A quantitative analysis method of terrain and geomagnetic complexity based on PointNetMLP network

Abstract

This paper discloses a method for quantitatively analyzing topographic and geomagnetic complexity based on the PointNetMLP network. Based on the concept of the PointNet network, the core idea of ​​PointNet is to directly process three-dimensional point cloud data without converting it to other formats (such as voxel grids or multi-view images). PointNetMLP is a key component in the PointNet network. It consists of multiple multi-layer perceptrons (MLPs) with shared weights and is used to extract local features from point cloud data. The method uses deep learning technology to automatically extract features from two types of maps (geomagnetic maps and topographic maps) and perform complexity quantification analysis, improving the accuracy and efficiency of the analysis. It can effectively process large-scale, high-dimensional topographic and geomagnetic data, output complexity scores for user reference, automatically extract features, and perform complexity quantification analysis, improving the accuracy and efficiency of the analysis.

Description

Technical Field

[0001] The present invention belongs to the field of underwater exploration technology, and specifically provides a method for quantitatively analyzing the complexity of terrain and geomagnetic fields based on a PointNetMLP network. Background Art

[0002] Complexity analysis of terrain and geomagnetic data, that is, judging the richness of terrain / geomagnetic features, is of great significance in the fields of geographic information systems, navigation, geological exploration, etc. Traditional methods rely on manual feature extraction and statistical analysis, which makes it difficult to process large-scale, high-dimensional data. In recent years, deep learning technology has made significant progress in image and point cloud data processing, especially the PointNet network has performed well in point cloud classification and segmentation tasks. However, the application of existing technologies in the quantitative analysis of the complexity of terrain and geomagnetic data is still insufficient. For example, the application with publication number CN107132521A discloses a patent application for a method for judging the correctness of terrain matching results in BSLAM. This method judges the terrain matching results through a multi-window consistency method to obtain accurate front-end closed-loop results.

[0003] However, this method only eliminates erroneous closed loops rather than directly obtaining accurate closed loop results. Moreover, the elimination of closed loops cannot meet the real-time requirements of online mapping and can only process terrain information.

[0004] Application publication number CN116124151A discloses a method for terrain matching-assisted navigation in the absence of satellite navigation. This method uses operations such as translation, convolution, altimetry, and multiplication with an elevation map to obtain the current estimated latitude and longitude, posterior standard deviation, map height standard deviation, and confidence level of the terrain match.

[0005] However, this method matches the sequence elevation values ​​with the entire prior map, does not consider the tidal range and attitude measurement error problems, and its application scope is only applicable to terrain data. Summary of the Invention

[0006] The purpose of the present invention is to provide a method for quantitative analysis of topographic and geomagnetic complexity based on the PointNetMLP network in order to solve the above problems.

[0007] The technical solution adopted by the present invention is as follows: a method for quantitative analysis of topographic and geomagnetic complexity based on a PointNetMLP network, the method comprising the following steps:

[0008] S1: Data preprocessing: preprocess and standardize the acquired terrain elevation data and geomagnetic intensity data to generate terrain and geomagnetic point cloud data;

[0009] S2: Feature extraction: Use the PointNet network to extract global and local features of point cloud data from pre-processed terrain and geomagnetic point cloud data;

[0010] S3: Complexity quantification: Complexity quantification analysis of the extracted features is performed through a multi-layer perceptron (MLP);

[0011] S4: Result output: Output the quantitative results of terrain / geomagnetic complexity.

[0012] In a preferred embodiment, in step S1, the terrain elevation can be obtained by using the digital elevation model (DEM) data of the terrain and using an interpolation algorithm to obtain the elevation value of a specific location;

[0013] The geomagnetic intensity can be calculated using the Earth's magnetic field model. The commonly used formula is:

[0014] B=B0(1+3sin 2 θ)

[0015] Where B is the geomagnetic intensity; B0 is the geomagnetic intensity at the Earth's equator (approximately 0.5 G, or 50 μT); and θ is the geographic latitude. Geomagnetic intensity is typically measured using magnetometers, including the following types: fluxgate magnetometers (highly accurate and suitable for a variety of environments), proton precession magnetometers (commonly used for field measurements, with accuracy reaching nanoT levels), optically pumped magnetometers, and superconducting magnetometers (for high-precision measurements).

