Spectrometer, noise processing method, concentration prediction method, device, equipment and medium

Abstract

The embodiment of the invention discloses a spectrograph, a noise processing method, a liquid concentration prediction method, a device, equipment and a medium, the spectrograph comprises a signal processor, the signal processor comprises a data preprocessing module, a noise modeling module, a noise suppression module and a data reconstruction module, the data preprocessing module is used for preprocessing the to-be-processed spectral data; the noise modeling module is used for estimating noise in the to-be-processed spectral data based on a preset noise model to obtain a noise estimation value of each sampling point so as to form a noise estimation curve; the noise suppression module is used for carrying out noise reduction processing on the preprocessed spectral data; and the data reconstruction module is used for performing noise restoration on the spectral data obtained after noise reduction based on the noise estimation curve, and correcting the spectral data after noise restoration through the correction coefficient to obtain corrected spectral data. By adopting the technical scheme, the problem of high noise level of the traditional chip-level spectrograph is solved.

Description

Technical Field

[0001] The embodiments of the present invention relate to the field of spectral detection technology, and in particular, to a spectrometer, a noise processing method for spectral data, a liquid concentration prediction method, a device, a equipment and a medium. Background Art

[0002] Spectral detection is a method of analyzing the composition, structure or properties of a substance by using its properties of light absorption, emission or scattering. Spectral detection has the advantages of high sensitivity, high resolution, non-destructive and non-polluting, and is widely used in chemistry, biology, medicine, agriculture, environment, food, pharmaceuticals and other fields.

[0003] Near Infrared spectroscopy (NIR) is a commonly used spectral detection method with a wavelength range of 700-2500nm. It can reflect the molecular vibration and rotation of substances, thereby obtaining the chemical composition and structural information of substances. Near infrared spectroscopy has the advantages of strong penetration, less interference, and fast speed, and is suitable for rapid and non-destructive analysis of complex samples.

[0004] Near-infrared spectrometer is the core equipment for near-infrared spectrum detection. Its main function is to decompose the light beam emitted by the near-infrared light source into spectra of different wavelengths after passing through the sample, and convert it into electrical signal output. The performance and quality of the near-infrared spectrometer directly affect the accuracy and stability of near-infrared spectrum detection.

[0005] With the development of science and technology and market demand, the miniaturization and integration of near-infrared spectrometers have attracted more and more attention. Miniaturized and integrated near-infrared spectrometers can be used in portable, online or embedded applications, expanding the application scenarios and fields of near-infrared spectral detection. However, when miniaturizing the near-infrared spectrometer to a chip-level spectrometer, the following new problems may be encountered: the chip-level spectrometer is small in size, and the sensor may face a higher noise level, which affects the accuracy and stability of spectral signal detection. Therefore, there is an urgent need for a chip-level spectrometer that can effectively suppress spectral data noise. Summary of the invention

[0006] Embodiments of the present invention provide a spectrometer, a noise processing method for spectral data, a liquid concentration prediction method, a device, an equipment and a medium, which are used to overcome the problem of high noise level of traditional chip-level spectrometers and improve the measurement performance and quality of chip-level spectrometers.

[0007] In a first aspect, the present invention provides a spectrometer, comprising: a light source, a grating, a sensor array and a signal processor, wherein:

[0008] a light source for emitting a near-infrared light beam;

[0009] A grating is used to receive a near-infrared spectrum formed after the near-infrared light beam passes through the sample to be tested, and decompose the near-infrared spectrum into spectral components of different wavelengths;

[0010] A sensor array is used to receive spectral components of different wavelengths, convert the spectral components of different wavelengths into electrical signals, and send the electrical signals to a signal processor as spectral data to be processed;

[0011] The signal processor is connected to the sensor array and includes: a data preprocessing module, a noise modeling module, a noise suppression module and a data reconstruction module, wherein:

[0012] A data preprocessing module, used for preprocessing the spectral data to be processed, wherein the preprocessing includes removing background noise and removing noise higher than a preset frequency;

[0013] A noise modeling module is used to estimate the noise of each sampling point in the spectral data to be processed based on a preset noise model, obtain the noise estimation value of each sampling point, and form a noise estimation curve;

[0014] A noise suppression module is used to reduce the noise of the preprocessed spectral data;

[0015] The data reconstruction module is used to perform noise repair on the spectral data obtained after noise reduction processing based on the noise estimation curve, and to correct the spectral data after noise repair through a correction coefficient to obtain corrected spectral data, wherein the correction coefficient is obtained by performing linear regression on the standard sample data.

