A passive magnetic anomaly detection positioning system based on a CNN neural network and a method thereof

Abstract

The application discloses a passive magnetic anomaly detection positioning system based on a CNN neural network and a method thereof, and belongs to the technical field of detection positioning systems. In order to solve the problems that the existing systems cannot automatically detect ferromagnetic substances, doctors are few but patients are many, personnel are dense and complex, it is difficult to manage, there is no way to conduct detailed examination on patients, manual detection is needed, detection efficiency is low, and the detection result is affected, the magnetic detection function is realized through an embedded device and multiple small single-axis magnetometers, the integrated degree is high, application in an actual hospital scene is extremely convenient, the abnormal signal detection method of the CNN neural network is used, the signal-to-noise ratio is improved, the interference of the geomagnetic field and the surrounding environment is excluded, the false alarm rate and the false detection rate are extremely low, and alarm is conducted in the form of sound, light, screen video and the like, the application is simple to use and high in stability, a special person does not need to be specially on duty, and a patient can use the device according to the prompt.

Description

TECHNICAL FIELD

[0001] The application relates to the technical field of detection positioning systems, in particular to a passive magnetic anomaly detection positioning system based on a CNN neural network and a method thereof. BACKGROUND

[0002] Magnetic anomaly target detection is a specific technology in the military field, and the magnetic measurement technology is based on the physical basic principle of induced magnetic field, and a magnetic sensor is used to induce a large ferromagnetic device entering a target area, such as a submarine, a ship and the like. The secondary field generated by the huge target changes the original magnetic field distribution. In addition, the magnetic measurement method is also a main method for deducing geological interfaces, exploring mineral resource distribution and oil and gas distribution, and analyzing the volume and burial depth of a mineral body by detecting the distortion and change of the geomagnetic field.

[0003] Nuclear magnetic resonance (MRI) is a medical examination technology that has just emerged in the past 20 years, and is indeed a leap in the field of radiation, and has provided great help for the diagnosis of many diseases.

[0004] However, magnetic resonance also has taboos. Because the main body of the machine is a huge magnet, therefore, any ferromagnetic material such as a mobile phone, a watch, a steel coin, a magnetic card, a lighter and the like is absolutely not allowed to appear in the scanning room. These things are not allowed to be taken into the scanning room, and once they are taken in, they will affect the detection result, and even cause more serious medical accidents.

[0005] And the radiology department of a hospital is a department under great pressure of patients, and patients for examination are almost continuous for 24 hours, doctors are few and patients are many, personnel are dense and complex, and it is difficult to manage. If each patient is checked one by one, it is impossible to operate. SUMMARY

[0006] The application aims to provide a passive magnetic anomaly detection positioning system based on a CNN neural network and a method thereof, realize magnetic detection function through an embedded device and a plurality of small single-axis magnetometers, have high integration, be extremely convenient to apply in an actual hospital scene, improve signal-to-noise ratio through an abnormal signal detection method of the CNN neural network, exclude the interference of the geomagnetic field and the surrounding environment, have extremely low false alarm rate and false detection rate, alarm through sound, light, screen video and the like, are simple and convenient to use, high in stability, do not need to be specially guarded by personnel, and can be used by patients according to prompts, and are convenient for later viewing and calling, so as to solve the problems in the background technology.

[0007] To achieve the above-mentioned purpose, the application provides the following technical scheme.

[0008] A passive magnetic anomaly detection positioning system based on CNN neural network, comprising: a detection door body for installing a perception layer module, and a user layer module is arranged at the side end of the detection door body;

[0009] And an application layer module for docking the perception layer module and the user layer module;

[0010] The perception layer module is used for acquisition and positioning, and the application layer module is used for receiving data collected by the perception layer module;

[0011] The application layer module drives the user layer module according to the collected data.

[0012] Further, the installation perception layer module includes a sensor module and a millimeter wave radar module, the sensor module is composed of a plurality of single-axis fluxgate sensors, and the single-axis fluxgate sensors are arranged in an array on the detection door body, and the millimeter wave radar module is arranged on the detection door body, and the signal data in the installation perception layer module is returned to the application layer module through the UART serial port.

[0013] Further, the application layer module includes a main chip, a CNN deep neural network signal detection module, an artificial intelligence signal analysis module and an abnormal magnetic target positioning module, the CNN deep neural network signal detection module, the artificial intelligence signal analysis module and the abnormal magnetic target positioning module are arranged in the main chip respectively, and the main chip is used for connecting with the installation perception layer module;

[0014] The CNN deep neural network signal detection module is used for detecting signals and transmitting the detected signals to the artificial intelligence signal analysis module for processing, and the artificial intelligence signal analysis module classifies and processes the signals through the CNN network.

