Underwater sensor network node positioning method

Abstract

The invention relates to an underwater sensor network node positioning method, which comprises the following steps: acquiring transparency data of each node in an underwater sensor network, and constructing an initial three-dimensional space distribution model of the transparency data in combination with depth, temperature and turbidity information; carrying out iterative optimization on the initial three-dimensional space distribution model by adopting a self-adaptive Kalman filtering algorithm to obtain an optimized transparency three-dimensional distribution model; performing drift compensation on the optimized transparency three-dimensional distribution model according to a mapping model of the sensor drift distance and the working duration and a time compensation function of the sensor drift distance; based on the transparency distribution model after drift compensation, the redundancy of transparency data of each node is judged, and dimension reduction compression is carried out on redundant data; and adopting a data compression algorithm based on dictionary learning to carry out adaptive compression on the transparency data after dimension reduction compression.

Description

Technical Field

[0001] The present application relates to the field of sensor technology, and in particular to a method for positioning nodes in an underwater sensor network. Background Art

[0002] In underwater sensor networks, the acquisition and processing of transparency data face multiple challenges. First, due to water movement and light scattering, the original transparency data is uncertain. The sensor drift caused by long-term operation affects the data accuracy. The limited radio wave transmission bandwidth limits the data transmission volume, which together affect the node positioning accuracy. To solve these problems, it is necessary to establish a data processing process.

[0003] Starting from the acquisition of raw data and the construction of the initial model, high-quality compressed transparency data is finally obtained through filtering optimization, redundancy judgment, dimensionality reduction compression, scattering compensation, motion compensation and drift compensation. These data will be input into the positioning algorithm based on particle swarm optimization, combined with the reflection and re-radiation signals of radio waves to achieve precise positioning of underwater nodes. The whole process needs to balance the relationship between data quality, compression efficiency and positioning accuracy to achieve the best node positioning effect under limited bandwidth.

[0004] The technical solution of this scheme to solve the above shortcomings is: by collecting transparency, depth, temperature and turbidity data of the underwater sensor network, building and optimizing the three-dimensional spatial distribution model of transparency, using adaptive Kalman filtering to iterate the model and perform drift compensation, reducing data redundancy through dimensionality reduction compression and dictionary learning algorithm, adapting to the transmission bandwidth, and using particle swarm optimization algorithm combined with radio wave signals to achieve high-precision positioning and dynamic update of nodes, thereby improving the accuracy and efficiency of underwater monitoring and node positioning. Summary of the invention

[0005] In order to solve the above problems existing in the prior art, the present application aims to provide a method for locating nodes in an underwater sensor network.

[0006] The present application discloses a method for locating underwater sensor network nodes, comprising:

[0007] S1. Obtain transparency data of each node in the underwater sensor network, obtain transparency data by combining depth, temperature and turbidity information, and establish an initial three-dimensional spatial distribution model of the transparency data;

[0008] The depth data is used to determine the vertical position of the node;

[0009] The temperature and turbidity data are used to calculate the transparency correction coefficient through the initial three-dimensional spatial distribution model to adjust the original transparency value;

[0010] The initial three-dimensional spatial distribution model adopts a voxel representation method, wherein the accuracy of the voxel is determined by the accuracy of the node position;

[0011] S2, using an adaptive Kalman filter algorithm to iteratively optimize the initial three-dimensional spatial distribution model in step S1 to adaptively adjust filter parameters;

[0012] Among them, the observation noise covariance matrix and the process noise covariance matrix are calculated to dynamically update the Kalman gain;

[0013] By calculating the mean square error of two consecutive iteration results, if the error is less than the preset threshold, the filtering result is judged to be converged. If the filtering result does not converge, the voxel accuracy is increased by one level and the filtering is performed again to obtain the optimized transparency three-dimensional spatial distribution model;

[0014] S3, by implanting a reference light source in the filter in step S2, regularly collecting raw data of transparency, establishing a mapping model of drift and time, and recording the intensity change of the reference light source to fit the functional relationship of the drift with time;

[0015] The optimized original transparency data is input into the mapping model to obtain the transparency data after drift compensation;

[0016] S4, judging the redundancy of transparency data of each node according to the transparency distribution model after drift compensation in step S3, and defining the neighborhood of the node as all nodes whose distance to the node is less than a preset threshold;

[0017] If the similarity between the transparency data of a node and the data of its neighboring nodes is higher than a preset threshold, the node data is considered redundant;

[0018] If the similarity between the transparency data of a node and the data of its neighboring nodes is lower than the preset threshold, the node is considered to have no data redundancy;

[0019] The principal component analysis method is used for dimensionality reduction and compression;

[0020] S5, using a data compression algorithm based on dictionary learning to adaptively compress the transparency data after dimensionality reduction compression in step S4, wherein the dictionary learning process includes dictionary initialization, sparse coding and dictionary update, and the final dictionary and sparse representation are obtained through iterative optimization; the overcompleteness and sparsity of the dictionary learning are determined by the compression rate;

[0021] S6, inputting the transparency data in step S5 into the particle swarm optimization positioning algorithm, minimizing the ranging error of each node by optimizing the objective function, and calculating the spatial position coordinates of each node in combination with the reflection and re-radiation signals of the radio waves;

[0022] The particle swarm optimization algorithm searches for the optimal solution by iteratively updating the position and velocity of particles;

[0023] By measuring the arrival time of radio waves and the strength of received signals, and combining the multi-point positioning algorithm to calculate the node position, the real-time positioning results of the underwater sensor network nodes are obtained.

[0024] Preferably, the step S1 includes: the underwater sensor network collects transparency, depth, temperature and turbidity data of each node, processes the transparency data through drift compensation, and uses the depth data to determine the three-dimensional position coordinates, which are iteratively updated through a particle swarm optimization algorithm to minimize the ranging error. At the same time, the temperature and turbidity data are used to calculate the transparency correction coefficient, adjust the original transparency value, and dynamically update the node position through a multi-point positioning algorithm in combination with radio wave signal measurement. Finally, a three-dimensional transparency distribution model is obtained through spatial interpolation and voxelization processing. If the voxel accuracy is insufficient, the node position is optimized until the threshold is met. Finally, the transparency distribution is displayed through three-dimensional visualization technology.

