Data processing method for high-power electromagnetic detection system

Abstract

The present application relates to the field of data processing, and specifically discloses a data processing method for a high-power electromagnetic detection system, which extracts multiple noise-reduced alternating electromagnetic field echo signals from multiple alternating electromagnetic field echo signals with different frequencies returned from crisis mines by adopting an automatic codec based on a deep neural network model of deep learning, and then extracts multiple observation values ​​from the multiple noise-reduced alternating electromagnetic field echo signals. The Clip model is used to complete the feature mining and fusion of the time domain and frequency domain of the noise-reduced alternating electromagnetic field echo signals, and further uses a convolutional neural network model with a three-dimensional convolution kernel to mine the correlation between the features of the multiple alternating electromagnetic field echo signals with different frequencies, so as to improve the accuracy of electromagnetic detection, thereby saving a lot of time and material resources in solving deep mineral exploration in crisis mines.

Description

Technical Field

[0001] The present application relates to the field of data processing, and more specifically, to a data processing method for a high-power electromagnetic detection system. Background Art

[0002] In recent years, my country's deep mineral exploration technology has developed rapidly, and the technical requirements for deep mineral exploration in various exploration projects have been increasing. Faced with the complex electromagnetic interference and deep detection depth requirements of mine exploration, the traditional electromagnetic method, which has been widely used in the past, has obvious shortcomings. First, most mines have complex electromagnetic interference caused by high-voltage lines, industrial and mining facilities, mining operations, underground pipelines, etc., making it difficult for traditional electromagnetic methods to obtain observation data with high signal-to-noise ratio in strong interference environments. Second, the low-frequency data obtained by traditional electromagnetic methods (reflecting deep structural information) are significantly affected by background electromagnetic field fluctuations, and the effective observation depth is relatively shallow.

[0003] After multiple rounds of mineral surveys, my country's shallow and surface large and medium-sized mines have been largely identified. With rapid socioeconomic development, resource consumption is increasing. Of the 415 large and medium-sized mines currently producing 25 major metal ores, 192 (46.2%) face varying degrees of resource crises. Deep prospecting in crisis mines is the primary goal for most mines in my country to increase reserves. However, crisis mines often feature frequent and intense human activity and complex electromagnetic environments. Deep prospecting in crisis mines, subject to strong electromagnetic interference, presents a major challenge in electromagnetic exploration. High-power electromagnetic detection systems, with their significant advantages in interference resistance and extended detection depth, are a key solution for deep prospecting in crisis mines. Currently, there is an increasingly urgent demand for high-power electromagnetic detection systems in China, particularly for instruments and equipment compatible with multiple electromagnetic detection methods in both the frequency and time domains. This represents a key development direction for electromagnetic exploration instruments and equipment.

[0004] Therefore, an optimized data processing solution for high-power electromagnetic detection systems is desired. Summary of the Invention

[0005] In order to solve the above technical problems, the present application is proposed. The embodiment of the present application provides a data processing method for a high-power electromagnetic detection system, which extracts multiple noise-reduced alternating electromagnetic field echo signals from multiple alternating electromagnetic field echo signals with different frequencies returned from crisis mines by adopting an automatic codec based on a deep neural network model of deep learning, and then extracts multiple observation values ​​from the multiple noise-reduced alternating electromagnetic field echo signals. The Clip model is used to complete the feature mining and fusion of the time domain and frequency domain of the noise-reduced alternating electromagnetic field echo signals, and further uses a convolutional neural network model with a three-dimensional convolution kernel to mine the correlation between the features of the multiple alternating electromagnetic field echo signals with different frequencies, so as to improve the accuracy of electromagnetic detection, thereby saving a lot of time and material resources in solving deep mineral exploration in crisis mines.

[0006] According to one aspect of the present application, a high-power electromagnetic detection system data processing method is provided, which includes:

[0007] Acquire multiple alternating electromagnetic field echo signals returned from the crisis mine to be detected, wherein the multiple alternating electromagnetic field echo signals have different frequencies;

[0008] Passing the plurality of alternating electromagnetic field echo signals through a noise reduction module based on an automatic codec to obtain a plurality of noise-reduced alternating electromagnetic field echo signals;

[0009] Extracting a plurality of observation values ​​of the plurality of noise-reduced alternating electromagnetic field echo signals respectively, wherein the plurality of observation values ​​include orthogonal electric and magnetic field components and impedance phase difference;

[0010] The waveform diagram of each of the noise-reduced alternating electromagnetic field echo signals and a plurality of observation values ​​of each of the noise-reduced alternating electromagnetic field echo signals are respectively passed through a Clip model to obtain a plurality of alternating magnetic field feature matrices, wherein the Clip model includes a parallel sequence encoder and an image encoder, and an image coding optimization module connected to the sequence encoder and the image encoder;

[0011] Aggregating the multiple alternating magnetic field feature matrices along the sample dimension into a three-dimensional input tensor and then obtaining a correlation feature map by using a convolutional neural network model with a three-dimensional convolution kernel;

[0012] Performing feature distribution modulation on the correlation feature map to obtain a classification feature map; and

[0013] The classification feature map is passed through a classifier to obtain a classification result, and the classification result is used to indicate whether there is a mineral deposit in the crisis mine to be detected.

[0014] In the above-mentioned high-power electromagnetic detection system data processing method, the multiple alternating electromagnetic field echo signals are passed through a noise reduction module based on an automatic codec to obtain multiple noise-reduced alternating electromagnetic field echo signals, including: using the encoder of the automatic encoder to extract multiple alternating electromagnetic field echo signal features from the multiple alternating electromagnetic field echo signals, wherein the encoder of the automatic encoder is a convolutional layer; and using the decoder of the automatic encoder to decode the multiple alternating electromagnetic field echo signal features to obtain the multiple noise-reduced alternating electromagnetic field echo signals, wherein the decoder of the automatic encoder is a deconvolution layer.

[0015] In the above-mentioned high-power electromagnetic detection system data processing method, the encoder of the automatic codec includes at least one convolution layer, and the decoder of the automatic codec includes at least one deconvolution layer. In the above-mentioned high-power electromagnetic detection system data processing method, the waveform diagram of each of the noise-reduced alternating electromagnetic field echo signals and the multiple observation values ​​of each of the noise-reduced alternating electromagnetic field echo signals are respectively passed through the Clip model to obtain multiple alternating magnetic field feature matrices, including: using the image encoder of the Clip model to perform deep convolution encoding on the waveform diagram of each of the noise-reduced alternating electromagnetic field echo signals to obtain a waveform feature vector; using the sequence encoder of the Clip model to perform multi-scale one-dimensional convolution encoding on the multiple observation values ​​of each of the noise-reduced alternating electromagnetic field echo signals to obtain an observation value feature vector; using the image coding optimization module of the Clip model to perform coding optimization on the waveform feature vector based on the observation value feature vector to obtain the alternating magnetic field feature matrix.

[0016] In the above-mentioned high-power electromagnetic detection system data processing method, the image encoder is a convolutional neural network model serving as a filter, and the sequence encoder is a multi-scale neighborhood feature extraction module, wherein the multi-scale neighborhood feature extraction module includes a first convolutional layer and a second convolutional layer connected in parallel, and a multi-scale feature fusion layer connected to the first convolutional layer and the second convolutional layer.

[0017] In the above-mentioned high-power electromagnetic detection system data processing method, the image coding optimization module using the Clip model performs coding optimization on the waveform feature vector based on the observation value feature vector to obtain the alternating magnetic field feature matrix, including: using the image coding optimization module using the Clip model to perform coding optimization on the waveform feature vector based on the observation value feature vector to obtain the alternating magnetic field feature matrix; wherein the formula is:

[0018]

[0019] Where V s represents the observed value feature vector, represents the transposed vector of the observed value feature vector, V represents the waveform feature vector, M b represents the alternating magnetic field characteristic matrix, Represents vector multiplication.

[0020] In the above-mentioned high-power electromagnetic detection system data processing method, the multiple alternating magnetic field feature matrices are aggregated into a three-dimensional input tensor along the sample dimension and then a convolutional neural network model using a three-dimensional convolution kernel is used to obtain an associated feature map, including: using each layer of the convolutional neural network model using the three-dimensional convolution kernel to perform the following on the input data in the forward pass of the layer: convolution processing on the input data to obtain a convolution feature map; pooling the convolution feature map based on the local feature matrix to obtain a pooled feature map; and nonlinear activation on the pooled feature map to obtain an activation feature map; wherein the output of the last layer of the convolutional neural network using the three-dimensional convolution kernel is the associated feature map, and the input of the first layer of the convolutional neural network using the three-dimensional convolution kernel is the three-dimensional input tensor.

