A magnetic anomaly detection method based on deep multi-instance learning
Through a deep multi-example learning method, the data is divided into bags by measurement trajectory, and the model is trained using labeled bags, the difficulties of existing magnetic anomaly detection technology in noise interference and model construction are solved, and more efficient and accurate magnetic anomaly detection is achieved.
Patent Information
- Application Number
- CN202411628579.6
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2024-11-14
- Publication Date
- 2025-04-01
AI Technical Summary
The existing magnetic anomaly detection technology has difficulties in noise interference and model construction, resulting in inaccuracy and inefficiency of detection.
Magnetic anomaly detection method based on deep multi-example learning is adopted, and magnetic anomaly detection is achieved by dividing the data into bags according to the measurement trajectory and using labeled bags to train the MI-Nets model.
It improves the accuracy and efficiency of magnetic anomaly detection, can effectively detect and locate magnetic anomalies in complex and high-dimensional data environments, and reduces labeling costs.
Smart Images
Figure CN119740153A_ABST
Abstract
Description
Technical Field
[0001] The present invention relates to the field of anomaly detection application technology, and in particular to a magnetic anomaly detection method based on deep multi-instance learning. Background Art
[0002] Magnetic anomaly detection is a method that uses magnetic anomaly changes to identify and detect specific targets or phenomena. This technology is widely used in geophysical exploration, military detection, archeology, infrastructure monitoring and other fields. Magnetic anomalies are usually caused by changes in geological structures, underground buried objects or other magnetic materials. By detecting these anomalies, the identification and positioning of targets can be achieved.
[0003] Although magnetic anomaly detection technology has made significant progress in many fields, it still faces many difficulties and challenges in practical applications. On the one hand, there is the interference of noise. Factors such as changes in the geomagnetic field itself, atmospheric interference, and electromagnetic interference will affect the accuracy of magnetic anomaly detection. These noises will mask the real magnetic anomaly signals and cause false detection or missed detection. On the other hand, there is the construction of the model. In magnetic anomaly detection, especially in the initial data collection stage, there is often a lack of accurate label data, which makes model training difficult.
[0004] Traditional anomaly detection methods mainly include statistical methods, distance-based methods, density-based methods, and machine learning methods. Statistical methods are sensitive to minor anomalies and noise, and are prone to false positives or false negatives. Distance-based methods are sensitive to the selection of parameters (such as K value, distance threshold, etc.), and improper parameter selection will affect the detection results. Density-based methods perform poorly when dealing with clusters of different densities, and are prone to misjudging sparse normal points as anomalies. Summary of the invention
[0005] In view of the above problems existing in the prior art, an embodiment of the present invention provides a magnetic anomaly detection method based on deep multi-instance learning, which is used for magnetic anomaly detection of seabed magnetic targets. First, data preparation is performed, and the data is divided according to the measurement track. Each measurement track is used as a bag, and each bag is marked with the presence or absence of magnetic foreign matter; then, the MI-Nets model is trained using the labeled bags, and the area to be detected is magnetized and measured; finally, the trained model is used to predict the newly measured data to achieve magnetic anomaly detection.
[0006] The embodiment of the present invention provides a magnetic anomaly detection method based on deep multi-instance learning, comprising:
[0007] Step 1: Use a large powered coil towed by a ship to magnetize the measurement area. After magnetization, use a magnetic field measuring instrument to collect magnetic field data of the target area to obtain a magnetic field data set, and perform data preprocessing on the collected magnetic field data set;
[0008] Step 2: Assign bag labels. According to the preset assignment scheme, assign labels to each bag. Bags containing magnetic anomalies are positively labeled, otherwise they are negatively labeled.
[0009] Step 3: Model architecture design and build a multi-instance learning model;
[0010] Step 4: Divide the newly collected magnetic field data into multiple bags according to the driving trajectory, extract features and input them into the trained model;
[0011] Step 5: Conduct field verification based on the prediction values and scoring values output by the trained model.
[0012] In some embodiments of the present invention, in step 1, the magnetization treatment of the measurement area is performed by towing a large energized coil by a ship, specifically:
[0013] Step 11: Use a ship to tow a 5m×5m rectangular energized coil, 1m from the seabed, and move along the set route to magnetize all areas where magnetic anomaly detection is performed.
