A fast quantum magnetic measurement method and device based on multi-layer perception

Abstract

The present application discloses a fast quantum magnetic measurement method and device based on multi-layer perception, which relates to the field of electrical quantum sensing. It includes: constructing a mixed distribution function model to be trained and its training set; based on the training set, training the mixed distribution function model to be trained to obtain a mixed distribution function model; performing microwave variable frequency scanning within a preset microwave sweep frequency range, recording the fluorescence intensity of the diamond NV color center corresponding to different microwave frequencies, and recording the points between the falling edge point of the first deviation from the fluorescence intensity peak and the rising edge point of the first deviation from the fluorescence intensity trough as valid points; based on the microwave frequency value corresponding to the valid point and the mixed distribution function model, fitting the mixed distribution function of the magnetic field to be measured; obtaining the frequency value of the trough position in the mixed distribution function of the magnetic field to be measured, and calculating the magnetic field intensity to be measured. This solves the problem that the existing optical detection magnetic resonance technology cannot meet the demand for high sampling rate of the power system in terms of sampling rate.

Description

Technical Field

[0001] The present application relates to the field of electrical quantum sensing technology, and in particular to a fast quantum magnetic measurement method and device based on multi-layer sensing. Background Art

[0002] In related technologies, the measurement of magnetic field intensity of diamond NV color center mainly adopts pulsed magnetometry. This method requires multi-channel pulse control of laser and microwave signals, and the operation steps are complicated and difficult to implement. In the process of magnetic field measurement, the NV color center will emit a red fluorescence signal under the combined action of laser polarization, microwave source and magnetic field to be measured. The key step to obtain magnetic field information is to find the sweep frequency of the microwave source at the moment of minimum value of the fluorescence signal. The frequency of microwaves determines whether the electron ratio in the electronic energy level of the NV color center can be changed, while the power of microwaves determines the speed of change of the electron ratio. Specifically, Figure 1 As shown in the figure, the 0-level electrons of the NV color center can transition to the +1 energy level and -1 energy level under microwave excitation. This transition is determined by the microwave frequency and the energy difference between the energy levels. By scanning the frequency difference between ms = 0 and ms = ±1, the electrons at ms = 0 can absorb energy and transition to the +1 energy level and -1 energy level. When the microwave frequency is equal to the resonant frequency, the electron transition will reduce the difference in the number of electrons at the energy levels.

[0003] When the NV color center is affected by laser and microwave at the same time, the electrons at the 0 energy level will emit stronger fluorescence, while the electrons at the +1 and -1 energy levels will emit weaker fluorescence. When the microwave frequency does not match, the electrons gather at the 0 energy level and the fluorescence intensity is higher; when the microwave frequency is close to the resonance frequency, some of the 0 energy level electrons transition to the +1 and -1 energy levels, and the fluorescence intensity decreases. Therefore, Figure 2 As shown, two peaks will appear during the frequency sweep, corresponding to the resonant frequencies of +1 and -1. The frequency difference between these peaks reflects the frequency of the external magnetic field. The splitting of the trough frequency and the center frequency on the waveform can be expressed by the following formula:

[0004]

[0005] In the above formula, f ±1 is the microwave resonance frequency of the NV color center, the center frequency f c is 2.87GHz, B is the external magnetic field strength, and θ is the angle between the external magnetic field and the sensitive direction of the NV color center. When the angle between the external magnetic field and the sensitive direction of the NV color center coincides, the magnetic field strength of 100Gs corresponds to a frequency splitting size of 280MHz (|ff c |).

[0006] In order to detect the resonance frequency or zero-field splitting size of the ground state energy level of the NV color center, continuous optical detection magnetic resonance (cw-ODMR) is a widely used method. Continuous optical detection magnetic resonance technology generally measures the difference between the two trough frequencies of ODMR by sweeping the microwave frequency within a certain range. The time T for a complete sampling rate is expressed by the following formula:

[0007] T=T L +T W +T D (2)

[0008] Where T L is the laser polarization time. According to different measurement methods, laser polarization can be divided into pulse irradiation and long-term irradiation. The polarization time is generally in the order of microseconds. W is the microwave modulation time. In the frequency sweep method, the entire frequency range is swept at intervals of a certain step length, which generally takes time in the order of ms to seconds. D is the fluorescence detection time, which is detected by photodetector, and the detection time is generally very short. From the above analysis, it can be seen that the measurement time of the sweep frequency method is T W It accounts for the largest proportion, reaching the s order of magnitude.