[0021] In the above-mentioned high-power electromagnetic detection system data processing method, the step of performing feature distribution modulation on the correlation feature map to obtain the classification feature map includes: flattening the correlation feature map using the following formula to obtain the classification feature map; wherein the formula is:

[0022]

[0023] where f i is the predetermined characteristic value of the associated characteristic graph, f j is another feature value other than the predetermined feature value of the associated feature map, is the mean of all eigenvalues ​​of the correlation feature graph, and N is the scale of the correlation feature graph, exp(·) represents the exponential operation of a value, and the exponential operation of the value represents the calculation of the natural exponential function value with the value as the power, f i ' is the feature value of the i-th position of the classification feature map.

[0024] In the above-mentioned high-power electromagnetic detection system data processing method, the classification feature map is passed through a classifier to obtain a classification result, and the classification result is used to indicate whether there are mineral deposits in the crisis mine to be detected, including: expanding the classification feature map into a classification feature vector based on a row vector or a column vector; using multiple fully connected layers of the classifier to fully connect encode the classification feature vector to obtain an encoded classification feature vector; and passing the encoded classification feature vector through the Softmax classification function of the classifier to obtain the classification result.

[0025] According to another aspect of the present application, a high-power electromagnetic detection system data processing system is provided, comprising:

[0026] A signal acquisition module is used to acquire multiple alternating electromagnetic field echo signals returned from the crisis mine to be detected, wherein the multiple alternating electromagnetic field echo signals have different frequencies;

[0027] a noise reduction module, configured to pass the plurality of alternating electromagnetic field echo signals through a noise reduction module based on an automatic codec to obtain a plurality of noise-reduced alternating electromagnetic field echo signals;

[0028] An observation value extraction module is used to respectively extract multiple observation values ​​of the multiple noise-reduced alternating electromagnetic field echo signals, wherein the multiple observation values ​​include orthogonal electric and magnetic field components and impedance phase difference;

[0029] an alternating magnetic field characteristic matrix generation module, configured to pass the waveform diagram of each of the noise-reduced alternating electromagnetic field echo signals and a plurality of observation values ​​of each of the noise-reduced alternating electromagnetic field echo signals through a Clip model to obtain a plurality of alternating magnetic field characteristic matrices, wherein the Clip model includes a parallel sequence encoder and an image encoder, and an image coding optimization module connected to the sequence encoder and the image encoder;

[0030] A convolution module is configured to aggregate the multiple alternating magnetic field feature matrices into a three-dimensional input tensor along a sample dimension and then obtain a correlation feature map by using a convolutional neural network model with a three-dimensional convolution kernel;

[0031] a feature distribution modulation module, configured to perform feature distribution modulation on the associated feature map to obtain a classification feature map; and

[0032] The classification result generating module is used to pass the classification feature map through a classifier to obtain a classification result, and the classification result is used to indicate whether there is a mineral deposit in the crisis mine to be detected.

[0033] According to another aspect of the present application, an electronic device is provided, comprising: a processor; and a memory, wherein computer program instructions are stored in the memory, and when the computer program instructions are executed by the processor, the processor executes the high-power electromagnetic detection system data processing method as described above.

[0034] According to another aspect of the present application, a computer-readable medium is provided, on which computer program instructions are stored. When the computer program instructions are executed by a processor, the processor executes the high-power electromagnetic detection system data processing method as described above.

[0035] Compared with the existing technology, the present application provides a data processing method for a high-power electromagnetic detection system, which extracts multiple noise-reduced alternating electromagnetic field echo signals from multiple alternating electromagnetic field echo signals with different frequencies returned from crisis mines by adopting an automatic codec based on a deep neural network model based on deep learning, and then extracts multiple observation values ​​from the multiple noise-reduced alternating electromagnetic field echo signals. The Clip model is used to complete the time domain and frequency domain feature mining and fusion of the noise-reduced alternating electromagnetic field echo signals, and further uses a convolutional neural network model with a three-dimensional convolution kernel to mine the correlation between the characteristics of the multiple alternating electromagnetic field echo signals with different frequencies, so as to improve the accuracy of electromagnetic detection, thereby saving a lot of time and material resources in solving deep mineral exploration in crisis mines. BRIEF DESCRIPTION OF THE DRAWINGS

[0036] The above and other purposes, features, and advantages of the present application will become more apparent through a more detailed description of the embodiments of the present application in conjunction with the accompanying drawings. The accompanying drawings are intended to provide a further understanding of the embodiments of the present application and constitute a part of the specification. Together with the embodiments of the present application, they are used to explain the present application and do not constitute a limitation of the present application. In the drawings, the same reference numerals generally represent the same components or steps.

[0037] Figure 1 This is a diagram of an application scenario of a high-power electromagnetic detection system data processing method according to an embodiment of the present application;

[0038] Figure 2 Flowchart of a data processing method for a high-power electromagnetic detection system according to an embodiment of the present application;

[0039] Figure 3 Schematic diagram of the architecture of a data processing method for a high-power electromagnetic detection system according to an embodiment of the present application;

[0040] Figure 4 Flowchart of the noise reduction process in the data processing method of the high-power electromagnetic detection system according to an embodiment of the present application;

[0041] Figure 5 This is a flowchart of Clip model encoding in the data processing method of a high-power electromagnetic detection system according to an embodiment of the present application;

[0042] Figure 6 This is a flowchart of convolutional neural network encoding in a high-power electromagnetic detection system data processing method according to an embodiment of the present application;

[0043] Figure 7 Flowchart of a classification process in a data processing method for a high-power electromagnetic detection system according to an embodiment of the present application;

[0044] Figure 8is a block diagram of a data processing system for a high-power electromagnetic detection system according to an embodiment of the present application;

[0045] Figure 9 is a block diagram of an electronic device according to an embodiment of the present application. DETAILED DESCRIPTION

[0046] Below, the exemplary embodiments according to the present application will be described in detail with reference to the accompanying drawings. Obviously, the described embodiments are only part of the embodiments of the present application, rather than all the embodiments of the present application, and it should be understood that the present application is not limited to the exemplary embodiments described herein.

[0047] Scenario Overview

[0048] As mentioned in the background, the demand for high-power electromagnetic detection systems is increasingly urgent in the domestic electromagnetic exploration field. In particular, there is an urgent need for instruments and equipment that can accommodate multiple electromagnetic detection methods in both the frequency and time domains. This is also the main direction of development for electromagnetic exploration equipment. Therefore, an optimized data processing solution for high-power electromagnetic detection systems is desired.

[0049] Correspondingly, the CSAMT (Controlled Source Audio Frequency Magnetotelluric Method) method uses a manually controllable excitation field source to transmit alternating electromagnetic fields of varying frequencies into the earth. The observation location is located at a considerable distance from the field source, generally greater than three to five times the exploration depth (depending on the target exploration depth and the observation device used). By observing the orthogonal electric and magnetic field components of different frequencies and their impedance phase differences, the apparent resistivity at different frequencies is calculated. Because electromagnetic fields of varying frequencies have different skin depths, higher frequencies have shallower skin depths, and vice versa. Skin depth is closely related to the conductivity of the underground geological body: the greater the conductivity, the shallower the skin depth, and vice versa. Therefore, the apparent resistivity and phase at different frequencies reflect geoelectric information at different depths.

[0050] Based on this, in the technical solution of the present application, multiple alternating electromagnetic field echo signals with different frequencies returned from crisis mines are used as input data. The key lies in how to establish the correlation between the frequency domain and time domain of each alternating electromagnetic field echo signal and the correlation between the multiple alternating electromagnetic fields, so as to improve the accuracy of mineral detection in crisis mines.

[0051] In recent years, deep learning and neural networks have been widely used in fields such as computer vision, natural language processing, and text signal processing. Furthermore, deep learning and neural networks have demonstrated capabilities approaching or even surpassing those of humans in areas such as image classification, object detection, semantic segmentation, and text translation.

[0052] The development of deep learning and neural networks has provided new solutions and solutions for mining the complex correlations between the frequency domain and time domain of each alternating electromagnetic field echo signal, as well as between the multiple alternating electromagnetic fields. Those skilled in the art will appreciate that a deep neural network model based on deep learning can be trained using appropriate strategies, such as a gradient descent backpropagation algorithm, to adjust the parameters of the deep neural network model so that it can simulate complex nonlinear correlations between objects. This is clearly suitable for simulating and establishing the complex correlations between the frequency domain and time domain of each alternating electromagnetic field echo signal, as well as between the multiple alternating electromagnetic fields.