[0015] Further, the user layer module includes a buzzer, an LED lamp and a large screen, and the buzzer, the LED lamp and the large screen are connected with the application layer module respectively.

[0016] Further, the artificial intelligence signal analysis module realizes signal analysis through software, analyzes the signals through the CNN neural network based on the floating point operation resources in the main chip, and judges whether there is a magnetic anomaly signal.

[0017] A method of a passive magnetic anomaly detection positioning system based on CNN neural network, comprising the following steps:

[0018] S1: The single-axis fluxgate sensors in the perception layer module form an array, the sampling frequency is adjusted to 20Hz-200Hz, and the magnetic field change in the space is collected;

[0019] S2: The millimeter wave radar module detects the human passing situation in real time. The antenna emission angle FOV is 40*80, and when a human body is detected, the CNN deep neural network signal detection module in the back-end application layer module is triggered for processing;

[0020] S3: The millimeter wave radar module continuously tracks the real-time position of the human body movement during the detection trigger process;

[0021] S4: The magnetic anomaly signal collected by the sensor module is returned to the main chip through the UART serial port for calculation;

[0022] S5: The artificial intelligence signal analysis module in the application layer module analyzes the signal through the CNN neural network to determine whether there is a magnetic anomaly signal;

[0023] S6: When the magnetic anomaly is actually detected, the buzzer, LED light and large screen alarm through sound, light and screen video.

[0024] Further, in S5, the artificial intelligence signal analysis module has two models when classifying and processing signals, including a single-channel model and a multi-channel model. The single-channel model includes the following steps:

[0025] S501: Use one-dimensional CNN instead of traditional FFT to transform the signal into a two-dimensional signal;

[0026] S502: Use two-dimensional CNN for channel expansion and feature extraction;

[0027] S503: Use GRU for historical information tracking and connection;

[0028] S504: Use the fully connected layer FC for probability estimation and output.

[0029] Further, the multi-channel model expands the dimension based on the single-channel model, increases the model structure, utilizes the coupling relationship of multi-channel information, and increases the complexity of the model structure. Increasing the model requires model training, and the structure model training requires data simulation. The data simulation for magnetic anomaly detection includes background noise simulation and magnetic anomaly signal simulation, and the specific process of data simulation is as follows:

[0030] Select a recorded background noise segment, and generalize based on the recorded noise. The generalization operation includes superimposing Gaussian noise, signal flipping and gain randomization;

[0031] The simulation of the magnetic anomaly signal includes pure simulation and simulation based on the recorded magnetic anomaly:

[0032] The pure simulation is mainly based on the generation of standard orthogonal basis functions and the generalization means similar to the superposition of Gaussian noise, signal flipping and gain randomization, and the real recorded magnetic anomaly is selected and generalized based on the recorded magnetic anomaly signal;

[0033] If the magnetic anomaly signal is a single-channel anomaly, the multi-channel signal is expanded according to the relative orientation and physical principle; if the magnetic anomaly signal is a multi-channel anomaly, the data can be generalized according to the distance;

[0034] Set a suitable signal-to-noise ratio, superimpose the background noise and the magnetic anomaly signal, and avoid discontinuity at both ends of the magnetic anomaly when superimposing;

[0035] The gain of the generated signal is generalized;

[0036] Obtain the final detection signal, and generate the corresponding detection target according to the magnetic anomaly time period as the training target of the model.

[0037] Further, the CNN deep neural network signal detection module is provided with an anomaly detection module, the anomaly detection module includes a tracker module, a function library module, a strong sequence constructor module, a detector module, a response module, a decision module and a control module, the tracker module is used for tracking the process derived from the program running, and the calling sequence of the process to the system is obtained, the function library module is a database for storing system functions, the strong sequence constructor module is used for making up the defect of fixed sliding window size, and is composed of a function set and a strong sequence, the detector module divides the process sequence into normal or abnormal, the response module interrupts the connection through a firewall, a router and the like, and kills the abnormal process through a system command, the decision module is used for making a decision on the abnormal condition reported in the preprocessor and the detector, and the control module is used for constructing and updating the strong function set and setting the threshold in the detector and the preprocessor.

[0038] Further, the data preprocessing process of the tracker in the anomaly detection module includes the following steps:

[0039] S100: Further decompose the original data according to the need, remove the error data, and convert the symbol field that cannot be processed by the detector into a numerical value;

[0040] S200: Perform normalization processing to reduce the adverse effects on network training caused by too large differences in field values between records;

[0041] S300: Check whether the related items in the "calling function set" in a unit time exceed the preset threshold.

