Deep Learning Enhancement Method for Geomagnetic Signal Data Characteristics

Abstract

The present invention relates to the technical field of signal processing, in particular to a deep learning enhancement method for geomagnetic signal data features, including: an acquisition step, including: collecting geomagnetic field signals; a preprocessing step, including: preprocessing the geomagnetic field signals to obtain a time series signal with variable data volume; a feature extraction step, including: extracting data features from the time series signal to obtain a data feature extraction result of geomagnetic signal features; an enhancement step, including: performing deep learning processing on the data feature extraction result of the geomagnetic signal features to obtain a data enhancement result of geomagnetic field features. Through deep learning enhancement, the representation ability of geomagnetic signal features is significantly improved, making the key features in the signal more obvious and facilitating subsequent identification and classification; the adaptive noise management mechanism enables the method to have strong resistance to environmental interference and can effectively extract useful signals in complex environments.

Description

Technical Field

[0001] The present invention relates to the technical field of signal processing, and particularly to a deep learning enhancement method for geomagnetic signal data features, which is applied to fields such as geomagnetic signal acquisition, processing, feature extraction, and classification. Background Art

[0002] Geomagnetic signals are the manifestation forms of the Earth's magnetic field in specific regions, which contain rich geological, environmental, and spatial location information. In recent years, geomagnetic signals have been widely applied in fields such as navigation and positioning, mineral exploration, geological survey, and military reconnaissance. However, due to problems such as weak signals, susceptibility to interference, and unclear features commonly existing in geomagnetic signals, how to effectively extract and enhance the features of geomagnetic signals has become the focus and difficulty in the research of this field.

[0003] Traditional geomagnetic signal processing methods mainly rely on technologies such as filtering, denoising, and Fourier transform. Although they can improve the signal quality to a certain extent, when dealing with geomagnetic signals in complex environments, traditional methods often perform poorly. Especially when geomagnetic signals are affected by multiple factors such as electromagnetic interference, equipment errors, and environmental changes, traditional methods are difficult to effectively separate and extract useful signal features.

[0004] With the development of deep learning technology, applying it to the field of signal processing has become a trend. In the prior art, there have been some attempts to process geomagnetic signals based on methods such as convolutional neural networks (CNNs) and recurrent neural networks (RNNs). However, these methods usually directly apply deep learning algorithms to the original signals, lacking a customized processing flow for the characteristics of geomagnetic signals, resulting in problems such as limited model generalization ability, insufficient feature extraction, and sensitivity to noise.

[0005] In addition, existing geomagnetic signal processing methods are mostly single algorithms or fixed processes, lacking adaptability and being difficult to handle geomagnetic signals with different qualities and characteristics. At the same time, the existing methods have limited enhancement effects on geomagnetic signal features and cannot meet the requirements of high-precision geomagnetic applications.

[0006] Therefore, there is an urgent need for a deep learning enhancement method designed for the characteristics of geomagnetic signals and having an adaptive processing ability to improve the representation ability and classification accuracy of geomagnetic signal features. Summary of the Invention

[0007] Aiming at the problems existing in the prior art, the present invention provides a deep learning enhancement method for geomagnetic signal data features. This method combines adaptive preprocessing, multi-mode segmented processing, progressive deep learning, and a feature verification mechanism, and can effectively improve the representation ability and classification accuracy of geomagnetic signal features.

[0008] The object of the present invention is to provide a method capable of effectively processing geomagnetic signals of different qualities, which can automatically adjust the processing strategy according to the signal characteristics, including intelligent noise management and optimized segmented processing, so as to improve the accuracy of subsequent feature extraction and classification.

[0009] Another object of the present invention is to provide a progressive deep learning architecture, which can effectively improve the model training effect and generalization ability through three-subset division and progressive training strategy, and realize high-quality enhancement of geomagnetic signal features.

[0010] Another object of the present invention is to provide a verification mechanism based on feature distribution, which can automatically evaluate the feature quality and decide whether further enhancement is needed to ensure the reliability and stability of the processing results.

[0011] The present invention proposes a deep learning enhancement method for geomagnetic signal data features, including:

[0012] An acquisition step, including: collecting geomagnetic field signals;

[0013] A preprocessing step, including: preprocessing the geomagnetic field signals to obtain time series signals with variable data volume;

[0014] A feature extraction step, including: extracting data features from the time series signals to obtain a data feature extraction result of geomagnetic signal features;

[0015] An enhancement step, including: performing deep learning processing on the data feature extraction result of the geomagnetic signal features to obtain a data enhancement result of geomagnetic field features;

[0016] The preprocessing step specifically includes:

[0017] Performing moving average calculation, normalization and detrending processing on the geomagnetic field signals in sequence, wherein the moving average calculation is used to reduce high-frequency noise in the geomagnetic field signals, the normalization is used to convert the geomagnetic field signal data into a standard normal distribution, and the detrending is used to remove the linear trend in the geomagnetic field signals;

[0018] Performing standardized denoising data processing on the processed geomagnetic field signals, and judging whether to increase noise and interference according to the noise and interference intensity;

[0019] Performing non-overlapping segmented processing on the processed geomagnetic field signals;

[0020] The noise and interference intensity is obtained by calculating the variance of the geomagnetic field signals; when the variance is less than 0.01, it is determined that no noise and interference need to be increased; when the variance is greater than or equal to 0.01, it is determined that noise and interference need to be increased.