[0053] Specifically, in the technical solution of the present application, first, a plurality of alternating electromagnetic field echo signals returned from the crisis mine to be detected are obtained, and the plurality of alternating electromagnetic field echo signals have different frequencies. Next, considering that during the acquisition process of the plurality of alternating electromagnetic field echo signals, the electromagnetic interference around the crisis mine to be detected will cause noise in the acquired alternating electromagnetic field echo signals, thereby affecting the feature extraction of the alternating electromagnetic field echo signals and reducing the accuracy of electromagnetic detection. Therefore, in the technical solution of the present application, the plurality of alternating electromagnetic field echo signals are further subjected to electromagnetic signal denoising in a noise reduction module based on an automatic codec to obtain a plurality of denoised alternating electromagnetic field echo signals. In particular, here, the automatic encoder includes an encoder and a decoder, wherein the encoder of the automatic codec includes at least one convolutional layer, and the decoder of the automatic codec includes at least one deconvolution layer.

[0054] Then, considering that for the multiple noise-reduced alternating electromagnetic field echo signals, since the alternating electromagnetic field echo signals are time-domain signals, although the time-domain signals are more intuitive in terms of feature dominance in time correlation, the alternating electromagnetic field echo signals of the crisis mine are relatively weak and are easily interfered with by external electromagnetic fields, resulting in low feature extraction accuracy for the alternating electromagnetic field echo signals, which in turn affects the accuracy of high-power electromagnetic detection of crisis mines. However, the characteristics of frequency domain signals are different from those of time domain signals. Converting the alternating electromagnetic field echo signals to the frequency domain can determine whether a crisis mine has mineral deposits through the implicit feature distribution information of the alternating electromagnetic field echo signals in the frequency domain. However, the feature dominance of the alternating electromagnetic field echo signals is not intuitive and ignores the temporal correlation characteristics. Therefore, in the technical solution of the present application, a combination of the implicit features of the alternating electromagnetic field echo signals in the time domain and the frequency domain is used to detect and determine whether a crisis mine has mineral deposits.

[0055] Specifically, considering that the denoised alternating electromagnetic field echo signal contains more characteristic information and these characteristic information are correlated, when performing frequency domain feature extraction of the signal, in order to fully mine the correlated characteristic information of the characteristic distribution of the denoised alternating electromagnetic field echo signal in the frequency domain, so as to improve the accuracy of detecting whether there are mineral deposits in the crisis mine, multiple observation values ​​of the multiple denoised alternating electromagnetic field echo signals are extracted respectively, and the multiple observation values ​​include orthogonal electric and magnetic field components and impedance phase difference.

[0056] Then, in order to further improve the feature extraction of the denoised alternating electromagnetic field echo signal and improve the accuracy of mineral deposit detection in crisis mines, the Clip model is further used to complete the time domain and frequency domain feature mining and fusion of the denoised alternating electromagnetic field echo signal. Specifically, the waveform diagram of each denoised alternating electromagnetic field echo signal and multiple observation values ​​of each denoised alternating electromagnetic field echo signal are respectively passed through the Clip model to obtain multiple alternating magnetic field feature matrices. Here, the Clip model includes a parallel sequence encoder and image encoder, as well as an image coding optimization module connected to the sequence encoder and the image encoder.

[0057] That is, the waveforms of the denoised alternating electromagnetic field echo signals are subjected to deep convolutional encoding in the image encoder of the Clip model, so that the convolutional neural network model of the image encoder, which acts as a filter, extracts the time-domain implicit feature distribution information of the waveforms of the denoised alternating electromagnetic field echo signals, thereby obtaining multiple waveform feature vectors. Subsequently, the multiple observation values ​​of the denoised alternating electromagnetic field echo signals are subjected to multi-scale one-dimensional convolutional encoding in the sequence encoder of the Clip model to obtain multiple observation value feature vectors. In particular, here, the sequence encoder uses a multi-scale neighborhood feature extraction module to perform feature mining on the multiple observation values ​​to extract the correlation feature information of different scales between the multiple observation values, that is, the frequency-domain multi-scale correlation feature distribution information of the denoised alternating electromagnetic field echo signals, thereby obtaining multiple observation value feature vectors. It is worth mentioning that the multi-scale neighborhood feature extraction module includes a first convolutional layer and a second convolutional layer connected in parallel, and a multi-scale feature fusion layer connected to the first convolutional layer and the second convolutional layer.

[0058] Furthermore, the image coding optimization module of the Clip model is used to perform coding optimization on the multiple waveform feature vectors based on the multiple observation value feature vectors to obtain the multiple alternating magnetic field feature matrices. It should be understood that here, the image coding optimization module of the Clip model is used to perform joint coding optimization of the time domain features and frequency domain features of the alternating electromagnetic field echo signal, so as to perform feature optimization expression of the time domain feature distribution of the alternating electromagnetic field echo signal based on the frequency domain feature distribution of the alternating electromagnetic field echo signal, thereby obtaining the multiple alternating magnetic field feature matrices.

[0059] Then, in order to mine the correlation between the characteristics of the multiple alternating electromagnetic field echo signals with different frequencies to improve the accuracy of electromagnetic detection, the multiple alternating magnetic field feature matrices are further aggregated into a three-dimensional input tensor along the sample dimension, and then feature mining is performed in a convolutional neural network model using a three-dimensional convolution kernel to extract the correlation feature distribution information between the multiple alternating electromagnetic field echo signal features in the three-dimensional input tensor to obtain a correlation feature graph. Then, the correlation feature graph is used as a classification feature graph to perform classification processing in a classifier to obtain a classification result for indicating whether there are mineral deposits in the crisis mine to be detected. In this way, the mineral deposits of the crisis mine can be accurately detected intelligently, so as to save a lot of time and material resources when solving deep mineral exploration in crisis mines.

[0060] In particular, in the technical solution of the present application, since each alternating magnetic field feature matrix is ​​obtained by using the Clip model to obtain the waveform diagram of the denoised alternating electromagnetic field echo signal and the observation value of the denoised alternating electromagnetic field echo signal, it collectively expresses the signal waveform semantic coding features and the observation value association coding features. By aggregating the multiple alternating magnetic field feature matrices into a three-dimensional input tensor along the sample dimension and then obtaining the association feature graph using a convolutional neural network model with a three-dimensional convolution kernel, the association feature graph can fully express the cross-correlation features of the signal waveform semantic coding features and the observation value association coding features in the intra-sample dimension and the cross-sample dimension.

[0061] However, since the feature distribution of the correlation feature graph needs to include feature distributions expressing cross-correlation features in multiple different dimensions, after the correlation feature graph is expanded into a feature vector in the classifier, the fitting burden between the correlation feature graph and the weight matrix of the classifier is heavy, thereby affecting the training speed of the classifier and the accuracy of the classification results.

[0062] Therefore, the applicant of this application flattens the class representation of the associated feature graph, which is specifically expressed as follows:

[0063]

[0064] fi is the predetermined characteristic value of the associated characteristic graph, f j is another feature value other than the predetermined feature value of the associated feature map, is the mean of all eigenvalues ​​of the association feature map, and N is the scale of the association feature map, that is, width times height times number of channels.

[0065] Here, the class representation flattening of the associated feature graph will flatten the finite polyhedron manifold used for the class representation of the feature distribution in the high-dimensional feature space, while maintaining the inherent distance between the planes of the manifold and avoiding intersection based on spatial intuition. In essence, it is to decompose the finite polyhedron manifold into a cubic lattice based on the intersection of right-angled faces and the intersection of the same vertices, so as to obtain the flat "slice" continuity of the class plane to enhance the fitting performance of the associated feature graph for the weight matrix of the classifier. In this way, the training speed of the associated feature graph for classification by the classifier and the accuracy of the classification results are improved. In this way, the mineral deposits of the crisis mines can be accurately detected intelligently, so as to save a lot of time and material resources when solving the deep prospecting of crisis mines.

[0066] Based on this, the present application proposes a data processing method for a high-power electromagnetic detection system, which includes: obtaining multiple alternating electromagnetic field echo signals returned from a crisis mine to be detected, wherein the multiple alternating electromagnetic field echo signals have different frequencies; passing the multiple alternating electromagnetic field echo signals through a noise reduction module based on an automatic codec to obtain multiple noise-reduced alternating electromagnetic field echo signals; extracting multiple observation values ​​of the multiple noise-reduced alternating electromagnetic field echo signals respectively, wherein the multiple observation values ​​include orthogonal electric and magnetic field components and impedance phase differences; and generating a waveform diagram of each of the noise-reduced alternating electromagnetic field echo signals and a multiple waveform diagram of each of the noise-reduced alternating electromagnetic field echo signals. The observation values ​​are respectively passed through the Clip model to obtain multiple alternating magnetic field feature matrices, wherein the Clip model includes a parallel sequence encoder and an image encoder, and an image coding optimization module connected to the sequence encoder and the image encoder; the multiple alternating magnetic field feature matrices are aggregated into a three-dimensional input tensor along the sample dimension and then a convolutional neural network model with a three-dimensional convolution kernel is used to obtain an associated feature map; the feature distribution of the associated feature map is modulated to obtain a classification feature map; and the classification feature map is passed through a classifier to obtain a classification result, and the classification result is used to indicate whether there is a mineral deposit in the crisis mine to be detected.

