A data set construction method for multi-view magnetic anomaly signal intelligent identification

Abstract

This invention discloses a dataset construction method for intelligent identification of magnetic anomaly signals from multiple perspectives, relating to the field of intelligent identification technology. The method includes the following steps: collecting positive and negative sample data and performing noise reduction processing to generate a sample dataset, wherein positive samples correspond to magnetic anomaly signals and negative samples correspond to magnetic signals; performing multi-view transformation on the sample dataset to generate a dataset used for training an intelligent identification algorithm model for magnetic anomaly signals, thereby improving the effectiveness and detection rate of the trained intelligent identification algorithm model and reducing the false alarm rate and missed alarm rate. The dataset constructed by this invention fully reflects various characteristics of magnetic anomaly signals, enabling the intelligent identification algorithm to fully extract and learn the characteristics of magnetic anomaly signals from multiple perspectives and dimensions. This not only improves the effectiveness of the intelligent identification algorithm model but also increases the detection rate and reduces the false alarm rate and missed alarm rate.

Description

Technical Field

[0001] This invention relates to the field of intelligent recognition technology, and more specifically, to a dataset construction method for intelligent recognition of multi-view magnetic anomaly signals. Background Technology

[0002] The identification of magnetic anomaly signals is crucial for magnetic target detection. Currently, traditional methods for detecting magnetic anomaly signals suffer from high false alarm and false false alarm rates. To improve the detection rate of magnetic anomaly signals and reduce the false alarm and false alarm rates, many researchers are constantly exploring new detection algorithms.

[0003] In recent years, driven by deep learning and artificial intelligence technologies, intelligent inspection algorithms have achieved leapfrog development, deeply integrating into the core business processes of various industries, from industrial quality inspection to medical diagnosis, and from autonomous driving to security monitoring. They are widely used in fields such as quality inspection, equipment condition monitoring, and autonomous driving.

[0004] The implementation of intelligent detection algorithms relies on a large amount of sample data. Model training and feature extraction are used to automatically extract sample features for detection and identification. In practical applications of magnetic anomaly detection, interpreting magnetic anomaly signals often requires combining signal features from multiple perspectives, including time-frequency maps, waveform width, and peak-to-peak value, to reduce false alarm and missed alarm rates. Therefore, model training based on a single-view dataset can only learn and extract signal features from that single perspective, limiting the model's ability to detect and identify magnetic anomaly signals. Intelligent detection requires constructing a multi-view magnetic anomaly signal dataset to extract and display features from multiple dimensions, making the intelligent identification algorithm model closer to human interpretation. However, using a dataset containing only a single perspective for model training and testing results in the model only extracting signal features from that single perspective, offering little improvement in detection and identification performance. Therefore, there is an urgent need for a dataset construction method for training and testing multi-view intelligent magnetic anomaly signal identification algorithms. This method involves constructing a multi-view dataset and constraining the magnetic anomaly signal during detection from multiple perspectives, thereby improving the detection rate and reducing the false alarm rate. Summary of the Invention

[0005] To address the problems of limited dataset features and low model accuracy in existing magnetic anomaly detection algorithms, this invention aims to provide a dataset construction method for intelligent identification of magnetic anomaly signals from multiple perspectives, thereby improving model recognition performance and reducing the false alarm rate and missed alarm rate of magnetic anomaly signals.

[0006] To achieve the above technical objectives, this application provides a dataset construction method for intelligent identification of multi-view magnetic anomaly signals, comprising the following steps: Collect positive and negative sample data, perform noise reduction processing, and generate a sample dataset, where positive samples correspond to magnetic anomaly signals and negative samples correspond to magnetic signals; The sample dataset is transformed from multiple perspectives to generate a dataset for training the intelligent identification algorithm model for magnetic anomaly signals. This improves the effectiveness and detection rate of the trained intelligent identification algorithm model for magnetic anomaly signals, and reduces the false alarm rate and missed alarm rate.