[0009] Due to different application scenarios, magnetic measurement has different requirements for sampling rate. In many measurement scenarios, there are higher requirements for sampling rate. For example, in power systems, the sampling rate of AC current needs to be at least 4kHz, and the sampling rate of DC current needs to be at least 10kHz. However, the existing continuous optical detection magnetic resonance technology is limited by the sweep speed, and the sampling rate can only reach the Hz level, which is difficult to meet these requirements, limiting its application in power systems or other scenarios with high sampling rate requirements. Summary of the invention

[0010] The purpose of this application is to provide a fast quantum magnetic measurement method and device based on multi-layer sensing, so as to solve the problem that the existing optical detection magnetic resonance technology cannot meet the high sampling rate requirements of the power system in terms of sampling rate.

[0011] To achieve the above objectives, this application adopts the following technical solutions:

[0012] On the one hand, the present application provides a fast quantum magnetic measurement method based on multi-layer perception, comprising:

[0013] Constructing a mixed distribution function model to be trained and its training set, wherein the training set includes microwave frequency, fluorescence intensity and parameters of a mixed distribution function, wherein the mixed distribution function is used to fit the relationship between microwave frequency and fluorescence intensity of diamond NV color centers under different magnetic field intensities;

[0014] Based on the training set, the mixed distribution function model to be trained is trained to obtain a mixed distribution function model;

[0015] Perform microwave frequency conversion scanning within a preset microwave frequency sweep range, record the fluorescence intensity of the diamond NV color center corresponding to different microwave frequencies, and record the point between the falling edge point that first deviates from the fluorescence intensity peak and the rising edge point that first deviates from the fluorescence intensity trough as a valid point;

[0016] Based on the microwave frequency value corresponding to the effective point and the mixed distribution function model, a mixed distribution function of the magnetic field to be measured is obtained by fitting;

[0017] The frequency value of the trough position in the mixed distribution function of the magnetic field to be measured is obtained, and the intensity of the magnetic field to be measured is calculated.

[0018] On the other hand, the present application also provides a fast quantum magnetic measurement device based on multi-layer perception, comprising:

[0019] A model building module, used to build a mixed distribution function model to be trained and its training set, wherein the training set includes microwave frequency, fluorescence intensity and parameters of the mixed distribution function, and the mixed distribution function is used to fit the relationship between microwave frequency and fluorescence intensity of diamond NV color center under different magnetic field intensities;

[0020] A model training module, used for training the mixed distribution function model to be trained based on the training set to obtain a mixed distribution function model;

[0021] A frequency conversion scanning module is used to perform microwave frequency conversion scanning within a preset microwave frequency sweep range, record the fluorescence intensity of the diamond NV color center corresponding to different microwave frequencies, and record the point between the falling edge point that first deviates from the peak value of the fluorescence intensity and the rising edge point that first deviates from the trough value of the fluorescence intensity as a valid point;

[0022] A function inversion module, used for fitting a mixed distribution function of the magnetic field to be measured based on the microwave frequency value corresponding to the effective point and the mixed distribution function model;

[0023] The magnetic field calculation module is used to obtain the frequency value of the trough position in the mixed distribution function of the magnetic field to be measured and calculate the intensity of the magnetic field to be measured.

[0024] On the other hand, the present application also provides an electronic device, including:

[0025] A processor; and a memory arranged to store computer executable instructions, which when executed cause the processor to perform any one of the steps of the above method.

[0026] Based on the above technical solution, this application can achieve the following technical effects:

[0027] This application proposes a fast quantum magnetic measurement method based on multi-layer perception, which optimizes the microwave frequency sweeping method of continuous optical detection of magnetic resonance by means of "OMDR spectrum distribution function fitting + a small number of point measurements + function inversion". Specifically, according to the spectral line distribution characteristics measured by the system and the range of the field to be measured, the function distribution curve obtained by neural network fitting is adopted; by frequency sweeping at very few points, through the combination of light intensity and distribution function, fast waveform inversion is achieved, and then the trough frequency point is quickly determined. Thereby, without adding any devices, the microwave modulation time and data processing time are shortened, and the sampling efficiency of the overall quantum measurement is effectively improved. At the same time, key frequency information is retained. BRIEF DESCRIPTION OF THE DRAWINGS

[0028] Figure 1 It is a schematic diagram of the energy level structure of the diamond NV color center provided by the background technology of this application;

[0029] Figure 2 It is a schematic diagram of the diamond NV color center ODMR spectrum distribution and the characterization of the center frequency and trough frequency provided by the background technology of this application;

[0030] Figure 3 This is a flow chart of a fast quantum magnetic measurement method based on multi-layer sensing provided in one embodiment of the present application;

[0031] Figure 4 This is a schematic diagram of a multi-layer perceptron neural network model architecture provided by an embodiment of the present application;

[0032] Figure 5 is a schematic diagram of microwave frequency scanning comparison provided by an embodiment of the present application;