[0025] Preferably, the step S2 includes: iteratively optimizing the initial transparency distribution model using an adaptive Kalman filter algorithm, dynamically updating the Kalman gain and filter parameters, judging convergence by the mean square error, and if the threshold is not reached, improving the voxel accuracy and iterating again until convergence or reaching the maximum number of iterations, to obtain the final transparency three-dimensional distribution model;

[0026] Based on this model, the transparency of voxels in the rendered scene is determined, and a ray casting algorithm is used for semi-transparent rendering to output high-resolution images or video sequences.

[0027] Preferably, the step S3 includes: collecting sensor data in real time and applying a drift time compensation function, correcting the original transparency data to obtain drift compensated data, and then establishing an initial distribution model, iteratively optimizing the model using an adaptive Kalman filter algorithm, dynamically updating the Kalman gain and filter parameters, judging convergence by the mean square error, and if the threshold is not reached, improving the voxel accuracy and iterating again until convergence or the maximum number of iterations is reached, to obtain a final transparency three-dimensional distribution model, using this model to determine the voxel transparency of the rendered scene, rendering using a ray casting algorithm, and merging with the original scene image to obtain a semi-transparent visualization result.

[0028] Preferably, the step S4 includes: by determining the node neighborhood and calculating the similarity of the transparency data, identifying and reducing the dimension of the redundant data for compression rate to meet the transmission bandwidth requirement, and then encoding and sending the node transparency data, after the receiving end decodes and recovers the data, reconstructing the transparency distribution model, supporting subsequent applications including three-dimensional visualization, using the ray casting algorithm and the optimized model to perform scene rendering, and obtaining a semi-transparent effect diagram.

[0029] Preferably, the step S5 includes: setting a target compression rate and determining dictionary parameters, initializing the dictionary using the K-SVD algorithm, then performing sparse coding of the orthogonal matching pursuit algorithm and dictionary update of the least squares method, iterating until the dictionary converges, the dictionary is used for dimension reduction processing of transparency data, sparse representation coefficients are obtained, and Huffman coding is used for compression, data transmission packets are generated and sent via radio waves, the receiving end decodes and recovers the data, and the transparency distribution model is reconstructed.

[0030] Preferably, the step S6 includes: the underwater sensor network nodes collect original transparency data, obtain correction data through drift compensation processing, input the data into the particle swarm optimization algorithm for iterative update to minimize the ranging error, combine the radio wave signal measurement, calculate the node position through the multi-point positioning algorithm until the accuracy threshold is met, update the node position coordinates in real time, form a dynamic positioning mechanism, and display it through three-dimensional visualization technology, combine the depth, temperature and turbidity information, establish a three-dimensional spatial distribution model of transparency data, adjust the original transparency value, and use the voxel accuracy to meet the node position accuracy requirements.

[0031] The advantage of the underwater sensor network node positioning method described in the present application is that, first, a three-dimensional spatial distribution model of transparency data is established, and correction is performed in combination with depth, temperature and turbidity information. In view of data uncertainty, an adaptive Kalman filter algorithm is used for optimization. In order to solve the sensor drift problem, a drift compensation model is established by implanting a reference light source. According to the transparency distribution characteristics, the data redundancy is judged and dimensionality reduction compression is performed. An algorithm based on dictionary learning is used to adaptively compress the data to adapt to the transmission bandwidth limitation. Finally, a particle swarm optimization positioning algorithm is used in combination with radio wave signals to achieve real-time and accurate positioning of sensor nodes.

[0032] The present invention effectively improves the data processing accuracy and transmission efficiency of the underwater sensor network, and provides reliable support for the analysis of the spatial distribution characteristics of underwater targets. BRIEF DESCRIPTION OF THE DRAWINGS

[0033] Figure 1 This is the process of a method for locating underwater sensor network nodes described in this application Figure 1 ;

[0034] Figure 2 This is the process of a method for locating underwater sensor network nodes described in this application Figure 2 . DETAILED DESCRIPTION

[0035] like Figure 1-Figure 2 As shown, a method for locating an underwater sensor network node described in the present application includes:

[0036] S1. Obtain transparency data of each node in the underwater sensor network, obtain transparency data by combining depth, temperature and turbidity information, and establish an initial three-dimensional spatial distribution model of the transparency data;

[0037] The depth data is used to determine the vertical position of the node;

[0038] The temperature and turbidity data are calculated using the initial three-dimensional spatial distribution model to calculate the transparency correction coefficient, and the original transparency value is adjusted. The initial three-dimensional spatial distribution model is represented by voxelization, where the accuracy of the voxel is determined by the accuracy of the node position;

[0039] S2, using an adaptive Kalman filter algorithm to iteratively optimize the initial three-dimensional spatial distribution model in step S1, and adaptively adjust the filter parameters;

[0040] The Kalman gain is dynamically updated by calculating the observation noise covariance matrix and the process noise covariance matrix;

[0041] By calculating the mean square error of two consecutive iteration results, when the error is less than the preset threshold, the filtering result is judged to be converged. If the filtering result does not converge, the voxel accuracy is increased by one level and the filtering is performed again to obtain the optimized transparency three-dimensional spatial distribution model;

[0042] S3, by implanting a reference light source in the filter in step S2, regularly collecting reference data, establishing a mapping model between drift and time, and fitting a functional relationship between drift and time by recording the intensity change of the reference light source;

[0043] The optimized transparency data is input into the mapping model to obtain the transparency data after drift compensation;

[0044] S4, judging the redundancy of transparency data of each node according to the transparency distribution model after drift compensation in step S3, and defining the neighborhood of the node as all nodes whose distance to the node is less than a preset threshold;

[0045] If the similarity between the transparency data of a node and the data of its neighboring nodes is higher than a preset threshold, the node data is considered redundant;