[0021] Preferably, the non-overlapping segmentation process includes:

[0022] Statistically processing the variance value of the processed geomagnetic field signal;

[0023] Calculating the number of valid segments of the geomagnetic field signal with a variance value less than 0.01;

[0024] When the number of valid segments is greater than or equal to the threshold, performing non-overlapping segmentation on the geomagnetic field signal in an equal-time segmentation manner;

[0025] When the number of valid segments is less than the threshold, performing non-overlapping segmentation on the geomagnetic field signal in an equal-variance segmentation manner.

[0026] Preferably, the data volume of the time series signal obtained in the preprocessing step includes at least one of 160,000, 10,000, and 1,600; where:

[0027] When the data volume is 160,000, it includes a time series signal with 2,048 consecutive geomagnetic signal data points as a group;

[0028] When the data volume is 10,000, it includes a time series signal with 250 consecutive geomagnetic signal data points as a group;

[0029] When the data volume is 1,600, it includes a time series signal with 100 consecutive geomagnetic signal data points as a group.

[0030] Preferably, in the data feature extraction in the feature extraction step, a random forest algorithm or a deep learning algorithm is used, and it is judged whether the time series signal needs to add noise and interference according to the obtained feature vector or data distribution histogram.

[0031] Preferably, the enhancement step specifically includes:

[0032] Dividing the data feature extraction result into three partial subsets, where the first partial subset accounts for 15%, the second partial subset accounts for 25%, and the remaining data subset is used as the third partial subset accounting for 60%;

[0033] Inputting the second partial subset into a deep learning neural network for non-linear classification to obtain a vector of the geomagnetic signal features of the second partial subset;

[0034] Inputting the first partial subset into the trained deep learning neural network for non-linear classification;

[0035] Inputting the third partial subset into the further trained deep learning neural network for non-linear classification.

[0036] Preferably, the method further includes:

[0037] Calculate the distribution histogram after vector classification of the geomagnetic signal characteristics of the three partial subsets;

[0038] Compare the distribution histogram with the data feature extraction results in the feature extraction step;

[0039] When the standard deviation between any ratio and the data feature extraction results is greater than or equal to 10%, add noise and interference to all the data feature extraction results, and use deep learning algorithms for data augmentation;

[0040] When the standard deviation between all ratios and the data feature extraction results is less than 10%, it is determined that no noise and interference need to be added.

[0041] Preferably, the non-linear classification by the deep learning neural network includes:

[0042] Perform binary classification, ternary classification or multi-classification processing on the vector classification of the geomagnetic signal characteristics;

[0043] Use deep learning DNN for processing to obtain a trained deep learning classification model;

[0044] Classify newly input vectors through the trained deep learning classification model.

[0045] Preferably, the method further includes:

[0046] Input the data augmentation results of the geomagnetic field characteristics into a fully convolutional neural network, where the fully convolutional neural network includes:

[0047] Use a 1×1×1 convolutional kernel to extract data channel features, enhance channel dimension features, and add a Sigmoid activation layer after the convolutional layer;

[0048] Perform rotation transformation in the two-dimensional plane and use max pooling for two-dimensional dimensionality reduction;

[0049] Perform batch normalization on the data and set a feature mapping layer;

[0050] Perform downsampling and high-density feature enhancement in the target area and classify through a fully connected layer.

[0051] The deep learning enhancement method for the geomagnetic signal data features provided by the present invention has the following beneficial effects:

[0052] 1. Improve the representation ability of geomagnetic signal features: Through deep learning enhancement, the representation ability of geomagnetic signal features is significantly improved, making the key features in the signal more obvious, which is convenient for subsequent recognition and classification;

[0053] 2. Enhance the ability to resist noise interference: The adaptive noise management mechanism enables the method to have strong resistance to environmental interference and can effectively extract useful signals in complex environments;

[0054] 3. Improve classification accuracy: The progressive training method and algorithm fusion improve the classification accuracy, making the classification results based on enhanced features more reliable;

[0055] 4. Achieve efficient dimensionality reduction: The dedicated network architecture effectively reduces the dimension while retaining important signal information, reducing the computational complexity;

[0056] 5. Have the ability of adaptive processing: The method can automatically adjust the processing strategy according to the signal characteristics and adapt to signals with different qualities and characteristics;

[0057] 6. Enhance the generalization ability: The overall method has good generalization ability for geomagnetic signals with various characteristics and can be applied to different scenarios;

