A method for compressing and storing seabed multi-beam bathymetric data applied to underwater robots

Abstract

The present invention belongs to the technical field of multi-beam bathymetry, and specifically relates to a method for compressing and storing seabed multi-beam bathymetry data applied to an underwater robot. The present invention includes the underwater robot acquiring raw data through multi-beam sonar; solving seabed terrain data according to posture information; processing terrain data in blocks according to terrain entropy; selecting a suitable network architecture according to prior terrain training results; using an artificial neural network to regress each piece of terrain as a training set; and replacing the original bathymetry data with a trained model to achieve data compression. Compared with directly storing three-dimensional point cloud data, the present invention only stores a small amount of model parameters, thereby greatly reducing the storage space consumed, and at the same time can achieve rapid terrain reconstruction.

Description

Technical Field

[0001] The invention belongs to the technical field of multi-beam bathymetry, and in particular relates to a method for compressing and storing seabed multi-beam bathymetry data applied to an underwater robot. Background Art

[0002] Due to the obstruction of seawater, it is difficult to directly observe the topographic changes of the seabed. Therefore, people often use multi-beam sonar for bathymetry, and underwater robots (Autonomous Underwater Vehicle, AUV) are widely used in seabed topography measurement as one of the best tools for people to explore the ocean. Because the average depth of the ocean is 3795 meters, and the range of multi-beam sonar is limited. Therefore, compared with surface vessels, underwater robots have a wider working range and can obtain seabed topography measurement data with higher resolution. When AUV performs seabed topography measurement tasks underwater, due to its limited internal storage space, if the raw data returned by the multi-beam bathymetry sonar is directly stored without processing, the work efficiency of AUV will be greatly reduced. Therefore, it is very necessary to use appropriate methods to compress multi-beam bathymetry data while maintaining accuracy as much as possible.

[0003] At present, the research on seabed multi-beam bathymetry of AUV mainly focuses on the multi-beam bathymetry instrument itself and methods such as multi-beam bathymetry data filtering and processing. There is no relevant method for the compression and storage of seabed multi-beam bathymetry data applied to underwater robots.

[0004] The patent application with the publication date of August 3, 2016, publication number CN105825011A, and invention name “A parallel thinning method for multi-beam bathymetric discrete point cloud data”. This method achieves data compression by gridding multi-beam data. However, this method only fuses the data points around the grid, and does not further compress the regional terrain after gridding.

[0005] The patent application with the publication date of May 5, 2017, publication number CN107229594A, and invention name “A method and system for filtering trend surfaces of multi-beam bathymetric data”. This method divides the multi-beam measurement area into grids, fits the bathymetric points in the grid into a second-order trend surface using the least square median method, and then realizes the filtering of the measurement data through gross error calculation. This method is an optimized filtering method that improves the data quality and does not involve data storage. Summary of the invention

[0006] The object of the present invention is to provide a method for compressing and storing seabed multi-beam bathymetric data applied to an underwater robot.

[0007] A method for compressing and storing seabed multi-beam bathymetric data applied to an underwater robot comprises the following steps:

[0008] Step 1: Divide the area where the underwater robot performs the seabed depth measurement task into multiple subtask areas;

[0009] Step 2: The underwater robot performs the bathymetric task in the subtask area to obtain the underwater robot's position information, attitude information and corresponding multi-beam bathymetric raw data;

[0010] Step 3: The multi-beam bathymetric data is resolved into a three-dimensional point cloud based on the body coordinate system of the underwater robot, and then the seabed topography data of the geodetic coordinate system is obtained through coordinate transformation;

[0011] Step 3: Put the data in step 2 into the buffer pool, and determine whether the number of data measurement lines in the buffer pool reaches the preset scale; if not, return to step 2 to continue to obtain data in the current subtask area; otherwise, execute step 4;

[0012] The cache pool Data is:

[0013] Data = {X; Y; Z};

[0014]

[0015]

[0016]