[0042] Compared with the prior art, the present application has the following advantages:

[0043] 1. The passive magnetic anomaly detection and positioning system and method based on the CNN neural network, the sensor module is composed of a plurality of single-axis fluxgate sensors, and the single-axis fluxgate sensors are arranged in an array on the detection door body, and the millimeter wave radar module is arranged on the detection door body, the signal data in the installation perception layer module is returned to the application layer module through the UART serial port, the high-precision fluxgate sensor array is placed on both sides or a single side of the door, the magnetic sensors are respectively installed on the left and right sides of the door, and there are 3-5 on the left and right sides, 4 are used in this example, the distance between two of them is 30-50cm, 30cm is used, and there are 8 magnetic sensors in total, the magnetic field change in the space is collected, the millimeter wave radar module can detect the passing of the human body in the facing direction in real time, the antenna emission angle FOV is 40*80, the detection distance can reach 3-5m, when a human body is detected to pass, the CNN deep neural network signal detection module in the rear-end main chip is triggered to process, and the real-time position of the human body movement is also tracked in this process, the fluxgate sensor has high precision, the fluxgate sensor is used, and even a small magnetic field fluctuation can be detected, the magnetic field fluctuation within 1nT has been almost completely submerged in the interference of the geomagnetic field, the magnetic detection function is realized through the embedded device and a plurality of small single-axis magnetometers, the integration is high, and it is extremely convenient to apply in the actual hospital scene.

[0044] 2. The passive magnetic anomaly detection and positioning system and method based on the CNN neural network, the main chip is used for connecting with the installation perception layer module, the CNN deep neural network signal detection module is used for detecting signals and transmitting the detected signals to the artificial intelligence signal analysis module for processing, the artificial intelligence signal analysis module classifies and processes the signals through the CNN network, the artificial intelligence signal analysis module realizes signal analysis through software, the artificial intelligence signal analysis module analyzes the signals through the CNN neural network based on the floating point operation resources in the main chip, judges whether there is a magnetic anomaly signal, the signal processing algorithm is advanced, the signal-to-noise ratio is improved through the abnormal signal detection method of the CNN neural network, the interference of the geomagnetic field and the surrounding environment is excluded, and the false alarm rate and the false detection rate are extremely low.

[0045] 3. The passive magnetic anomaly detection and localization system and method based on CNN neural network proposed in this invention connects a buzzer, LED lights, and a large screen to the application layer module. Real-time detection results are displayed and tracked on the large screen, alerting patients to avoid related risks. For example, if a patient has a mobile phone in their waist pocket, a red dot will be marked on the screen at the patient's waist area, an alarm sound will be emitted by the buzzer, and the LED lights will change color to remind the user of the risks associated with MRI detection. This passive approach does not generate any magnetic field radiation, does not affect the normal operation of the MRI machine, and does not... It can affect the patient's health and examination results. It is easy to use, highly stable, and does not require special supervision. Patients can use the device themselves by following the prompts. The device calculates the presence or absence of magnetic anomalies through the CNN neural network and determines the specific location through the abnormal magnetic target positioning module. It is then compared with the specific location of the human body calculated by the millimeter wave module to obtain the specific location of the magnetic anomaly. When a magnetic anomaly is actually detected, it will alarm through sound, light, and screen video when no one is watching. It is easy to use, highly stable, and does not require special supervision. Patients can use the device themselves by following the prompts.

[0046] 4. The passive magnetic anomaly detection and localization system and method based on CNN neural network proposed in this invention, after implementing the entire detection system, uses the collected normal data to train the system, complete some system database initialization and threshold preset work, and more accurately and comprehensively describes the normal behavior of the program. Under the same false detection rate, the detection rate of this model is higher than that of simply using a detector. By introducing concepts such as functional subsequence, strong sequence set, function call set, and function reference number, the behavior of the program during runtime is fully explored. Furthermore, due to the introduction of strong sequence, the impact of the sliding window size setting on the detection rate is reduced. Attached Figure Description

[0047] Figure 1 This is an overall module diagram of the present invention;

[0048] Figure 2 This is an overall flowchart of the present invention;

[0049] Figure 3 This is a schematic diagram of the installation sensing layer module structure of the present invention;

[0050] Figure 4 This is a block diagram of the application layer module of the present invention;

[0051] Figure 5 This is a block diagram of the user layer module of the present invention;

[0052] Figure 6 This is an overall flowchart of the present invention;

[0053] Figure 7 Data simulation flowchart of the present application;

[0054] Figure 8 Abnormality detection module flowchart of the present application.