[0058] 7. Improve the signal-to-noise ratio: The preprocessing and enhancement techniques improve the signal-to-noise ratio of the signal, making the useful signal more prominent;

[0059] 8. Optimize the computational efficiency: The architecture design considers the computational efficiency of geomagnetic signal processing and reduces the computational resource requirements while ensuring the processing effect. BRIEF DESCRIPTION OF THE DRAWINGS

[0060] Figure 1 is a schematic flow chart of the deep learning enhancement method for the geomagnetic signal data characteristics of the present invention;

[0061] Figure 2 is a detailed schematic flow chart of the preprocessing step of the present invention;

[0062] Figure 3 is a schematic flow chart of the non-overlapping segmentation processing of the present invention;

[0063] Figure 4 is a schematic flow chart of the data feature extraction step of the present invention;

[0064] Figure 5 is a schematic diagram of the three-subset division and progressive learning of the present invention;

[0065] Figure 6 is a schematic diagram of the fully convolutional neural network architecture of the present invention;

[0066] Figure 7 is a schematic diagram of the RBF neural network fusion calculation of the present invention;

[0067] Figure 8 is a comparison diagram of the processing effects of the method of the present invention applied to typical geomagnetic signals. DETAILED DESCRIPTION OF THE INVENTION

[0068] Please refer to the attached Figure 1-8 , and the present invention will be further described in detail below in conjunction with the accompanying drawings and specific embodiments.

[0069] Next, the technical solutions in the embodiments of the present invention will be clearly and completely described in conjunction with the accompanying drawings in the embodiments of the present invention. Obviously, the described embodiments are only a part of the embodiments of the present invention, rather than all the embodiments. All other embodiments obtained by those of ordinary skill in the art based on the embodiments of the present invention without creative efforts shall fall within the protection scope of the present invention.

[0070] As Figure 1 shown, the present invention provides a deep learning enhancement method for the characteristics of geomagnetic signal data, and the method includes the following steps:

[0071] 1. Acquisition step S1: Collect geomagnetic field signals;

[0072] 2. Preprocessing step S2: Preprocess the geomagnetic field signals to obtain a time series signal with variable data volume;

[0073] 3. Feature extraction step S3: Extract data features from the time series signal to obtain the data feature extraction result of the geomagnetic signal features;

[0074] 4. Enhancement step S4: Perform deep learning processing on the data feature extraction result of the geomagnetic signal features to obtain the data enhancement result of the geomagnetic field features.

[0075] In the acquisition step S1, the present invention uses a professional geomagnetic sensor to collect geomagnetic field signals. Preferably, a high-precision three-axis magnetometer is used for collection to obtain geomagnetic field signal data in three dimensions of x, y, and z. The collection frequency is preferably 10 - 100 Hz to ensure capturing the subtle changes of the geomagnetic field. In a preferred embodiment, the collection frequency is set to 50 Hz, and this frequency can avoid generating too much redundant data while ensuring signal integrity.

[0076] In the preprocessing step S2, the geomagnetic field signals collected in step S1 are preprocessed, and the geomagnetic field signals are processed to obtain a time series signal with variable data volume through preprocessing. As Figure 2 shown, this preprocessing step specifically includes:

[0077] Step S201: Perform moving average calculation, normalization, and detrending processing on the geomagnetic field signals collected in step S1 in sequence.

[0078] The moving average calculation is used to reduce the high-frequency noise in the collected geomagnetic field signals. Preferably, the moving average calculation uses the following formula:

[0079] ,

[0080] Among them, is the original geomagnetic field signal value, is the signal value after moving average, is the moving window size, represents the index of the signal point being currently processed, represents the indices of the sampling points within the moving window. In an embodiment of the present invention, takes a value of 5. This window size can effectively smooth short-term noise without overly losing signal details. The normalization uses a standardization formula to convert the geomagnetic field signal data collected in step S1 into a standard normal distribution. The standardization formula is as follows:

[0081] ,

[0082] Among them, is the signal value after moving average, is the signal mean, is the signal standard deviation, is the standardized signal value. The standardization process makes signals of different amplitudes comparable, facilitating subsequent processing. The detrending uses moving average filtering to remove the linear trend in the geomagnetic field signal collected in step S1. The detrending process uses the following formula:

[0083] ,

[0084] Among them, is the standardized signal value, is the long-window moving average value, is the detrended signal value. In a preferred embodiment, the long-window size is taken as 10% of the signal length. This setting can effectively capture the long-term trend of the signal without affecting local features.

[0085] Step S202: Perform standardized denoising data processing on the geomagnetic field signal after the moving average calculation, normalization, and detrending in step S201. Determine whether noise and interference need to be added to the geomagnetic field signal after the moving average calculation, normalization, and detrending in step S201 according to the noise and interference intensity. When the determination result is that noise and interference need to be added, then add noise and interference to the geomagnetic field signal after the moving average calculation, normalization, and detrending in step S201; when the determination result is that noise and interference do not need to be added, then proceed to step S3.