[0016] In a second aspect, an embodiment of the present invention further provides a noise processing method for spectral data, comprising:

[0017] Receive the spectral data to be processed, and pre-process the spectral data to be processed, the pre-processing including removing background noise and removing noise higher than a preset frequency, wherein the spectral data to be processed is the electrical signal data obtained after the sample to be tested is irradiated by the light source in the spectrometer to form a near-infrared spectrum, and the near-infrared spectrum passes through the grating and sensor array in the spectrometer in sequence;

[0018] The noise of each sampling point in the spectral data to be processed is estimated based on a preset noise model to obtain a noise estimation value of each sampling point to form a noise estimation curve;

[0019] Perform noise reduction on the preprocessed spectral data;

[0020] Based on the noise estimation curve, the spectral data obtained after the noise reduction process is subjected to noise repair, and the spectral data after the noise repair is corrected by a correction coefficient to obtain corrected spectral data, wherein the correction coefficient is obtained by performing linear regression on the standard sample data.

[0021] Optionally, the preset noise model is a Gaussian white noise model. Accordingly, the noise of each sampling point in the spectral data to be processed is estimated based on the preset noise model to obtain the noise estimation value of each sampling point, including:

[0022] Calculate the noise mean and noise variance of each sampling point in the background spectrum data, and use the obtained noise mean and noise variance as parameter values ​​of the preset noise model;

[0023] According to the parameter values, the noise of each sampling point in the spectral data to be processed is estimated based on the maximum likelihood estimation method to obtain the noise estimation value of each sampling point.

[0024] Optionally, the preprocessed spectral data is subjected to noise reduction, including:

[0025] Based on the wavelet transform algorithm, the preprocessed spectral data is decomposed into wavelet coefficients of different scales and frequencies;

[0026] Determine a threshold value according to the noise estimation curve to perform soft threshold processing on the wavelet coefficients to obtain the wavelet coefficients after soft threshold processing;

[0027] Based on the least mean square error LMS algorithm, the wavelet coefficients after soft threshold processing are filtered to obtain the wavelet coefficients after filtering;

[0028] The wavelet coefficients after filtering are reconstructed to obtain the signal after noise reduction.

[0029] Optionally, based on the noise estimation curve, noise repair is performed on the spectral data obtained after the noise reduction process, including:

[0030] Subtract the spectral data of each sampling point after noise reduction processing from the noise estimation value of the corresponding sampling point in the noise estimation curve to obtain the spectral data after noise repair;

[0031] Accordingly, the noise-repaired spectral data is corrected by correction coefficients, including:

[0032] The noise-repaired spectral data is corrected using the correction coefficient according to the following formula:

[0033] y_corrected = a × y_repaired + b

[0034] Among them, y_corrected represents the corrected spectral data, y_repaired represents the spectral data after noise repair, and a and b are correction coefficients.

[0035] In a third aspect, an embodiment of the present invention further provides a liquid concentration prediction method, comprising:

[0036] After the liquid to be tested is subjected to near-infrared spectrum detection by the spectrometer provided by any embodiment of the present invention, a signal to be tested after noise processing output by the spectrometer is obtained;

[0037] The signal to be measured is input into the trained preset neural network model to obtain the liquid concentration information corresponding to the signal to be measured, wherein the preset neural network model is trained in the following way:

[0038] Placing each group of sample liquids in turn in a spectrometer for near-infrared spectrum detection to obtain sample spectrum data corresponding to each group of sample liquids after noise processing;

[0039] The spectral data of each group of samples and their corresponding concentration information are used as training samples and input into the preset neural network model for training. When the objective function of the preset neural network model reaches convergence, the training of the preset neural network model is completed. The preset neural network model is represented by the following formula:

[0040]

[0041] Among them, y is the signal to be tested after noise processing, W1 and W2 are the weight matrices of the neural network, b1 and b2 are the bias vectors of the neural network, and σ is the activation function of the preset neural network model.

[0042] In a fourth aspect, an embodiment of the present invention further provides a noise processing device for spectral data, comprising:

[0043] A data preprocessing module is used to receive the spectral data to be processed and preprocess the spectral data to be processed, wherein the preprocessing includes removing background noise and removing noise higher than a preset frequency, wherein the spectral data to be processed is the electrical signal data obtained after the sample to be tested is irradiated by the light source in the spectrometer to form a near-infrared spectrum, and the near-infrared spectrum passes through the grating and sensor array in the spectrometer in sequence;

[0044] A noise modeling module is used to estimate the noise of each sampling point in the spectral data to be processed based on a preset noise model, obtain the noise estimation value of each sampling point, and form a noise estimation curve;

[0045] A noise suppression module is used to reduce the noise of the preprocessed spectral data;

[0046] The data reconstruction module is used to perform noise repair on the spectral data obtained after noise reduction processing based on the noise estimation curve, and to correct the spectral data after noise repair through a correction coefficient to obtain corrected spectral data, wherein the correction coefficient is obtained by performing linear regression on the standard sample data.