[0014] In some embodiments of the present invention, in step 1, after the magnetization is completed, a magnetic field measuring instrument is used to collect magnetic field data of the target area to obtain a magnetic field data set, specifically:
[0015] Step 12: Fix the magnetic measurement equipment on the ship and collect magnetic field data in a large area according to the predetermined zigzag trajectory to obtain r measurement trajectories {M1, M2, …, M r};
[0016] Each measurement trajectory contains s points, that is, M i ={m1,m2,…,m s}, where m i =(x i ,y i ,z i ,t i ,c i ) is the data of the i-th point, (x i ,y i ,z i ) is the measured latitude, longitude and depth information, t i To measure the time at this point, c i is the magnetic field value at that point;
[0017] The magnetic field data set is represented by N.
[0018] In some embodiments of the present invention, in step 1, the data preprocessing of the collected magnetic field data is specifically:
[0019] Clean the collected data to remove obvious noise and erroneous data.
[0020] In some embodiments of the present invention, step 2 comprises:
[0021] For each bag {M1,M2,…,M r}, assign labels {0,1}, where 0 means there is no magnetic anomaly in the bag and 1 means there is a magnetic anomaly in the bag;
[0022] The bag with all assigned labels is {M′1,M′2,…,M′ r}, where M′ i ={M i ,flag i}, flag i =0 or 1.
[0023] In some embodiments of the present invention, in step 3, the method specifically includes:
[0024] Step 31: Construct the input layer to accept the features of the data in each bag. The input is represented as X = {n1, n2, ..., n t}, where n i =(x i ,y i ,z i ,t i ,c i );
[0025] Step 32: Extract features through convolutional layers and pooling layers;
[0026] Step 33: For {M′1,M′2,…,M′ r}, use the fully connected layer score i =σ(W s ·M i +b s ) is used as the scoring layer to score and reflect the probability of abnormality of the bag, where W s and b s are the weights and biases of the fully connected layer, and σ is the sigmoid activation function;
[0027] Step 34: Use softmax to convert the score i Convert to weight And perform weighted aggregation
[0028] Step 35: Classify the aggregated bag features and output the predicted label of the bag flag′=σ(W0·bag feature +b0);
[0029] Step 3.6: Use the labeled data {M′1,M′2,…,M′ r}Train the model, use the cross entropy loss function, and use the back propagation algorithm for parameter optimization
[0030] Where N is the total number of bags, y i is the true label of the i-th bag, y' i is the predicted label of the i-th bag; the model parameters are updated by gradient descent: Where η is the learning rate.
[0031] In some embodiments of the present invention, in step 32, specifically:
[0032] The number of convolution kernels k = 64, the size of each convolution kernel m = 3, and the output feature is represented as h i =ReLU(W·n i +b), where W and b are the weight and bias of the convolution kernel, and ReLU is the activation function;
[0033] Perform pooling operation Further extract high-level features through the fully connected layer Where W h and b h are the weights and biases of the fully connected layers.
[0034] In some embodiments of the present invention, in step 4, the method specifically includes:
[0035] Step 41: Bag the newly collected magnetic field data new Use the trained model to predict y new =σ(W0·bag new +b0);
[0036] Among them, y new =1, it means that there is an anomaly in the bag; otherwise, there is no anomaly in the bag;
[0037] Step 42: Use the fully connected layer score new =σ(W s ·bag new +b s ) as the scoring layer, scoring, calculating the probability of the bag being abnormal new .
[0038] In some embodiments of the present invention, in step 5, the method is specifically:
[0039] According to y newand score new value for field verification.
[0040] Compared with the prior art, the magnetic anomaly detection method based on deep multi-instance learning provided by the embodiment of the present invention has the beneficial effect that each measurement curve is regarded as a bag, and each bag contains multiple measurement points. Unlike traditional supervised learning, the label of the bag is known, but the label of each measurement point in the bag is unknown. Since the abnormal points of magnetic anomaly detection generally appear continuously, the "bag" processing method can greatly improve the accuracy and efficiency of magnetic anomaly detection.
[0041] The present invention does not rely on the characteristics of a single measurement point, but takes into account the characteristics of the entire "bag", which makes it perform better when processing data with complex abnormal patterns.
[0042] The present invention only needs to label the "bags" instead of labeling each measurement point. Therefore, the present invention can significantly reduce the labeling cost. It has great advantages for a large number of data sets, especially for situations where there are few abnormal data and measurement points are difficult to label.