[0017] in, They represent the horizontal coordinate of the survey line, the vertical coordinate of the survey line, and the terrain depth value obtained in the i-th measurement, and their sizes are all n×1; n represents the number of survey points contained in the survey line; t represents the number of data survey lines in the current buffer pool;

[0018] Step 4: Calculate the entropy value H of the terrain data in the cache pool and determine whether it reaches the output threshold H thre ; If H exceeds the output threshold H thre , then execute step 5; otherwise, return to step 2 to continue to obtain the data of the current subtask area;

[0019]

[0020]

[0021] Among them, z ij is the depth value of measuring point j in measuring line i, is the mean terrain depth of the cache pool;

[0022] Step 5: Take the data in the cache pool and input it into the constructed artificial neural network for regression training. When the parameters converge, output the obtained neural network model parameters;

[0023] The artificial neural network comprises an input layer, a hidden layer and an output layer. The input of the artificial neural network is the horizontal coordinate and the vertical coordinate of the survey line, and the output is the terrain depth value. The number of nodes in the middle layer of the artificial neural network is positively correlated with the entropy value of the terrain data.

[0024] Step 6: Store the trained artificial neural network model parameters to replace the original point cloud data to achieve compressed storage of the data;

[0025] Step 7: If the bathymetric task of all subtask areas is not completed, return to step 2 to perform bathymetric measurement of other subtask areas; when it is necessary to obtain bathymetric data of a certain location, obtain the corresponding artificial neural network model parameters according to the subtask area to which the location belongs, and input the horizontal and vertical coordinates of the location into the corresponding artificial neural network model to obtain the terrain depth value.

[0026] Furthermore, the output threshold H thre The number of nodes in the middle layer of the artificial neural network is obtained by pre-training with prior seabed topography data. The pre-training steps are as follows:

[0027] Step 2.1: Obtain N prior terrain blocks in the subtask area and calculate the terrain data entropy value T of each prior terrain block E,a , and get T E ={T E,a |a=1,2,...,N};

[0028] Step 2.2: Calculate the entropy value T of terrain data one by one E,a The error e of terrain fitting in the artificial neural network model corresponding to different numbers of intermediate layer nodes b ab , and record the time consumed c ab , we get E = {e ab |a=1,2,...,N,b=1,.2...,M}, C={c ab |a=1,2,...,N,b=1,.2...,M}; where M is the maximum number of intermediate layer nodes that the artificial neural network can accommodate;

[0029] Step 2.3: Normalize E and C respectively, calculate the harmonic mean F, and select the a corresponding to the minimum harmonic mean F. * ,b * As a result, the calculation is as follows:

[0030]

[0031]

[0032] Step 2.4: Set the output threshold H thre for Set the number of nodes in the middle layer of the artificial neural network to b * .

[0033] The beneficial effects of the present invention are:

[0034] Compared with directly storing three-dimensional point cloud data, the present invention only stores a small amount of model parameters, thereby greatly reducing the consumed storage space and at the same time being able to quickly achieve terrain reconstruction. BRIEF DESCRIPTION OF THE DRAWINGS

[0035] Figure 1 To obtain the output threshold H in the present invention thre Pre-training flowchart with the number of nodes in the middle layer of the artificial neural network.

[0036] Figure 2 It is the overall flow chart of the present invention. DETAILED DESCRIPTION

[0037] The present invention is further described below in conjunction with the accompanying drawings.

[0038] The present invention relates to a method for compressing and storing seabed multi-beam bathymetric data applied to an underwater robot. The method can compress the multi-beam bathymetric data while ensuring accuracy, thereby reducing the occupation of AUV storage space and improving its exploration efficiency.