[0086] The noise and interference intensity is obtained by calculating the variance of the geomagnetic field signal. When determining whether to add noise and interference to the geomagnetic field signal after the moving average calculation, normalization, and detrending in step S2, if the variance is less than 0.01, it is determined that the geomagnetic field signal after the moving average calculation, normalization, and detrending in step S2 does not require adding noise and interference; when the variance is greater than or equal to 0.01, it is determined that noise and interference need to be added.

[0087] The present invention determines the signal quality through the variance threshold of 0.01, and the selection of this threshold is based on a large amount of experimental data analysis. When the variance is less than 0.01, it indicates that the signal fluctuation is small and the features are not obvious enough, and at this time, there is no need to add noise and interference; when the variance is greater than or equal to 0.01, it indicates that the signal fluctuation is large and the features may be masked by noise, and appropriate noise and interference need to be added for data enhancement to improve the robustness of the model.

[0088] Step S203: Perform non-overlapping segmentation processing on the geomagnetic field signal obtained in step S202.

[0089] Combined Figure 3 As shown, the non-overlapping segmentation processing includes:

[0090] Step S2031: Statistically analyze the variance value of the geomagnetic field signal processed in step S202;

[0091] Step S2032: Calculate the number of valid segments of the geomagnetic field signal with a variance value less than 0.01. When the number of valid segments is greater than or equal to the threshold, the non-overlapping segmentation processing of the geomagnetic field signal is performed in an equal-time segmentation manner; when the number of valid segments is less than the threshold, the non-overlapping segmentation processing of the geomagnetic field signal is performed in an equal-variance value segmentation manner.

[0092] Preferably, the threshold is set to 30% of the total number of segments. For example, if the signal is divided into 100 segments, the threshold is 30. The selection of this threshold is based on statistical analysis to ensure that the segmentation processing can adapt to different types of geomagnetic signals. When the proportion of the number of valid segments is relatively high, the signal is relatively stable as a whole, and equal-time segmentation is suitable; when the proportion of the number of valid segments is relatively low, the signal fluctuates greatly, and equal-variance value segmentation is suitable to better capture the signal features.

[0093] Step S2033: Perform non-overlapping segmentation processing on the geomagnetic field signal processed in step S202 to obtain the time-series segmented signal of the geomagnetic signal.

[0094] The data volume of the time-series signal obtained by the preprocessing step includes at least one of 160,000, 10,000, and 1,600. Specifically, it includes:

[0095] 1. When the data volume is 160,000, a time series signal with 2,048 consecutive geomagnetic signal data points as a group;

[0096] 2. When the data volume is 10,000, a time series signal with 250 consecutive geomagnetic signal data points as a group;

[0097] 3. When the data volume is 1,600, a time series signal with 100 consecutive geomagnetic signal data points as a group.

[0098] The present invention designs three time series signals with different data volumes, which are suitable for different application scenarios and computing resource conditions. A data volume of 160,000 is suitable for scenarios that require high-precision analysis and can capture subtle changes in the signal; a data volume of 10,000 is suitable for general application scenarios, achieving a balance between accuracy and computing efficiency; a data volume of 1,600 is suitable for scenarios with limited resources or those that require fast processing, with a slightly lower accuracy but a faster processing speed.

[0099] In the feature extraction step S3, as Figure 4 shown, data feature extraction is performed on the time series signal in step S2, and the data feature extraction algorithm is used to perform data feature extraction on the time series signal to obtain the data feature extraction result of the geomagnetic signal feature.

[0100] The data feature extraction in the feature extraction step uses a random forest algorithm or a deep learning algorithm, and it is determined whether the time series signal needs to add noise and interference according to the obtained feature vector or data distribution histogram.

[0101] In a preferred embodiment of the present invention, the implementation of the random forest algorithm includes the following steps:

[0102] 1. Randomly select n samples from the original training set by the Bootstrap method to construct a decision tree;

[0103] 2. Randomly select m features at each node and select the optimal feature for splitting from them;

[0104] 3. Each tree grows to the maximum extent without pruning;

[0105] 4. Repeat the above steps to generate multiple decision trees to form a random forest;

[0106] 5. For the new time series signal, pass it into each decision tree for prediction, and obtain the final prediction result through voting or averaging.

[0107] Preferably, the number of decision trees is set to 100, and the number of feature selections m is set to the square root of the total number of features. These parameter settings can avoid overfitting while ensuring the performance of the model.

[0108] When using a deep learning algorithm for feature extraction, it is preferred to use a convolutional neural network (CNN) structure. The basic structure of a CNN includes:

[0109] 1. Input layer: Receives the preprocessed time series signal;

[0110] 2. Convolutional layer: Uses multiple convolutional kernels to extract signal features;

[0111] 3. Pooling layer: Reduces the feature dimension and improves the robustness of the model;

[0112] 4. Fully connected layer: Maps the features to the feature space;

[0113] 5. Output layer: Outputs the extracted feature vector.