[0043] The present invention can not only detect abnormal "bags", but also locate specific abnormal points, which is particularly important for magnetic anomaly detection that requires precise positioning. BRIEF DESCRIPTION OF THE DRAWINGS
[0044] Figure 1 A flow chart of a magnetic anomaly judgment algorithm in a magnetic anomaly detection method based on deep multi-instance learning provided in an embodiment of the present invention;
[0045] Figure 2 A flow chart of a magnetic anomaly detection method based on deep multi-instance learning provided in an embodiment of the present invention. DETAILED DESCRIPTION
[0046] In order to enable those skilled in the art to better understand the technical solution of the present invention, the present invention is described in detail below in conjunction with the accompanying drawings and specific implementation methods.
[0047] Various aspects and features of the present application are described herein with reference to the accompanying drawings.
[0048] These and other characteristics of the present application will become apparent from the following description of a preferred form of embodiment given as a non-limiting example with reference to the accompanying drawings.
[0049] It should also be understood that, although the present application has been described with reference to some specific examples, those skilled in the art will be able to realize many other equivalent forms of the present application that have the features described in the claims and are therefore within the scope of protection defined thereby.
[0050] The above and other aspects, features and advantages of the present application will become more apparent in view of the following detailed description when taken in conjunction with the accompanying drawings.
[0051] Specific embodiments of the present application are described hereinafter with reference to the accompanying drawings; however, it should be understood that the embodiments applied for are merely examples of the present application, which may be implemented in a variety of ways. Well-known and / or repeated functions and structures are not described in detail to determine the true intent based on the user's historical operations and to avoid unnecessary or redundant details that make the present application unclear. Therefore, the specific structural and functional details applied for herein are not intended to be limiting, but are merely used as the basis and representative basis for the claims to teach those skilled in the art to use the present application in a variety of ways with substantially any suitable detailed structure.
[0052] This specification may use the phrases "in one embodiment," "in another embodiment," "in yet another embodiment," or "in other embodiments," all of which may refer to one or more of the same or different embodiments according to the present application.
[0053] The embodiment of the present invention provides a magnetic anomaly detection method based on deep multi-instance learning, which adopts multi-instance learning as an anomaly detection method, provides a new perspective for anomaly detection, and improves the robustness and accuracy of detection by processing uncertain labels and utilizing bag structures. In complex and high-dimensional data environments, multi-instance learning has demonstrated its strong application potential. Applying multi-instance learning to magnetic anomaly detection, combining the characteristics of magnetic field data and the advantages of multi-instance learning in processing uncertain label data, can significantly improve the effect and accuracy of magnetic anomaly detection. This method has broad application prospects in geophysical exploration, military detection, archaeology and other fields.
[0054] like Figure 1 and Figure 2 As shown, the embodiment of the present invention provides a magnetic anomaly detection method based on deep multi-instance learning, which specifically includes the following steps:
[0055] Step 1: Use a large powered coil towed by a ship to magnetize the measurement area. After magnetization, use a magnetic field measuring instrument to collect magnetic field data of the target area to obtain a magnetic field data set, and perform data preprocessing on the collected magnetic field data set;
[0056] Step 2: Assign bag labels. According to the preset assignment scheme, assign labels to each bag. Bags containing magnetic anomalies are positively labeled, otherwise they are negatively labeled.
[0057] Step 3: Model architecture design and build a multi-instance learning model;
[0058] Step 4: Divide the newly collected magnetic field data into multiple bags according to the driving trajectory, extract features and input them into the trained model;
[0059] Step 5: Conduct field verification based on the prediction values and scoring values output by the trained model.
[0060] In order to facilitate the understanding of the above technical solution, the following is a detailed description with reference to the accompanying drawings and specific examples, as follows:
[0061] Step 1: Data collection. First, use a ship to tow a large energized coil and travel along the seabed to magnetize the measurement area. After magnetization, use a magnetic field measuring instrument to collect magnetic field data in the target area to obtain a data set N. Perform data preprocessing on the collected magnetic field data set N.
[0062] Step 11: Use a ship to tow a 5m×5m rectangular energized coil, 1m above the seabed, in a zigzag pattern to magnetize all areas where magnetic anomaly detection is performed.
[0063] Step 12: Fix the magnetic measurement equipment on the ship and collect magnetic field data in a large area according to the predetermined zigzag trajectory to obtain r measurement trajectories {M1, M2, …, M r Each measurement trajectory contains s points, namely M i ={m1,m2,…,m s}, where m i =(x i ,y i ,z i ,t i ,c i ) is the data of the i-th point, (x i ,y i ,z i ) is the measured latitude, longitude and depth information, t i To measure the time at this point, c i is the magnetic field value at that point.
[0064] Step 13: Clean the collected data to remove obvious noise and erroneous data.