[0046] If the similarity between the transparency data of a node and the data of its neighboring nodes is lower than the preset threshold, the node is considered to have no data redundancy;

[0047] The principal component analysis method is used for dimensionality reduction and compression;

[0048] S5, using a data compression algorithm based on dictionary learning to adaptively compress the transparency data after dimensionality reduction compression in step S4, wherein the dictionary learning process includes three steps: dictionary initialization, sparse coding and dictionary update, and the final dictionary and sparse representation are obtained through iterative optimization; the overcompleteness and sparsity of the dictionary learning are used to obtain the compressed transparency data from the compression ratio;

[0049] S6, inputting the transparency data in step S5 into the particle swarm optimization positioning algorithm, minimizing the ranging error of each node by optimizing the objective function, and calculating the spatial position coordinates of each node in combination with the reflection and re-radiation signals of the radio waves;

[0050] The particle swarm optimization algorithm searches for the optimal solution by iteratively updating the position and velocity of particles;

[0051] By measuring the arrival time of radio waves and the strength of received signals, and combining the multi-point positioning algorithm to calculate the node position, the real-time positioning results of the underwater sensor network nodes are obtained.

[0052] like Figure 1-Figure 2 As shown, in step S1, the transparency data of each node in the underwater sensor network is obtained, the transparency data is obtained by combining the information of depth, temperature and turbidity, and an initial three-dimensional spatial distribution model of the transparency data is established;

[0053] The depth data is used to determine the vertical position of the node;

[0054] The temperature data and turbidity data are used to calculate the transparency correction coefficient through the initial three-dimensional spatial distribution model to adjust the original transparency value;

[0055] The initial three-dimensional spatial distribution model is represented by voxelization, wherein the accuracy of the voxel is determined by the accuracy of the node position.

[0056] Further, in step S1, the original transparency, depth, temperature and turbidity data of each node in the underwater sensor network are obtained, and drift compensation processing is performed on the transparency data to obtain corrected transparency data;

[0057] For each node, the vertical position coordinates of the node in three-dimensional space are determined according to the depth data, and the node position coordinates are input into the particle swarm optimization algorithm for iterative update to minimize the ranging error of each node and obtain the optimized node spatial position;

[0058] For each node, the temperature data and turbidity data are input into the pre-established transparency correction coefficient regression model to obtain the transparency correction coefficient of the node, which is used to adjust the original transparency value;

[0059] For each node, the corrected transparency value is multiplied by the transparency correction coefficient to obtain the corrected node transparency value as the final measurement result of the node transparency;

[0060] Obtain the reflected signal and re-radiated signal of the radio wave, measure the arrival time of the radio wave and the received signal strength, fuse the measurement results with the output results of the particle swarm optimization algorithm, calculate the node position through the multi-point positioning algorithm, and dynamically update the spatial position coordinates of the node;

[0061] According to the optimized three-dimensional spatial coordinates of each node and the corrected transparency value, a spatial interpolation method is used to perform spatial interpolation of transparency data in three-dimensional space to obtain continuous three-dimensional transparency distribution data;

[0062] According to the preset voxel accuracy, the continuous three-dimensional transparency distribution data is voxelized to obtain a three-dimensional voxel representation model of the transparency data. If the voxel accuracy does not meet the preset threshold, the node position accuracy is improved and the execution is returned to perform node position optimization.

[0063] If the voxel accuracy meets the preset threshold, the three-dimensional voxel representation model of the transparency data is output to form a three-dimensional spatial distribution model of the transparency of the underwater sensor network nodes and obtain the real-time distribution results of the transparency;

[0064] Using three-dimensional visualization technology, the three-dimensional transparency distribution model of underwater sensor network nodes is visualized, which intuitively displays the spatial distribution characteristics of underwater transparency and provides data support for subsequent underwater target analysis.

[0065] Specifically, in step S1, the original transparency, depth, temperature and turbidity data of each node in the underwater sensor network are first obtained, the transparency data is drift compensated, and the drift curve is fitted using the least squares method to obtain the corrected transparency data;

[0066] Then, for each node, the vertical position coordinates of the node in three-dimensional space are determined according to the depth data, and the node position coordinates are input into the particle swarm optimization algorithm. The number of particles is set to 50, the number of iterations is set to 100, and the search space range is set to 100m×100m×100m. The ranging error of each node is minimized to obtain the optimized node spatial position;

[0067] Then, the temperature data and turbidity data are input into the pre-established multiple linear regression model to obtain the transparency correction coefficient of the node and the determination coefficient R of the regression model. 2 is 0.85, which is used to adjust the original transparency value;

[0068] Then, the corrected transparency value is multiplied by the transparency correction coefficient to obtain the corrected node transparency value as the final measurement result of the node transparency;

[0069] At the same time, the reflected and re-radiated radio wave signals with a frequency of 2.4 GHz are obtained, the arrival time of the radio waves and the received signal strength are measured, the output results of the Kalman filter algorithm and the particle swarm optimization algorithm are fused, the three-dimensional positioning of the node is realized through the Chan algorithm, and the spatial position coordinates of the node are dynamically updated;

[0070] According to the optimized three-dimensional spatial coordinates of each node and the corrected transparency value, the Kriging interpolation method is used to interpolate the transparency data in the three-dimensional space of 100m×100m×100m, with an interpolation accuracy of 1m, to obtain continuous three-dimensional transparency distribution data;

[0071] Finally, according to the preset voxel accuracy of 0.5m, the continuous three-dimensional transparency distribution data is voxelized, and the octree algorithm is used for voxel division to obtain a three-dimensional voxel representation model of the transparency data;

[0072] If the voxel accuracy does not meet the preset threshold, the node position accuracy is increased to 0.1m, and the node position optimization step is returned to execute;

[0073] If the voxel accuracy meets the preset threshold, the three-dimensional voxel representation model of the transparency data is output to form a three-dimensional spatial distribution model of the transparency of the underwater sensor network nodes and obtain the real-time distribution results of the transparency;

[0074] Finally, the Unity 3D engine is used to visualize the transparency three-dimensional distribution model of the underwater sensor network nodes. Appropriate lighting and materials are set to intuitively display the spatial distribution characteristics of underwater transparency, providing data support for subsequent underwater target analysis.