[0067] Figure 1 FIG is an application scenario diagram of the data processing method for a high-power electromagnetic detection system according to an embodiment of the present application. Figure 1 As shown, in this application scenario, a signal sensor (e.g., Figure 1The method shown in S1) obtains multiple alternating electromagnetic field echo signals returned from the mine to be detected, wherein the multiple alternating electromagnetic field echo signals have different frequencies. Then, the above signals are input to a server (for example, a server) that is equipped with a data processing algorithm for a high-power electromagnetic detection system. Figure 1 In S2), the server can process the above-mentioned input signal using the high-power electromagnetic detection system data processing algorithm to generate a classification result indicating whether there is a mineral deposit in the crisis mine to be detected.

[0068] After introducing the basic principles of the present application, various non-limiting embodiments of the present application will be described in detail with reference to the accompanying drawings.

[0069] Exemplary Methods

[0070] Figure 2 FIG. 1 is a flow chart of a data processing method for a high-power electromagnetic detection system according to an embodiment of the present application. Figure 2 As shown, the data processing method of the high-power electromagnetic detection system according to the embodiment of the present application includes the following steps: S110, obtaining a plurality of alternating electromagnetic field echo signals returned from the crisis mine to be detected, wherein the plurality of alternating electromagnetic field echo signals have different frequencies; S120, passing the plurality of alternating electromagnetic field echo signals through a noise reduction module based on an automatic codec to obtain a plurality of noise-reduced alternating electromagnetic field echo signals; S130, respectively extracting a plurality of observation values ​​of the plurality of noise-reduced alternating electromagnetic field echo signals, wherein the plurality of observation values ​​include orthogonal electric and magnetic field components and impedance phase difference; S140, generating a waveform diagram of each of the noise-reduced alternating electromagnetic field echo signals and a waveform diagram of each of the noise-reduced alternating electromagnetic field echo signals. Multiple observation values ​​of the signal are respectively passed through the Clip model to obtain multiple alternating magnetic field feature matrices, wherein the Clip model includes a parallel sequence encoder and an image encoder, and an image coding optimization module connected to the sequence encoder and the image encoder; S150, the multiple alternating magnetic field feature matrices are aggregated into a three-dimensional input tensor along the sample dimension and then a convolutional neural network model with a three-dimensional convolution kernel is used to obtain an associated feature map; S160, the associated feature map is subjected to feature distribution modulation to obtain a classification feature map; and, S170, the classification feature map is passed through a classifier to obtain a classification result, and the classification result is used to indicate whether there is a mineral deposit in the crisis mine to be detected.

[0071] Figure 3 FIG. 1 is a schematic diagram of the architecture of a data processing method for a high-power electromagnetic detection system according to an embodiment of the present application. Figure 3As shown in the network structure, first, multiple alternating electromagnetic field echo signals returned from the crisis mine to be detected are obtained, and the multiple alternating electromagnetic field echo signals have different frequencies; then, the multiple alternating electromagnetic field echo signals are passed through a noise reduction module based on an automatic codec to obtain multiple noise-reduced alternating electromagnetic field echo signals; multiple observation values ​​of the multiple noise-reduced alternating electromagnetic field echo signals are extracted respectively, and the multiple observation values ​​include orthogonal electric and magnetic field components and impedance phase differences; then, the waveform diagram of each noise-reduced alternating electromagnetic field echo signal and the multiple observation values ​​of each noise-reduced alternating electromagnetic field echo signal are respectively obtained through C A Clip model is used to obtain multiple alternating magnetic field feature matrices, wherein the Clip model includes a parallel sequence encoder and an image encoder, and an image coding optimization module connected to the sequence encoder and the image encoder; then, the multiple alternating magnetic field feature matrices are aggregated into a three-dimensional input tensor along the sample dimension and then a convolutional neural network model with a three-dimensional convolution kernel is used to obtain an associated feature map; the associated feature map is subjected to feature distribution modulation to obtain a classification feature map; and further, the classification feature map is passed through a classifier to obtain a classification result, and the classification result is used to indicate whether there is a mineral deposit in the crisis mine to be detected.

[0072] Specifically, in step S110, multiple alternating electromagnetic field echo signals are obtained from the mine under investigation, each of which has different frequencies. It should be understood that the CSAMT (Controlled Source Audio Frequency Magnetotelluric Method) method uses a manually controllable excitation field source to transmit alternating electromagnetic fields of different frequencies to the earth. The observation location is located far away from the field source, generally greater than 3 to 5 times the exploration depth (depending on the target exploration depth and the observation device used). By observing the orthogonal electric and magnetic field components of different frequencies and their impedance phase difference, the apparent resistivity at different frequencies is calculated. Because electromagnetic fields of different frequencies have different skin depths, the higher the frequency, the shallower the skin depth, and vice versa. The skin depth is closely related to the conductivity of the underground geological body: the better the conductivity, the shallower the skin depth, and vice versa. Therefore, the apparent resistivity and phase of different frequencies reflect the geoelectric information at different depths. Therefore, in the technical solution of the present application, a plurality of alternating electromagnetic field echo signals returned from the crisis mine to be detected can be acquired by a signal sensor in the alternating electromagnetic field, and the plurality of alternating electromagnetic field echo signals have different frequencies.

[0073] Specifically, in step S120, the multiple alternating electromagnetic field echo signals are passed through a noise reduction module based on an automatic codec to obtain multiple noise-reduced alternating electromagnetic field echo signals. Considering that during the acquisition process of the multiple alternating electromagnetic field echo signals, the electromagnetic interference around the crisis mine to be detected will cause noise in the acquired alternating electromagnetic field echo signals, thereby affecting the feature extraction of the alternating electromagnetic field echo signals and reducing the accuracy of electromagnetic detection. Therefore, in the technical solution of the present application, the multiple alternating electromagnetic field echo signals are further subjected to electromagnetic signal noise reduction in a noise reduction module based on an automatic codec to obtain multiple noise-reduced alternating electromagnetic field echo signals. In particular, here, the automatic encoder is a three-layer feedforward neural network composed of an encoder and a decoder, which belongs to an unsupervised learning method, wherein the encoder of the automatic codec includes at least one convolutional layer, and the decoder of the automatic codec includes at least one deconvolution layer. The encoder maps the input data from a high-dimensional space to a low-dimensional feature space to achieve compressed representation of the input data and extract feature vectors. At the same time, the decoder reconstructs as many low-dimensional features of the input data as possible.

[0074] Figure 4 FIG. 1 is a flow chart of the noise reduction process in the data processing method of the high-power electromagnetic detection system according to an embodiment of the present application. Figure 4 As shown, in the noise reduction process, it includes: S210, using the encoder of the autoencoder to extract multiple alternating electromagnetic field echo signal features from the multiple alternating electromagnetic field echo signals, wherein the encoder of the autoencoder is a convolution layer; and, S220, using the decoder of the autoencoder to decode the multiple alternating electromagnetic field echo signal features to obtain the multiple noise-reduced alternating electromagnetic field echo signals, wherein the decoder of the autoencoder is a deconvolution layer.

[0075] Specifically, in step S130, multiple observation values ​​of the multiple noise-reduced alternating electromagnetic field echo signals are extracted, each of which includes orthogonal electric and magnetic field components and impedance phase differences. Considering that the multiple noise-reduced alternating electromagnetic field echo signals are time-domain signals, while time-domain signals are more intuitive in terms of feature dominance in time correlation, the alternating electromagnetic field echo signals of the crisis mine are relatively weak and are easily interfered with by external electromagnetic fields, resulting in low feature extraction accuracy for the alternating electromagnetic field echo signals, which in turn affects the accuracy of high-power electromagnetic detection of crisis mines. However, the characteristics of frequency domain signals are different from those of time domain signals. Converting the alternating electromagnetic field echo signals to the frequency domain can determine whether a crisis mine contains mineral deposits based on the implicit feature distribution information of the alternating electromagnetic field echo signals in the frequency domain. However, the dominance of the alternating electromagnetic field echo signals is not intuitive, and the temporal correlation characteristics are ignored. Therefore, in the technical solution of the present application, a combination of the implicit features of the alternating electromagnetic field echo signal in the time domain and the frequency domain is used to detect and judge whether there are mineral deposits in the crisis mine. Specifically, considering that there is a lot of characteristic information in the alternating electromagnetic field echo signal after noise reduction, and these characteristic information are correlated, therefore, when performing frequency domain feature extraction of the signal, in order to fully mine the associated characteristic information of the characteristic distribution of the alternating electromagnetic field echo signal in the frequency domain after noise reduction, so as to improve the accuracy of detecting whether there are mineral deposits in the crisis mine, multiple observation values ​​of the multiple alternating electromagnetic field echo signals after noise reduction are extracted respectively, and the multiple observation values ​​include orthogonal electric and magnetic field components and impedance phase difference.