[0055] Figure 9 Abnormality detection module flowchart of the present application.

[0056] In the figure: 1, install the perception layer module; 11, sensor module; 111, single-axis fluxgate sensor; 12, millimeter wave radar module; 2, detection door body; 3, user layer module; 31, buzzer; 32, LED lamp; 33, large screen; 4, application layer module; 41, main chip; 42, CNN deep neural network signal detection module; 421, abnormality detection module; 4211, tracker module; 4212, function library module; 4213, strong sequence constructor module; 4214, detector module; 4215, response module; 4216, decision module; 4217, control module; 43, artificial intelligence signal analysis module; 44, abnormal magnetic target positioning module. DETAILED DESCRIPTION

[0057] The technical solutions in the embodiments of the present application will be clearly and completely described below with reference to the drawings in the embodiments of the present application. Obviously, the described embodiments are only part of the embodiments of the present application, rather than all the embodiments. Based on the embodiments in the present application, all other embodiments obtained by those skilled in the art without creative work fall within the scope of protection of the present application.

[0058] Please refer to Figures 1-2 A passive magnetic abnormality detection and positioning system based on a CNN neural network, comprising a detection door body 2 for installing a perception layer module 1, and a user layer module 3 arranged at the side end of the detection door body 2;

[0059] And an application layer module 4 for docking the perception layer module 1 and the user layer module 3;

[0060] The perception layer module 1 is used for collection and positioning, and the application layer module 4 is used for receiving the data collected by the perception layer module 1;

[0061] The application layer module 4 drives the user layer module 3 according to the collected data.

[0062] In order to solve the technical problem that ferromagnetic substances will affect the detection results during the existing magnetic resonance detection, and even cause more serious medical accidents, and automatic detection of ferromagnetic substances cannot be achieved, please refer to Figure 3 The embodiment provides the following technical solutions:

[0063] The installation perception layer module 1 comprises a sensor module 11 and a millimeter wave radar module 12, the sensor module 11 is composed of a plurality of single-axis fluxgate sensors 111, and the single-axis fluxgate sensors 111 are arranged in an array on the detection door body 2, and the millimeter wave radar module 12 is arranged on the detection door body 2, signal data in the installation perception layer module 1 is returned to the application layer module 4 through a UART serial port, the high-precision fluxgate sensor array is placed on both sides or a single side of the door, and the magnetic sensors are respectively installed on the left and right sides of the door, 3-5 on the left and right sides, 4 are used in this example, the distance between two of them is 30-50 cm, 30 cm is used, and a total of 8 magnetic sensors are used to collect the magnetic field changes in the space, the millimeter wave radar module 12 can detect the passing of the human body in the facing direction in real time, the antenna emission angle FOV is 40°*80°, the detection distance can reach 3-5 m, when a human body is detected to pass, the CNN deep neural network signal detection module 42 in the rear-end main chip 41 is triggered to process, and the real-time position of the human body movement is also tracked in this process, the fluxgate sensor has high precision, the fluxgate sensor is used, and even a small magnetic field fluctuation can be detected, a magnetic field fluctuation of 1nT is almost completely submerged in the interference of the geomagnetic field, the integration is high, the magnetic detection function is realized by embedding an embedded device and a plurality of small single-axis magnetometers, and the integration is high, which is extremely convenient to apply in an actual hospital scene.

[0064] In order to solve the technical problems that doctors are few and patients are many, personnel are dense and complex, it is difficult to manage, and there is no way to perform detailed examination on patients in the prior art, please refer to Figure 4 The embodiment provides the following technical scheme:

[0065] The application layer module 4 comprises a main chip 41, a CNN deep neural network signal detection module 42, an artificial intelligence signal analysis module 43 and an abnormal magnetic target positioning module 44, the CNN deep neural network signal detection module 42, the artificial intelligence signal analysis module 43 and the abnormal magnetic target positioning module 44 are arranged in the main chip 41 respectively, the main chip 41 is used for being connected with the installation perception layer module 1, the CNN deep neural network signal detection module 42 is used for detecting a signal and transmitting the detected signal to the artificial intelligence signal analysis module 43 for processing, the artificial intelligence signal analysis module 43 classifies and processes the signal through a CNN network, the artificial intelligence signal analysis module 43 realizes signal analysis through software, the artificial intelligence signal analysis module 43 analyzes the signal through the CNN neural network based on the floating point operation resource in the main chip 41, judges whether there is a magnetic abnormal signal, the signal processing algorithm is advanced, the signal-to-noise ratio is improved through the abnormal signal detection method of the CNN neural network, the interference of the geomagnetic field and the surrounding environment is excluded, and the false alarm rate and the false detection rate are extremely low.