[0047] Optionally, the preset noise model is a Gaussian white noise model, and the noise modeling module is specifically used to:

[0048] Calculate the noise mean and noise variance of each sampling point in the background spectrum data, and use the obtained noise mean and noise variance as parameter values ​​of the preset noise model;

[0049] According to the parameter value, the noise of each sampling point in the spectral data to be processed is estimated based on the maximum likelihood estimation method to obtain the noise estimation value of each sampling point.

[0050] Optional, noise suppression module, specifically used for:

[0051] Decomposing the preprocessed spectral data into wavelet coefficients of different scales and frequencies based on a wavelet transform algorithm;

[0052] Perform soft threshold processing on the wavelet coefficients to obtain the wavelet coefficients after soft threshold processing;

[0053] Based on the minimum mean square error LMS algorithm, filtering is performed on the wavelet coefficients after the soft threshold processing to obtain the wavelet coefficients after the filtering processing;

[0054] The wavelet coefficients after filtering are reconstructed to obtain the signal after noise reduction.

[0055] Optional, data reconstruction module, including:

[0056] A noise restoration unit, used for subtracting the spectral data of each sampling point after noise reduction processing from the noise estimation value of the corresponding sampling point in the noise estimation curve to obtain the spectral data after noise restoration;

[0057] The data correction unit is used to correct the noise-repaired spectral data using a correction coefficient according to the following formula:

[0058] y_corrected = a × y_repaired + b

[0059] Among them, y_corrected represents the corrected spectral data, y_repaired represents the spectral data after noise repair, and a and b are correction coefficients.

[0060] In a fifth aspect, an embodiment of the present invention further provides a device for predicting liquid concentration, characterized in that it includes:

[0061] A signal acquisition module for obtaining a signal to be tested, which is used to obtain a signal to be tested after noise processing is performed on the liquid to be tested by the spectrometer provided by any embodiment of the present invention after near-infrared spectrum detection;

[0062] The liquid concentration determination module is used to input the signal to be measured into the trained preset neural network model to obtain the liquid concentration information corresponding to the signal to be measured, wherein the preset neural network model is trained in the following way:

[0063] Placing each group of sample liquids in turn in a spectrometer for near-infrared spectrum detection to obtain sample spectrum data corresponding to each group of sample liquids after noise processing;

[0064] The spectral data of each group of samples and their corresponding concentration information are used as training samples and input into the preset neural network model for training. When the objective function of the preset neural network model reaches convergence, the training of the preset neural network model is completed, wherein the preset neural network model is represented by the following formula:

[0065]

[0066] Among them, y is the signal to be tested after noise processing, W1 and W2 are the weight matrices of the neural network, b1 and b2 are the bias vectors of the neural network, and σ is the activation function of the preset neural network model.

[0067] In a sixth aspect, an embodiment of the present invention further provides a computing device, including:

[0068] A memory storing executable program code;

[0069] a processor coupled to the memory;

[0070] The processor calls the executable program code stored in the memory to execute the liquid concentration prediction method provided by any embodiment of the present invention.

[0071] In a seventh aspect, an embodiment of the present invention further provides a computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the noise processing method for spectral data provided by any embodiment of the present invention.

[0072] In an eighth aspect, an embodiment of the present invention further provides a computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the liquid concentration prediction method provided by any embodiment of the present invention.

[0073] The technical solution provided by the embodiment of the present invention can remove background noise and high-frequency noise in the spectral data to be processed by using a spectrometer for near-infrared spectrum detection through a data preprocessing module in a signal processor. The preprocessed spectral data can be subjected to noise reduction and filtering through a noise suppression module, and appropriate thresholds and parameters can be determined according to a noise estimation curve for noise removal through a data reconstruction module. The spectral data after noise removal can be subjected to noise repair using a noise estimation curve, and data correction can be performed in combination with known standard sample data to improve the accuracy of the measurement results. By integrating the signal processor provided by this embodiment into a spectrometer, the problem of poor measurement accuracy caused by high noise levels during the measurement process of a chip-level spectrometer is avoided, and the accuracy and stability of the chip-level spectrometer during near-infrared spectrum detection are improved.