[0033] Figure 6 It is a schematic diagram of a specific frequency range of the OMDR spectrum distribution function and each distribution fitting provided in an embodiment of the present application;

[0034] Figure 7 is a schematic diagram of a diamond NV color center magnetic measurement device provided in one embodiment of the present application;

[0035] Figure 8 This is a schematic diagram of a fast quantum magnetic measurement device based on multi-layer sensing provided in one embodiment of the present application. DETAILED DESCRIPTION

[0036] The present application is further described in detail below in conjunction with the accompanying drawings and specific embodiments. The advantages and features of the present application will become clearer according to the following description. It should be noted that the accompanying drawings are all in a very simplified form and are not in precise proportions, and are only used to conveniently and clearly assist in explaining the purpose of the embodiments of the present application.

[0037] It should be noted that, in order to clearly explain the content of the present application, the present application specifically cites multiple embodiments to further illustrate different implementation methods of the present application, wherein the multiple embodiments are enumerated rather than exhaustive. In addition, for the sake of brevity of explanation, the contents mentioned in the previous embodiments are often omitted in the subsequent embodiments. Therefore, the contents not mentioned in the subsequent embodiments can refer to the previous embodiments accordingly.

[0038] Example 1

[0039] Please refer to Figure 3 , Figure 3 The figure shows a flow chart of a fast quantum magnetic measurement method based on multi-layer sensing provided by this embodiment. The method may specifically include the following steps:

[0040] Step 101, constructing a mixed distribution function model to be trained and its training set, wherein the training set includes microwave frequency, fluorescence intensity and parameters of a mixed distribution function, and the mixed distribution function is used to fit the relationship between microwave frequency and fluorescence intensity of diamond NV color centers under different magnetic field intensities.

[0041] It should be noted that one implementation of step 101 may be:

[0042] A combined model of Gaussian distribution function, Lorentz distribution function and gamma distribution function is used to describe the morphology of ODMR spectrum. These distribution functions are combined by mixing weight coefficients to form a comprehensive probability density function, which is used to fit the relationship between microwave frequency and fluorescence intensity. A general neural network model is designed to learn the relationship between microwave frequency and fluorescence intensity, which can be specifically:

[0043] Initialize the parameters of the mixed distribution function model, including: Figure 4 As shown, the input layer is a neuron, and the input microwave frequency x i ; Two hidden layers, each with 10 neurons, using ReLU activation function; The output layer has three branches, each branch corresponds to a subnetwork, used to estimate different parameters of the mixture distribution function: Branch 1 outputs λ 1 , μ, σ, respectively represent the weight, mean and standard deviation of the Gaussian distribution function; branch 2 outputs λ 2 , x 0 ,γ, respectively represent the weight, center position and width of the Lorentz distribution function; branch 3 outputs λ 3 ,θ,k,represent the weight, shape and scale of the gamma distribution function, respectively.

[0044] Among them, the joint probability density function of the mixed distribution can be expressed as:

[0045] P(x i )=(λ 1 fGaussian (x i )+λ 2 f Lorentzian (x i )+λ 3 f Gamma (x i )) (3)

[0046] The definitions of each part are as follows:

[0047]

[0048] In the formula, f Gaussian (x i ) is the estimated fluorescence intensity based on the Gaussian distribution function, f Lorentzian (x i ) is the estimated fluorescence intensity based on the Lorentz distribution function, f Gamma (x i ) is the estimated fluorescence intensity based on the gamma distribution function, x i is the microwave frequency,

[0049] I i is the real fluorescence intensity, λ 1 , 2 , 3 are the weights of Gaussian distribution function, Lorentz distribution function and gamma distribution function respectively, μ and σ are the mean and standard deviation of Gaussian distribution function respectively, x 0 and γ are the center position and width of the Lorentz distribution function, θ and k are the shape and scale of the gamma distribution function, is the sample index of the training set, and n is the total number of samples in the training set.

[0050] Network design for mixed distribution function model:

[0051] First hidden layer:

[0052] H 1 =ReLU(W 1 x+b 1 ) (7)

[0053] Among them, W 1 is the weight matrix from the input layer to the first hidden layer, which determines how the input features affect the neurons in the first hidden layer; x is the input vector, that is, the original data features received by the model; b 1 is the bias vector of the first hidden layer, which is used to adjust the activation threshold of neurons.

[0054] Second hidden layer:

[0055] H 2 =ReLU(W 2 H1 +b 2 ) (8)

[0056] Among them, W 2 is the weight matrix from the first hidden layer to the second hidden layer, which determines how the output of the first hidden layer affects the neurons of the second hidden layer; H 1 is the output vector of the first hidden layer, the result after the activation function; b 2 is the bias vector of the second hidden layer, which is used to adjust the activation threshold of the neurons in the second hidden layer.