[0007] Preferably, when collecting positive and negative sample data, the positive and negative sample data are collected according to the sample time length and sampling frequency, and the sample length of the negative samples and the positive samples are kept the same.

[0008] Preferably, when performing multi-view transformation to obtain the dataset, positive and negative samples are visualized as waveforms and saved as images. Specifically, the visualization processing of positive and negative samples removes redundant information and deletes the coordinate information of the X and Y axes, retaining only the coordinate axes and curves to form waveforms of magnetic anomaly signals and background noise, which are used to generate the dataset. This allows the model to learn and extract the differences between positive and negative samples in the images during the training process.

[0009] Preferably, when performing multi-view conversion to obtain the dataset, the complete magnetic anomaly signal waveform is found from the waveform diagram corresponding to the positive sample and feature extraction is performed. The waveform width, peak-to-peak value, and peak-to-peak ratio of the magnetic anomaly signal waveform are extracted and output in text format to generate the dataset, so that the model can learn and extract various features of the positive sample waveform during the training process.

[0010] Preferably, when performing multi-view transformation to obtain the dataset, positive and negative samples are converted into time-frequency maps, and the time-frequency maps are saved as images to generate the dataset. This allows the model to accurately learn and extract the difference between the time-frequency maps of magnetic anomaly signals and background noise during training. Specifically, the X and Y axis coordinate ranges of the time-frequency maps are adjusted to ensure that the frequencies of the signal concentration are fully displayed.

[0011] Preferably, when performing multi-view transformation to obtain the dataset, each positive and negative sample is linearly combined with three unrelated orthogonal basis functions of the orthogonal basis to calculate the energy value, which is used to form the dataset, so that the trained model learns and extracts the magnitude and distribution of the orthogonal basis energy values ​​corresponding to the positive and negative samples.

[0012] Preferably, when performing multi-view transformation to obtain the dataset, the orthogonal basis energy values ​​are visualized to form a waveform, which is used to form the dataset, so that the trained model can learn and extract the features in the orthogonal basis energy waveform corresponding to the positive and negative samples.

[0013] Based on the same inventive concept, this invention also discloses a dataset construction system for intelligent identification of multi-view magnetic anomaly signals, comprising: The data acquisition module is used to collect positive and negative sample data, perform noise reduction processing, and generate a sample dataset, where positive samples correspond to magnetic anomaly signals and negative samples correspond to magnetic signals. The dataset construction module is used to perform multi-view transformation on the sample dataset to generate a dataset for training the intelligent magnetic anomaly signal recognition algorithm model. This improves the effectiveness and detection rate of the trained intelligent magnetic anomaly signal recognition algorithm model and reduces the false alarm rate and missed alarm rate.

[0014] The present invention discloses the following technical effects: The dataset constructed by this invention fully reflects the various characteristics of magnetic anomaly signals, enabling intelligent identification algorithms to fully extract and learn the characteristics of magnetic anomaly signals from multiple perspectives and dimensions. This not only improves the effectiveness of the intelligent identification algorithm model but also increases the detection rate and reduces the false alarm rate and missed alarm rate. Attached Figure Description

[0015] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the drawings used in the embodiments will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0016] Figure 1 These are examples of positive and negative samples that undergo only filtering as described in this invention. (a) is a positive sample, and (b) is a negative sample. Figure 2 These are examples of waveforms for positive and negative samples as described in this invention. (a) represents a positive sample, and (b) represents a negative sample. Figure 3 This is the output of the feature extraction result of the positive sample as described in this invention; Figure 4 This is an example of a positive and negative sample time-frequency diagram according to the present invention. Wherein, (a) represents a positive sample, and (b) represents a negative sample; Figure 5 This refers to the orthogonal basis energy calculated from the positive and negative samples and orthogonal basis functions described in this invention. Where (a) represents a positive sample and (b) represents a negative sample; Figure 6 This is a graph of the orthogonal basis energy obtained by calculating the positive and negative samples and orthogonal basis functions according to the present invention. Wherein, (a) represents a positive sample, and (b) represents a negative sample; Figure 7 This is a schematic diagram of the method described in this invention. Detailed Implementation