[0114] In the CNN structure, the preferred sizes of the convolutional kernels are 3, 5, and 7, and the preferred number of convolutional layers is 3 - 5 layers. These settings can capture signal features at different scales.

[0115] After processing by the random forest algorithm or the deep learning algorithm, a feature vector or a data distribution histogram will be obtained. The present invention determines whether to add noise and interference to the time series signal based on these results. The judgment criteria are as follows:

[0116] 1. When the feature distribution is too concentrated (entropy value is less than 0.5), it indicates that the features are not rich enough, and noise and interference need to be added;

[0117] 2. When the discrimination between features is insufficient (the sum of the weights of the main features is less than 0.7), noise and interference need to be added;

[0118] 3. When the dimensional effectiveness of the feature vector is insufficient (the proportion of effective dimensions is less than 0.6), noise and interference need to be added.

[0119] The thresholds of these judgment criteria are determined based on a large number of experiments and data analyses, and can effectively evaluate the quality of feature extraction and guide subsequent processing.

[0120] In enhancement step S4, and in combination with Figure 5 as shown, the enhancement step specifically includes:

[0121] Step S401: Divide the data feature extraction result of the geomagnetic signal features obtained by data feature extraction in step S3 into three partial subsets. Among them, the first partial subset accounts for 15%, the second partial subset accounts for 25%, and the remaining data subset is used as the third partial subset, accounting for 60%;

[0122] This three - subset division ratio (15% - 25% - 60%) is one of the innovation points of the present invention. After a large number of experimental verifications, this ratio allocation can improve the generalization ability of the model while ensuring the training effect. The first subset (15%) is used for fine - tuning, the second subset (25%) is used for initial training, and the third subset (60%) is used for global optimization. The three work together to achieve the optimization of the model performance.

[0123] Step S402: Input the second subset into the deep - learning neural network for non - linear classification to obtain the vector of the geomagnetic signal features of the second subset;

[0124] Step S403: Input the first subset into the trained deep - learning neural network for non - linear classification to obtain the vector of the geomagnetic signal features of the first subset;

[0125] Step S404: Input the third subset into the further trained deep - learning neural network for non - linear classification to obtain the vector of the geomagnetic signal features of the third subset;

[0126] The method further includes:

[0127] Step S405: Calculate the distribution histogram of the vectors of the geomagnetic signal features of the three subsets after classification;

[0128] Step S406: Compare the distribution histogram with the data feature extraction results in the feature extraction step;

[0129] Step S407: When the standard deviation between any ratio and the data feature extraction results is greater than or equal to 10%, add noise and interference to all of the data feature extraction results and perform data augmentation using a deep - learning algorithm; when the standard deviation between all ratios and the data feature extraction results is less than 10%, it is determined that no noise and interference need to be added.

[0130] The present invention uses a standard deviation threshold of 10% for judgment. This threshold is the optimal value determined based on a large amount of experimental data. When the standard deviation is greater than or equal to 10%, it indicates that the feature distribution has changed significantly, and noise and interference need to be added to enhance the adaptability of the model; when the standard deviation is less than 10%, it indicates that the feature distribution is relatively stable and no additional enhancement is required.

[0131] The non - linear classification by the deep - learning neural network includes:

[0132] Step S4021: Perform binary classification, ternary classification, or multi - classification on the classification of the vector of the geomagnetic signal features;

[0133] Step S4022: Process using deep learning DNN to obtain a trained deep learning classification model;

[0134] Step S4023: Classify the newly input vector through the trained deep learning classification model.

[0135] In a preferred embodiment of the present invention, the structure of the DNN model is designed as follows:

[0136] 1. Input layer: The number of nodes is equal to the dimension of the feature vector;

[0137] 2. Hidden layer: Adopt 3 - 5 hidden layers, and the number of nodes in each layer is 128, 64, 32, 16, 8 respectively (gradually decreasing from the input layer to the output layer);

[0138] 3. Output layer: The number of nodes is equal to the number of classification categories;

[0139] 4. Activation function: The ReLU activation function is used in the hidden layer, and the Softmax activation function is used in the output layer;

[0140] 5. Optimizer: Adam optimizer, the initial learning rate is set to 0.001, and the learning rate decay strategy is adopted;

[0141] 6. Loss function: Cross - entropy loss function.

[0142] The training process of the DNN model is as follows:

[0143] 1. Use the second subset (25%) for initial training, with the number of iterations being 100;

[0144] 2. Use the first subset (15%) for fine - tuning, with the number of iterations being 50;

[0145] 3. Use the third subset (60%) for global optimization, with the number of iterations being 200.

[0146] This progressive training strategy can give full play to the advantages of the three - subset division, gradually optimize the model performance, and improve the feature representation ability.