[0075] like Figure 1-Figure 2 As shown, in step S2, in view of the uncertainty of transparency data, an adaptive Kalman filter algorithm is used to iteratively optimize the initial three-dimensional spatial distribution model in step S1 to adaptively adjust the filter parameters;

[0076] Among them, the observation noise covariance matrix and the process noise covariance matrix are calculated to dynamically update the Kalman gain;

[0077] By calculating the mean square error of the results of two consecutive iterations, if the error is less than the preset threshold, the filtering result is judged to be converged. If the filtering result does not converge, the voxel accuracy is improved by one level and the filtering is performed again to obtain the optimized transparency three-dimensional spatial distribution model.

[0078] Further, in step S2, in step 1, an initial distribution model of transparency data is obtained, and an adaptive Kalman filter algorithm is used to iteratively optimize the initial distribution model in view of the uncertainty of the transparency data;

[0079] Step 2: In each iterative optimization process, the Kalman gain is dynamically updated by calculating the observation noise covariance matrix and the process noise covariance matrix, and the filtering parameters of the Kalman filter are adaptively adjusted;

[0080] Step 3: Calculate the mean square error between the current iteration result and the previous iteration result, and determine whether the mean square error is less than a preset convergence threshold. If so, it is considered that the filtering result has converged, and the current iteration result is used as the final transparency three-dimensional distribution model.

[0081] Step 4: if the mean square error is greater than or equal to a preset convergence threshold, the voxel accuracy is increased by one level, and the initial distribution model of the transparency data is obtained under the increased voxel accuracy, and the process returns to step 2 to perform adaptive Kalman filter iterative optimization processing again;

[0082] Repeat steps 2 to 4 until the filtering result converges or reaches the preset maximum number of iterations, and use the result of the last iterative optimization as the final transparency three-dimensional distribution model;

[0083] According to the finally obtained transparency three-dimensional distribution model, the transparency value of each voxel in the scene to be rendered is determined for subsequent volume rendering processing;

[0084] The ray casting algorithm is used in combination with the optimized transparency 3D distribution model to render the scene to be rendered, and a realistic semi-transparent rendering effect is obtained;

[0085] The rendered semi-transparent effect image is post-processed to further enhance the realism and visual impact of the rendering effect by adjusting the brightness, contrast, and color balance parameters;

[0086] The post-processed semi-transparent rendering effect image is output as a high-resolution image or video sequence for subsequent display, analysis and application.

[0087] Specifically, in step S2, firstly, an initial distribution model of transparency data is obtained, and in view of the uncertainty of transparency data, an adaptive Kalman filter algorithm is used to iteratively optimize the initial distribution model;

[0088] In each iterative optimization process, the Kalman gain is dynamically updated by calculating the observation noise covariance matrix and the process noise covariance matrix, including setting the observation noise covariance matrix to a diagonal matrix with diagonal elements of 0.01 and the process noise covariance matrix to the unit matrix multiplied by 0.001, and adaptively adjusting the filtering parameters of the Kalman filter;

[0089] Then, the mean square error between the current iteration result and the previous iteration result is calculated to determine whether the mean square error is less than the preset convergence threshold 0.0001. If so, the filtering result is considered to have converged, and the current iteration result is used as the final transparency three-dimensional distribution model;

[0090] If the mean square error is greater than or equal to 0.0001, the voxel accuracy is increased from 1m to 0.5m. Under the improved voxel accuracy, the initial distribution model of the transparency data is re-obtained, and the adaptive Kalman filter iterative optimization process is continued. The iterative process is repeated until the filtering result converges or the preset maximum number of iterations of 50 times is reached;

[0091] According to the finally obtained transparency three-dimensional distribution model, the transparency value of each voxel in the scene to be rendered is determined for subsequent volume rendering processing;

[0092] The ray casting algorithm is used in combination with the optimized transparency 3D distribution model to render the scene to be rendered. The ray step size is set to 0.01 and the number of ray sampling points is set to 500 to obtain a realistic semi-transparent rendering effect.

[0093] The rendered semi-transparent effect image was post-processed by adjusting the brightness by +10%, the contrast by +20%, and the color balance RGB components by +5% to further enhance the realism and visual impact of the rendering effect.

[0094] Finally, the post-processed semi-transparent rendering effect image is output as a 4K resolution PNG image sequence for subsequent display, analysis and application.

[0095] like Figure 1-Figure 2 As shown, in step S3, in order to solve the drift problem caused by the long-term operation of the sensor, a reference light source is implanted in the filter in step S2, and the original data of transparency is collected regularly to establish a mapping model of drift and time. By recording the intensity change of the reference light source, the functional relationship of the drift with time is fitted;

[0096] The optimized original transparency data is input into the mapping model to obtain the transparency data after drift compensation.

[0097] Further, in step S3, in step 1, the original transparency data collected by the sensor in real time is obtained, and according to the current cumulative working time of the sensor, the corresponding drift compensation value is calculated by the drift time compensation function, and the original transparency data is subjected to drift compensation correction to obtain the transparency data after drift compensation;

[0098] Step 2: Obtain an initial distribution model of the transparency data based on the transparency data after drift compensation, and use an adaptive Kalman filter algorithm to iteratively optimize the initial distribution model in view of the uncertainty of the transparency data;

[0099] Step 3: In each iterative optimization process, the Kalman gain is dynamically updated by calculating the observation noise covariance matrix and the process noise covariance matrix, and the filtering parameters of the Kalman filter are adaptively adjusted to obtain the current iterative result;

[0100] Step 4: Calculate the mean square error between the current iteration result and the previous iteration result, and determine whether the mean square error is less than a preset convergence threshold. If so, it is considered that the filtering result has converged, and the current iteration result is used as the final transparency three-dimensional distribution model;

[0101] Step 5: If the mean square error is greater than or equal to a preset convergence threshold, the voxel accuracy is increased by one level, and the initial distribution model of the transparency data is obtained under the increased voxel accuracy, and the process returns to step 2 to perform adaptive Kalman filter iterative optimization again;

[0102] Repeat steps 2 to 5 until the filtering result converges or reaches the preset maximum number of iterations, and use the result of the last iterative optimization as the final transparency three-dimensional distribution model;

[0103] According to the finally obtained transparency three-dimensional distribution model, the transparency value of each voxel in the scene to be rendered is determined for subsequent volume rendering processing;

[0104] The ray casting algorithm is used in combination with the optimized transparency 3D distribution model to render the scene to be rendered, and a realistic semi-transparent rendering effect is obtained;

[0105] The rendered semi-transparent effect image is fused with the original scene image, and the final semi-transparent visualization result is generated through image synthesis technology, which is provided to users for visualization analysis and decision support.