[0017] To make the objectives, technical solutions, and advantages of the embodiments of this application clearer, the technical solutions of the embodiments of this application will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of this application, and not all embodiments. The components of the embodiments of this application described and shown in the accompanying drawings can generally be arranged and designed in various different configurations. Therefore, the following detailed description of the embodiments of this application provided in the accompanying drawings is not intended to limit the scope of the claimed application, but merely represents selected embodiments of this application. All other embodiments obtained by those skilled in the art based on the embodiments of this application without inventive effort are within the scope of protection of this application.

[0018] like Figures 1-7 As shown, this invention provides a dataset construction method for intelligent identification of multi-view magnetic anomaly signals. It belongs to a dataset construction method for training and testing intelligent identification algorithms for multi-view magnetic anomaly signals, and is mainly applicable to the construction of model training and testing datasets for intelligent identification algorithms for magnetic anomaly signals. When constructing the dataset, a basic dataset with a single feature and only filtering is first constructed. Then, starting from this dataset, the magnetic anomaly signal features in the dataset are displayed from multiple perspectives, forming a dataset containing multi-view features of magnetic anomaly signals. Specifically, the method includes the following steps: Step 1: Based on the collected data, sample selection is performed. During this process, the magnetic anomaly signal is filtered and denoised to minimize the influence of other background noise, obtaining samples of length d (where d represents the time length). If the sampling frequency is 10Hz, then the sample at this point contains 10d data points. Figure 1 As shown.

[0019] Step 2: For the samples from Step 1, visualize the samples as waveforms and save them as images in JPG or TIFF format. For example... Figure 2 As shown.

[0020] Step 3: For the samples from Step 1, visualize the samples and extract features from the magnetic anomaly signal portion of the samples, including features such as the waveform width, peak-to-peak ratio, peak width-to-width ratio, number of peaks, and center frequency. For example... Figure 3 As shown.

[0021] Step 4: For the samples in Step 1, convert them into time-frequency graphs and save the time-frequency graphs as image files, such as .jpg or tiff. Figure 4 As shown.

[0022] Step 5: For the samples in Step 1, based on the theory of orthogonal basis detection algorithm, linearly combine the samples with three uncorrelated orthogonal basis functions to calculate the energy functions, obtaining energy 1, energy 2, energy 3, and energy 2 / energy 1 and energy 3 / energy 1. For example... Figure 5 As shown.

[0023] Step 6: Visualize the five features obtained from the convolution in Step 5 and save them as images in .jpg or tiff format, etc. For example... Figure 6 As shown.

[0024] Example: This invention provides a dataset construction method for intelligent identification of multi-view magnetic anomaly signals. This invention constructs an intelligent identification algorithm dataset based on measured data. During the measurement process, data is collected through cooperation between the data acquisition platform and the target platform. However, due to the difficulty of experimental coordination and the need for substantial financial support, the amount of data collected is limited. Given this limited data, the key research objective is to effectively and efficiently utilize the data to bring greater value to subsequent applications.

[0025] Improving the detection algorithm's ability to identify magnetic anomaly signals using existing data is particularly important in the field of magnetic detection. In particular, it is crucial to use current data to construct an effective and high-quality dataset for training and testing intelligent identification algorithms and subsequent improvements based on this dataset. This will enhance the detection rate of intelligent identification algorithms and reduce false alarm and missed alarm rates.

[0026] Step 1: Construct a dataset from the collected data by combining the time of encounter between the data acquisition platform and the target. If the sampling frequency is 10Hz, extract samples of length d (where d represents the time length) based on the time when the magnetic anomaly signal appears, resulting in 10d samples. Positive samples must contain the complete magnetic anomaly waveform, while negative samples must have the same length as the positive samples. After acquiring the sample data, perform filtering and noise reduction processing. The result is as follows: Figure 1 As shown, the sample here consists of 10 data points in txt text format.