[0147] The method further includes: inputting the data augmentation result of the geomagnetic field features into a fully convolutional neural network, where the fully convolutional neural network includes:

[0148] 1. Use a 1×1×1 convolutional kernel to extract data channel features, enhance the channel - dimension features, and add a Sigmoid activation layer after the convolutional layer;

[0149] 2. Perform a rotation transformation in the two - dimensional plane and use max - pooling for two - dimensional dimensionality reduction;

[0150] 3. Batch normalize the data and set the feature mapping layer;

[0151] 4. Downsample and enhance high-density features in the target area, and classify through the fully connected layer.

[0152] As Figure 6 shown, the fully convolutional neural network structure designed by the present invention is optimized for the characteristics of geomagnetic signals and has the following advantages:

[0153] 1. 1×1×1 convolutional kernel design: This design can effectively extract the mutual relationship between data channels, enhance the channel dimension features, and is suitable for the processing of multi-dimensional geomagnetic signals;

[0154] 2. Sigmoid activation layer: Adding a Sigmoid activation layer after the convolutional layer can highlight important features, suppress secondary features, and improve the feature representation ability of the network;

[0155] 3. Two-dimensional rotation transformation: By performing rotation transformation in the two-dimensional plane, the network is made invariant to signal direction changes and the robustness of the model is enhanced;

[0156] 4. Max pooling dimensionality reduction: Use max pooling for two-dimensional dimensionality reduction operations to reduce computational complexity while retaining important features;

[0157] 5. Batch normalization processing: Perform batch normalization processing on the data to accelerate network convergence and improve training stability;

[0158] 6. Feature mapping layer design: Set the feature mapping layer to perform feature mapping on the target area to highlight key features;

[0159] 7. Downsampling and high-density feature enhancement: Downsample and enhance high-density features in the target area to improve the accuracy and richness of feature representation;

[0160] 8. Fully connected layer classification: Perform final classification through the fully connected layer to achieve high-precision feature classification.

[0161] The mathematical expression of the fully convolutional neural network is as follows:

[0162] For the input feature map X, the output feature map of the l-th layer The calculation formula is

[0163] ,

[0164] where is the convolutional kernel parameter of the -th layer, is the output feature map of the -th layer, used as the input of the first layer, is the bias term, Denotes a convolution operation, is an activation function. For the 1×1×1 convolution operation, its mathematical expression is:

[0165] ,

[0166] where, denotes the spatial position of the feature map, c denotes the channel index, and C is the number of input channels. For the Sigmoid activation function, its mathematical expression is:

[0167] ,

[0168] For the batch normalization operation, its mathematical expression is:

[0169] ,

[0170] ,

[0171] where, and are the mean and variance within the batch respectively, is a small constant to avoid division by zero, and are learnable parameters.

[0172] As Figure 7 shown, the present invention also provides a fusion calculation method based on an RBF neural network for further improving the accuracy of feature classification. The method includes the following steps:

[0173] 1. Using the data augmentation result as the input of the RBF neural network;

[0174] 2. Performing a non - linear transformation through the radial basis function;

[0175] 3. Combining the probability value and the classification result to obtain the final classification output.

[0176] The structure of the RBF neural network includes: an input layer, a hidden layer, and an output layer. A transition layer is added between the input layer and the hidden layer to optimize the calculation process.

[0177] The mathematical expression of the RBF neural network is as follows:

[0178] For the input vector X, the output of the hidden layer node is:

[0179] ,

[0180] where, is the center of the th hidden layer node, is the spread parameter.

[0181] The calculation formula of the output layer is as follows:

[0182] ,

[0183] where, is the weight, is the bias.

[0184] In the preferred embodiment of the present invention, the parameters of the RBF neural network are set as follows:

[0185] 1. Number of hidden layer nodes: selected as 1.5 times the input dimension to ensure sufficient expressive ability;

[0186] 2. Center point selection: Use the K-means clustering algorithm to determine the center point position;

[0187] 3. Spread parameter: determined by the average distance method, , where is the maximum distance and m is the number of hidden layer nodes;

[0188] 4. Weight initialization: Use the He initialization method.

[0189] The training of the RBF neural network adopts a two-stage strategy:

[0190] 1. The first stage: Determine the hidden layer parameters (center point and spread parameter);

[0191] 2. The second stage: Train the output layer weights and use the gradient descent method to minimize the mean square error.

[0192] Through the fusion calculation of the RBF neural network, the probability value and the classification result can be organically combined to improve the accuracy and stability of classification. Experiments show that compared with using only DNN classification, the RBF fusion calculation can increase the classification accuracy by 3-5 percentage points.

[0193] The present invention also provides a method for processing multi-dimensional geomagnetic signals, which can make full use of the multi-dimensional information of geomagnetic signals and improve the accuracy of feature extraction and classification.