[0065] Step 2: Bag label assignment: Based on existing knowledge or expert annotation, a label is assigned to each bag, where the bag containing magnetic anomalies is a positive label, otherwise it is a negative label.
[0066] Step 2.1: For each bag {M1,M2,…,M r}, assign labels {0,1}, where 0 means there is no magnetic anomaly in the bag, and 1 means there is a magnetic anomaly in the bag. The bag after assigning labels is {M′1,M′2,…,M′ r}, where M′ i ={M i ,flagi}, flag i =0 or 1.
[0067] Step 3: Design the model architecture and build a multi-instance learning model.
[0068] Step 3.1: Construct the input layer to accept the features of the data in each bag. The input is represented as X = {n1,n2,…,n t}, where n i =(x i ,y i ,z i ,t i ,c i ).
[0069] Step 3.2: Extract features through the convolution layer (Conv1D) and the pooling layer (Maxpooling1D). Select the number of convolution kernels k = 64, the size of each convolution kernel m = 3, and the output feature is represented as h i =ReLU(W·n i +b), where W and b are the weight and bias of the convolution kernel, and ReLU is the activation function. Then the pooling operation is performed Then the high-level features are further extracted through the fully connected layer Where W h and b h are the weights and biases of the fully connected layers.
[0070] Step 3.3: For {M′1,M′2,…,M′ r}, use the fully connected layer score i =σ(W s ·M i +b s ) is used as the scoring layer to score and reflect the probability of abnormality of the bag, where W s and b s are the weights and biases of the fully connected layer, and σ is the sigmoid activation function.
[0071] Step 3.4: Use softmax to convert the score i Convert to weight And perform weighted aggregation
[0072] Step 3.5: Classify the aggregated bag features and output the predicted label of the bag flag′=σ(W0·bag feature +b0).
[0073] Step 3.6: Use the labeled data {M′1,M′2,…,M′ r}Train the model, use the cross entropy loss function, and use the back propagation algorithm for parameter optimization Where N is the total number of bags, yi is the true label of the ith bag, and y'i is the predicted label of the ith bag. Update the model parameters by gradient descent: Where η is the learning rate. Step 4: Divide the newly collected magnetic field data into multiple bags according to the driving trajectory, extract features and input them into the trained model.
[0074] Step 4.1: For the newly collected magnetic field data bag new Use the trained model to predict y new =σ(W0·bag new +b0). Among them, y new =1, it means that there is an anomaly in the bag; otherwise, there is no anomaly in the bag.
[0075] Step 4.2: Use the fully connected layer score new =σ(W s ·bag new +b s ) as the scoring layer, scoring, calculating the probability of the bag being abnormal new .
[0076] Step 5: According to y new and score new value for field verification.
[0077] It can be seen from the above technical solution that the beneficial effect of the magnetic anomaly detection method based on deep multi-instance learning provided by the above embodiment of the present invention is that it regards each measurement curve as a bag, and each bag contains multiple measurement points. Unlike traditional supervised learning, the label of the bag is known, but the label of each measurement point in the bag is unknown. Since the abnormal points of magnetic anomaly detection generally appear continuously, the "bag" processing method can greatly improve the accuracy and efficiency of magnetic anomaly detection.
[0078] The present invention does not rely on the characteristics of a single measurement point, but takes into account the characteristics of the entire "bag", which makes it perform better when processing data with complex abnormal patterns.
[0079] The present invention only needs to label the "bags" instead of labeling each measurement point. Therefore, the present invention can significantly reduce the labeling cost. It has great advantages for a large number of data sets, especially for situations where there are few abnormal data and measurement points are difficult to label.
[0080] The present invention can not only detect abnormal "bags", but also locate specific abnormal points, which is particularly important for magnetic anomaly detection that requires precise positioning.
[0081] The above embodiments are only exemplary embodiments of the present invention and are not intended to limit the present invention. The protection scope of the present invention is defined by the claims. Those skilled in the art may make various modifications or equivalent substitutions to the present invention within the essence and protection scope of the present invention, and such modifications or equivalent substitutions shall also be deemed to fall within the protection scope of the present invention.
