CN116484177B - Motion-induced noise prediction elimination method for electromagnetic detection of flight platform - Google Patents

Abstract

The present invention proposes a method for predicting and eliminating motion-induced noise for electromagnetic detection of a flying platform, comprising: conducting a test flight for pure noise collection before the formal line survey flight, collecting a set of motion state data and corresponding pure noise data; training a wavelet neural network model based on the test flight motion state data and pure noise data; obtaining the motion state data and receiving data of the line survey flight; mapping the motion state data based on the wavelet neural network model to predict motion-induced noise; removing the predicted motion-induced noise from the receiving data, and completing the motion-induced noise elimination based on the electromagnetic method of the flying platform. The present invention directly uses the motion state measurement data to predict the motion-induced noise, and can be used for semi-aeronautical electromagnetic methods and full-aeronautical electromagnetic methods. It can remove the motion-induced noise of the same frequency as the effective signal, and can predict and eliminate the motion-induced noise of the flying platform with a shock absorbing device.

Description

A method for predicting and eliminating motion-induced noise for electromagnetic detection of flying platforms

Technical Field

The present invention belongs to the technical field of electromagnetic detection based on a flying platform, and in particular relates to a motion state data prediction and elimination of motion induced noise technology, which can be used for full-aerotronic methods and semi-aerotronic methods.

Background technique

Artificial source electromagnetic detection technology is an important branch of geophysical electromagnetic detection. It can be divided into ground, semi-aerial and full-aerial electromagnetic detection methods according to different construction spaces. Among them, semi-aerial and airborne electromagnetic methods are electromagnetic detection methods based on flying platforms, which measure the induced magnetic field in the air by carrying receiving systems on flying platforms such as helicopters, airships and drones. The semi-aerial electromagnetic detection method stimulates underground anomalies through a transmitting source laid on the ground. In the full-aerial electromagnetic method, the transmitting source and the receiving device are carried on the flying platform together, and the signals are transmitted and received in the air. Compared with the ground electromagnetic method, the electromagnetic detection system based on the flying platform is flexible and has strong adaptability to terrain. It can overcome the difficulties in geological resource exploration under complex terrain conditions in my country, and has the advantage of rapid detection at great depth.

Motion-induced noise is the most influential type of noise in all electromagnetic detection methods based on flight platforms. During the flight, the receiving coil is affected by wind speed, wind force, and the flight status of the aircraft, which will cause changes in the attitude angle, causing changes in the effective area of the receiving coil, so that the receiving coil cuts the geomagnetic field and causes changes in the induced magnetic flux. The strength of the geomagnetic field is 6× ^10-5 Tesla, which is 100,000 times stronger than the magnetic field strength of the electromagnetic detection system based on the flight platform. Therefore, any change in the angle of the receiving coil will cause a strong change in the induced electromotive force, introducing motion-induced noise in the received signal. Motion-induced noise reduces the signal-to-noise ratio and greatly affects the detection depth and detection accuracy of the system.

When a helicopter or drone is used as a carrier platform, the propeller rotation frequency is 350Hz-400Hz. The vibration of this frequency will be transmitted to the magnetic field sensor through the rope system, introducing vibration interference. At the same time, whether a helicopter, drone or airship is used as a carrier platform, the sudden change of terrain, wind speed, wind direction, or sudden airflow impact will cause the change of the flight platform's motion state to be transmitted to the magnetic field sensor through the rope system. At this time, the receiving coil itself will also cause high-frequency vibration due to the impact of airflow disturbance. Therefore, in general, when designing the system, a shock-absorbing structure will be added to the receiving coil to reduce the vibration frequency of the coil, which will make the motion state of the coil itself unpredictable.

The research on motion-induced noise removal technology can be roughly divided into three approaches.

The first approach is to design correction factors and filter coefficients based on the generation mechanism of motion-induced noise by observing the flight process, or directly use the measured attitude data to calculate and remove motion-induced noise. Liu Fubo and others from the Institute of Space Science and Technology of the Chinese Academy of Sciences directly used attitude measurement data to predict motion-induced noise through the rotation matrix method. Direct observation of the flight process has low observation accuracy and limited effect on suppressing motion-induced noise. Using attitude data to calculate and remove motion-induced noise is a good idea, however, this method cannot be used in systems with shock absorbers.