[0076] Specifically, in step S140, the waveforms of each of the denoised alternating electromagnetic field echo signals and the multiple observations of each of the denoised alternating electromagnetic field echo signals are respectively passed through a Clip model to obtain multiple alternating magnetic field feature matrices. The Clip model includes a parallel sequence encoder and an image encoder, as well as an image coding optimization module connected to the sequence encoder and the image encoder. It should be understood that in order to further improve the feature extraction of the denoised alternating electromagnetic field echo signals and thus improve the accuracy of mineral deposit detection in critical mines, the Clip model is further used to perform time-domain and frequency-domain feature mining and fusion of the denoised alternating electromagnetic field echo signals. Specifically, the waveforms of each of the denoised alternating electromagnetic field echo signals and the multiple observations of each of the denoised alternating electromagnetic field echo signals are respectively passed through a Clip model to obtain multiple alternating magnetic field feature matrices. Here, the Clip model includes a parallel sequence encoder and an image encoder, as well as an image coding optimization module connected to the sequence encoder and the image encoder. That is, the waveforms of the noise-reduced alternating electromagnetic field echo signals are subjected to deep convolution coding in the image encoder of the Clip model, so that the convolutional neural network model of the image encoder, which serves as a filter, extracts time-domain implicit feature distribution information about the waveforms of the noise-reduced alternating electromagnetic field echo signals, thereby obtaining a plurality of waveform feature vectors. In other words, the waveforms of the noise-reduced alternating electromagnetic field echo signals are subjected to deep convolution coding using the image encoder of the Clip model to obtain waveform feature vectors. More specifically, each layer of the image encoder using the Clip model performs the following steps on the input data during the forward pass: convolution of the input data to obtain a convolution feature map; global mean pooling of the convolution feature map based on a feature matrix to obtain a pooled feature map; and nonlinear activation of the pooled feature map to obtain an activation feature map. The output of the last layer of the Clip model image encoder is the waveform feature vector, and the input of the first layer of the Clip model image encoder is the waveform of the denoised alternating electromagnetic field echo signal. Subsequently, the multiple observations of each denoised alternating electromagnetic field echo signal are subjected to multi-scale one-dimensional convolutional encoding in the sequence encoder of the Clip model to obtain multiple observation feature vectors. Specifically, the sequence encoder employs a multi-scale neighborhood feature extraction module to perform feature mining on the multiple observations, extracting correlation feature information at different scales between the multiple observations, namely, frequency-domain multi-scale correlation feature distribution information of each denoised alternating electromagnetic field echo signal, thereby obtaining multiple observation feature vectors.It is worth mentioning that the multi-scale neighborhood feature extraction module includes a first convolution layer and a second convolution layer in parallel, and a multi-scale feature fusion layer connected to the first convolution layer and the second convolution layer. Specifically, the multiple observation values ​​of the denoised alternating electromagnetic field echo signals are subjected to multi-scale one-dimensional convolution encoding in the sequence encoder of the Clip model to obtain multiple observation value feature vectors, including: passing the multiple observation values ​​of the denoised alternating electromagnetic field echo signals through the first convolution layer of the trained sequence encoder of the Clip model to obtain a first neighborhood scale observation value feature vector, wherein the first convolution layer has a first one-dimensional convolution kernel of a first length; passing the multiple observation values ​​of the denoised alternating electromagnetic field echo signals through the second convolution layer of the trained sequence encoder of the Clip model to obtain a second neighborhood scale observation value feature vector, wherein the second convolution layer has a second one-dimensional convolution kernel of a second length, and the first length is different from the second length; and cascading the first neighborhood scale observation value feature vector and the second neighborhood scale observation value feature vector to obtain the observation value feature vector. More specifically, the method of passing the multiple observation values ​​of the denoised alternating electromagnetic field echo signal through the first convolutional layer of the sequence encoder of the trained Clip model to obtain a first neighborhood-scale observation value feature vector includes: using the first convolutional layer of the sequence encoder of the trained Clip model to perform one-dimensional convolution encoding on the multiple observation values ​​of the denoised alternating electromagnetic field echo signal using the following formula to obtain a first neighborhood-scale observation value feature vector; wherein the formula is:.

[0077]

[0078] Wherein, a is the width of the first convolution kernel in the x direction, F(a) is the first convolution kernel parameter vector, G(xa) is the local vector matrix operated with the convolution kernel function, w is the size of the first convolution kernel, and X represents multiple observation values ​​of the denoised alternating electromagnetic field echo signal; and the method of passing the multiple observation values ​​of the denoised alternating electromagnetic field echo signal through the second convolution layer of the sequence encoder of the trained Clip model to obtain a second neighborhood-scale observation value feature vector comprises: using the second convolution layer of the sequence encoder of the trained Clip model to perform one-dimensional convolution encoding on the multiple observation values ​​of the denoised alternating electromagnetic field echo signal using the following formula to obtain the second neighborhood-scale observation value feature vector; wherein, the formula is:

[0079]

[0080] Wherein, b is the width of the second convolution kernel in the x direction, F(b) is the second convolution kernel parameter vector, G(xb) is the local vector matrix operated with the convolution kernel function, m is the size of the second convolution kernel, and X represents multiple observation values ​​of the alternating electromagnetic field echo signal after noise reduction; more specifically, the cascading of the first neighborhood scale observation value feature vector and the second neighborhood scale observation value feature vector to obtain the observation value feature vector includes: fusing the first neighborhood scale observation value feature vector and the second neighborhood scale observation value feature vector using the following formula to obtain the observation value feature vector; wherein, the formula is: V c =Concat[V1, V2], where V1 represents the first neighborhood scale observation value feature vector, V2 represents the second neighborhood scale observation value feature vector, Concat[·,·] represents the cascade function, V c Represents the observation value feature vector. Further, the image coding optimization module of the Clip model is used to perform coding optimization on the multiple waveform feature vectors based on the multiple observation value feature vectors to obtain the multiple alternating magnetic field feature matrices. It should be understood that here, the image coding optimization module of the Clip model is used to perform joint coding optimization of the time domain features and frequency domain features of the alternating electromagnetic field echo signal, so as to perform feature optimization expression on the time domain feature distribution based on the frequency domain feature distribution of the alternating electromagnetic field echo signal, thereby obtaining the multiple alternating magnetic field feature matrices. In a specific example of the present application, the image coding optimization module of the Clip model is used to perform coding optimization on the waveform feature vector based on the observation value feature vector to obtain the alternating magnetic field feature matrix; wherein, the formula is: Where V s represents the observed value feature vector, represents the transposed vector of the observed value feature vector, V represents the waveform feature vector, M b represents the alternating magnetic field characteristic matrix, Represents vector multiplication.

[0081] Figure 5 Flowchart of Clip model encoding in the data processing method of the high-power electromagnetic detection system according to the embodiment of the present application. Figure 5As shown, in the Clip model encoding process, it includes: S310, using the image encoder of the Clip model to perform deep convolution encoding on the waveform diagram of each of the noise-reduced alternating electromagnetic field echo signals to obtain a waveform feature vector; S320, using the sequence encoder of the Clip model to perform multi-scale one-dimensional convolution encoding on multiple observation values ​​of each of the noise-reduced alternating electromagnetic field echo signals to obtain an observation value feature vector; S330, using the image encoding optimization module of the Clip model to perform encoding optimization on the waveform feature vector based on the observation value feature vector to obtain the alternating magnetic field feature matrix.

[0082] Specifically, in step S150, the multiple alternating magnetic field feature matrices are aggregated along the sample dimension into a three-dimensional input tensor and then a convolutional neural network model using a three-dimensional convolution kernel is used to obtain a correlation feature map. It should be understood that in order to mine the correlation relationship between the multiple alternating electromagnetic field echo signal features with different frequencies to improve the accuracy of electromagnetic detection, the multiple alternating magnetic field feature matrices are further aggregated along the sample dimension into a three-dimensional input tensor and then feature mining is performed in a convolutional neural network model using a three-dimensional convolution kernel to extract the correlation feature distribution information between the multiple alternating electromagnetic field echo signal features in the three-dimensional input tensor to obtain the correlation feature map. Specifically, a convolution module is used to perform three-dimensional convolution encoding on the input data using a three-dimensional convolution kernel to obtain a convolution feature map; a pooling module is used to perform pooling processing on the convolution feature map to obtain a pooled feature map; and an activation module is used to perform nonlinear activation on the eigenvalues ​​of each position of the pooled feature map to obtain an associated feature map; wherein, the input of the first layer of the convolutional neural network using the three-dimensional convolution kernel is a three-dimensional input tensor aggregated by aggregating the multiple alternating magnetic field feature matrices along the sample dimension, and the output of the last layer of the convolutional neural network using the three-dimensional convolution kernel is the associated feature map.