[0027] The following processing is performed on the above dataset: Step 2: Visualize the text format from Step 1, removing redundant information and deleting the X and Y axis coordinates, retaining only the axes and curves to create waveforms of the magnetic anomaly signal and background noise. Save this as a JPG or TIFF image. Perform this processing on both positive and negative samples. The results are as follows: Figure 2 As shown. During the training process, the model is made to learn and extract the differences between positive and negative samples in the image.

[0028] Step 3: Visualize the text format from Step 1. Extract the complete magnetic anomaly signal waveform from the visualized image, including its width, peak-to-peak value, and peak-to-peak ratio. Output the extracted information as a text file, where each line corresponds to a positive sample. The results are as follows: Figure 3 As shown, this step only processes positive samples, enabling the model to learn and extract various features of the positive sample waveforms during training.

[0029] Step 4: Convert the text format from Step 1 to a time-frequency graph. Adjust the X and Y axis coordinates of the image to ensure that the frequencies of the signal focus are fully displayed, and the shape is clear, complete, and not compact. Save it as a JPG or TIFF image. Perform this processing on both positive and negative samples. The result is as follows: Figure 4 As shown, this allows the model to accurately learn and extract the difference between the time-frequency plot of the magnetic anomaly signal and the time-frequency plot of the background noise during model training.

[0030] Step 5: Process the text-formatted dataset from Step 1. Linearly combine each sample with three uncorrelated orthogonal basis functions of the orthogonal basis to calculate the energy value, i.e., the square root of the sum of squares of the energy, energy 1, energy 2, energy 3, and energy 2 / energy 1, energy 3 / energy 1, in txt text format. Each column in this text corresponds to one of the features mentioned above, i.e., the original filtered data, the square root of the sum of squares of the energy, energy 1, energy 2, energy 3, and energy 2 / energy 1, energy 3 / energy 1. The results are as follows: Figure 5 As shown. This processing is required for both positive and negative samples. This step allows the model to learn and extract the numerical magnitude and distribution of the orthogonal basis energy corresponding to the positive and negative samples.

[0031] Step 6: Visualize the orthogonal basis energy values ​​from Step 5 as waveforms, i.e., energy 1 corresponds to one waveform, energy 2 corresponds to one waveform, and so on, with one sample corresponding to six waveforms. Save the results as JPG or TIFF image formats, as shown below. Figure 6 As shown in the diagram. This step enables the model to learn and extract features from the orthogonal basis energy waveforms corresponding to the positive and negative samples.

[0032] The dataset obtained through the above steps contains the characteristics of various magnetic anomaly signals involved in human interpretation. By training and testing the model in the intelligent recognition algorithm using the above dataset, the recognition ability of the model can be effectively enhanced, the detection rate of magnetic anomaly signals of the intelligent recognition algorithm can be improved, and the false alarm rate and missed alarm rate can be reduced.

[0033] This invention is described with reference to flowchart illustrations and / or block diagrams of methods, apparatus (systems), and computer program products according to embodiments of the invention. It will be understood that each block of the flowchart illustrations and / or block diagrams, and combinations of blocks in the flowchart illustrations and / or block diagrams, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, special-purpose computer, embedded processor, or other programmable data processing apparatus to produce a machine, such that the instructions, which execute via the processor of the computer or other programmable data processing apparatus, generate instructions for implementing the flowchart illustrations and / or block diagrams. Figure 1 One or more processes and / or boxes Figure 1 A device that provides the functions specified in one or more boxes.

[0034] In the description of this invention, it should be understood that the terms "first" and "second" are used for descriptive purposes only and should not be construed as indicating or implying relative importance or implicitly specifying the number of indicated technical features. Therefore, a feature defined as "first" or "second" may explicitly or implicitly include one or more of that feature. In the description of this invention, "a plurality of" means two or more, unless otherwise explicitly specified.

[0035] Obviously, those skilled in the art can make various modifications and variations to this invention without departing from its spirit and scope. Therefore, if these modifications and variations fall within the scope of the claims of this invention and their equivalents, this invention also intends to include these modifications and variations.

