value is +1, the node i is an event node, and when z i And when the value is-1, the node i is a non-event node.
In the step S5-2, the support vector machine is used to train the classification hyperplane, and the training samples are input into a classification decision function, wherein the classification decision function is as follows:
Figure GDA0004045363670000101
wherein N is sv Representing the number of support vectors, alpha i The Lagrangian multiplier corresponding to the support vector, K (,) is a Gaussian kernel function. Let sigmoid [ f (x)]The obtained classification decision function represents a classification hyperplane, i.e., an estimation result of the position of the fluid boundary.
The vertical cascade learning strategy described in step S5-2 is directed to a set of backbone cells that have a parent-child relationship. As shown in fig. 2, the backbone cells 9, 18, 36 are in parent-child relationship with each other, forming a set of three-layered longitudinal cascade structure. As shown in fig. 3, the backbone cell 9 is divided into 4 learning communities, LC i_1,LC i_2,LC i_3,LC i_4, respectively. Since the backbone cell 18 is a child cell of the backbone cell 9, the backbone cell 18 can be divided into 2 learning communities LC i+1_1, LC i+1_2. Because backbone cell 36 is a child of backbone cell 18, backbone cell 36 can be divided into 1 learning communities LC i +2_1. After the four learning communities of the backbone cell 9 trained out the local support vectors SV i_1, SV i_2, SV i_3, SV i_4, the four local support vectors are sent to the learning communities LC i+1_1, LC i+1_2 of their child cells 18. The two learning communities combine the local training data set and the received local support vectors, train the local support vectors SV i+1_1 and SV i+1_2, and then send the local support vectors SV i+1_1 and SV i+1_2 to the learning community LC i+2_1 of the child cell 36. The learning community LC i+2_1 combines the local training data set and the received local support vector, trains the local support vector SV i+2_1 itself, and then feeds back SV i+2_1 to the four learning communities of the backbone cell 9. And iterating the process until the support vector at the top end of the longitudinal cascade structure converges.
The lateral cascade learning strategy described in step S5-2 is directed to a set of backbone unit cells that have sibling relationships. As shown in fig. 2, the backbone cells 20, 21, 22 are sibling to each other, forming a set of laterally cascaded structures. The area of backbone cells 20, 21, 22 and the number of learning communities are the same. As shown in fig. 4, each backbone cell is divided into one learning community, i.e., backbone cell 20 is divided into learning community LC i_1, backbone cell 21 is divided into learning community LC i_2, and backbone cell 22 is divided into learning community LC i_3. The learning community LC i_1 of the backbone cell 20 trains out the local support vector SV i_1, and sends SV i_1 to the learning community LC i_2 in its sibling cell 21. The learning community LC i_2 combines the local training data set and the received local support vector, trains the local support vector SV i_2 itself, and then transmits SV i_2 to the learning community LC i_3 of its sibling cell 22. The learning community LC i_3 combines the local training data set and the received local support vector, trains the local support vector SV i_3 itself, and then feeds back SV i_3 to the learning community LC i_1 of the backbone cell 20. And iterating the process until the support vector at the top end of the transverse cascade structure is converged.
The foregoing is merely a preferred embodiment of the present invention and it should be noted that modifications and adaptations to those skilled in the art may be made without departing from the principles of the present invention, which are intended to be comprehended within the scope of the present invention.

Claims (9)