The second approach is to remove motion-induced noise based on the characteristics of motion-induced noise using signal-noise separation and data reconstruction. Representative works include: Buselli et al., considering the low-frequency characteristics of avionics motion-induced noise, proposed the use of high-pass filters to remove motion-induced noise; Li Nan used wavelet decomposition and wavelet threshold method to remove noise; Zhu Kaiguang et al., based on the correlation between noise and signal, proposed the use of principal component analysis for denoising; Liu Fubo et al., considering the periodicity of effective signals, proposed the empirical mode decomposition method; Li Yuan et al., integrating the periodicity of effective signals with the low-frequency characteristics of noise, reconstructed the high-order empirical mode quantities obtained by empirical mode decomposition and performed high-pass filtering, and then reconstructed the signal together with the low-order empirical modes to preserve the effective signal as much as possible.

However, when the signal-to-noise distinction is not strong, the signal-to-noise separation often fails, which can lead to the following results: 1) Part of the effective signal is damaged along with the removal of motion-induced noise; 2) The effective signal is kept intact, but the motion-induced noise is not removed cleanly; 3) The motion-induced noise is not removed cleanly, and the effective signal is damaged. More typically, when the motion-induced noise is at the same frequency as the effective signal, such methods are often useless.

The third approach is to treat the late signal as pure motion-induced noise data, fit it and extend it to the full-wave half-cycle signal, and then subtract the fitted noise data from the original signal. This method assumes that after the transmission signal is turned off, the secondary field signal gradually weakens to near zero, which is the annihilation of the motion-induced noise. Therefore, the late data after the annihilation point can be treated as pure motion-induced noise and fitted and extended to calculate the motion-induced noise of the entire half-cycle. However, motion-induced noise has the characteristics of non-stationarity, and it is not appropriate to extend the late signal to the entire half-cycle. In addition, the amount of late data after the annihilation point is small, and the noise source is complex and extremely unstable, which will inevitably lead to an unstable fitting process and severe fitting distortion.

Motion-induced noise is caused by the change of the motion state of the receiving coil. Therefore, we can calculate and eliminate the motion-induced noise by measuring the motion state of the receiving coil. However, the sampling frequency limit of the motion state measurement sensor makes it impossible to predict the high-frequency motion-induced noise. This part of the noise is often very strong and mixed with the effective signal frequency. Therefore, in the electromagnetic system based on the flight platform, a shock-absorbing structure is usually added to the receiving coil to reduce the vibration frequency of the coil, which will make the motion state of the coil itself unmeasurable. At this time, the motion state sensor is installed outside the shock-absorbing device, and the measured motion state is not the motion state of the coil, but it is still closely related to the motion state of the receiving coil. At the same time, the motion state of the receiving coil is related to wind force and speed, the overall structure of the system and the flight conditions. Its motion process is constrained by the overall structure of the system, and has certain internal relationships and constraints with the overall motion state of the system. For the same set of avionics systems, the internal relationships and constraints are the same. However, these internal relationships and constraints are too complex to be expressed by formula derivation. How to make full use of the measured motion state, solve the motion-induced noise through this internal relationship and constraint conditions, and complete the prediction and elimination of the motion-induced noise of the electromagnetic detection system based on the flight platform with a shock-absorbing device is the main research content of the present invention.