[0057] Assume that the output layer consists of three branch networks, estimating the parameters of the mixture distribution respectively:

[0058] Branch 1 output:

[0059] (λ 1 ,μ,σ)=W o1 H 2 +b o1 (9)

[0060] Among them, W o1 is the weight matrix from the second hidden layer to the output layer of branch 1, responsible for mapping the hidden layer features to the output parameters of branch 1; b o1 is the bias vector of the output layer of branch 1, which is used to adjust the baseline value of the output parameter.

[0061] Branch 2 output:

[0062] (λ 2 ,x 0 ,γ)=W o2 H 2 +b o2 (10)

[0063] Among them, W o2 is the weight matrix from the second hidden layer to the output layer of branch 2, responsible for mapping the hidden layer features to the output parameters of branch 2; b o2 is the bias vector of the output layer of branch 2, which is used to adjust the baseline value of the output parameter.

[0064] Branch 3 output:

[0065] (λ 3 ,k,θ)=W o3 H 2 +b o3 (11)

[0066] Among them, W o3 is the weight matrix from the second hidden layer to the output layer of branch 3, responsible for mapping the hidden layer features to the output parameters of branch 3; b o3is the bias vector of the output layer of branch 3, which is used to adjust the baseline value of the output parameter.

[0067] The above parameters together constitute the core part of the neural network. The weight matrix and bias vector are the parameters that need to be learned in model training. They determine how the input is mapped to the output.

[0068] Step 102: Based on the training set, the mixed distribution function model to be trained is trained to obtain a mixed distribution function model.

[0069] It should be noted that one implementation of step 102 may be:

[0070] S21. Assume that the fluorescence intensity is estimated It is expressed by the above joint probability density function:

[0071]

[0072] Construct the loss function as the true fluorescence intensity I and the estimated fluorescence intensity The average of the sum of squared errors between:

[0073]

[0074] The above expression represents the true fluorescence intensity I and the estimated fluorescence intensity By minimizing the loss function, the model parameters can be optimized to obtain the optimal fluorescence intensity estimation.

[0075] S22, parameter update: Use gradient descent method to update parameters. The update rules are as follows:

[0076]

[0077] Among them, η represents the learning rate, is the partial derivative of the loss function with respect to the weight parameters, α represents all the weight parameters of the neural network model, and is iteratively updated.

[0078] S23, repeating the process of steps S21-S22 until the value of the mean square error converges or reaches a preset maximum number of iterations, thereby obtaining a mixed distribution function model.

[0079] Step 103, perform microwave frequency conversion scanning within a preset microwave frequency sweep range, record the diamond NV color center fluorescence intensity corresponding to different microwave frequencies, and record the point between the falling edge point that first deviates from the fluorescence intensity peak and the rising edge point that first deviates from the fluorescence intensity trough as a valid point.

[0080] It should be noted that one implementation of step 103 may be:

[0081] According to the calibrated upper limit of the magnetic field measurement of the sensor, the resonance frequency f corresponding to +1 and -1 is calculated using formula (1): +1 and f -1 , f +1 Multiply by the proportionality factor k, f -1 Divide by the proportional coefficient k, and get the sweep frequency range [f min ,f max ];

[0082] The frequency of the RF microwave sweep is from f min Initially, the frequency sweep step is set to f 0 , and then increase the frequency in the preset sweep range, perform microwave frequency scanning, and record the fluorescence intensity of the diamond NV color center. Figure 5 As shown, the falling edge point p of the first deviation from the peak value of fluorescence intensity is recorded. 1 The first deviation from the fluorescence intensity trough appears at the rising edge point p n The points between are recorded as N valid points, and the microwave frequency values ​​corresponding to the positions of the N valid points are obtained.

[0083] Further, according to the initial microwave frequency value f of the microwave frequency conversion scanning min , frequency sweep step length f 0 , the time required for each step t 0 and p n The frequency value corresponding to the point is frequency f p , the time to complete a frequency sweep cycle is calculated as:

[0084] The total time of one magnetic measurement is T 1 for:

[0085] T 1 =T L +T W (15)

[0086] General T L Compared to T W Can be ignored, T 1 ≈T W .

[0087] Based on this, the frequency sweeping time of the present application is different from the traditional time T 2 The ratio is:

[0088]

[0089] Among them, f 0′ is the step length of the traditional frequency sweep, f 0 <f 0′ Within the sensor’s calibration parameters, as the magnetic field strength increases, f p -fmin , the smaller the value, the more obvious the advantage of this application over the traditional microwave modulation method. Therefore, by scanning a very small number of points, the trough frequency can be quickly determined, thereby greatly compressing the time consumption of the microwave modulation stage and improving the sampling rate.