Claims

1. A method for constructing a dataset for intelligent identification of multi-view magnetic anomaly signals, characterized in that, Includes the following steps: Collect positive and negative sample data, perform noise reduction processing, and generate a sample dataset, where positive samples correspond to magnetic anomaly signals and negative samples correspond to magnetic signals; The sample dataset is transformed from multiple perspectives to generate a dataset for training the intelligent identification algorithm model for magnetic anomaly signals. This improves the effectiveness and detection rate of the trained intelligent identification algorithm model for magnetic anomaly signals and reduces the false alarm rate and missed alarm rate.

2. The dataset construction method for intelligent identification of multi-view magnetic anomaly signals according to claim 1, characterized in that: When collecting positive and negative sample data, the positive and negative sample data are collected according to the sample time length and sampling frequency, and the sample length of negative samples is kept the same as that of positive samples.

3. The dataset construction method for intelligent identification of multi-view magnetic anomaly signals according to claim 1, characterized in that: When performing multi-view transformation to obtain the dataset, positive and negative samples are visualized as waveforms and saved as images. In this process, the positive and negative samples are visualized to remove redundant information and delete the X and Y axis coordinate information, retaining only the coordinate axes and curves to form waveforms of magnetic anomaly signals and background noise. These waveforms are used to generate the dataset, enabling the model to learn and extract the differences between positive and negative samples in the images during the model training process.

4. The dataset construction method for intelligent identification of multi-view magnetic anomaly signals according to claim 3, characterized in that: When performing multi-view transformation to obtain the dataset, the complete magnetic anomaly signal waveform is found from the waveform diagram corresponding to the positive sample and feature extraction is performed. The waveform width, peak-to-peak value, and peak-to-peak ratio of the magnetic anomaly signal waveform are extracted and output in text format to generate the dataset. This enables the model to learn and extract various features of the positive sample waveform during the training process.

5. The dataset construction method for intelligent identification of multi-view magnetic anomaly signals according to claim 1, characterized in that: When performing multi-view transformation to obtain the dataset, positive and negative samples are converted into time-frequency maps, and the time-frequency maps are saved as images to generate the dataset. This allows the model to accurately learn and extract the difference between the time-frequency maps of magnetic anomaly signals and background noise during training. The X and Y coordinate ranges of the time-frequency maps are adjusted to ensure that the frequencies of the signal set are fully displayed.

6. The dataset construction method for intelligent identification of multi-view magnetic anomaly signals according to claim 1, characterized in that: When performing multi-view transformation to obtain the dataset, each positive and negative sample is linearly combined with three unrelated orthogonal basis functions of the orthogonal basis to calculate the energy value, which is used to form the dataset. This allows the trained model to learn and extract the magnitude and distribution of the orthogonal basis energy values ​​corresponding to the positive and negative samples.

7. The dataset construction method for intelligent identification of multi-view magnetic anomaly signals according to claim 6, characterized in that: When performing multi-view transformation to obtain the dataset, the orthogonal basis energy values ​​are visualized to form a waveform, which is used to form the dataset, so that the trained model can learn and extract the features in the orthogonal basis energy waveform corresponding to the positive and negative samples.

8. A dataset construction system for intelligent identification of multi-view magnetic anomaly signals, used to implement the dataset construction method for intelligent identification of multi-view magnetic anomaly signals as described in claim 1, characterized in that, include: The data acquisition module is used to collect positive and negative sample data, perform noise reduction processing, and generate a sample dataset, where positive samples correspond to magnetic anomaly signals and negative samples correspond to magnetic signals. The dataset construction module is used to perform multi-view transformation on the sample dataset to generate a dataset for training the intelligent identification algorithm model of magnetic anomaly signals. This is used to improve the effectiveness and detection rate of the trained intelligent identification algorithm model of magnetic anomaly signals, and reduce the false alarm rate and missed alarm rate.

Images