Claims
1. A magnetic anomaly detection method based on deep multi-instance learning, characterized in that: include: Step 1: Use a large powered coil towed by a ship to magnetize the measurement area. After magnetization, use a magnetic field measuring instrument to collect magnetic field data of the target area to obtain a magnetic field data set, and perform data preprocessing on the collected magnetic field data set; Step 2: Assign bag labels. According to the preset assignment scheme, assign labels to each bag. Bags containing magnetic anomalies are positively labeled, otherwise they are negatively labeled. Step 3: Model architecture design and build a multi-instance learning model; Step 4: Divide the newly collected magnetic field data into multiple bags according to the driving trajectory, extract features and input them into the trained model; Step 5: Conduct field verification based on the prediction values and scoring values output by the trained model.
2. The magnetic anomaly detection method based on deep multi-instance learning according to claim 1 is characterized in that: In step 1, the measurement area is magnetized by towing a large powered coil by a ship, specifically: Step 11: Use a ship to tow a 5m×5m rectangular energized coil, 1m from the seabed, and move along the set route to magnetize all areas where magnetic anomaly detection is performed.
3. The magnetic anomaly detection method based on deep multi-instance learning according to claim 2 is characterized in that: In step 1, after the magnetization is completed, a magnetic field measuring instrument is used to collect magnetic field data of the target area to obtain a magnetic field data set, specifically: Step 12: Fix the magnetic measurement equipment on the ship and collect magnetic field data in a large area according to the predetermined zigzag trajectory to obtain r measurement trajectories {M1, M2, …, M r }; Each measurement trajectory contains s points, that is, M i ={m1,m2,…,m s }, where m i =(x i ,y i ,z i ,t i ,c i ) is the data of the i-th point, (x i ,y i ,z i ) is the measured latitude, longitude and depth information, t i To measure the time at this point, c i is the magnetic field value at that point; The magnetic field data set is represented by N.
4. The magnetic anomaly detection method based on deep multi-instance learning according to claim 3 is characterized in that: In step 1, the collected magnetic field data is preprocessed, specifically: Clean the collected data to remove obvious noise and erroneous data.
5. The magnetic anomaly detection method based on deep multi-instance learning according to claim 4 is characterized in that: Step 2 includes: For each bag {M1,M2,…,M r }, assign labels {0,1}, where 0 means there is no magnetic anomaly in the bag and 1 means there is a magnetic anomaly in the bag; The bag with all assigned labels is {M′1,M′2,…,M′ r }, where M′ i ={M i ,flag i }, flag i =0 or 1.
6. The magnetic anomaly detection method based on deep multi-instance learning according to claim 5 is characterized in that: In step 3, the method specifically includes: Step 31: Construct the input layer to accept the features of the data in each bag. The input is represented as X = {n1, n2, ..., n t }, where n i =(x i ,y i ,z i ,t i ,c i ); Step 32: Extract features through convolutional layers and pooling layers; Step 33: For {M′1,M′2,…,M′ r }, use the fully connected layer score i =σ(W s ·M i +b s ) is used as the scoring layer to score and reflect the probability of abnormality of the bag, where W s and b s are the weights and biases of the fully connected layer, and σ is the sigmoid activation function; Step 34: Use softmax to convert the score i Convert to weight And perform weighted aggregation Step 35: Classify the aggregated bag features and output the predicted label flag of the bag ′ =σ(W0·bag feature +b0); Step 3.6: Use the labeled data {M′1,M′2,…,M′ r }Train the model, use the cross entropy loss function, and use the back propagation algorithm for parameter optimization Where N is the total number of bags, y i is the true label of the i-th bag, y' i is the predicted label of the i-th bag; the model parameters are updated by gradient descent: Where η is the learning rate.
7. The magnetic anomaly detection method based on deep multi-instance learning according to claim 6 is characterized in that: In step 32, specifically: The number of convolution kernels k = 64, the size of each convolution kernel m = 3, and the output feature is represented as h i =ReLU(W·n i +b), where W and b are the weight and bias of the convolution kernel, and ReLU is the activation function; Perform pooling operation Further extract high-level features through the fully connected layer Where W h and b h are the weights and biases of the fully connected layers.
8. The magnetic anomaly detection method based on deep multi-instance learning according to claim 7 is characterized in that: In step 4, the method specifically includes: Step 41: Bag the newly collected magnetic field data bag new Use the trained model to predict y new =σ(W0·bag new +b0); Among them, y new =1, it means that there is an anomaly in the bag; otherwise, there is no anomaly in the bag; Step 42: Use the fully connected layer score new =σ(W s ·bag new +b s ) as the scoring layer, scoring, calculating the probability of the bag being abnormal new .
9. The magnetic anomaly detection method based on deep multi-instance learning according to claim 8 is characterized in that: In step 5, the method is specifically: According to y new and score new value for field verification.