[0090] Step 104: fitting a mixed distribution function of the magnetic field to be measured based on the microwave frequency value corresponding to the effective point and the mixed distribution function model.

[0091] It should be noted that one implementation of step 104 may be:

[0092] like Figure 6 As shown, using the trained mixed distribution function model, through N effective points, the parameters of the mixed distribution function are inverted to determine the frequency point f corresponding to the first trough T .

[0093] Step 105: Obtain the frequency value of the trough position in the mixed distribution function of the magnetic field to be measured, and calculate the intensity of the magnetic field to be measured.

[0094] It should be noted that one implementation of step 105 may be:

[0095] The frequency point f T Substituting into formula (1), the intensity B of the magnetic field to be measured at this time can be calculated.

[0096] In summary, this application proposes a fast quantum magnetic measurement method based on multi-layer perception, which optimizes the microwave frequency scanning method of continuous optical detection of magnetic resonance by the method of "OMDR spectrum distribution function fitting + a small number of point measurements + function inversion". Specifically, according to the spectral line distribution characteristics measured by the system and the range of the field to be measured, the function distribution curve obtained by neural network fitting is adopted; by frequency scanning of very few points, through the combination of light intensity and distribution function, fast waveform inversion is achieved, and then the trough frequency point is quickly determined. Thereby, without adding any devices, the microwave modulation time and data processing time are shortened, and the sampling efficiency of the overall quantum measurement is effectively improved. At the same time, key frequency information is retained.

[0097] Example 2

[0098] This embodiment provides another fast quantum magnetic measurement method based on multi-layer sensing.

[0099] Build as Figure 7 The diamond NV color center magnetic measurement device shown includes:

[0100] 1. Quantum magnetic sensing system: Using diamond NV color center as the core magnetic sensor, it can perform high-precision measurement of weak magnetic fields at room temperature. The quantum control technology can effectively eliminate environmental noise, thereby significantly improving the sensitivity and accuracy of the measurement;

[0101] 2. Laser and optical path system: Use a 532nm laser to couple the laser to the NV color center sample through an optical fiber to achieve efficient excitation of the NV color center. By optimizing the laser power and stability, ensure that the quantum state has a high polarization rate and excellent control fidelity to improve the accuracy and repeatability of the experiment;

[0102] 3. Microwave system: including a high-stability microwave source and its modulation system, which are used to precisely modulate the electron spin state of the NV color center to obtain accurate magnetic field information. The system includes key components such as microwave power amplifier, circulator and microwave antenna to ensure the stability and accuracy of microwave signals;

[0103] 4. ODMR spectrum fast fitting device: A combined model of mixed Gaussian distribution, Lorentz distribution and gamma distribution is used to describe the morphology of the ODMR spectrum, and a neural network is used to quickly fit the ODMR spectrum. A general neural network model is designed to learn the relationship between frequency and fluorescence intensity, and through the combination of light intensity and mixed distribution function, fast waveform inversion is achieved to obtain the ODMR trough frequency point;

[0104] 5. Data acquisition and processing system: Based on field programmable gate array (FPGA) and high-speed data acquisition card, it processes and analyzes the fluorescence signal obtained from the NV color center in real time, uses the ODMR spectrum distribution function fitted by neural network, inverts key parameters, obtains the trough frequency of the ODMR spectrum from the mixed distribution function, and then calculates the magnetic field strength to be measured.

[0105] In this embodiment, the fast quantum magnetic measurement method based on multi-layer sensing specifically includes the following steps:

[0106] Step 201, the mixed distribution function model training phase, the specific steps are:

[0107] The first step is to initialize the network parameters: the input layer is a neuron, and the input microwave frequency is x i ; Two hidden layers, each with 10 neurons, using ReLU activation function; The output layer has three branches, each branch corresponds to a subnetwork, used to estimate different parameters of the mixture distribution: Branch 1 outputs λ 1 ,μ,σ;Branch 2 outputs λ 2 , x 0 ,γ; branch 3 outputs λ 3 ,θ,k.

[0108] Among them, the joint probability density function of the mixed distribution can be expressed as:

[0109] P(x i )=(λ 1 f Gaussian (x i )+λ 2 f Lorentzian (x i )+λ 3 f Gamma (x i )) (3)

[0110] The definitions of each part are as follows:

[0111]

[0112] Step 2: Network design:

[0113] First hidden layer:

[0114] H 1 =ReLU(W 1 x+b 1 ) (7)

[0115] Second hidden layer:

[0116] H 2 =ReLU(W 2 H 1 +b 2 ) (8)

[0117] Assume that the output layer consists of three branch networks, which estimate the parameters of the mixture distribution respectively.