Electromagnetic methods based on flying platforms include time-domain full-aircraft electromagnetic method, frequency-domain full-aircraft electromagnetic method and semi-aircraft electromagnetic method. Figure 1 is a typical device in the time-domain full-aircraft electromagnetic method: a co-centered time-domain aircraft electromagnetic device. The co-centered time-domain aircraft electromagnetic device has a pod mainly composed of a transmitting coil, a compensation coil and a receiving coil. The three coils are connected by a rope, and the motion states of the three are not exactly the same. The sensors (three-axis attitude, three-axis acceleration, and three-component velocity sensors) for measuring motion state data in the co-centered time-domain electromagnetic system are installed on the outside of the transmitting coil with a shock-absorbing device, and are also installed on the compensation coil (as shown in Figure 4) or the transmitting coil. The method described in this patent can be used in the above-mentioned cases. Figure 2 is a typical device in the frequency-domain full-aircraft electromagnetic method. The sensor for measuring motion state data is often installed in the pod. Figure 3 is a typical device of the semi-aeronautical transient electromagnetic method. The device only receives signals in the air, and the sensor for measuring motion state data is installed on a receiving device with a shock-absorbing structure.

Summary of the invention

To solve the above technical problems, the present invention proposes a technology for predicting and eliminating motion-induced noise for electromagnetic detection of flying platforms. The motion state data is used to predict and eliminate motion-induced noise through a wavelet neural network. The wavelet neural network method is used to establish a set of complex nonlinear mapping networks to describe the intrinsic relationship and constraints between the motion of the receiving coil and the overall structure of the system, flight conditions, flight acceleration, etc. In a system with a shock-absorbing device outside the transmitting coil, the network can also be used to simulate the damping process of the shock-absorbing device.

A method for predicting and eliminating motion-induced noise for electromagnetic detection of a flying platform, comprising:

Before the formal survey flight, a pure noise collection test flight with the launch turned off shall be carried out in the normal survey flight mode. The flight time shall not be less than half an hour, and a set of motion state data and pure noise data shall be collected.

Based on the experimental flight motion state data and pure noise data, training a wavelet neural network model;

Obtain the motion status data and receiving data of the survey line flight;

Mapping the motion state data based on the wavelet neural network model to predict motion induced noise;

Obtain the motion status data and receiving signals of the survey line flight;

Mapping the motion state data based on the wavelet neural network model to predict motion induced noise;

The predicted motion induced noise is removed from the received data, thereby completing the motion induced noise elimination based on the electromagnetic method of the flight platform.

Optionally, the flying platform electromagnetic method is a semi-aeronautical electromagnetic method or a fully aeronautical electromagnetic method; the flying platform mounting tool is a helicopter, an airship or a UAV; the motion state data includes three-axis attitude, three-axis acceleration and three-component velocity; the motion induced noise is the noise caused by the receiving coil cutting the geomagnetic field.

Optionally, in the time domain airborne electromagnetic system, the sensor for detecting the motion state data is installed on the compensation coil, the receiving coil or the shock absorbing device wrapping the receiving coil, and the receiving coil and the compensation coil are hard-connected or soft-connected, wherein the hard connection is a hard bracket and the soft connection is a soft rope; in the frequency domain airborne electromagnetic system, the sensor for detecting the motion state data is installed on the integral pod; in the semi-airborne electromagnetic system, the sensor for detecting the motion state data is installed on the receiving coil or the shock absorbing device wrapping the receiving coil.

Optionally, the test flight is a straight-line flight of no less than half an hour conducted in a normal line-surveying flight mode. The difference from the normal line-surveying flight is that the transmitting device is turned off during the test flight, so there is no valid signal input into the receiving device.

Optionally, when the wavelet neural network model is trained using the experimental flight motion state data and pure noise data, its output is pure noise data Y=[Y ₁ ,Y ₂ ,…Y _i ,…,Y _q ], where _Yi is the pure noise data at time _ti .

When using the experimental flight motion state data and pure noise data to train the wavelet neural network model, its input is the motion state data:

X＝[X ₁ ,X ₂ ,…X _i ,…,X _q ]

Where θ is the attitude angle, a is the acceleration, and v is the velocity. Their first subscripts represent the axis. For example, for the attitude angle θ, the first subscript is 1 for the attitude angle rotating around the x-axis, 2 for the attitude angle rotating around the y-axis, and 3 for the attitude angle rotating around the z-axis. m, n, and p represent the number of sampling points of the attitude angle, acceleration, and velocity between k seconds before time t and k seconds after time t. _Xi represents the motion state parameters related to the noise _Yi at time _t .