[0194] In this embodiment, the geomagnetic signals include the following dimensions:

[0195] 1. x-dimension information: the component of the geomagnetic field in the x-axis direction;

[0196] 2. y-dimension information: the component of the geomagnetic field in the y-axis direction;

[0197] 3. z-dimension information: the component of the geomagnetic field in the z-axis direction;

[0198] 4. xy - dimensional information: composite information of the x and y dimensions;

[0199] 5. yz - dimensional information: composite information of the y and z dimensions;

[0200] 6. zx - dimensional information: composite information of the z and x dimensions.

[0201] The processing methods of multi - dimensional information include the following three modes:

[0202] 1. Three - dimensional data mode: Collect information on the three dimensions of x, y, and z simultaneously;

[0203] 2. Composite - dimension mode: Collect information on the three composite dimensions of xy, yz, and zx;

[0204] 3. Hybrid mode: Collect multiple dimensions from the above - mentioned dimensions simultaneously.

[0205] For multi - dimensional data, the present invention adopts the following feature calculation methods:

[0206] 1. Mean calculation:

[0207] ,

[0208] where, represents the value of the d - th dimension of the i - th data point, represents the mean of the d - th dimension, and N represents the total number of data points.

[0209] 2. Standard - deviation calculation:

[0210] ,

[0211] where, represents the standard deviation of the d - th dimension.

[0212] 3. Information - entropy calculation:

[0213] ,

[0214] where, represents the information entropy of the d - th dimension, represents the probability distribution of the d - th dimension. 4. Eigen - value calculation:

[0215] ,

[0216] where, represents the eigen - value of the d - th dimension.

[0217] Through the above - mentioned methods, rich feature information can be extracted from multi - dimensional geomagnetic signals, providing a more comprehensive input for subsequent deep - learning processing and further improving the effects of feature enhancement and classification.

[0218] The method of the present invention has been verified in actual geomagnetic signal processing, Figure 8 showing a comparison of application effects. The verification experiment used geomagnetic signals collected in three different environments:

[0219] 1. Low-noise environment: an indoor laboratory environment with little electromagnetic interference;

[0220] 2. Medium-noise environment: an urban street environment with a certain degree of electromagnetic interference;

[0221] 3. High-noise environment: an industrial area environment with severe electromagnetic interference.

[0222] The verification results show that the method of the present invention exhibits excellent performance in all three environments:

[0223] 1. Low-noise environment: The signal-to-noise ratio is increased by 40%, and the feature classification accuracy rate reaches 98.5%;

[0224] 2. Medium-noise environment: The signal-to-noise ratio is increased by 55%, and the feature classification accuracy rate reaches 95.2%;

[0225] 3. High-noise environment: The signal-to-noise ratio is increased by 70%, and the feature classification accuracy rate reaches 92.1%.

[0226] Compared with the traditional method, the present invention has significant advantages in processing geomagnetic signals in a high-noise environment, and the feature classification accuracy rate is increased by 15 - 20 percentage points, indicating that the method of the present invention has strong anti-noise ability and adaptability.

[0227] In addition, the method of the present invention also shows good stability under different data scales:

[0228] 1. Data volume of 160,000: The processing time is about 3 seconds, and the classification accuracy rate is 98.0%;

[0229] 2. Data volume of 10,000: The processing time is about 0.8 seconds, and the classification accuracy rate is 96.5%;

[0230] 3. Data volume of 1,600: The processing time is about 0.2 seconds, and the classification accuracy rate is 94.0%.

[0231] This indicates that the method of the present invention can adapt to different data scales and processing speed requirements while ensuring high accuracy.

[0232] The present invention provides a deep learning enhancement method for geomagnetic signal data features. This method realizes high-quality enhancement of geomagnetic signal features through adaptive preprocessing, multi-mode segmented processing, three-subset progressive learning, and feature verification mechanisms. Compared with the traditional method, the present invention has the following significant advantages:

[0233] 1. Adaptive processing ability: It can automatically adjust the processing strategy according to the signal characteristics to adapt to signals with different qualities and characteristics;

[0234] 2. Efficient feature extraction: Through a variety of algorithms and optimized structures, improve the efficiency and quality of feature extraction;

[0235] 3. Progressive learning architecture: Adopt a three - subset partition and progressive training strategy to improve the model performance and generalization ability;

[0236] 4. Verification feedback mechanism: Ensure the reliability of the processing results through feature distribution comparison and verification feedback;

[0237] 5. Multi - dimensional collaborative processing: It can process multi - dimensional geomagnetic signals to improve the comprehensiveness and accuracy of feature representation.

[0238] The method of the present invention has broad application prospects in many fields such as geomagnetic navigation, geological exploration, space positioning, military reconnaissance, etc. It can provide higher - quality feature representation and classification results for these applications, and promote the development and application of related technologies.

[0239] The above - mentioned is only the specific implementation manner of the present invention, but the protection scope of the present invention is not limited thereto. Any person skilled in the art can easily think of changes or substitutions within the technical scope disclosed by the present invention, and all should be covered by the protection scope of the present invention. Therefore, the protection scope of the present invention should be subject to the protection scope of the claims.