[0083] Figure 6 FIG. 1 is a flow chart of convolutional neural network coding in a data processing method for a high-power electromagnetic detection system according to an embodiment of the present application. Figure 6 As shown, in the convolutional neural network encoding process, it includes: using each layer of the convolutional neural network model using the three-dimensional convolution kernel to perform the following on the input data in the forward pass of the layer: S410, convolution processing on the input data to obtain a convolution feature map; S420, pooling the convolution feature map based on the local feature matrix to obtain a pooled feature map; and, S430, performing nonlinear activation on the pooled feature map to obtain an activation feature map; wherein, the output of the last layer of the convolutional neural network using the three-dimensional convolution kernel is the associated feature map, and the input of the first layer of the convolutional neural network using the three-dimensional convolution kernel is the three-dimensional input tensor.

[0084] Specifically, in step S160, the correlation feature map is subjected to feature distribution modulation to obtain a classification feature map. In particular, in the technical solution of the present application, since each alternating magnetic field feature matrix is ​​obtained by using the Clip model to obtain the waveform diagram of the denoised alternating electromagnetic field echo signal and the observed value of the denoised alternating electromagnetic field echo signal, it collectively expresses the signal waveform semantic coding features and the observed value associated coding features, and by aggregating the multiple alternating magnetic field feature matrices along the sample dimension into a three-dimensional input tensor and then obtaining the correlation feature map by using a convolutional neural network model with a three-dimensional convolution kernel, the correlation feature map can fully express the cross-correlation features of the signal waveform semantic coding features and the observed value associated coding features in the sample dimension and the cross-sample dimension. However, since the feature distribution of the correlation feature map needs to include feature distributions that express cross-correlation features in multiple different dimensions, after the correlation feature map is expanded into a feature vector in the classifier, the fitting burden between the weight matrix of the classifier is heavy, thereby affecting the training speed of the classifier and the accuracy of the classification result. Therefore, the applicant of this application flattens the class representation of the associated feature graph, which is specifically expressed as follows:

[0085]

[0086] where f i is the predetermined characteristic value of the associated characteristic graph, f j is another feature value other than the predetermined feature value of the associated feature map, is the mean of all eigenvalues ​​of the correlation feature graph, and N is the scale of the correlation feature graph, exp(·) represents the exponential operation of a value, and the exponential operation of the value represents the calculation of the natural exponential function value with the value as the power, f i ' is the eigenvalue of the i-th position of the classification feature map. Here, the class representation flattening of the association feature map will flatten the finite polyhedron manifold used for the class representation of the feature distribution in the high-dimensional feature space, while maintaining the inherent distance between the planes of the manifold, and avoiding intersection based on spatial intuition. In essence, it is to decompose the finite polyhedron manifold into a cubic lattice based on the intersection of right-angled faces and the intersection of the same vertices, so as to obtain the flat "slice" continuity of the class plane, so as to enhance the fitting performance of the association feature map for the weight matrix of the classifier. In this way, the training speed of the association feature map for classification by the classifier and the accuracy of the classification results are improved. In this way, the mineral deposits of crisis mines can be accurately detected intelligently, so as to save a lot of time and material resources in solving deep mineral exploration in crisis mines.

[0087] Specifically, in step S170, the classification feature map is passed through a classifier to obtain a classification result, which is used to indicate whether the crisis mine to be detected has mineral deposits. In other words, the classification feature map is passed through a classifier for classification processing to obtain a classification result that indicates whether the crisis mine to be detected has mineral deposits. In this way, it is possible to accurately and intelligently detect the mineral deposits in crisis mines, saving a lot of time and material resources when solving deep mineral exploration in crisis mines. Specifically, the classifier includes multiple fully connected layers and a Softmax layer cascaded with the last fully connected layer of the multiple fully connected layers. In which, in the classification processing of the classifier, the classification feature map is first projected into a vector. For example, in a specific example, the classification feature map is expanded along the row vector or column vector into a classification feature vector; then, the classification feature vector is fully connected encoded multiple times using multiple fully connected layers of the classifier to obtain an encoded classification feature vector; further, the encoded classification feature vector is input into the Softmax layer of the classifier, that is, the encoded classification feature vector is classified using the Softmax classification function to obtain a first probability value of the encoded classification feature vector belonging to the existence of mineral deposits in the crisis mine to be detected and a second probability value of the encoded classification feature vector belonging to the absence of mineral deposits in the crisis mine to be detected; then, the label corresponding to the larger of the first probability value and the second probability value is determined as the classification result, that is, if the first probability value is greater than the second probability value, the classification result is that the crisis mine to be detected has mineral deposits, otherwise, the crisis mine to be detected does not have mineral deposits. More specifically, in a specific example of the present application, the step of passing the classification feature map through a classifier to obtain a classification result includes: using the classifier to process the classification feature map using the following formula to obtain a classification result, wherein the formula is: O = softmax{(W n ,B n ):…:(W1,B1)|Project(F)}, where Project(F) represents projecting the classification feature map into a vector, W1 to W n is the weight matrix of each fully connected layer, B1 to B n Represents the bias vector of each fully connected layer.

[0088] Figure 7 FIG. 1 is a flow chart of the classification process in the data processing method of the high-power electromagnetic detection system according to an embodiment of the present application. Figure 7As shown, the classification process includes: S510, expanding the classification feature map into a classification feature vector based on a row vector or a column vector; S520, using multiple fully connected layers of the classifier to fully connect encode the classification feature vector to obtain an encoded classification feature vector; and, S530, passing the encoded classification feature vector through the Softmax classification function of the classifier to obtain the classification result.

[0089] In summary, according to the embodiment of the present application, the data processing method of the high-power electromagnetic detection system is explained, which extracts multiple noise-reduced alternating electromagnetic field echo signals from multiple alternating electromagnetic field echo signals with different frequencies returned from the crisis mine by adopting an automatic codec based on a deep neural network model of deep learning, and then extracts multiple observation values ​​from the multiple noise-reduced alternating electromagnetic field echo signals. The Clip model is used to complete the feature mining and fusion of the time domain and frequency domain of the noise-reduced alternating electromagnetic field echo signals, and further uses a convolutional neural network model with a three-dimensional convolution kernel to mine the correlation between the characteristics of the multiple alternating electromagnetic field echo signals with different frequencies, so as to improve the accuracy of electromagnetic detection, thereby saving a lot of time and material resources in solving deep mineral exploration in crisis mines.

[0090] Exemplary Systems

[0091] Figure 8 FIG. 1 is a block diagram of a data processing system for a high-power electromagnetic detection system according to an embodiment of the present application. Figure 8 As shown, the high-power electromagnetic detection system data processing 300 according to the embodiment of the present application includes: a signal acquisition module 310; a noise reduction module 320; an observation value extraction module 330; an alternating magnetic field feature matrix generation module 340; a convolution module 350; a feature distribution modulation module 360; and a classification result generation module 370.

[0092] Among them, the signal acquisition module 310 is used to obtain multiple alternating electromagnetic field echo signals returned from the crisis mine to be detected, and the multiple alternating electromagnetic field echo signals have different frequencies; the noise reduction module 320 is used to pass the multiple alternating electromagnetic field echo signals through a noise reduction module based on an automatic codec to obtain multiple noise-reduced alternating electromagnetic field echo signals; the observation value extraction module 330 is used to respectively extract multiple observation values ​​of the multiple noise-reduced alternating electromagnetic field echo signals, and the multiple observation values ​​include orthogonal electric and magnetic field components and impedance phase differences; the alternating magnetic field characteristic matrix generation module 340 is used to respectively generate the waveform diagram of each of the noise-reduced alternating electromagnetic field echo signals and the multiple observation values ​​of each of the noise-reduced alternating electromagnetic field echo signals. A plurality of alternating magnetic field feature matrices are obtained through a Clip model, wherein the Clip model includes a parallel sequence encoder and an image encoder, and an image coding optimization module connected to the sequence encoder and the image encoder; the convolution module 350 is used to aggregate the plurality of alternating magnetic field feature matrices into a three-dimensional input tensor along the sample dimension and then obtain an associated feature map by using a convolutional neural network model with a three-dimensional convolution kernel; the feature distribution modulation module 360 ​​is used to perform feature distribution modulation on the associated feature map to obtain a classification feature map; and the classification result generation module 370 is used to pass the classification feature map through a classifier to obtain a classification result, and the classification result is used to indicate whether there is a mineral deposit in the crisis mine to be detected.