[0016] In a preferred embodiment, in step S1, the standardization process converts data of different features into the same scale or range, thereby eliminating the influence caused by different feature dimensions (units or scales of data).

[0017] In a preferred embodiment, in step S1, the standardization method includes:

[0018] ① Standardization (Z-score standardization):

[0019] Transform the data into a distribution with mean 0 and standard deviation 1.

[0020] formula:

[0021] Where: x is the original data, μ is the mean, and σ is the standard deviation.

[0022] Applicable scenario: When the data distribution is close to normal distribution.

[0023] ② Normalization (Min-Max Normalization):

[0024] Scale the data linearly to a fixed range (such as [0,1] or [-1,1]).

[0025] formula:

[0026] Where: x min is the minimum value of the data, x max is the maximum value of the data.

[0027] Applicable scenarios: When the data distribution is unknown or has obvious boundaries.

[0028] In a preferred embodiment, in step S2, the network structure of PointNet includes the following parts:

[0029] ① Input layer:

[0030] The input is point cloud data, usually represented as an N×3 matrix, where N is the number of points and 3 represents the three-dimensional coordinates (x, y, z) of each point.

[0031] ② Shared Multilayer Perceptron (MLP):

[0032] Feature extraction is performed on each point, and the coordinates of each point are mapped to a high-dimensional feature space using an MLP with shared weights.

[0033] ③Maximum pooling layer:

[0034] The maximum pooling operation is used to aggregate the features of all points to generate a global feature vector. This step ensures that the network is invariant to the order of the input points.

[0035] ④Fully connected layer:

[0036] The global feature vector is input into the fully connected layer for further feature transformation.

[0037] ⑤Output layer:

[0038] Depending on the task, the output layer can be a classification layer (output category probability, such as inputting a point cloud and outputting the category of the object) or a segmentation layer (outputting the category label of each point).

[0039] In a preferred embodiment, in step S3, the MLP is a feedforward neural network composed of multiple fully connected layers. It is one of the most basic neural network structures in deep learning and is widely used in classification, regression and other machine learning tasks.

[0040] In a preferred embodiment, in step S3, the MLP consists of the following parts:

[0041] a. Input Layer: Receives input data. Each input node corresponds to a feature. The number of nodes in the input layer is equal to the dimension of the input feature.

[0042] b. Hidden Layers: Contains one or more fully connected layers, each consisting of multiple neurons (nodes). Each neuron performs a linear transformation (weighted summation) on the input, and then introduces nonlinearity through an activation function.

[0043] c. Output Layer: Outputs the final prediction results.

[0044] The number of nodes depends on the task type:

[0045] Binary classification task: 1 node (using Sigmoid activation function).

[0046] Multi-classification task: C nodes (using Softmax activation function, C is the number of categories).

[0047] Regression task: 1 node (no activation function or using linear activation function).

[0048] d. Activation Function:

[0049] Introducing nonlinearity enables the network to learn complex patterns.

[0050] Common activation functions:

[0051] ReLU (Rectified Linear Unit): f(x)=max(0,x)

[0052] Sigmoid:

[0053] Tanh:

[0054] Softmax (for multi-classification output layer).

[0055] In a preferred embodiment, in step S3, the working principle of the MLP includes:

[0056] a. Forward Propagation:

[0057] Input data is passed from the input layer to the output layer through the hidden layers.

[0058] The calculation formula for each layer is:

[0059] z=W x+b

[0060] a=f(z)

[0061] Where: x is the input, W is the weight matrix, b is the bias vector, f is the activation function, and a is the output.

[0062] b. Loss Function: measures the difference between the predicted value and the true value.

[0063] Common loss functions:

[0064] Mean Squared Error (MSE, for regression tasks):

[0065] Cross-Entropy Loss (for classification tasks):

[0066] c. Backpropagation: Calculate the gradient of the loss function with respect to each parameter using the chain rule.

[0067] Update weights and biases using gradient descent:

[0068] where η is the learning rate.

[0069] d. Training process: Repeat forward propagation, loss calculation, backpropagation and parameter update until the model converges.