[0074] The innovative features of the embodiments of the present invention include:

[0075] 1. Before testing the sample to be tested, the noise signal is collected and a noise model is established so as to perform background noise correction during sample measurement, thereby ensuring the accuracy of sample measurement, which is one of the inventive points of the present invention.

[0076] 2. By using a neural network model to establish a correlation between spectral intensity and the concentration of the liquid to be tested, the neural network model can be directly used to obtain the corresponding concentration of the liquid to be tested after the spectrometer outputs the spectral intensity data, thereby achieving a fast and accurate prediction of the liquid concentration, which is one of the inventive points of the present invention. BRIEF DESCRIPTION OF THE DRAWINGS

[0077] In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the drawings required for use in the embodiments or the description of the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying creative work.

[0078] Figure 1 A structural block diagram of a spectrometer provided in Embodiment 1 of the present invention;

[0079] Figure 2 A flowchart of a noise processing method for spectral data provided in the second embodiment of the present invention;

[0080] Figure 3 A flow chart of a liquid concentration prediction method provided in Embodiment 3 of the present invention;

[0081] Figure 4 A structural block diagram of a noise processing device for spectral data provided in Embodiment 4 of the present invention;

[0082] Figure 5 A structural block diagram of a liquid concentration measuring device provided in Embodiment 5 of the present invention;

[0083] Figure 6 A schematic diagram of the structure of a computing device provided in Embodiment 6 of the present invention. DETAILED DESCRIPTION

[0084] The following will be combined with the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0085] It should be noted that the terms "including" and "having" and any variations thereof in the embodiments of the present invention and the accompanying drawings are intended to cover non-exclusive inclusions. For example, a process, method, system, product or device including a series of steps or units is not limited to the listed steps or units, but may optionally include steps or units not listed, or may optionally include other steps or units inherent to these processes, methods, products or devices.

[0086] The embodiments of the present invention disclose a spectrometer, a noise processing method for spectral data, a liquid concentration prediction method, a device, an equipment and a medium, which are described in detail below.

[0087] Embodiment 1

[0088] Figure 1 A structural block diagram of a spectrometer provided in Embodiment 1 of the present invention is shown in FIG. Figure 1 As shown, the spectrometer includes: a light source 110, a grating 120, a sensor array 130 and a signal processor 140, wherein:

[0089] The light source 110 is used to emit a near-infrared light beam.

[0090] The grating 120 is used to receive the near-infrared spectrum formed after the near-infrared light beam passes through the sample to be tested, and decompose the near-infrared spectrum into spectral components of different wavelengths, which can be projected onto the sensor array 130. The sensor array 130 can be composed of a plurality of photodiodes, each of which corresponds to a wavelength range and converts the received spectral components into electrical signals for output.

[0091] The sensor array 130 is used to receive spectral components of different wavelengths, convert the spectral components of different wavelengths into electrical signals, and send the electrical signals to the signal processor 140 as spectral data to be processed.

[0092] For example, the sensor array may include 8 photodiodes corresponding to the following 8 wavelength ranges, as shown in the following table:

[0093] Table 1 Wavelength range corresponding to the sensor array

[0094]

[0095]

[0096] The eight photodiodes in the sensor array convert spectral components of different wavelengths into electrical signals and output electrical signals of eight sampling points, for example, y = (10, 12, 15, 18, 20, 22, 25, 28). These electrical signals reflect the spectral intensity values ​​of the sample to be tested at different wavelengths.

[0097] The signal processor 140 is connected to the sensor array and includes: a data preprocessing module 141, a noise modeling module 142, a noise suppression module 143 and a data reconstruction module 144. The functions of each module are described in detail below:

[0098] (1) Data preprocessing module 141

[0099] The data preprocessing module 141 is used to preprocess the spectral data to be processed, and the preprocessing includes removing background noise and removing noise higher than a preset frequency.

[0100] Optionally, the spectral data to be processed may be subjected to baseline correction and smoothing filtering to remove background noise and high-frequency signals. There are many methods for baseline correction and smoothing filtering. For example, a polynomial fitting method may be used to implement baseline correction, and a Savitzky-Golay algorithm (a convolution fitting algorithm based on least squares) may be used for smoothing filtering. This embodiment does not specifically limit the preprocessing method.