[0093] In one example, in the high-power electromagnetic detection system data processing system 300, the noise reduction module 320 is further configured to: extract multiple alternating electromagnetic field echo signal features from the multiple alternating electromagnetic field echo signals using the encoder of the autoencoder, wherein the encoder of the autoencoder is a convolutional layer; and decode the multiple alternating electromagnetic field echo signal features using the decoder of the autoencoder to obtain the multiple noise-reduced alternating electromagnetic field echo signals, wherein the decoder of the autoencoder is a deconvolutional layer. The encoder of the autoencoder / decoder includes at least one convolutional layer, and the decoder of the autoencoder / decoder includes at least one deconvolutional layer.

[0094] In one example, in the high-power electromagnetic detection system data processing system 300, the alternating magnetic field feature matrix generation module 340 is further configured to: use the image encoder of the Clip model to perform deep convolution encoding on the waveform graph of each of the noise-reduced alternating electromagnetic field echo signals to obtain a waveform feature vector; use the sequence encoder of the Clip model to perform multi-scale one-dimensional convolution encoding on multiple observation values ​​of each of the noise-reduced alternating electromagnetic field echo signals to obtain an observation value feature vector; and use the image encoding optimization module of the Clip model to perform encoding optimization on the waveform feature vector based on the observation value feature vector to obtain the alternating magnetic field feature matrix. The image encoder is a convolutional neural network model serving as a filter, and the sequence encoder is a multi-scale neighborhood feature extraction module, wherein the multi-scale neighborhood feature extraction module includes a first convolution layer and a second convolution layer connected in parallel, and a multi-scale feature fusion layer connected to the first convolution layer and the second convolution layer. More specifically, the image coding optimization module using the Clip model performs coding optimization on the waveform feature vector based on the observation value feature vector to obtain the alternating magnetic field feature matrix, including: using the image coding optimization module using the Clip model to perform coding optimization on the waveform feature vector based on the observation value feature vector to obtain the alternating magnetic field feature matrix; wherein the formula is:

[0095]

[0096] Where V s represents the observed value feature vector, represents the transposed vector of the observed value feature vector, V represents the waveform feature vector, M b represents the alternating magnetic field characteristic matrix, Represents vector multiplication.

[0097] In one example, in the above-mentioned high-power electromagnetic detection system data processing system 300, the convolution module 350 is further used to: use each layer of the convolutional neural network model using the three-dimensional convolution kernel to perform the following on the input data in the forward pass of the layer: convolution processing on the input data to obtain a convolution feature map; pooling based on the local feature matrix on the convolution feature map to obtain a pooled feature map; and, nonlinear activation on the pooled feature map to obtain an activation feature map; wherein the output of the last layer of the convolutional neural network using the three-dimensional convolution kernel is the correlation feature map, and the input of the first layer of the convolutional neural network using the three-dimensional convolution kernel is the three-dimensional input tensor.

[0098] In one example, in the high-power electromagnetic detection system data processing system 300, the feature distribution modulation module 360 ​​is further configured to flatten the correlation feature map using the following formula to obtain the classification feature map; wherein the formula is:

[0099]

[0100] where f i is the predetermined characteristic value of the associated characteristic graph, f j is another feature value other than the predetermined feature value of the associated feature map, is the mean of all eigenvalues ​​of the correlation feature graph, and N is the scale of the correlation feature graph, exp(·) represents the exponential operation of a value, and the exponential operation of the value represents the calculation of the natural exponential function value with the value as the power, f i ' is the feature value of the i-th position of the classification feature map.

[0101] In one example, in the above-mentioned high-power electromagnetic detection system data processing system 300, the classification result generation module 370 is further used to: expand the classification feature map into a classification feature vector based on a row vector or a column vector; use multiple fully connected layers of the classifier to fully connect encode the classification feature vector to obtain an encoded classification feature vector; and pass the encoded classification feature vector through the Softmax classification function of the classifier to obtain the classification result.

[0102] In summary, according to the embodiment of the present application, the high-power electromagnetic detection system data processing system 300 is explained, which extracts multiple noise-reduced alternating electromagnetic field echo signals from multiple alternating electromagnetic field echo signals with different frequencies returned from crisis mines by adopting an automatic codec based on a deep neural network model of deep learning, and then extracts multiple observation values ​​from the multiple noise-reduced alternating electromagnetic field echo signals. The Clip model is used to complete the time domain and frequency domain feature mining and fusion of the noise-reduced alternating electromagnetic field echo signals, and further uses a convolutional neural network model with a three-dimensional convolution kernel to mine the correlation between the characteristics of the multiple alternating electromagnetic field echo signals with different frequencies, so as to improve the accuracy of electromagnetic detection, thereby saving a lot of time and material resources in solving deep mineral exploration in crisis mines.

[0103] As described above, the high-power electromagnetic detection system data processing system according to the embodiments of the present application can be implemented in various terminal devices. In one example, the high-power electromagnetic detection system data processing system 300 according to the embodiments of the present application can be integrated into the terminal device as a software module and / or hardware module. For example, the high-power electromagnetic detection system data processing system 300 can be a software module in the terminal device's operating system, or it can be an application developed for the terminal device. Of course, the high-power electromagnetic detection system data processing system 300 can also be one of the terminal device's many hardware modules.

[0104] Alternatively, in another example, the high-power electromagnetic detection system data processing system 300 and the terminal device may also be separate devices, and the high-power electromagnetic detection system data processing system 300 may be connected to the terminal device via a wired and / or wireless network and transmit interactive information in accordance with an agreed data format.

[0105] Exemplary electronic devices

[0106] Below, reference Figure 9 To describe the electronic device according to the embodiment of the present application.

[0107] Figure 9 The figure shows a block diagram of an electronic device according to an embodiment of the present application.

[0108] like Figure 9 As shown, the electronic device 10 includes one or more processors 11 and a memory 12 .

[0109] The processor 11 may be a central processing unit (CPU) or other forms of processing units having data processing capabilities and / or instruction execution capabilities, and may control other components in the electronic device 10 to perform desired functions.

[0110] The memory 12 may include one or more computer program products, which may include various forms of computer-readable storage media, such as volatile memory and / or non-volatile memory. The volatile memory may include, for example, random access memory (RAM) and / or cache memory. The non-volatile memory may include, for example, read-only memory (ROM), a hard disk, flash memory, etc. One or more computer program instructions may be stored on the computer-readable storage medium, and the processor 11 may execute the program instructions to implement the functions of the high-power electromagnetic detection system data processing method of each embodiment of the present application described above and / or other desired functions. Various contents such as classification feature maps may also be stored in the computer-readable storage medium.

[0111] In one example, the electronic device 10 may further include an input device 13 and an output device 14 , and these components are interconnected via a bus system and / or other forms of connection mechanisms (not shown).

[0112] The input device 13 may include, for example, a keyboard, a mouse, and the like.

[0113] The output device 14 can output various information to the outside, including classification results, etc. The output device 14 can include, for example, a display, a speaker, a printer, a communication network and a remote output device connected thereto, and the like.

[0114] Of course, to simplify, Figure 9 Only some of the components related to the present application in the electronic device 10 are shown, and components such as a bus, an input / output interface, etc. are omitted. In addition, the electronic device 10 may further include any other appropriate components according to specific application scenarios.

[0115] Exemplary computer program products and computer-readable storage media

[0116] In addition to the above-mentioned methods and devices, an embodiment of the present application may also be a computer program product, which includes computer program instructions, which, when executed by a processor, enable the processor to perform the steps in the functions of the high-power electromagnetic detection system data processing method according to various embodiments of the present application described in the above "Exemplary Method" section of this specification.

[0117] The computer program product may be written in any combination of one or more programming languages ​​to implement the program code for performing the operations of the embodiments of the present application, including object-oriented programming languages ​​such as Java, C++, and conventional procedural programming languages ​​such as "C" or similar programming languages. The program code may be executed entirely on the user's computing device, partially on the user's computing device, as a standalone software package, partially on the user's computing device and partially on a remote computing device, or entirely on a remote computing device or server.

[0118] In addition, an embodiment of the present application may also be a computer-readable storage medium having computer program instructions stored thereon, which, when executed by a processor, enables the processor to execute the steps in the functions of the high-power electromagnetic detection system data processing method according to various embodiments of the present application described in the above "Exemplary Method" section of this specification.