1. A fluid boundary extraction method based on classified hyperplane search in an underwater acoustic sensor network is characterized by comprising the following steps:
s1: fluid coverage area search calibration phase
S1-1: the sensor nodes are clustered according to geographic positions, each cluster at least comprises 4 nodes, and different monitoring areas are formed according to the node clustering; the sensor node covered by the fluid target can sense the existence of the fluid, and is called an event node, and a cluster where the event node is located is called an event cluster;
s1-2: all event nodes send notification messages to the cluster heads of the event cluster i, the cluster heads determine the maximum and minimum values of the transverse coordinates and the longitudinal coordinates in all event nodes in the cluster according to the collected event node IDs and the coordinates, the maximum and minimum values are respectively marked as Xmin_ia, xmax_ib, ymin_ic and Ymax_id, and a local quadruple quadricycle_i= [ L_ia, R_ib, D_ic and U_id ] is generated according to the node coordinates corresponding to the maximum and minimum values of the transverse coordinates and the longitudinal coordinates, and the quadruple is used for representing the coverage range of fluid in the up, down, left and right directions in the cluster;
s1-3: each event cluster in the range of the underwater acoustic sensor network exchanges respective local quaternions, and each event cluster updates own quaternion according to the coordinate value in the received quaternion until the local quaternion of all event clusters in the network is not changed any more; at this time, the four-tuple maintained by the event cluster in the network represents the coverage range of the fluid in four directions of the whole network, namely the up direction, the down direction, the left direction and the right direction; determining an external rectangle capable of containing the fluid coverage area by using the quadruple;
s2: elliptical envelope determination stage
S2-1: determining an elliptical envelope according to the length-width ratio of the circumscribed rectangle; if the aspect ratio of the circumscribed rectangle is more than or equal to 2, determining the circumscribed ellipse of the rectangle as an envelope; if the aspect ratio of the circumscribed rectangle is smaller than 2, determining that the circumscribed circle of the rectangle is taken as an envelope;
s3: annular region dividing stage
S3-1: setting the number of layers of a desired annular region as H, and uniformly dividing H concentric ellipses by taking an elliptical envelope as an outermost layer to obtain H annular regions;
s3-2: dividing the annular areas of each layer according to the organization mode of the full binary tree; the ith annular area is uniformly divided into 2 i-1 A sub-region; eventually, 2 will be contained within the elliptical envelope H -1 sub-partition of different sizes; all the sub-partitions are called as unit cells, and are positioned at the outermost layer 2 H-1 The individual unit cells are called leaf unit cells;
s4: backbone cell selection
S4-1: selecting a backbone unit cell from the unit cells formed in step S3-2, wherein the conditions for satisfying the backbone unit cell are: (1) leaf cells containing event nodes; (2) the device contains event nodes, and at least one of the left child unit cells and the right child unit cells has no event node; both of the above conditions are satisfied and one is selected as a backbone unit cell;
s5: fluid boundary extraction stage
S5-1: all sensor nodes in each backbone unit cell are distributed and organized according to geographic positions to form a learning community; the number of nodes contained in each learning community is the same, so that the learning communities with the same scale but different numbers are arranged in different backbone units; each learning community still structurally maintains the organization form of a full binary tree;
s5-2: and designing a longitudinal cascade type learning strategy and a transverse cascade type learning strategy by using father-son type and brother type adjacency relations maintained by using a full binary tree by each learning community, enabling nodes in different learning communities to take own position information and whether fluid is monitored as samples, and performing cooperative training in a support vector machine to obtain a classification hyperplane for maximizing the interval between event nodes and non-event nodes, and taking the classification hyperplane as an extraction result of the fluid boundary.
2. The fluid boundary extraction method based on classified hyperplane search in an underwater acoustic sensor network according to claim 1, wherein the method comprises the following steps: the mode of updating the quadruple of each event cluster is to compare the coordinates of the local quadruple with the received quadruple in the four directions of the upper, lower, left and right, and update the quadruple of the local quadruple by the node coordinates corresponding to the maximum and minimum values of the horizontal and vertical coordinates obtained after comparison.
3. The fluid boundary extraction method based on classified hyperplane search in an underwater acoustic sensor network according to claim 1, wherein the method comprises the following steps: in the elliptical envelope determination stage, since the ellipse becomes a circle when the eccentricity is 0, a circular envelope can be regarded as a special case of an elliptical envelope.