[0066] To solve the technical problem that the existing structure needs manual detection, resulting in low detection efficiency and affecting the detection result, please refer to Figure 5 The embodiment provides the following technical scheme:

[0067] The user layer module 3 includes a buzzer 31, an LED lamp 32 and a large screen 33, and the buzzer 31, the LED lamp 32 and the large screen 33 are connected with the application layer module 4 respectively, so that the real-time detection result is displayed on the large screen 33 and tracked in real time, and the patient is reminded to avoid relevant risks, for example, if the patient has a mobile phone in the waist pocket, a red dot is marked on the waist position of the human body on the screen, the user is reminded through the alarm sound of the buzzer 31, and the user is reminded through the color change of the LED lamp 32, a passive scheme, which does not generate any magnetic field radiation, does not affect the normal operation of the nuclear magnetic resonance machine, and does not affect the patient's body and the detection result, is simple and convenient to use, has high stability, does not need to be specially guarded by personnel, and the patient can use the equipment according to the prompt, the existence of a magnetic anomaly target is calculated through the front CNN neural network, and the specific position is determined through the positioning module of the magnetic anomaly target, and then the specific position of the magnetic anomaly target is obtained by comparing the specific position of the human body obtained by the millimeter wave module, in the case that no one is in charge, when a magnetic anomaly is actually detected, the alarm is performed in the form of sound, light and screen video, which is simple and convenient to use, has high stability, does not need to be specially guarded by personnel, and the patient can use the equipment according to the prompt.

[0068] Please refer to Figure 6 A passive magnetic anomaly detection and positioning system based on a CNN neural network includes the following steps:

[0069] S1: The single-axis fluxgate sensor 111 in the perception layer module 1 forms an array, the sampling frequency is adjusted to 20Hz-200Hz, and the magnetic field change in the space is collected;

[0070] S2: The millimeter wave radar module 12 detects the passing situation of the human body in the facing direction in real time, the antenna emission angle FOV is 40°*80°, when a human body is detected to pass, the CNN deep neural network signal detection module 42 in the back-end application layer module 4 is triggered to process;

[0071] S3: The millimeter wave radar module 12 continuously tracks the real-time position of the human body motion in the detection trigger process;

[0072] S4: The magnetic anomaly signal collected by the sensor module 11 is returned to the main chip 41 through the UART serial port for calculation;

[0073] S5: The artificial intelligence signal analysis module 43 in the application layer module 4 analyzes the signal through the CNN neural network, and judges whether there is a magnetic anomaly signal.

[0074] S6: When the magnetic anomaly is actually detected, the buzzer 31, the LED lamp 32 and the large screen 33 alarm through sound, light and screen video.

[0075] Referring to Figure 7 , the artificial intelligence signal analysis module 43 has two models when classifying the signal, including a single-channel model and a multi-channel model, wherein the model structure of the single-channel model includes the following steps:

[0076] S501: using a one-dimensional CNN instead of a traditional FFT to transform the signal into a two-dimensional signal;

[0077] S502: using a two-dimensional CNN for channel expansion and feature extraction;

[0078] S503: using GRU for historical information tracking and connection;

[0079] S504: using a fully connected layer FC for probability estimation and output.

[0080] The model structure of the channel model is expanded in dimension based on the single-channel model, increasing the model structure, utilizing the coupling relationship of multi-channel information, and increasing the complexity of the model structure, wherein increasing the model requires model training, and the structure model training requires data simulation, and the data simulation of the magnetic anomaly detection includes simulation of background noise and simulation of magnetic anomaly signals, and the specific process of data simulation is as follows:

[0081] Select a recorded background noise segment, and generalize based on the recorded noise, and the generalization operation includes superimposing Gaussian noise, signal flipping and gain randomization;

[0082] The simulation of the magnetic anomaly signal includes pure simulation and simulation based on recorded magnetic anomaly:

[0083] The pure simulation is mainly based on the generation of standard orthogonal basis functions and the generalization methods similar to superimposing Gaussian noise, signal flipping and gain randomization, and the recorded magnetic anomaly is selected and generalized based on the recorded magnetic anomaly;

[0084] If the magnetic anomaly signal is a single-channel anomaly, then the multi-channel signal is expanded according to the relative orientation and physical principle; if the magnetic anomaly signal is a multi-channel anomaly, the data can be generalized according to the distance;

[0085] Set an appropriate signal-to-noise ratio to superimpose the background noise and the magnetic anomaly signal, and avoid discontinuity at both ends of the magnetic anomaly when superimposing;

[0086] Gain generalization is performed on the generated signal;

[0087] Obtain the final to be detected signal, and generate a corresponding detection target according to the magnetic anomaly time period as a training target of the model.