[0118] Branch 1 output:

[0119] (λ 1 ,μ,σ)=W o1 H 2 +b o1 (9)

[0120] Branch 2 output:

[0121] (λ 2 ,x 0 ,γ)=W o2 H 2 +b o2 (10)

[0122] Branch 3 output:

[0123] (λ 3 ,k,θ)=W o3 H 2 +b o3(11)

[0124] The third step is model training:

[0125] Assume that the estimated fluorescence intensity It is expressed by the above joint probability density function:

[0126]

[0127] Construct the loss function as the true fluorescence intensity I and the estimated fluorescence intensity The average of the sum of squared errors between:

[0128]

[0129] This expression represents the true fluorescence intensity I and the estimated fluorescence intensity By minimizing the loss function, the model parameters can be optimized to obtain the optimal fluorescence intensity estimation.

[0130] Step 4: Parameter update: Use gradient descent method to update parameters. The update rules are as follows:

[0131]

[0132] Among them, η represents the learning rate, and α represents all the weight parameters of the neural network model, which are iteratively updated.

[0133] Repeat the process of steps 3 and 4 until the mean square error converges or reaches the preset maximum number of iterations. Output the trained mixture distribution function model.

[0134] Step 202, spin activation stage: In this stage, a laser with a wavelength of 532 nm is emitted from a 532 nm laser source and irradiated on the diamond. The duration of this stage is T L ;

[0135] Step 203, resonance frequency calculation phase: Based on the calibrated upper limit of the magnetic field measurement of the sensor, the resonance frequencies f corresponding to +1 and -1 are calculated using formula (1) +1 and f -1 , f +1 Multiply by the proportionality factor k, f -1 Divide by the proportional coefficient k, and get the sweep frequency range [f min ,f max ];

[0136] Step 204, RF microwave frequency sweeping stage: Each RF microwave frequency sweep starts from f min or max At the beginning, in the sweep frequency interval [f min ,f max] to perform microwave frequency scanning and record the fluorescence intensity of the diamond NV color center. Record the first falling edge point p that deviates from the peak value of the fluorescence intensity. 1 Until the first rising edge point p n , that is, N effective points and the corresponding microwave frequency values, p n The frequency value corresponding to the point is recorded as f p At the same time, the frequency sweep step is f 0 , the time required for each step is t 0 , then the time to complete a frequency sweep cycle is:

[0137] Step 205, frequency inversion stage: using the neural network model trained in step 201, invert the parameters of the mixed distribution function through N effective points, and then determine the frequency point f corresponding to the first trough T ;

[0138] Step 206, magnetic field calculation phase: using formula (1) and f T Calculate the magnetic field strength B to be measured at this time.

[0139] Through the analysis of this process, the total time T of a magnetic measurement is 1 for:

[0140] T 1 =T L +T W (15)

[0141] General T L Compared to T W Can be ignored, T 1 ≈T W .

[0142] According to the general modulation scheme, the measurement time of one microwave stage is:

[0143]

[0144] The ratio of the frequency sweeping time of this application to the traditional time is:

[0145]

[0146] where f 0 <f 0′ , within the sensor's calibration parameters, as the magnetic field strength increases, f p -f min , the smaller the value, the more obvious the advantage.

[0147] From the analysis of the microwave modulation stage, it can be seen that the actual useful signal value is only related to the frequency value of the trough, and a large number of frequency sweeping steps can be omitted. Therefore, if the function distribution of the OMDR spectrum is analyzed, the trough frequency can be quickly determined by scanning a very small number of points and fitting the function, thereby greatly compressing the time consumption of the microwave modulation stage and improving the sampling rate.

[0148] Example 3

[0149] See also Figure 8 , Figure 8 The figure shows a schematic diagram of a fast quantum magnetic measurement device based on multi-layer sensing provided by this embodiment. The device shown specifically includes:

[0150] A model building module 301 is used to build a mixed distribution function model to be trained and its training set, wherein the training set includes microwave frequency, fluorescence intensity and parameters of the mixed distribution function, and the mixed distribution function is used to fit the relationship between microwave frequency and fluorescence intensity of diamond NV color centers under different magnetic field intensities;

[0151] A model training module 302 is used to train the mixed distribution function model to be trained based on the training set to obtain a mixed distribution function model;

[0152] The frequency conversion scanning module 303 is used to perform microwave frequency conversion scanning within a preset microwave frequency sweep range, record the fluorescence intensity of the diamond NV color center corresponding to different microwave frequencies, and record the point between the falling edge point that first deviates from the peak value of the fluorescence intensity and the rising edge point that first deviates from the trough value of the fluorescence intensity as a valid point;

[0153] A function inversion module 304 is used to fit the mixed distribution function of the magnetic field to be measured based on the microwave frequency value corresponding to the effective point and the mixed distribution function model;

[0154] The magnetic field calculation module 305 is used to obtain the frequency value of the trough position in the mixed distribution function of the magnetic field to be measured, and calculate the intensity of the magnetic field to be measured.