[0240] The above - mentioned is only the preferred embodiment of the present invention and is not used to limit the present invention. For those skilled in the art, the present invention can have various changes and modifications. 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 deep learning enhancement method for geomagnetic signal data characteristics, characterized in that, Including: An acquisition step, including: collecting geomagnetic field signals; A preprocessing step, including: preprocessing the geomagnetic field signals to obtain a time series signal with variable data volume; A feature extraction step, including: extracting data features from the time series signal to obtain a data feature extraction result of geomagnetic signal features; An enhancement step, including: performing deep learning processing on the data feature extraction result of the geomagnetic signal features to obtain a data enhancement result of geomagnetic field features; Specifically, the preprocessing step includes: Performing moving average calculation, normalization, and detrending processing on the geomagnetic field signals in sequence, where the moving average calculation is used to reduce high-frequency noise in the geomagnetic field signals, the normalization is used to convert the geomagnetic field signal data into a standard normal distribution, and the detrending is used to remove the linear trend in the geomagnetic field signals; Performing standardized denoising data processing on the processed geomagnetic field signals, and judging whether to add noise and interference according to the noise and interference intensity; Performing non-overlapping segmentation processing on the processed geomagnetic field signals; The noise and interference intensity is obtained by calculating the variance of the geomagnetic field signals; when the variance is less than 0.01, it is determined that no noise and interference need to be added; when the variance is greater than or equal to 0.01, it is determined that noise and interference need to be added; Specifically, the enhancement step includes: Dividing the data feature extraction result into three partial subsets, where the first partial subset accounts for 15%, the second partial subset accounts for 25%, and the remaining data subset is used as the third partial subset accounting for 60%; Inputting the second partial subset into a deep learning neural network for non-linear classification to obtain a vector of geomagnetic signal features of the second partial subset; Inputting the first partial subset into the trained deep learning neural network for non-linear classification; Inputting the third partial subset into the further trained deep learning neural network for non-linear classification; The method further includes: Calculating a distribution histogram of the vectors of the geomagnetic signal features of the three partial subsets after classification; Comparing the distribution histogram with the data feature extraction result in the feature extraction step; When the standard deviation between any ratio and the data feature extraction result is greater than or equal to 10%, adding noise and interference to all of the data feature extraction results and performing data enhancement using a deep learning algorithm; When the standard deviation between all ratios and the data feature extraction result is less than 10%, it is determined that no noise and interference need to be added.

2. The deep learning enhancement method for geomagnetic signal data characteristics according to claim 1, wherein The non-overlapping segmentation processing includes: Statistically calculating the variance value of the processed geomagnetic field signals; Calculating the number of valid segments of the geomagnetic field signals with a variance value less than 0.01; When the number of valid segments is greater than or equal to the threshold, performing non-overlapping segmentation processing on the geomagnetic field signals in an equal-time segmentation manner; When the number of valid segments is less than the threshold, performing non-overlapping segmentation processing on the geomagnetic field signals in an equal-variance value segmentation manner.

3. The deep learning enhancement method for geomagnetic signal data characteristics according to claim 1, characterized in that The data volume of the time series signal obtained in the preprocessing step includes at least one of 160,000, 10,000, and 1,600; where: When the data volume is 160,000, a time series signal with 2,048 consecutive geomagnetic signal data points as a group; When the data volume is 10,000, a time series signal with 250 consecutive geomagnetic signal data points as a group; When the data volume is 1,600, a time series signal with 100 consecutive geomagnetic signal data points as a group.

4. The deep learning enhancement method for geomagnetic signal data characteristics according to claim 1, characterized in that, In the data feature extraction in the feature extraction step, a random forest algorithm or a deep learning algorithm is used, and it is judged whether the time series signal needs to add noise and interference according to the obtained feature vector or data distribution histogram.

5. The deep learning enhancement method for geomagnetic signal data characteristics according to claim 1, characterized in that The non-linear classification by the deep learning neural network includes: Performing binary classification, ternary classification or multi-classification processing on the vector classification of the geomagnetic signal features; Using deep learning DNN for processing to obtain a trained deep learning classification model; Classifying the newly input vector through the trained deep learning classification model.

6. The deep learning enhancement method for geomagnetic signal data characteristics according to claim 1, wherein The method further includes: Inputting the data enhancement result of the geomagnetic field features into a fully convolutional neural network, where the fully convolutional neural network includes: Using a 1×1×1 convolutional kernel to extract data channel features, enhancing channel dimension features, and adding a Sigmoid activation layer after the convolutional layer; Performing a rotation transformation in the two-dimensional plane and using max pooling for two-dimensional dimensionality reduction; Performing batch normalization on the data and setting a feature mapping layer; Performing downsampling and high-density feature enhancement in the target area and classifying through a fully connected layer.

Images