Compared with the prior art, the present invention has the following advantages and technical effects:

For the same avionics system, the damping properties of the shock absorber are the same, and the motion of the receiving coil in the shock absorber has the same intrinsic relationship and constraints with the system structure, flight conditions, flight acceleration, etc. Through the method of the present invention, a set of complex nonlinear mapping networks are established using wavelet neural networks to describe the intrinsic relationship and constraints between the motion of the receiving coil and the overall structure of the system, flight conditions, flight acceleration, etc., and the damping process of the shock absorber can be simulated in a system with a shock absorber, and finally a mapping relationship (wavelet neural network) reflecting the system motion state to motion induced noise is obtained. The present invention directly uses motion state measurement data to map motion induced noise, and can remove the same frequency motion induced noise as the effective signal. It can predict and eliminate motion induced noise in a system with a shock absorber.

BRIEF DESCRIPTION OF THE DRAWINGS

The drawings constituting a part of this application are used to provide a further understanding of this application. The illustrative embodiments and descriptions of this application are used to explain this application and do not constitute an improper limitation on this application. In the drawings:

FIG1 is a schematic diagram of the detection principle of a central loop type device commonly used in a time domain full airborne electromagnetic detection device in the background art;

FIG2 is a schematic diagram of a typical device in a frequency domain full airborne electromagnetic detection method in the background art;

FIG3 is a schematic diagram of a typical device in a semi-aerial electromagnetic detection method in the background art;

FIG4 is a schematic diagram of a transmitting-receiving plane of an existing receiving coil of a center loop type device in the background art, wherein a damping device is installed on the receiving coil;

FIG5 is a schematic diagram of the basic structure of a neural network according to an embodiment of the present invention;

FIG6 is a schematic diagram of a single neuron model according to an embodiment of the present invention;

FIG. 7 is a schematic flow chart of a method for eliminating airborne electromagnetic motion-induced noise based on a wavelet neural network according to an embodiment of the present invention.

Detailed ways

It should be noted that, in the absence of conflict, the embodiments and features in the embodiments of the present application can be combined with each other. The present application will be described in detail below with reference to the accompanying drawings and in combination with the embodiments.

It should be noted that the steps shown in the flowcharts of the accompanying drawings can be executed in a computer system such as a set of computer executable instructions, and that, although a logical order is shown in the flowcharts, in some cases, the steps shown or described can be executed in an order different from that shown here.

As a type of machine learning algorithm, artificial neural network is a research hotspot in the field of artificial intelligence that emerged in the 1980s. This algorithm establishes networks with different connection structures by imitating the way the brain processes information. After specific training, such networks become complex nonlinear mapping functions that can achieve certain specific functions. The basic structure of a neural network is shown in Figure 5. It consists of an input layer, a hidden layer, and an output layer, reflecting a mapping relationship between input and output data. The hidden layer consists of a series of neurons (as shown in Figure 6, its activation function g(r) is replaced by the wavelet basis function Ψ). The output of the neuron is the weighted sum of all its inputs according to the corresponding weights, and then added to the bias to obtain the state function r, and finally the state function r is nonlinearly activated by a certain specific activation function g(r).

Wavelet neural network (WNN) is a neural network structure obtained by replacing the activation function of each neuron in the neural network with a wavelet function, and replacing the corresponding weights from the input layer to the hidden layer and the threshold of the hidden layer with the scale expansion factor and time translation factor of the wavelet function respectively. WNN combines the advantages of artificial neural network and wavelet analysis, which can make the network converge quickly and avoid falling into local optimality, and also has the characteristics of time-frequency local analysis.

For the damping process with a shock-absorbing structure, the movement of the receiving coil in the shock-absorbing structure is intrinsically related to the system structure, flight conditions, flight acceleration, and other constraints. The three-axis attitude, three-axis acceleration, three-component velocity and other motion state data of the measurement system are used as network input, and the pure motion-induced noise received during the corresponding period is not transmitted as output to train the model to obtain a set of neural network parameters. Finally, the motion state data collected in the survey line acquisition phase is used to map and predict the motion-induced noise through the neural network, and it is subtracted from the measured survey line data to obtain a denoised signal. According to this denoising idea, this embodiment proposes a four-stage workflow (as shown in Figure 7), using a wavelet neural network to predict and eliminate motion-induced noise.