[0155] Optionally, the device further comprises a magnetic measurement time calculation module, which is used to:

[0156] According to the initial microwave frequency value of the microwave variable frequency scanning, the frequency sweeping step length, the time required for each step length and the microwave frequency value corresponding to the rising edge point of the first deviation from the fluorescence intensity wave valley, the time of a single frequency sweeping working cycle is obtained;

[0157] Based on the time of the single frequency sweep duty cycle, the total time of one magnetic measurement is calculated.

[0158] Based on this, by scanning a very small number of points, the trough frequency can be quickly determined, thereby greatly compressing the time consumption of the microwave modulation stage and improving the sampling rate.

[0159] In summary, this application proposes a fast quantum magnetic measurement method based on multi-layer perception, which optimizes the microwave frequency scanning method of continuous optical detection of magnetic resonance by the method of "OMDR spectrum distribution function fitting + a small number of point measurements + function inversion". Specifically, according to the spectral line distribution characteristics measured by the system and the range of the field to be measured, the function distribution curve obtained by neural network fitting is adopted; by frequency scanning of very few points, through the combination of light intensity and distribution function, fast waveform inversion is achieved, and then the trough frequency point is quickly determined. Thereby, without adding any devices, the microwave modulation time and data processing time are shortened, and the sampling efficiency of the overall quantum measurement is effectively improved. At the same time, key frequency information is retained.

[0160] Example 4

[0161] In another feasible embodiment, this embodiment provides a device for fast quantum magnetic measurement based on multi-layer sensing, and the device may specifically include:

[0162] A processor; and a memory arranged to store computer executable instructions, which, when executed, cause the processor to perform the steps in any of the above method embodiments.

[0163] Example 5

[0164] In another feasible embodiment, this embodiment provides a storage medium for fast quantum magnetic measurement based on multi-layer sensing, and the storage medium may specifically include:

[0165] The storage medium stores a processing program for fast quantum magnetic measurement based on multi-layer sensing, and when the processing program for fast quantum magnetic measurement based on multi-layer sensing is executed by a processor, the steps in any of the above method embodiments are implemented.

[0166] The above is only an embodiment of the present application and is not intended to limit the present application. For those skilled in the art, the present application can be modified and varied in various ways. Any modification, equivalent replacement, improvement, etc. made within the spirit and principle of the present application should be included in the scope of the claims of the present application.

Claims

1. A fast quantum magnetic measurement method based on multi-layer perception, characterized in that: include: Constructing a mixed distribution function model to be trained and a training set thereof, wherein the training set includes microwave frequency, fluorescence intensity and parameters of a mixed distribution function, wherein the mixed distribution function is used to fit the functional relationship between microwave frequency and fluorescence intensity of diamond NV color centers under different magnetic field intensities; Based on the training set, the mixed distribution function model to be trained is trained to obtain a mixed distribution function model; Perform microwave frequency conversion scanning within a preset microwave frequency sweep range, record the fluorescence intensity of the diamond NV color center corresponding to different microwave frequencies, and record the point between the falling edge point that first deviates from the fluorescence intensity peak and the rising edge point that first deviates from the fluorescence intensity trough as a valid point; Based on the microwave frequency value corresponding to the effective point and the mixed distribution function model, a mixed distribution function of the magnetic field to be measured is obtained by fitting; The frequency value of the trough position in the mixed distribution function of the magnetic field to be measured is obtained, and the intensity of the magnetic field to be measured is calculated.

2. The method according to claim 1, characterized in that The step of constructing a mixed distribution function model to be trained includes: Initialize the parameters of the mixed distribution function model to be trained, wherein the mixed distribution function model to be trained includes an input layer, a hidden layer, and an output layer, specifically including: Initializing the number of neurons in the input layer to 1, and the input parameters include microwave frequency and fluorescence intensity; Initialize the number of neurons in the hidden layer to 10, and use the ReLU activation function; Initialize the output parameters of the first branch of the output layer to include parameters of a Gaussian distribution function, the output parameters of the second branch to include parameters of a Lorentzian distribution function, and the output parameters of the third branch to include parameters of a gamma distribution function.