[0088] In order to solve the technical problems that the detection system of the existing structure cannot be described in all directions, and the detection rate is affected by the size of the sliding window, please refer to Figures 8-9 The embodiment provides the following technical scheme:

[0089] The CNN deep neural network signal detection module 42 is provided with an anomaly detection module 421, and the anomaly detection module 421 comprises a tracker module 4211, a function library module 4212, a strong sequence constructor module 4213, a detector module 4214, a response module 4215, a decision module 4216 and a control module 4217. The tracker module 4211 is used for tracking a process derived from program running, and obtaining a call sequence of the process to the system. The function library module 4212 is a database used for storing system functions. The strong sequence constructor module 4213 is used for making up the defect of fixed size of the sliding window, and is composed of a function set and a strong sequence. The detector module 4214 divides the process sequence into normal or abnormal. The response module 4215 interrupts the connection through a firewall, a router and the like, and kills the abnormal process through a system command. The decision module 4216 is used for making a decision on the abnormal condition reported in the preprocessor and the detector. The control module 4217 is used for constructing and updating the strong function set, and setting the threshold in the detector and the preprocessor.

[0090] The preprocessing procedure of the data for the tracker in the anomaly detection module 421 comprises the following steps.

[0091] S100: According to the need, the original data is further decomposed, the error data is removed, and the symbol field that cannot be processed by the detector is converted into a numerical value.

[0092] S200: Normalization processing is performed, so as to reduce the adverse effect on the training of the network due to the too large difference in the field value between records.

[0093] S300: Whether the related items in the “call function set” in a unit time exceed a preset threshold value is viewed.

[0094] After the whole detection system is implemented, the system is trained by using the collected normal data, the initialization of some system databases and the presetting of the threshold value are completed, the normal behavior of the program is more accurately and comprehensively described, the detection rate of the model is higher than that of the detector in the same false detection rate, the behavior of the program running is fully mined by introducing the concepts of the functional subsequence, the strong sequence set, the call function set and the function reference number, and the influence of the setting of the size of the sliding window on the detection rate is reduced due to the introduction of the strong sequence.

[0095] In summary, the passive magnetic anomaly detection positioning system and method based on the CNN neural network provided by the application, the sensor module is composed of a plurality of single-axis fluxgate sensors, and the single-axis fluxgate sensors are arranged in an array on the detection door body, the millimeter wave radar module is arranged on the detection door body, the signal data in the installation perception layer module is returned to the application layer module through the UART serial port, the high-precision fluxgate sensor array is placed on both sides or a single side of the door, the magnetic sensors are respectively installed on the left and right sides of the door, and there are 3-5 on the left and right sides, 4 are used in this example, the distance between two of them is between 30-50cm, 30cm is used in this example, and there are 8 magnetic sensors in total, the magnetic field change in the space is collected, the millimeter wave radar module can detect the passing of the human body in the facing direction in real time, the antenna emission angle FOV is 40*80, the detection distance can reach 3-5m, when a human body is detected to pass, the CNN deep neural network signal detection module in the rear-end main chip is triggered to process, and the real-time position of the human body movement is also tracked in this process, the fluxgate sensor has high precision, the fluxgate sensor is used, and even a small magnetic field fluctuation can be detected, the magnetic field fluctuation within 1nT has been almost completely submerged in the interference of the geomagnetic field, the integration is high, the magnetic detection function is realized through the embedded device and a plurality of small single-axis magnetometers, the integration is high, and the application is extremely convenient in the actual hospital scene, the main chip is used for connecting with the installation perception layer module, the CNN deep neural network signal detection module is used for detecting signals and transmitting the detected signals to the artificial intelligence signal analysis module for processing, the artificial intelligence signal analysis module classifies and processes the signals through the CNN network, the artificial intelligence signal analysis module realizes signal analysis through software, the artificial intelligence signal analysis module analyzes the signals through the CNN neural network based on the floating point operation resources in the main chip, judges whether there is a magnetic anomaly signal, the signal processing algorithm is advanced, the signal-to-noise ratio is improved through the abnormal signal detection method of the CNN neural network, the interference of the geomagnetic field and the surrounding environment is excluded, the false alarm rate and the false detection rate are extremely low, the buzzer, the LED lamp and the large screen are connected with the application layer module respectively, the real-time detection result is displayed on the large screen and tracked in real time, the patient is reminded to avoid relevant risks, for example, if the patient has a mobile phone in the waist pocket, a red dot is marked on the waist position of the human body on the screen, the buzzer emits an alarm sound to remind the user, and the LED lamp changes color to remind the user to pay attention to the MRI detection risk, the passive scheme does not generate any magnetic field radiation, does not affect the normal operation of the nuclear magnetic resonance machine, and does not affect the patient's body and the test result, is simple to use and has high stability, does not need to be specially guarded by personnel, and the patient can use the equipment according to the prompt, the existence of the magnetic anomaly target is calculated through the front CNN neural network, the specific position is determined through the abnormal magnetic target positioning module, and the specific magnetic anomaly target position is obtained by comparing the specific position of the human body obtained by the millimeter wave module, and in the case of no one watching,When the magnetic anomaly is actually detected, the alarm is given through sound, light, screen video and the like, the use is simple, the stability is high, a special person does not need to be specially on duty, the patient can use the equipment according to the prompt, after the whole detection system is realized, the normal data collected are used to train the system, complete the initialization of some system databases and the threshold preset work, more accurately describe the normal behavior of the program in all directions, the application of the model is higher than the simple use of the detector under the same false detection rate, the behavior of the program running time is fully mined by introducing the concepts of function sub-sequence, strong sequence set, calling function set, function reference number and the like, and the influence of the setting of the size of the sliding window on the detection rate is reduced due to the introduction of the strong sequence.