[0106] Specifically, in step S3, the sensor collects raw transparency data in real time, including collecting data 10 times per second. According to the fact that the sensor has been working for 720 hours, the drift compensation value is calculated to be 7.2% by fitting function y=0.01x (where x is the working time in hours; y is the drift amount in transparency percentage). The raw data is multiplied by the compensation coefficient 0.928 to obtain the compensated transparency data.

[0107] The initial distribution model is constructed based on the compensated data, and the adaptive Kalman filter is used for iterative optimization. Each iteration calculates the observation noise covariance matrix Q and the process noise covariance matrix R, and dynamically adjusts the Kalman gain using the Q / R ratio to obtain the optimization result.

[0108] Calculate the mean square error between the current result and the previous result. If it is less than the threshold value of 0.01, it is considered to be converged and the three-dimensional distribution model is output; otherwise, the voxel accuracy is increased by one level, including from 1 cm to 0.5 cm, the initial model is re-obtained and it is returned to continue iterating until convergence or the maximum number of iterations is 100;

[0109] The transparency value of each voxel in the final 1024*1024*1024 3D model is used for subsequent rendering;

[0110] Using the ray casting algorithm, setting the ray step size to 0.01, accumulating the transparency and calculating the color of each ray, rendering a semi-transparent effect image with a resolution of 1920*1080;

[0111] The rendering is synthesized with the original image through Alpha blending, and the blending factor is set to the transparency of the rendering. That is, the part with transparency 0 fully displays the original image, and the part with transparency 1 fully displays the rendering result, to generate the final visual image.

[0112] like Figure 1-Figure 2 As shown, in step S4, according to the transparency distribution model after drift compensation in step S3, the redundancy of transparency data of each node is determined, and the neighborhood of the node is defined as all nodes whose distance to the node is less than a preset threshold;

[0113] If the similarity between the transparency data of a node and the data of its neighboring nodes is higher than a preset threshold, the node data is considered redundant;

[0114] If the similarity between the transparency data of a node and the data of its neighboring nodes is lower than the preset threshold, the node is considered to have no data redundancy;

[0115] The transparency data is compressed by reducing the dimension using a principal component analysis method, and the compression rate is determined by the transmission bandwidth limitation of the radio waves.

[0116] Further, in step S4, a transparency distribution model after drift compensation is obtained, and for each node, its neighborhood range is determined, where the neighborhood is defined as all nodes whose distance from the node is less than a preset distance threshold;

[0117] Calculate the similarity between the transparency data of each node and the transparency data of all nodes in its neighborhood. If the similarity is higher than the preset similarity threshold, it is determined that the node data is redundant.

[0118] For nodes with redundant data, the principal component analysis method is used to reduce the dimension and compress the data. By decomposing the transparency data, the main characteristic components are extracted to achieve data compression.

[0119] According to the limitation of radio wave transmission bandwidth, a compression rate threshold is determined. When the compression rate of node data is higher than the threshold, it is considered that the compressed data meets the transmission bandwidth requirement.

[0120] Encode the compressed node transparency data to generate data transmission packets, which are sent via radio waves to achieve efficient transmission of transparency data between nodes;

[0121] At the receiving end, after obtaining the transmission packet, decoding is performed to obtain the compressed node transparency data, and then the same principal component analysis method as the sending end is used to reduce the dimension of the data and restore the original transparency data;

[0122] Based on the restored node transparency data, the transparency distribution model is reconstructed for subsequent business applications, including 3D visualization and line-of-sight analysis, providing efficient and accurate transparency information support;

[0123] According to the reconstructed transparency distribution model, the transparency value of each voxel in the scene to be rendered is determined for subsequent volume rendering processing;

[0124] The ray casting algorithm is used in combination with the optimized transparency three-dimensional distribution model to render the scene to be rendered and obtain a realistic semi-transparent rendering effect.

[0125] Specifically, in step S4, when obtaining the transparency distribution model after drift compensation, the distance threshold of the node neighborhood range is set to 10 meters, that is, all nodes with a distance less than 10 meters from the target node are defined as the neighborhood of the node;

[0126] Then, the cosine similarity algorithm is used to calculate the similarity of transparency data between each node and the nodes in its neighborhood. If the similarity is higher than 0.8, the node data is considered redundant;

[0127] For nodes with redundant data, principal component analysis is used for dimensionality reduction and compression. By eigenvalue decomposition, the first three principal components are selected to achieve a data compression rate of 70%.

[0128] According to the radio wave transmission bandwidth limit of 10Mbps, the compression rate threshold is determined to be 75%, and the compressed data meets the transmission requirements;

[0129] The compressed node transparency data is Huffman-encoded to generate data transmission packets, which are sent via radio waves;

[0130] After receiving the transmission packet, the receiving end performs Huffman decoding to obtain the compressed transparency data, and then uses the same principal component analysis method as the sending end to reconstruct the feature vector to restore the original transparency data;

[0131] Based on the restored node transparency data, the transparency distribution model is reconstructed using the trilinear interpolation algorithm and applied to subsequent 3D visualization and line-of-sight analysis services.