[0070] In summary, due to the adoption of the above technical solution, the beneficial effects of the present invention are:

[0071] In this paper, based on the concept of the PointNet network, three-dimensional point cloud data is directly processed without converting it to other formats. PointNetMLP is used to extract local features from point cloud data. The present invention uses deep learning technology to automatically extract features of two types of maps: geomagnetic maps and topographic maps, and conducts complexity quantification analysis. This method automatically extracts topographic / geomagnetic features and conducts complexity quantification analysis through deep learning technology, improving the accuracy and efficiency of the analysis. The method includes four steps: data preprocessing, feature extraction, complexity quantification, and result output. It can effectively process large-scale, high-dimensional topographic and geomagnetic data, and outputs a complexity score for user reference. BRIEF DESCRIPTION OF THE DRAWINGS

[0072] Figure 1 This is a system architecture diagram of the present invention;

[0073] Figure 2 This is the PointNet network structure diagram in the present invention;

[0074] Figure 3 This is the MLP network structure diagram in the present invention. DETAILED DESCRIPTION

[0075] In order to make the purpose, technical solutions and advantages of the present invention more clearly understood, the present invention will be further described in detail below with reference to the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are only used to explain the present invention and are not intended to limit the present invention.

[0076] Example:

[0077] Reference Figure 1-3 ,

[0078] A method for quantitative analysis of topographic and geomagnetic complexity based on a PointNetMLP network, comprising the following steps:

[0079] S1: Data preprocessing: preprocess and standardize the acquired terrain elevation data and geomagnetic intensity data to generate terrain and geomagnetic point cloud data;

[0080] S2: Feature extraction: Use the PointNet network to extract global and local features of point cloud data from pre-processed terrain and geomagnetic point cloud data;

[0081] S3: Complexity quantification: Complexity quantification analysis of the extracted features is performed through a multi-layer perceptron (MLP);

[0082] S4: Result output: Output the quantitative results of terrain / geomagnetic complexity.

[0083] In step S1, the terrain elevation can be obtained by using the digital elevation model (DEM) data of the terrain and using an interpolation algorithm to obtain the elevation value of a specific location;

[0084] The geomagnetic intensity can be calculated using the Earth's magnetic field model. The commonly used formula is:

[0085] B=B0(1+3sin 2 θ)

[0086] Where B is the geomagnetic intensity; B0 is the geomagnetic intensity at the Earth's equator (approximately 0.5 G, or 50 μT); and θ is the geographic latitude. Geomagnetic intensity is typically measured using magnetometers, including the following types: fluxgate magnetometers (highly accurate and suitable for a variety of environments), proton precession magnetometers (commonly used for field measurements, with accuracy reaching nanoT levels), optically pumped magnetometers, and superconducting magnetometers (for high-precision measurements).

[0087] In step S1 , the standardization process converts data of different features to the same scale or range, thereby eliminating the effects caused by different feature dimensions (units or scales of data).

[0088] In step S1, the standardization method includes:

[0089] ① Standardization (Z-score standardization):

[0090] Transform the data into a distribution with mean 0 and standard deviation 1.

[0091] formula:

[0092] Where: x is the original data, μ is the mean, and σ is the standard deviation.

[0093] Applicable scenario: When the data distribution is close to normal distribution.

[0094] ② Normalization (Min-Max Normalization):

[0095] Scale the data linearly to a fixed range (such as [0,1] or [-1,1]).

[0096] formula:

[0097] Where: x min is the minimum value of the data, x max is the maximum value of the data.

[0098] Applicable scenarios: When the data distribution is unknown or has obvious boundaries.

[0099] In step S2, the network structure of PointNet includes the following parts:

[0100] ① Input layer:

[0101] The input is point cloud data, usually represented as an N×3 matrix, where N is the number of points and 3 represents the three-dimensional coordinates (x, y, z) of each point.

[0102] ② Shared Multilayer Perceptron (MLP):

[0103] Feature extraction is performed on each point, and the coordinates of each point are mapped to a high-dimensional feature space using an MLP with shared weights.

[0104] ③Maximum pooling layer:

[0105] The maximum pooling operation is used to aggregate the features of all points to generate a global feature vector. This step ensures that the network is invariant to the order of the input points.