[0096] The above merely describes preferred specific embodiments of the present application, but the protection scope of the present application is not limited thereto, any person skilled in the art can make equivalent replacements or changes according to the technical solution and the inventive concept of the present application within the technical range disclosed by the present application, which should be covered within the protection scope of the present application.

Claims

1. A passive magnetic anomaly detection and localization system based on a CNN neural network, characterized in that, include: The detection gate (2) is used to install the sensing layer module (1), and the user layer module (3) is provided on the side of the detection gate (2). And an application layer module (4) for interfacing the perception layer module (1) and the user layer module (3); The perception layer module (1) is used for data acquisition and positioning, and the application layer module (4) is used to receive the data acquired by the perception layer module (1). The application layer module (4) drives the user layer module (3) based on the collected data; The perception layer module (1) includes a sensor module (11) and a millimeter-wave radar module (12). The sensor module (11) is composed of multiple single-axis fluxgate sensors (111), and the single-axis fluxgate sensors (111) are arranged in an array on the detection gate (2). The millimeter-wave radar module (12) is set on the detection gate (2). The signal data in the perception layer module (1) is transmitted back to the application layer module (4) through the UART serial port. The application layer module (4) includes a main chip (41), a CNN deep neural network signal detection module (42), an artificial intelligence signal analysis module (43), and an abnormal magnetic target localization module (44). The CNN deep neural network signal detection module (42), the artificial intelligence signal analysis module (43), and the abnormal magnetic target localization module (44) are respectively located in the main chip (41). The main chip (41) is used to connect with the perception layer module (1). The CNN deep neural network signal detection module (42) is used to detect signals and transmit the detected signals to the artificial intelligence signal analysis module (43) for processing. The artificial intelligence signal analysis module (43) classifies the signals through the CNN network. The user layer module (3) includes a buzzer (31), an LED light (32) and a large screen (33), and the buzzer (31), LED light (32) and large screen (33) are respectively connected to the application layer module (4); The artificial intelligence signal analysis module (43) performs signal analysis through software. Based on the floating-point operation resources inside the main chip (41), the artificial intelligence signal analysis module (43) analyzes the signal through a CNN neural network to determine whether there is a magnetic anomaly signal. Furthermore, the CNN neural network employs a single-channel model or a multi-channel model, wherein the single-channel model includes: using a one-dimensional CNN to transform the signal into a two-dimensional signal; using a two-dimensional CNN for channel expansion and feature extraction; using a GRU for historical information tracking and connection; and using a fully connected layer (FC) for probability estimation and output.

2. The system according to claim 1, characterized in that, The millimeter-wave radar module (12) has an antenna emission angle FOV of 40°×80° and a detection distance of 3-5m. It is used to detect the passage of human bodies in real time and track the real-time position of human movement.

3. The system according to claim 1, characterized in that, The sampling frequency of the single-axis fluxgate sensor (111) is adjusted to 20Hz to 200Hz, and the array spacing is 30-50cm.