4. The fluid boundary extraction method based on classified hyperplane search in an underwater acoustic sensor network according to claim 1, wherein the method comprises the following steps: in the step S3-2, the specific dividing mode for dividing the annular area of each layer according to the organization mode of the full binary tree is as follows:
the innermost annular region, i.e., the first annular region, is referred to as the root region;
the second layer annular region is uniformly divided into 2 regions with equal area as left and right sub-regions of the root region;
the annular region of the third layer is uniformly divided into 4 sub-partitions with equal areas, and the sub-partitions are respectively used as left and right sub-partitions of two partitions in the second layer;
the dividing process is equivalent to the process of generating left and right child nodes by father nodes in the full binary tree, the H layer where the elliptical envelope is located is continuously executed, and the layer is evenly divided into 2 H-1 Sub-partitions; eventually, 2 will be contained within the elliptical envelope H -1 sub-partition of different sizes.
5. The fluid boundary extraction method based on classified hyperplane search in an underwater acoustic sensor network according to claim 1, wherein the method comprises the following steps: the density of all the sensor nodes of the backbone unit cells in the step S5-1 is uniformly distributed according to the geographical position distribution, and the specific distribution mode is as follows:
the unit cell areas of brother relations are the same, the node distribution quantity is the same, and the learning community quantity is the same;
the area of the father unit cell is 2 times that of the child unit cell, the node distribution number of the father unit cell is 2 times that of the child unit cell, and the learning community number of the father unit cell is 2 times that of the child unit cell.
6. The fluid boundary extraction method based on classified hyperplane search in an underwater acoustic sensor network according to claim 1, wherein the method comprises the following steps: the node in the learning community in the step S5-2 takes the self-position information and whether the fluid is monitored as samples, namely each schoolSensor nodes in the community will be in their own Cartesian coordinates (x i ，y i ) For inputting, perceiving label z i Packaging the output into a training sample; wherein z is i E { +1, -1}, when z i When the value is +1, the node i is an event node, and when z i And when the value is-1, the node i is a non-event node.
7. The fluid boundary extraction method based on classified hyperplane search in an underwater acoustic sensor network according to claim 1, wherein the method comprises the following steps: in the step S5-2, the support vector machine is used for training the classification hyperplane, the training sample is input into a classification decision function, and the classification decision function is as follows:
Figure FDA0004045363660000041
wherein N is sv Representing the number of support vectors, alpha i The Lagrangian multiplier corresponding to the support vector is provided, and K (,) is a Gaussian kernel function; let sigmoid [ f (x)]The obtained classification decision function represents a classification hyperplane, i.e., an estimation result of the position of the fluid boundary.
8. The fluid boundary extraction method based on classified hyperplane search in an underwater acoustic sensor network according to claim 1, wherein the method comprises the following steps: the longitudinal cascade learning strategy in the step S5-2 is aimed at a set of backbone cells with father-son relations; starting from the top end of the longitudinal cascade structure, the father backbone unit cell transmits the local support vectors generated by training in each learning community to the child backbone unit cell; the child backbone unit cell combines training samples of all school communities with the received local support vector, trains the local support vector of the child backbone unit cell, and then continuously transmits the local vector downwards; repeating the above process until the bottom end of the longitudinal cascade structure; feeding back the local support vector at the bottom end of the cascade structure to the top end of the cascade structure; and iterating the process until the support vector at the top end of the longitudinal cascade structure converges.
9. The fluid boundary extraction method based on classified hyperplane search in an underwater acoustic sensor network according to claim 1, wherein the method comprises the following steps: the transverse cascade learning strategy in the step S5-2 aims at backbone unit cell sets with brother relations; starting from the beginning of the horizontal cascade structure, the backbone cells send the local support vectors generated by training in each learning community to the brother backbone cells; the brother backbone unit cell combines the training samples of all school communities with the received local support vectors, trains the local support vectors of the brother backbone unit cell, and then continuously transmits the local vectors backwards; repeating the above process until the end of the transverse cascade structure; feeding back the local support vector at the tail end of the cascade structure to the beginning of the cascade structure; and iterating the process until the support vector at the top end of the transverse cascade structure is converged.
CN202010818276.6A 2020-08-14 2020-08-14 Fluid boundary extraction method based on classified hyperplane search in underwater acoustic sensor network Active CN111970716B (en)