[0039] Step 1: Divide the area where the underwater robot performs the seabed depth measurement task into multiple subtask areas;

[0040] For each subtask area, obtain the prior seabed topography data for pre-training, and then set the output threshold H thre The number of nodes in the middle layer of the artificial neural network, the pre-training steps are as follows:

[0041] Step 1.1: Obtain N prior terrain blocks in the subtask area and calculate the terrain data entropy value T of each prior terrain block E,a , and get T E ={T E,a |a=1,2,...,N};

[0042] Step 1.2: Calculate the entropy value T of terrain data one by one E,a The error e of terrain fitting in the artificial neural network model corresponding to different numbers of intermediate layer nodes b ab , and record the time consumed c ab , we get E = {e ab|a=1,2,...,N,b=1,.2...,M}, C={c ab |a=1,2,...,N,b=1,.2...,M}; where M is the maximum number of intermediate layer nodes that the artificial neural network can accommodate;

[0043] Step 1.3: Normalize E and C respectively, calculate the harmonic mean F, and select the a corresponding to the minimum harmonic mean F. * ,b * As a result, the calculation is as follows:

[0044]

[0045]

[0046] Step 1.4: Set the output threshold H thre for Set the number of nodes in the middle layer of the artificial neural network to b * .

[0047] For each subtask area, perform steps 2 to 6:

[0048] Step 2: The underwater robot performs the bathymetric task in the subtask area to obtain the underwater robot's position information, attitude information and corresponding multi-beam bathymetric raw data;

[0049] Step 3: The multi-beam bathymetric data is resolved into a three-dimensional point cloud based on the body coordinate system of the underwater robot, and then the seabed topography data of the geodetic coordinate system is obtained through coordinate transformation;

[0050] Step 3: Put the data in step 2 into the buffer pool, and determine whether the number of data measurement lines in the buffer pool reaches the preset scale; if not, return to step 2 to continue to obtain data in the current subtask area; otherwise, execute step 4;

[0051] The cache pool Data is:

[0052] Data = {X; Y; Z};

[0053]

[0054]

[0055]

[0056] in, They represent the horizontal coordinate of the survey line, the vertical coordinate of the survey line, and the terrain depth value obtained in the i-th measurement, and their sizes are all n×1; n represents the number of survey points contained in the survey line; t represents the number of data survey lines in the current buffer pool;

[0057] Step 4: Calculate the entropy value H of the terrain data in the cache pool and determine whether it reaches the output threshold H thre ; If H exceeds the output threshold H thre , then execute step 5; otherwise, return to step 2 to continue to obtain the data of the current subtask area;

[0058]

[0059]

[0060] Among them, z ij is the depth value of measuring point j in measuring line i, is the mean terrain depth of the cache pool;

[0061] Step 5: Take the data in the cache pool and input it into the constructed artificial neural network for regression training. When the parameters converge, output the obtained neural network model parameters;

[0062] The artificial neural network can adopt a standard feedforward artificial neural network, including an input layer, a hidden layer, and an output layer. The input data is the horizontal coordinate and the vertical coordinate of the survey line, and the output data is the terrain depth value. The number of nodes in the middle layer of the artificial neural network is positively correlated with the terrain entropy value. The activation function adopts the Sigmoid function, and the gradient descent method adopts the Levenberg-Marquardt method.

[0063] Step 6: Store the trained artificial neural network model parameters to replace the original point cloud data to achieve compressed storage of the data;

[0064] Step 7: If the bathymetric task of all subtask areas is not completed, return to step 2 to perform bathymetric measurement of other subtask areas; when it is necessary to obtain bathymetric data of a certain location, obtain the corresponding artificial neural network model parameters according to the subtask area to which the location belongs, and input the horizontal and vertical coordinates of the location into the corresponding artificial neural network model to obtain the terrain depth value.

[0065] The present invention is a method for compressing and storing seabed multi-beam bathymetric data applied to an underwater robot, including the underwater robot acquiring raw data through multi-beam sonar; solving seabed terrain data according to posture information; processing terrain data in blocks according to terrain entropy; selecting a suitable network architecture according to prior terrain training results; using each piece of terrain as a training set to regress it using an artificial neural network; and replacing the original bathymetric data with a trained model to achieve data compression. Compared with directly storing three-dimensional point cloud data, the present invention only stores a small amount of model parameters, thereby greatly reducing the storage space consumed, and at the same time can achieve rapid terrain reconstruction.