[0119] The computer-readable storage medium can adopt any combination of one or more readable media. The readable medium can be a readable signal medium or a readable storage medium. The readable storage medium can, for example, include but is not limited to a system, device or component of electricity, magnetism, light, electromagnetic, infrared, or semiconductor, or any combination thereof. More specific examples (non-exhaustive list) of readable storage media include: an electrical connection with one or more wires, a portable disk, a hard disk, a random access memory (RAM), a read-only memory (ROM), an erasable programmable read-only memory (EPROM or flash memory), an optical fiber, a portable compact disk read-only memory (CD-ROM), an optical storage device, a magnetic storage device, or any suitable combination thereof.

[0120] The basic principles of the present application have been described above in conjunction with specific embodiments. However, it should be noted that the advantages, strengths, and effects mentioned in this application are merely illustrative and not restrictive, and it should not be assumed that these advantages, strengths, and effects are required of each embodiment of this application. In addition, the specific details disclosed above are merely illustrative and facilitating understanding, and are not restrictive. The above details do not limit this application to necessarily being implemented using the above specific details.

[0121] The block diagrams of the devices, devices, equipment, and systems involved in this application are merely illustrative examples and are not intended to require or imply that they must be connected, arranged, or configured in the manner shown in the block diagrams. As will be appreciated by those skilled in the art, these devices, devices, equipment, and systems can be connected, arranged, or configured in any manner. Words such as "include," "comprise," "have," and the like are open-ended words, meaning "including but not limited to," and can be used interchangeably therewith. The words "or" and "and" used herein refer to the words "and / or" and can be used interchangeably therewith, unless the context clearly indicates otherwise. The word "such as" used herein refers to the phrase "such as but not limited to," and can be used interchangeably therewith.

[0122] It should also be noted that in the apparatus, device, and method of the present application, each component or each step can be decomposed and / or recombined, and such decomposition and / or recombination should be regarded as equivalent solutions of the present application.

[0123] The above description of the disclosed aspects is provided to enable any person skilled in the art to make or use the present application. Various modifications to these aspects will be readily apparent to those skilled in the art, and the general principles defined herein may be applied to other aspects without departing from the scope of the present application. Therefore, the present application is not intended to be limited to the aspects shown herein, but rather to be accorded the widest scope consistent with the principles and novel features disclosed herein.

[0124] The above description has been provided for the purpose of illustration and description. Furthermore, this description is not intended to limit the embodiments of the present application to the forms disclosed herein. Although a number of example aspects and embodiments have been discussed above, those skilled in the art will recognize certain variations, modifications, alterations, additions, and sub-combinations thereof.

Claims

1. A data processing method for a high-power electromagnetic detection system, characterized in that: include: Acquire multiple alternating electromagnetic field echo signals returned from the crisis mine to be detected, wherein the multiple alternating electromagnetic field echo signals have different frequencies; Passing the plurality of alternating electromagnetic field echo signals through a noise reduction module based on an automatic codec to obtain a plurality of noise-reduced alternating electromagnetic field echo signals; Extracting a plurality of observation values ​​of the plurality of noise-reduced alternating electromagnetic field echo signals respectively, wherein the plurality of observation values ​​include orthogonal electric and magnetic field components and impedance phase difference; The waveform diagram of each of the noise-reduced alternating electromagnetic field echo signals and a plurality of observation values ​​of each of the noise-reduced alternating electromagnetic field echo signals are respectively passed through a Clip model to obtain a plurality of alternating magnetic field feature matrices, wherein the Clip model includes a parallel sequence encoder and an image encoder, and an image coding optimization module connected to the sequence encoder and the image encoder; Aggregating the multiple alternating magnetic field feature matrices along the sample dimension into a three-dimensional input tensor and then obtaining a correlation feature map by using a convolutional neural network model with a three-dimensional convolution kernel; Performing feature distribution modulation on the correlation feature map to obtain a classification feature map; as well as The classification feature map is passed through a classifier to obtain a classification result, and the classification result is used to indicate whether there is a mineral deposit in the crisis mine to be detected.

2. The data processing method of a high-power electromagnetic detection system according to claim 1, characterized in that: The step of passing the plurality of alternating electromagnetic field echo signals through a noise reduction module based on an automatic codec to obtain a plurality of noise-reduced alternating electromagnetic field echo signals comprises: Extracting a plurality of alternating electromagnetic field echo signal features from the plurality of alternating electromagnetic field echo signals using an encoder of the autoencoder, wherein the encoder of the autoencoder is a convolutional layer; and The decoder of the autoencoder is used to decode the multiple alternating electromagnetic field echo signal features to obtain the multiple noise-reduced alternating electromagnetic field echo signals, wherein the decoder of the autoencoder is a deconvolution layer.

3. The data processing method of a high-power electromagnetic detection system according to claim 2, characterized in that: The encoder of the autocodec includes at least one convolutional layer, and the decoder of the autocodec includes at least one deconvolutional layer.

4. The data processing method of a high-power electromagnetic detection system according to claim 3, characterized in that: The waveform diagrams of the noise-reduced alternating electromagnetic field echo signals and the multiple observation values ​​of the noise-reduced alternating electromagnetic field echo signals are respectively subjected to a Clip model to obtain multiple alternating magnetic field characteristic matrices, including: Using the image encoder of the Clip model, deep convolution encoding is performed on the waveform graph of each of the noise-reduced alternating electromagnetic field echo signals to obtain a waveform feature vector; Using the sequence encoder of the Clip model, multi-scale one-dimensional convolution encoding is performed on the multiple observation values ​​of each of the noise-reduced alternating electromagnetic field echo signals to obtain an observation value feature vector; The image coding optimization module of the Clip model is used to perform coding optimization on the waveform feature vector based on the observation value feature vector to obtain the alternating magnetic field feature matrix.

5. The data processing method of a high-power electromagnetic detection system according to claim 4, characterized in that: The image encoder is a convolutional neural network model as a filter, and the sequence encoder is a multi-scale neighborhood feature extraction module, wherein the multi-scale neighborhood feature extraction module includes a first convolution layer and a second convolution layer in parallel, and a multi-scale feature fusion layer connected to the first convolution layer and the second convolution layer.

6. The data processing method of a high-power electromagnetic detection system according to claim 5, characterized in that: The image coding optimization module using the Clip model performs coding optimization on the waveform feature vector based on the observation value feature vector to obtain the alternating magnetic field feature matrix, comprising: using the image coding optimization module using the Clip model to perform coding optimization on the waveform feature vector based on the observation value feature vector to obtain the alternating magnetic field feature matrix; The formula is: Where V s represents the observed value feature vector, represents the transposed vector of the observed value feature vector, V represents the waveform feature vector, M b represents the alternating magnetic field characteristic matrix, Represents vector multiplication.

7. The data processing method of a high-power electromagnetic detection system according to claim 6, characterized in that: The method comprises: aggregating the plurality of alternating magnetic field feature matrices into a three-dimensional input tensor along the sample dimension and obtaining a correlation feature map by using a convolutional neural network model using a three-dimensional convolution kernel, comprising: performing the following on the input data in the forward pass of each layer of the convolutional neural network model using the three-dimensional convolution kernel: Perform convolution on the input data to obtain a convolution feature map; Performing pooling on the convolutional feature map based on a local feature matrix to obtain a pooled feature map; and Performing nonlinear activation on the pooled feature map to obtain an activated feature map; The output of the last layer of the convolutional neural network using the three-dimensional convolution kernel is the associated feature map, and the input of the first layer of the convolutional neural network using the three-dimensional convolution kernel is the three-dimensional input tensor.

8. The data processing method of a high-power electromagnetic detection system according to claim 7, characterized in that: The performing feature distribution modulation on the association feature map to obtain a classification feature map includes: The association feature map is characterized and flattened using the following formula to obtain the classification feature map; Wherein, the formula is: where f i is the predetermined characteristic value of the associated characteristic graph, f j is another feature value other than the predetermined feature value of the associated feature map, is the mean of all eigenvalues ​​of the correlation feature graph, and N is the scale of the correlation feature graph, exp(·) represents the exponential operation of a value, and the exponential operation of the value represents the calculation of the natural exponential function value with the value as the power, f i ' is the feature value of the i-th position of the classification feature map.

9. The data processing method of a high-power electromagnetic detection system according to claim 8, characterized in that: The classification feature map is passed through a classifier to obtain a classification result, wherein the classification result is used to indicate whether there is a mineral deposit in the crisis mine to be detected, including: Expanding the classification feature map into a classification feature vector based on the row vector or the column vector; Performing full-connection encoding on the classification feature vector using multiple fully-connected layers of the classifier to obtain an encoded classification feature vector; and The encoded classification feature vector is passed through the Softmax classification function of the classifier to obtain the classification result.

Images