1. Data collection. Before the formal survey flight, a pure noise collection test flight with the transmission turned off is carried out in the normal survey flight mode. The flight is a straight-line flight lasting no less than half an hour, so as to collect a set of motion state data and reception data; the reception data does not contain effective signals, but only contains motion-induced noise caused by the coil cutting the geomagnetic field, which is pure noise data.

After collecting pure noise data, turn on the transmission and perform normal flight along the designed flight line to collect motion status data and receive data.

2. Construct a wavelet neural network model, use the motion state data collected during the test flight as input and pure noise data as output to perform WNN training.

3. Using the trained WNN, the motion state data collected by the survey line flight are taken as input to predict the motion-induced noise corresponding to the survey line data where the Earth response and motion-induced noise coexist.

4. Subtract the predicted motion-induced noise from the corresponding survey line data.

During the data acquisition phase, in order to better utilize the intrinsic relationship between the motion state of the transmitting plane and the motion-induced noise, thereby calculating the motion-induced noise data and separating the effective signal, we installed as many sensors as possible on the transmitting coil plane to obtain more relevant motion state data: three-axis attitude sensor, three-axis acceleration sensor, and three-axis velocity sensor.

The second stage of sample set training is considered a critical step:

(1) constructing the wavelet neural network model: setting the number of elements contained in each layer of the wavelet neural network model according to the characteristics of the signal;

(2) Initializing the wavelet neural network model: randomly initializing the scale factor and the translation factor, determining the input and output connection weights of the wavelet function, and setting the learning rate;

(3) Predicting output: obtaining the input value of each node of the hidden layer in the wavelet neural network model; inserting the input value of each node of the hidden layer into the parent wavelet basis function to calculate the output value of the hidden layer; using the output connection weight to calculate the output value of the wavelet neural network model;

(4) Update parameters: calculate the network prediction deviation; modify the scale factor, translation factor and weight of the neural network connection of the wavelet function based on the network prediction deviation; evaluate the quality of the network output and complete the iteration.

For our input data, since the motion-induced noise is related to the change of posture, the shock-absorbing device has damping motion. Therefore, when predicting the motion-induced noise at time t, this embodiment must consider the motion state data between k seconds before time t and k seconds after time t.

The output of this embodiment is pure noise data Y=[Y ₁ ,Y ₂ ,…Y _i ,…,Y _q ], where _Yi is the pure noise data at time _ti ;

The input of this embodiment is motion state data:

X＝[X ₁ ,X ₂ ,… _Xi ,…,x _q ]

Where θ is the attitude angle, a is the acceleration, and v is the velocity. Their first subscripts represent the axis. For example, for the attitude angle θ, the first subscript is 1, which represents the attitude angle rotating around the x-axis, 2, which represents the attitude angle rotating around the y-axis, and 3, which represents the attitude angle rotating around the z-axis. m, n, and p represent the number of sampling points of the attitude angle, acceleration, and velocity between k seconds before time t and k seconds after time t. If their sampling rates are the same, m, n, and p are equal. If the sampling rates are different, the three are not equal. _Xi represents the motion state parameter related to the noise _Yi at time _t .

This embodiment proposes to turn off the transmission signal before line measurement, and collect a set of system motion state data (three-axis attitude, three-axis acceleration, three-component velocity) and corresponding pure noise data under normal flight operation. The effective signal and the motion induced noise are not coupled and have an additive relationship. Therefore, we can use the mapping relationship between the motion state measurement data and the motion induced noise under this operation to the line data containing the effective signal.

The mapping relationship between the planar motion state measurement data of the transmitting coil and the corresponding motion induced noise data in this embodiment includes the damping process of the shock absorbing device, the intrinsic relationship and constraints between the motion of the receiving coil in the shock absorbing device and the system structure, flight conditions, flight acceleration, etc. The mapping relationship is extremely complex, so it is proposed to use a wavelet neural network to train and describe the mapping relationship, and to train the network through the motion state measurement data and pure noise data to obtain network parameters.