[0066] The above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. For those skilled in the art, the present invention may have various modifications and variations. Any modification, equivalent replacement, improvement, etc. made within the spirit and principle of the present invention shall be included in the protection scope of the present invention.

Claims

1. A method for compressing and storing seabed multi-beam bathymetric data applied to an underwater robot, characterized in that: The following steps are involved: Step 1: Divide the area where the underwater robot performs the seabed depth measurement task into multiple subtask areas; Step 2: The underwater robot performs the bathymetric task in the subtask area to obtain the underwater robot's position information, attitude information and corresponding multi-beam bathymetric raw data; Step 3: The multi-beam bathymetric data is resolved into a three-dimensional point cloud based on the body coordinate system of the underwater robot, and then the seabed topography data of the geodetic coordinate system is obtained through coordinate transformation; Step 3: Put the data in step 2 into the buffer pool, and determine whether the number of data measurement lines in the buffer pool reaches the preset scale; if not, return to step 2 to continue to obtain data in the current subtask area; otherwise, execute step 4; The cache pool Data is: Data = {X; Y; Z}; in, They represent the horizontal coordinate of the survey line, the vertical coordinate of the survey line, and the terrain depth value obtained in the i-th measurement, and their sizes are all n×1; n represents the number of survey points contained in the survey line; t represents the number of data survey lines in the current buffer pool; Step 4: Calculate the entropy value H of the terrain data in the cache pool and determine whether it reaches the output threshold H thre ; If H exceeds the output threshold H thre , then execute step 5; otherwise, return to step 2 to continue to obtain the data of the current subtask area; Among them, z ij is the depth value of measuring point j in measuring line i, is the mean terrain depth of the cache pool; Step 5: Take the data in the cache pool and input it into the constructed artificial neural network for regression training. When the parameters converge, output the obtained neural network model parameters; The artificial neural network comprises an input layer, a hidden layer and an output layer. The input of the artificial neural network is the horizontal coordinate and the vertical coordinate of the survey line, and the output is the terrain depth value. The number of nodes in the middle layer of the artificial neural network is positively correlated with the entropy value of the terrain data. Step 6: Store the trained artificial neural network model parameters to replace the original point cloud data to achieve compressed storage of the data; Step 7: If the bathymetric task of all subtask areas is not completed, return to step 2 to perform bathymetric measurement of other subtask areas; when it is necessary to obtain bathymetric data of a certain location, obtain the corresponding artificial neural network model parameters according to the subtask area to which the location belongs, and input the horizontal and vertical coordinates of the location into the corresponding artificial neural network model to obtain the terrain depth value.

2. The method for compressing and storing seabed multi-beam bathymetric data applied to an underwater robot according to claim 1, characterized in that: The output threshold H thre The number of nodes in the middle layer of the artificial neural network is obtained by pre-training with prior seabed topography data. The pre-training steps are as follows: Step 2.1: Obtain N prior terrain blocks in the subtask area and calculate the terrain data entropy value T of each prior terrain block E,a , and get T E ={T E,a |a=1,2,...,N}; Step 2.2: Calculate the entropy value T of terrain data one by one E,a The error e of terrain fitting in the artificial neural network model corresponding to different numbers of intermediate layer nodes b ab , and record the time consumed c ab , we get E = {e ab |a=1,2,...,N,b=1,.2...,M}, C={c ab |a=1,2,...,N,b=1,.2...,M}; where M is the maximum number of intermediate layer nodes that the artificial neural network can accommodate; Step 2.3: Normalize E and C respectively, calculate the harmonic mean F, and select the a corresponding to the minimum harmonic mean F. * ,b * As a result, the calculation is as follows: Step 2.4: Set the output threshold H thre for Set the number of nodes in the middle layer of the artificial neural network to b * .

Images