[0106] ④Fully connected layer:

[0107] The global feature vector is input into the fully connected layer for further feature transformation.

[0108] ⑤Output layer:

[0109] Depending on the task, the output layer can be a classification layer (output category probability, such as inputting a point cloud and outputting the category of the object) or a segmentation layer (outputting the category label of each point).

[0110] In step S3, the MLP is a feedforward neural network consisting of multiple fully connected layers. It is one of the most fundamental neural network structures in deep learning and is widely used in classification, regression, and other machine learning tasks.

[0111] In step S3, the MLP consists of the following parts:

[0112] a. Input Layer: Receives input data. Each input node corresponds to a feature. The number of nodes in the input layer is equal to the dimension of the input feature.

[0113] b. Hidden Layers: Contains one or more fully connected layers, each consisting of multiple neurons (nodes). Each neuron performs a linear transformation (weighted summation) on the input, and then introduces nonlinearity through an activation function.

[0114] c. Output Layer: Outputs the final prediction results.

[0115] The number of nodes depends on the task type:

[0116] Binary classification task: 1 node (using Sigmoid activation function).

[0117] Multi-classification task: C nodes (using Softmax activation function, C is the number of categories).

[0118] Regression task: 1 node (no activation function or using linear activation function).

[0119] d. Activation Function:

[0120] Introducing nonlinearity enables the network to learn complex patterns.

[0121] Common activation functions:

[0122] ReLU (Rectified Linear Unit): f(x)=max(0,x)

[0123] Sigmoid:

[0124] Tanh:

[0125] Softmax (for multi-classification output layer).

[0126] In step S3, the working principle of MLP includes:

[0127] a. Forward Propagation:

[0128] Input data is passed from the input layer to the output layer through the hidden layers.

[0129] The calculation formula for each layer is:

[0130] z=W x+b

[0131] a=f(z)

[0132] Where: x is the input, W is the weight matrix, b is the bias vector, f is the activation function, and a is the output.

[0133] b. Loss Function: measures the difference between the predicted value and the true value.

[0134] Common loss functions:

[0135] Mean Squared Error (MSE, for regression tasks):

[0136] Cross-Entropy Loss (for classification tasks):

[0137] c. Backpropagation: Calculate the gradient of the loss function with respect to each parameter using the chain rule.

[0138] Update weights and biases using gradient descent:

[0139]

[0140] where η is the learning rate.

[0141] d. Training process: Repeat forward propagation, loss calculation, backpropagation and parameter update until the model converges.

[0142] From the above, we can see that the present invention, based on the concept of the PointNet network, directly processes three-dimensional point cloud data without converting it to other formats. PointNetMLP is used to extract local features from point cloud data. The present invention uses deep learning technology to automatically extract features of two types of maps: geomagnetic maps and topographic maps, and perform complexity quantification analysis. Deep learning technology is used to automatically extract topographic / geomagnetic features and perform complexity quantification analysis, thereby improving the accuracy and efficiency of the analysis. The method includes four steps: data preprocessing, feature extraction, complexity quantification, and result output. It can effectively process large-scale, high-dimensional topographic and geomagnetic data and output complexity scores for user reference.

[0143] It should be noted that, in this document, relational terms such as first and second, etc., are used only to distinguish one entity or operation from another entity or operation, and do not necessarily require or imply any actual relationship or order between these entities or operations. Moreover, the term "comprises" or any other variant thereof is intended to cover non-exclusive inclusion, so that a process, method, article or device that includes a series of elements includes not only those elements, but also other elements that are not explicitly listed, or also includes elements that are inherent to such process, method, article or device. In the absence of further restrictions, an element defined by the sentence "comprises a ..." does not exclude the presence of other identical elements in the process, method, article or device that includes the element.

[0144] The above embodiments are only used to illustrate the technical solutions of the present invention, rather than to limit the same. Although the present invention has been described in detail with reference to the aforementioned embodiments, those skilled in the art should understand that they can still modify the technical solutions described in the aforementioned embodiments, or make equivalent replacements for some of the technical features therein. However, these modifications or replacements do not deviate the essence of the corresponding technical solutions from the spirit and scope of the technical solutions of the various embodiments of the present invention.