[0132] According to the reconstructed transparency distribution model, the scene is divided into voxels of 0.5 m × 0.5 m × 0.5 m using the voxelization method, and the transparency value of each voxel is determined;

[0133] Finally, the ray casting algorithm is used, the ray step size is set to 0.1 meters, and the scene is rendered to obtain a realistic translucent effect.

[0134] like Figure 1-Figure 2 As shown, in step S5, a data compression algorithm based on dictionary learning is used to adaptively compress the transparency data after dimensionality reduction compression in step S4. The dictionary learning process includes dictionary initialization, sparse coding and dictionary update, and the final dictionary and sparse representation are obtained through iterative optimization; the overcompleteness and sparsity of the dictionary learning are determined by the compression rate, and the compressed transparency data is obtained.

[0135] Further, in step S5, the transparency data to be compressed is obtained, a target compression ratio is determined according to the transmission bandwidth limitation of the radio wave, and overcompleteness and sparsity parameters of the dictionary are determined according to the target compression ratio;

[0136] The K-SVD algorithm is used to initialize the dictionary to obtain an initial dictionary, and the initial dictionary is used as the input of the dictionary learning algorithm;

[0137] For the initial dictionary, the orthogonal matching pursuit algorithm is used for sparse coding to obtain a sparse coefficient matrix, which is used as the input for dictionary update;

[0138] The dictionary is updated using the least squares method to obtain an updated dictionary, and whether the dictionary converges is determined. If not, the updated dictionary is used as the input for a new round of iterative optimization, and the sparse coding step is returned to continue iterative optimization.

[0139] If the dictionary converges, the updated dictionary is used as the final dictionary, and the transparency data to be compressed is subjected to dimensionality reduction processing to obtain the transparency data after dimensionality reduction;

[0140] According to the final dictionary and sparse coefficient matrix, the sparse representation of the transparency data after dimension reduction is performed to obtain the sparse representation coefficient;

[0141] According to the sparse representation coefficient, an entropy coding algorithm such as Huffman coding is used to further compress the sparse representation coefficient to obtain the final compressed transparency data;

[0142] Encode the compressed node transparency data to generate data transmission packets, which are sent via radio waves to achieve efficient transmission of transparency data between nodes;

[0143] At the receiving end, after obtaining the transmission packet, decoding is performed to obtain the compressed node transparency data. Then, the same principal component analysis method as the sending end is used to reduce the dimension of the data, restore the original transparency data, and reconstruct the transparency distribution model for subsequent business applications.

[0144] Specifically, in step S5, the transparency data to be compressed is obtained, and according to the transmission bandwidth limit of radio waves being 10 Mbps, the target compression rate is determined to be 80%, and according to the target compression rate, the overcompleteness of the dictionary is determined to be 1.5 times, and the sparsity parameter is determined to be 0.1;

[0145] The K-SVD algorithm is used to initialize the dictionary. By performing SVD decomposition on the transparency data, the left singular vectors corresponding to the first 100 singular values ​​are selected as the atoms of the initial dictionary, and an initial dictionary of size 100×100 is obtained.

[0146] For the initial dictionary, the orthogonal matching pursuit algorithm is used for sparse coding, the sparse representation error threshold is set to 0.01, the number of iterations is 50, and a 100×1000 sparse coefficient matrix is ​​obtained;

[0147] The dictionary is updated using the least squares method. By solving the optimization problem of minimizing the reconstruction error, each atom of the dictionary is updated to obtain the updated dictionary. It is then determined whether the change in the dictionary atom is less than 0.001. If so, the dictionary is considered to have converged.

[0148] The transparency data to be compressed is subjected to dimensionality reduction processing, and the principal component analysis method is used to select the first 10 principal components with a cumulative contribution rate of 99% to obtain the transparency data after dimensionality reduction;

[0149] According to the final dictionary and sparse coefficient matrix, the transparency data after dimension reduction is sparsely represented, and the sparse representation coefficients are obtained by solving the L1 norm minimization problem;

[0150] According to the sparse representation coefficients, the Huffman coding algorithm is used to construct an optimal binary tree, encode the coefficients, and obtain compressed transparency data with an average coding length of 5 bits;

[0151] The compressed node transparency data is encoded using Manchester encoding to generate a data transmission packet, which is sent via radio waves with a frequency of 2.4 GHz;

[0152] At the receiving end, Manchester decoding is used to obtain the transmission packet, Huffman decoding is performed to obtain sparse representation coefficients, and then the sparse coefficients are sparsely reconstructed using a dictionary to obtain transparency data after dimensionality reduction;

[0153] Finally, the principal component analysis method is used to recover the data, restore the original transparency data, and reconstruct the transparency distribution model, which is applied to three-dimensional visualization and line of sight analysis business scenarios.

[0154] like Figure 1-Figure 2 As shown, in step S6, the transparency data in step S5 is input into the particle swarm optimization positioning algorithm, and the ranging error of each node is minimized by optimizing the objective function, and the spatial position coordinates of each node are calculated by combining the reflection and re-radiation signals of the radio waves;

[0155] The particle swarm optimization algorithm searches for the optimal solution by iteratively updating the position and velocity of particles;

[0156] By measuring the arrival time of radio waves and the strength of received signals, and combining the multi-point positioning algorithm to calculate the node position, the real-time positioning results of the underwater sensor network nodes are obtained. The positioning accuracy is determined by the accuracy requirements of the node position. The positioning results are presented in a three-dimensional visualization method to analyze the spatial distribution characteristics of underwater targets.