Priority Applications (1)

Application Number Priority Date Filing Date Title
CN202010818276.6A CN111970716B (en) 2020-08-14 2020-08-14 Fluid boundary extraction method based on classified hyperplane search in underwater acoustic sensor network

Applications Claiming Priority (1)

Application Number Priority Date Filing Date Title
CN202010818276.6A CN111970716B (en) 2020-08-14 2020-08-14 Fluid boundary extraction method based on classified hyperplane search in underwater acoustic sensor network

Publications (2)

Publication Number Publication Date
CN111970716A CN111970716A (en) 2020-11-20
CN111970716B true CN111970716B (en) 2023-06-20

Family

ID=73365761

Family Applications (1)

Application Number Title Priority Date Filing Date
CN202010818276.6A Active CN111970716B (en) 2020-08-14 2020-08-14 Fluid boundary extraction method based on classified hyperplane search in underwater acoustic sensor network

Country Status (1)

Country Link
CN (1) CN111970716B (en)

Families Citing this family (2)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN113179135B (en) * 2021-04-23 2023-09-26 河海大学 Fluid boundary extraction method based on information entropy maximization in underwater acoustic sensor network
CN113179136B (en) * 2021-04-23 2023-11-28 河海大学 Fluid boundary tracking method based on virtual node migration in underwater acoustic sensor network

Citations (1)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN101765095A (en) * 2009-12-14 2010-06-30 浙江大学 Tracking method for wireless sensor network moving target based on mixed cluster

Family Cites Families (2)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN104219704B (en) * 2014-09-24 2017-09-29 河海大学常州校区 Monitoring poisonous gas and method for tracing based on double layer grid model in wireless sensor network
CN108364031B (en) * 2018-03-20 2020-07-21 盘锦中录油气技术服务有限公司 Logging interpretation plate classification boundary dividing method based on support vector machine

Patent Citations (1)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN101765095A (en) * 2009-12-14 2010-06-30 浙江大学 Tracking method for wireless sensor network moving target based on mixed cluster

Also Published As

Publication number Publication date
CN111970716A (en) 2020-11-20

Similar Documents

Publication Publication Date Title
CN111970716B (en) Fluid boundary extraction method based on classified hyperplane search in underwater acoustic sensor network
CN109711295A (en) A kind of remote sensing image offshore Ship Detection
CN110502006A (en) A kind of Mine Abandoned Land mobile robot complete coverage path planning method
CN107422736A (en) A kind of unmanned boat independently makes a return voyage system and its method of work
CN111970715B (en) Fluid boundary extraction method based on virtual backbone node connection in underwater acoustic sensor network
CN105787501A (en) Vegetation classification method capable of automatically selecting features in power transmission line corridor area
CN113179135B (en) Fluid boundary extraction method based on information entropy maximization in underwater acoustic sensor network
CN115294452B (en) Rotary SAR ship target detection method based on bidirectional characteristic pyramid network
CN110889198A (en) Multi-factor joint learning-based dead reckoning probability distribution prediction method and system
CN106872970A (en) A kind of multiple target based on differential evolution becomes data transfer rate tracks of device and its method
Spiliopoulos et al. Knowledge extraction from maritime spatiotemporal data: An evaluation of clustering algorithms on Big Data
CN111132003B (en) UWSN sensor node positioning method based on dynamic path planning
CN116467565A (en) Enteromorpha green tide plaque optimal search area forecasting method
Yin et al. An enhanced lightweight convolutional neural network for ship detection in maritime surveillance system
CN114331056A (en) Online collaborative detection task planning method based on probability map dynamic planning
Gonzalez et al. Semi-automatic extraction of ship lanes and movement corridors from AIS data
Bye et al. An improved receding horizon genetic algorithm for the tug fleet optimisation problem
CN115718905A (en) VTS system-oriented multi-sensor information fusion method
CN113179136B (en) Fluid boundary tracking method based on virtual node migration in underwater acoustic sensor network
CN115439745A (en) Navigation mark carrying type monitoring system and method for ship image characteristics
Bose et al. Forecasting the evolution of north Atlantic hurricanes: a deep learning approach
Luo et al. Multi-unmanned vehicle for region traversal search based on ant colony algorithm
KR20220071038A (en) Devices and Methods for Detection and Prediction of Cold Water Mass Using Artificial Intelligence and Satellite Imaging
Zhang et al. Density-weighted ant colony algorithm for ship trajectory reconstruction
Graser et al. Extracting patterns from large movement datasets

Legal Events

Date Code Title Description
PB01 Publication
PB01 Publication
SE01 Entry into force of request for substantive examination
SE01 Entry into force of request for substantive examination
GR01 Patent grant
GR01 Patent grant