Claims

1. A quantitative analysis method for topographic and geomagnetic complexity based on the PointNetMLP network, characterized by: The method comprises the following steps: S1: Data preprocessing: preprocess and standardize the acquired terrain elevation data and geomagnetic intensity data to generate terrain and geomagnetic point cloud data; S2: Feature extraction: Use the PointNet network to extract global and local features of point cloud data from pre-processed terrain and geomagnetic point cloud data; S3: Complexity quantification: Complexity quantification analysis of the extracted features is performed using a multi-layer perceptron. S4: Result output: Output the quantitative results of topographic / geomagnetic complexity; In step S2, the network structure of PointNet includes the following parts: ① Input layer: The input is point cloud data, usually represented as an N×3 matrix, where N is the number of points and 3 represents the three-dimensional coordinates (x, y, z) of each point; ② Shared Multi-layer Perceptron: Extract features from each point and use a shared-weight MLP to map the coordinates of each point to a high-dimensional feature space; ③Maximum pooling layer: Use the maximum pooling operation to aggregate the features of all points and generate a global feature vector; this step ensures that the network is invariant to the order of input points; ④Fully connected layer: The global feature vector is input into the fully connected layer for further feature transformation; ⑤Output layer: Depending on the task, the output layer is a classification layer or a segmentation layer.

2. The method for quantitatively analyzing topographic and geomagnetic complexity based on the PointNetMLP network according to claim 1, wherein: In step S1, the terrain elevation is obtained by using the digital elevation model data of the terrain and using an interpolation algorithm to obtain the elevation value of a specific location; The geomagnetic intensity is calculated using the Earth's magnetic field model. The commonly used formula is: B=B0(1+3sin 2 i) Where: B is the geomagnetic intensity; B0 is the geomagnetic intensity at the Earth's equator; θ is the geographic latitude.

3. The method for quantitatively analyzing topographic and geomagnetic complexity based on the PointNetMLP network according to claim 1, wherein: In step S1, the standardization process converts data of different features into the same scale or range, thereby eliminating the influence caused by different feature dimensions.

4. The method for quantitatively analyzing topographic and geomagnetic complexity based on the PointNetMLP network according to claim 1, wherein: In step S1, the standardization method includes: ① Standardization: Transform the data into a distribution with a mean of 0 and a standard deviation of 1; official: Where: x is the original data, μ is the mean, σ is the standard deviation; ② Normalization: Scale the data linearly to a fixed range; official: Where: x min is the minimum value of the data, x max is the maximum value of the data.

5. The method for quantitatively analyzing topographic and geomagnetic complexity based on the PointNetMLP network according to claim 1, wherein: In step S3, the MLP consists of the following parts: a. Input layer: Receives input data, where each input node corresponds to a feature. The number of nodes in the input layer is equal to the dimension of the input feature. b. Hidden layer: Contains one or more fully connected layers, each consisting of multiple neurons; each neuron performs a linear transformation on the input and then introduces nonlinearity through an activation function; c. Output layer: outputs the final prediction results; The number of nodes depends on the task type: Binary classification task: 1 node; Multi-classification task: C nodes; Regression task: 1 node; d. Activation function: Introducing nonlinearity to enable the network to learn complex patterns; Common activation functions: ReLU: f(x) = max(0,x) Sigmoid: Fishy: Softmax.

6. The method for quantitatively analyzing topographic and geomagnetic complexity based on the PointNetMLP network according to claim 1, wherein: In step S3, the working principle of MLP includes: a. Forward propagation: Input data is passed from the input layer to the output layer through the hidden layer; The calculation formula for each layer is: z=W x+b; a=f(z); Where: x is the input, W is the weight matrix, b is the bias vector, f is the activation function, and a is the output; b. Loss function: measures the difference between the predicted value and the true value; The loss functions include: Mean Square Error: Cross Entropy Loss: c. Back propagation: Calculate the gradient of the loss function with respect to each parameter using the chain rule; Update weights and biases using gradient descent: Where η is the learning rate; d. Training process: Repeat forward propagation, loss calculation, backpropagation and parameter update until the model converges.

Images