3. The method according to claim 2, characterized in that The step of training the mixed distribution function model to be trained based on the training set to obtain the mixed distribution function model comprises: S21, constructing a loss function based on a mean square error between the true fluorescence intensity and the estimated fluorescence intensity, wherein the estimated fluorescence intensity is characterized by a parameter of the mixed distribution function; S22, iteratively updating the weight parameters of the mixed distribution function model to be trained by using a gradient descent method; S23. Repeat the above steps S21-S22 until the mean square error converges or reaches a preset number of iterations, and output a mixed distribution function model.

4. The method according to claim 3, characterized in that The loss function is constructed based on the mean square error between the true fluorescence intensity and the estimated fluorescence intensity, wherein the estimated fluorescence intensity is characterized by the parameters of the mixed distribution function, including: The loss function is constructed by the following formula: In the formula, f Gaussian (x i ) is the estimated fluorescence intensity based on the Gaussian distribution function, f Lorentzian (x i ) is the estimated fluorescence intensity based on the Lorentz distribution function, f Gamma (x i ) is the estimated fluorescence intensity based on the gamma distribution function, x i is the microwave frequency, I i is the true fluorescence intensity, λ1, λ2, λ3 are the weights of Gaussian distribution function, Lorentz distribution function and gamma distribution function, μ and σ are the mean and standard deviation of Gaussian distribution function, x0 and γ are the center position and width of Lorentz distribution function, θ and k are the shape and scale of gamma distribution function, i is the sample index of training set, and n is the total number of samples in training set.

5. The method according to claim 3, characterized in that: The iterative updating of the parameters of the mixed distribution function by using the gradient descent method comprises: Iterative update is performed using the following formula: In the formula, η is the learning rate, α is the weight parameter of the mixed distribution function model, is the partial derivative of the loss function with respect to the weight parameter.

6. The method according to claim 1, characterized in that Before performing microwave frequency conversion scanning within the preset microwave frequency sweep range, recording the fluorescence intensity of the diamond NV color center corresponding to different microwave frequencies, and recording the point between the falling edge point that first deviates from the peak value of the fluorescence intensity wave and the rising edge point that first deviates from the trough value of the fluorescence intensity wave as a valid point, the method further includes: Based on the upper limit of the sensor's magnetic field measurement, the microwave resonance frequency of the diamond NV color center is calculated; Based on the microwave resonance frequency and the preset proportionality coefficient, a preset microwave frequency sweep range is obtained.

7. The method according to claim 6, characterized in that After performing microwave frequency conversion scanning within the preset microwave frequency sweep range, recording the fluorescence intensity of the diamond NV color center corresponding to different microwave frequencies, and recording the point between the falling edge point that first deviates from the fluorescence intensity peak and the rising edge point that first deviates from the fluorescence intensity trough as a valid point, the method further includes: According to the initial microwave frequency value of the microwave variable frequency scanning, the frequency sweeping step length, the time required for each step length and the microwave frequency value corresponding to the rising edge point of the first deviation from the fluorescence intensity wave valley, the time of a single frequency sweeping working cycle is obtained; Based on the time of the single frequency sweep duty cycle, the total time of one magnetic measurement is calculated.

8. The method according to claim 1, characterized in that The fitting of the mixed distribution function of the magnetic field to be measured based on the microwave frequency value corresponding to the effective point and the mixed distribution function model comprises: Inputting the microwave frequency value corresponding to the effective point into the mixed distribution function model, and inverting to obtain the parameters of a new mixed distribution function; Based on the parameters of the new mixed distribution function, a mixed distribution function of the magnetic field to be measured is obtained by fitting.

9. A fast quantum magnetic measurement device based on multi-layer perception, characterized in that: include: A model building module, used to build a mixed distribution function model to be trained and its training set, wherein the training set includes microwave frequency, fluorescence intensity and parameters of the mixed distribution function, and the mixed distribution function is used to fit the relationship between microwave frequency and fluorescence intensity of diamond NV color center under different magnetic field intensities; A model training module, used for training the mixed distribution function model to be trained based on the training set to obtain a mixed distribution function model; A frequency conversion scanning module is used to perform microwave frequency conversion scanning within a preset microwave frequency sweep range, record the fluorescence intensity of the diamond NV color center corresponding to different microwave frequencies, and record the point between the falling edge point that first deviates from the peak value of the fluorescence intensity and the rising edge point that first deviates from the trough value of the fluorescence intensity as a valid point; A function inversion module, used for fitting a mixed distribution function of the magnetic field to be measured based on the microwave frequency value corresponding to the effective point and the mixed distribution function model; The magnetic field calculation module is used to obtain the frequency value of the trough position in the mixed distribution function of the magnetic field to be measured and calculate the intensity of the magnetic field to be measured.

10. An electronic device, characterized in that: include: A processor, a memory arranged to store computer executable instructions, which when executed cause the processor to perform the steps of the method as claimed in any one of claims 1 to 8.

Images