4. The system according to claim 1, characterized in that, The multi-channel model expands the dimensions of the single-channel model, increasing the model structure, and includes a data simulation process. This process includes: selecting a segment of recorded background noise for generalization, with generalization operations including Gaussian noise superposition, signal inversion, and gain randomization; simulating the magnetic anomaly signal using both pure simulation and methods based on recorded magnetic anomalies; setting an appropriate signal-to-noise ratio and superimposing the background noise with the magnetic anomaly signal; performing gain generalization on the generated signal; obtaining the final signal to be detected and generating corresponding detection targets based on the magnetic anomaly time period, which serve as the training targets for the model.

5. The system according to claim 1, characterized in that, The CNN deep neural network signal detection module (42) includes an anomaly detection module (421). The anomaly detection module (421) includes a tracker module (4211), a function library module (4212), a strong sequence constructor module (4213), a detector module (4214), a response module (4215), a decision module (4216), and a control module (4217). The tracker module (4211) is used to track processes derived from the program during runtime and obtain the call sequence of these processes to the system. The function library module (4212) The database is used to store system functions. The strong sequence builder module (4213) is used to compensate for the defect of fixed sliding window size. It consists of function set and strong sequence. The detector module (4214) divides the process sequence into normal and abnormal. The response module (4215) interrupts the connection through firewall and router and kills abnormal processes through system commands. The decision module (4216) is used to make decisions on the abnormal situations reported in preprocessing and detector. The control module (4217) is used to construct and update strong function set and set thresholds in detector and preprocessing.

6. A passive magnetic anomaly detection and localization method based on the system described in any one of claims 1-5, characterized in that, Includes the following steps: S1: The single-axis fluxgate sensor (111) in the sensing layer module (1) forms an array, and the sampling frequency is adjusted to 20Hz~200Hz to collect the changes in the magnetic field in the space; S2: The millimeter-wave radar module (12) detects the passage of human bodies in the facing direction in real time. The antenna transmission angle FOV is 40°×80°. When a human body is detected, the CNN deep neural network signal detection module (42) inside the back-end application layer module (4) will be triggered for processing. S3: The millimeter-wave radar module (12) continuously tracks the real-time position of human movement during the detection of the trigger; S4: The magnetic anomaly signal collected by the sensor module (11) will be transmitted back to the main chip (41) via the UART serial port for calculation; S5: The artificial intelligence signal analysis module (43) in the application layer module (4) analyzes the signal through the CNN neural network to determine whether there is a magnetic anomaly signal; The CNN neural network employs a single-channel model or a multi-channel model, wherein the single-channel model includes the following steps: S501: Use a one-dimensional CNN to transform a signal into a two-dimensional signal; S502: Use a 2D CNN for channel expansion and feature extraction; S503: Use GRU for historical information tracking and communication; S504: Probability estimation and output are performed using a fully connected layer (FC). S6: When a magnetic anomaly is actually detected, the buzzer (31), LED light (32) and large screen (33) will sound an alarm through sound, light and screen video.

7. The method according to claim 6, characterized in that, The multi-channel model expands the dimensions of the single-channel model, increasing the model structure, and includes a data simulation process. This process includes: selecting a segment of recorded background noise for generalization, with generalization operations including Gaussian noise superposition, signal inversion, and gain randomization; simulating the magnetic anomaly signal using both pure simulation and methods based on recorded magnetic anomalies; setting an appropriate signal-to-noise ratio and superimposing the background noise with the magnetic anomaly signal; performing gain generalization on the generated signal; obtaining the final signal to be detected and generating corresponding detection targets based on the magnetic anomaly time period, which serve as the training targets for the model.

8. The method according to claim 6, characterized in that, The CNN deep neural network signal detection module (42) is equipped with an anomaly detection module (421), which includes a tracker module (4211), a function library module (4212), a strong sequence builder module (4213), a detector module (4214), a response module (4215), a decision module (4216), and a control module (4217).

9. The method according to claim 8, characterized in that, The preprocessing procedure for tracker data in the anomaly detection module (421) includes the following steps: S100: Further decompose the raw data as needed, remove erroneous data, and convert symbolic fields that the detector cannot process into numerical values; S200: Perform normalization processing to reduce the adverse effects on network training caused by excessive differences in field values ​​between records; S300: Check whether the number of related items in the function set called within a unit of time exceeds the preset threshold.

10. The method according to claim 6, characterized in that, The array spacing of the single-axis fluxgate sensor (111) is 30-50cm, and the sampling frequency is 20Hz to 200Hz.

Images