[0157] Further, in step S6, original transparency data of the underwater sensor network node is obtained, and drift compensation processing is performed on the data to obtain corrected transparency data;

[0158] The corrected transparency data is input into the particle swarm optimization algorithm, and the position and velocity of the particles are iteratively updated to search for the optimal solution and minimize the ranging error of each node;

[0159] Obtain the reflected signal and re-radiated signal of the radio wave, combine the output results of the particle swarm optimization algorithm, and comprehensively calculate the spatial position coordinates of each node;

[0160] Measure the arrival time of radio waves and the received signal strength, fuse the measurement results with the output of the particle swarm optimization algorithm, and calculate the node position through the multi-point positioning algorithm;

[0161] According to the accuracy requirements of the node position, determine the accuracy threshold of the positioning result. If the positioning accuracy does not reach the threshold, return to continue iterative optimization until the accuracy requirements are met;

[0162] Dynamically update the spatial position coordinates of each node to form a real-time update mechanism for the node position of the underwater sensor network and obtain the real-time positioning results of the node;

[0163] Using 3D visualization technology, the real-time positioning results of underwater sensor network nodes are visualized, and the spatial distribution characteristics of underwater targets are intuitively displayed to provide data support for subsequent analysis;

[0164] Obtain transparency data from each node in the underwater sensor network, and build an initial three-dimensional spatial distribution model of transparency data by combining depth, temperature, and turbidity information;

[0165] The depth data is used to determine the vertical position of the node. The temperature and turbidity data are used to calculate the transparency correction coefficient through a pre-established regression model to adjust the original transparency value. The accuracy of the voxel is determined by the accuracy requirement of the node position.

[0166] Specifically, in step S6, the original transparency data of the underwater sensor network nodes are processed by drift compensation to obtain corrected transparency data, including the original transparency of node A is 0.8, and after correction by the drift compensation coefficient of 1.2, the corrected transparency is 0.96;

[0167] The corrected transparency data is input into the particle swarm optimization algorithm, the number of particles is set to 100, the number of iterations is set to 50, the position and velocity of the particles are updated iteratively, and the optimal solution is searched to minimize the ranging error of each node, including reducing the ranging error between node A and node B from 2.5m to 0.8m;

[0168] Obtain the reflected signal and re-radiated signal of the radio wave, and combine the output results of the particle swarm optimization algorithm to comprehensively calculate the spatial position coordinates of each node, including the position coordinates of node A (10.2, 5.8, -20.3);

[0169] Measure the arrival time of radio waves and the received signal strength, perform Kalman filtering on the measurement results and the output results of the particle swarm optimization algorithm, and calculate the node position through the Chan algorithm multi-point positioning algorithm. The fused positioning result of node A is (10.1, 5.9, -20.5);

[0170] According to the accuracy requirements of the node position, the accuracy threshold of the positioning result is determined to be 1m. If the positioning accuracy does not reach the threshold, the particle swarm optimization algorithm step is returned to continue iterative optimization until the accuracy requirements are met;

[0171] The spatial position coordinates of each node are dynamically updated to form a real-time update mechanism for the node position of the underwater sensor network. The node position is updated every 10 seconds to obtain the real-time positioning result of the node.

[0172] The Unity 3D engine is used to visualize the real-time positioning results of underwater sensor network nodes in three dimensions, intuitively displaying the spatial distribution characteristics of underwater targets, including displaying the location and movement trajectory of nodes in a three-dimensional scene;

[0173] Obtain transparency data of each node in the underwater sensor network, and combine depth, temperature and turbidity information to establish an initial three-dimensional spatial distribution model of transparency data, including node A with a transparency of 0.96, a depth of 50m, a temperature of 15°C, and a turbidity of 10NTU;

[0174] The depth data were used to determine the vertical position of the node, and the temperature and turbidity data were used to calculate the transparency correction coefficient through a pre-established multivariate linear regression model, including correction factors of 0.02 / °C and -0.01 / NTU. The original transparency value was adjusted, and the voxel accuracy was set to 1m to generate a three-dimensional spatial distribution model of transparency.

[0175] In the description of the present application, it should be understood that the directions or positional relationships indicated by directional words such as "front, back, up, down, left, right", "lateral, vertical, horizontal" and "top, bottom" are usually based on the directions or positional relationships shown in the drawings, and are only for the convenience of describing the present application and simplifying the description. Unless otherwise specified, these directional words do not indicate or imply that the device or element referred to must have a specific direction or be constructed and operated in a specific direction, and therefore cannot be understood as limiting the scope of protection of the present application.

[0176] For those skilled in the art, various other corresponding changes and deformations can be made according to the technical solutions and concepts described above, and all of these changes and deformations should fall within the protection scope of the claims of this application.

Claims

1. A method for locating underwater sensor network nodes, characterized in that: include: S1. Obtain transparency data of each node in the underwater sensor network, combine the information of depth, temperature and turbidity to obtain transparency data, and establish an initial three-dimensional spatial distribution model of the transparency data; The depth data is used to determine the vertical position of the node; The temperature data and turbidity data are used to calculate the transparency correction coefficient through the initial three-dimensional spatial distribution model to adjust the original transparency value; The initial three-dimensional spatial distribution model adopts a voxel representation method, wherein the accuracy of the voxel is determined by the accuracy of the node position; S2, using an adaptive Kalman filter algorithm to iteratively optimize the initial three-dimensional spatial distribution model in step S1 to adaptively adjust filter parameters; Among them, the observation noise covariance matrix and the process noise covariance matrix are calculated to dynamically update the Kalman gain; By calculating the mean square error of the results of two consecutive iterations, if the error is less than the preset threshold, the filtering result is judged to be converged. If the filtering result does not converge, the voxel accuracy is improved by one level and the filtering is performed again to obtain the optimized transparency three-dimensional spatial distribution model. S3, by implanting a reference light source in the filter in step S2, regularly collecting raw data of transparency, establishing a mapping model of drift and time, and recording the intensity change of the reference light source to fit the functional relationship of the drift with time; The optimized original transparency data is input into the mapping model to obtain the transparency data after drift compensation; S4, judging the redundancy of transparency data of each node according to the transparency distribution model after drift compensation in step S3, and defining the neighborhood of the node as all nodes whose distance to the node is less than a preset threshold; If the similarity between the transparency data of a node and the data of its neighboring nodes is higher than a preset threshold, the node data is considered redundant; If the similarity between the transparency data of a node and the data of its neighboring nodes is lower than the preset threshold, the node is considered to have no data redundancy; Using principal component analysis method to reduce the dimension of the transparency data; S5, using a data compression algorithm based on dictionary learning to adaptively compress the transparency data after dimensionality reduction compression in step S4, wherein the dictionary learning process includes dictionary initialization, sparse coding and dictionary update, and the final dictionary and sparse representation are obtained through iterative optimization; the overcompleteness and sparsity of the dictionary learning are determined by the compression rate; S6, inputting the transparency data in step S5 into the particle swarm optimization positioning algorithm, minimizing the ranging error of each node by optimizing the objective function, and calculating the spatial position coordinates of each node in combination with the reflection and re-radiation signals of the radio waves; The particle swarm optimization algorithm searches for the optimal solution by iteratively updating the position and velocity of particles; By measuring the arrival time of radio waves and the strength of received signals, and combining the multi-point positioning algorithm to calculate the node position, the real-time positioning results of the underwater sensor network nodes are obtained.