This embodiment proposes to use the motion state measurement data of the survey line as input, use the trained wavelet neural network model to predict the motion induced noise, and subtract it from the survey line data. This method is effective for the same-frequency motion induced noise as the effective signal.

The advantages of this embodiment are:

For the same avionics system, the damping properties of the shock absorber are the same, and the movement of the receiving coil in the shock absorber has the same intrinsic relationship and constraints as the system structure, flight conditions, flight acceleration, etc. Through the method of this patent, the wavelet neural network is used to train and map the intrinsic relationship and constraints, which can fully utilize this intrinsic relationship and constraints for denoising and improve the denoising accuracy. This method directly maps motion-induced noise with motion state measurement data, and can remove the same-frequency motion-induced noise as the effective signal. It can predict and eliminate motion-induced noise for systems with shock absorbers.

The above are only preferred specific implementations of the present application, but the protection scope of the present application is not limited thereto. Any changes or substitutions that can be easily thought of by a person skilled in the art within the technical scope disclosed in the present application should be included in the protection scope of the present application. Therefore, the protection scope of the present application should be based on the protection scope of the claims.

Claims (3)

1. A method for predicting and eliminating motion-induced noise for electromagnetic detection of a flying platform, characterized by comprising: Before the formal survey flight, a test flight for pure noise collection is carried out to collect a set of motion state data and pure noise data; The test flight is a straight-line flight of no less than half an hour in the normal line-finding flight mode. The difference from the normal line-finding flight is that the transmitter is turned off during the test flight and there is no effective signal input to the receiving device. Based on the experimental flight motion state data and pure noise data, training a wavelet neural network model; When the wavelet neural network model is trained using the experimental flight motion state data and pure noise data, its output is pure noise data Y = [Y ₁ ,Y ₂ ,…Y _i ,…,Y _q ], where _Yi is the pure noise data at time _ti ; When using the experimental flight motion state data and pure noise data to train the wavelet neural network model, its input is the motion state data: X＝[X ₁ ,X ₂ ,…X _i ,…,X _q ] Among them, θ is the attitude angle, a is the acceleration, v is the velocity, and their first subscripts represent the axial direction. For example, for the attitude angle θ, the first subscript is 1, which represents the attitude angle rotating around the x-axis, 2 represents the attitude angle rotating around the y-axis, and 3 represents the attitude angle rotating around the z-axis. m, n, and p represent the number of sampling points of the attitude angle, acceleration, and velocity between k seconds before time t and k seconds after time t. _Xi represents the motion state parameters related to the noise _Yi at time _t . Obtain the motion status data and receiving data of the survey line flight; Mapping the motion state data based on the wavelet neural network model to predict motion induced noise; The predicted motion induced noise is removed from the received data, thereby completing the motion induced noise elimination based on the electromagnetic method of the flight platform. 2. The method for predicting and eliminating motion-induced noise for electromagnetic detection of a flying platform according to claim 1 is characterized in that: the flying platform electromagnetic method is a semi-aeronautical electromagnetic method or a fully aeronautical electromagnetic method; the flying platform hanging tool is a helicopter, an airship or a UAV; the motion state data includes three-axis attitude, three-axis acceleration and three-component velocity; the motion-induced noise is the noise caused by the receiving coil cutting the geomagnetic field. 3. The motion-induced noise prediction and elimination method for electromagnetic detection of a flying platform according to claim 1 is characterized in that: in a time-domain airborne electromagnetic system, a sensor for detecting motion state data is installed on a compensation coil, a receiving coil or a shock absorbing device wrapping a receiving coil, and a hard connection or a soft connection is formed between the receiving coil and the compensation coil, wherein the hard connection is a hard bracket and the soft connection is a soft rope; in a frequency-domain airborne electromagnetic system, a sensor for detecting motion state data is installed on an integral pod; in a semi-airborne electromagnetic system, a sensor for detecting motion state data is installed on a receiving coil or a shock absorbing device wrapping a receiving coil.