2. According to claim 1, a method for locating underwater sensor network nodes, characterized in that: In step S1, it includes: obtaining the original transparency, depth, temperature and turbidity data of each node in the underwater sensor network, performing drift compensation processing, fitting the drift curve using the least square method, and obtaining the corrected transparency data; Determine the vertical position coordinates of the node in three-dimensional space according to the depth data, input the node position coordinates into the particle swarm optimization algorithm, and obtain the optimized node spatial position; Input the temperature data and turbidity data into a pre-established multivariate linear regression model to obtain the transparency correction coefficient of the node, then multiply the corrected transparency value by the transparency correction coefficient to obtain the corrected node transparency value, and simultaneously obtain the radio wave reflection signal and re-radiation signal with a frequency of 2.4 GHz, measure the arrival time of the radio wave and the received signal strength, and use them to dynamically update the spatial position coordinates of the node; According to the optimized three-dimensional spatial coordinates of each node and the corrected transparency value, the Kriging interpolation method is used to obtain continuous three-dimensional transparency distribution data.

3. The method for locating underwater sensor network nodes according to claim 1, characterized in that: In step S2, it includes: obtaining an initial distribution model of transparency data, iteratively optimizing the initial distribution model, calculating the mean square error between the current iteration result and the previous iteration result, outputting a final three-dimensional transparency distribution model, and determining the transparency value of each voxel in the scene to be rendered according to the final three-dimensional transparency distribution model.

4. The method for locating underwater sensor network nodes according to claim 1, characterized in that: In step S3, it includes: acquiring the original transparency data collected by the sensor in real time, calculating the corresponding drift compensation value through the drift time compensation function, performing drift compensation correction on the original transparency data to obtain transparency data after drift compensation, and acquiring the initial distribution model of transparency data based on the transparency data after drift compensation.

5. The method for locating underwater sensor network nodes according to claim 1, characterized in that: In step S4, it includes: calculating the similarity between the transparency data of each node and the transparency data of all nodes in its neighborhood, so as to determine data redundancy; If the similarity is higher than the preset similarity threshold, it is determined that the node data is redundant; If the similarity is not higher than the preset similarity threshold, it is determined that the node data does not have redundancy; For nodes with redundant data, the principal component analysis method is used to perform dimensionality reduction and compression processing, and the main characteristic components are extracted by feature decomposing the transparency data.

6. The method for locating underwater sensor network nodes according to claim 1, characterized in that: In step S5, it includes: obtaining transparency data to be compressed, and determining overcompleteness and sparsity parameters of the dictionary according to the target compression ratio; Use K-SVD algorithm to initialize the dictionary and obtain the initial dictionary; The dictionary is updated using the least square method to obtain an updated dictionary; According to the final dictionary and sparse coefficient matrix, the transparency data after dimension reduction is sparsely represented to obtain the sparse representation coefficients.

7. The underwater sensor network node positioning method according to claim 2, characterized in that: In step S6, it includes: obtaining original transparency data of the underwater sensor network node, performing drift compensation processing on the data to obtain corrected transparency data, and inputting the corrected transparency data into the particle swarm optimization algorithm; By iteratively updating the position and velocity of particles, searching for the optimal solution, minimizing the ranging error of each node, and obtaining the reflected signal and re-radiated signal of the radio wave, combined with the output results of the particle swarm optimization algorithm, to calculate the spatial position coordinates of each node; Measuring the arrival time of radio waves and the received signal strength, and fusing the measurement results with the output results of the particle swarm optimization algorithm to calculate the node position through the multi-point positioning algorithm; The accuracy threshold of the positioning result is determined according to the accuracy of the node position.

8. The method for locating underwater sensor network nodes according to claim 7, characterized in that: If the accuracy threshold of the positioning result does not reach the threshold value of 1m, return to the following steps: The corrected transparency data is input into the particle swarm optimization algorithm, and the particle position and velocity are iteratively updated to search for the optimal solution and minimize the ranging error of each node to continue iterative optimization; If the accuracy threshold of the positioning result reaches the threshold of 1m, no return will be made.

9. The underwater sensor network node positioning method according to claim 7, characterized in that: The method acquires transparency data of each node in the underwater sensor network, combines the depth, temperature and turbidity information, and establishes an initial three-dimensional spatial distribution model of the transparency data; The depth data is used to determine the vertical position of the node; The temperature information and the turbidity information are used to calculate a transparency correction coefficient through a pre-established regression model, which is used to adjust the original transparency value.

10. The underwater sensor network node positioning method according to claim 9, characterized in that: The transparency data of each node in the underwater sensor network is obtained, and the initial three-dimensional spatial distribution model of the transparency data is established by combining the information of depth, temperature and turbidity, including the transparency of node A is 0.96, the depth is 50m, the temperature is 15°C, and the turbidity is 10NTU; The depth data is used to determine the vertical position of the node; The temperature data and the turbidity information are used to calculate the transparency correction coefficient through a pre-established multivariate linear regression model, including correction coefficients of 0.02 / °C and -0.01 / NTU, which are used to adjust the original transparency value. The voxel accuracy is set to 1m to obtain a three-dimensional spatial distribution model of transparency.

Images