[0101] (2) Noise Modeling Module 142

[0102] The noise modeling module 142 is used to estimate the noise of each sampling point in the spectral data to be processed based on a preset noise model, obtain the noise estimation value of each sampling point, and form a noise estimation curve.

[0103] Among them, background spectrum data refers to the spectrum data under the environmental background recorded by the instrument before measuring the sample (mainly ambient stray light, and light emitted by other light sources in non-required wavelength ranges). These data are automatically collected by the spectrometer so that background noise correction can be performed during sample measurement to ensure the accuracy of sample measurement. In the signal processor, a noise model is established by analyzing multiple background spectrum data, and the noise of each sampling point is estimated according to the noise model to obtain a noise estimation curve. Optionally, the preset noise model can be a Gaussian white noise model.

[0104] Specifically, the noise modeling module is specifically configured as follows: calculating the mean and variance of the background spectral data, and using the obtained noise mean and noise variance values ​​as parameter values ​​of the preset noise model; estimating the noise of each sampling point in the spectral data to be processed based on the parameter values ​​and the maximum likelihood estimation method to obtain the noise estimation value of each sampling point.

[0105] For example, three background spectrum data are collected:

[0106] y_1=(0,-0.01,-0.02,-0.03,-0.01,-0.02,-0.01,-0.03);

[0107] y_2=(-0.01,-0.01,-0.03,-0.02,-0.01,-0.03,-0.02,-0.01);

[0108] y_3=(-0.02,-0.01,-0.03,-0.02,-0.01,-0.03,-0.02,-0.01).

[0109] For each sampling point in the above three sets of background spectral data, the data in each column of y1, y2 and y3 are added and then divided by 3 to obtain the noise mean of each sampling point. The variance of the data in each column of y1, y2 and y3 is calculated respectively, and then the average is taken to obtain the noise variance of each sampling point, as shown in the following table.

[0110] Table 2 Noise mean and variance table for each sampling point

[0111]

[0112]

[0113] Since the noise mean is close to zero and the noise variance is equal in each frequency band, it can be considered that the noise obeys the Gaussian white noise model.

[0114] Then, the maximum likelihood estimation method can be used to estimate the noise of each sampling point to obtain a noise estimation curve.

[0115] For example, if the spectral data to be processed is: y = [10, 12, 15, 18, 20, 22, 25, 28], based on the above parameter values ​​(mean and variance) and combined with the maximum likelihood estimation method, the noise estimation value of each sampling point can be obtained as follows:

[0116] n=[0.01,-0.01,-0.03,-0.02,-0.01,-0.03,-0.02,-0.02].

[0117] The noise estimation value of each sampling point mentioned above is expressed in a graphical form to obtain a noise estimation curve.

[0118] (3) Noise suppression module 143

[0119] The noise suppression module 143 is used to perform noise reduction on the pre-processed spectral data, which can be implemented by the following steps a to d:

[0120] a. Based on the wavelet transform algorithm, the preprocessed spectral data is decomposed into wavelet coefficients of different scales and frequencies.

[0121] Optionally, the db4 wavelet (a discrete wavelet basis function) can be used to perform three-layer wavelet decomposition on the electrical signal to obtain the following wavelet coefficients:

[0122] cA3=[0.71,-0.71];

[0123] cD3=[-0.71,-0.71];

[0124] cD2=[-0.71,-0.71,-0.71,-0.71];

[0125] cD1=[-0.71,-0.71,-0.71,-0.71,-0.71,-0.71,-0.71,-0.71];

[0126] b. Perform soft threshold processing on the wavelet coefficients to obtain the wavelet coefficients after soft threshold processing.

[0127] Among them, the soft threshold processing is to compare the size of the wavelet coefficient with the threshold. If the wavelet coefficient is less than the threshold, the wavelet coefficient is set to 0. If the wavelet coefficient is greater than or equal to the threshold, the wavelet coefficient is retained. Among them, there are many ways to determine the threshold. For example, a fixed threshold can be selected according to an empirical value, or a multi-scale mean absolute deviation threshold method can be used to determine the threshold by calculating the mean absolute deviation of the wavelet coefficient and multiplying it by an empirical parameter, or an adaptive threshold method can be used to determine the threshold of the wavelet transform. This embodiment does not specifically limit the calculation method of the threshold.

[0128] For example, an appropriate threshold of 0.5 can be used to perform soft threshold processing on the wavelet coefficients to obtain the following processed wavelet coefficients:

[0129] cA3=[0.21,-0.21];

[0130] cD3=[-0.21,-0.21];

[0131] cD2=[-0.21,-0.21,-0.21,-0.21];

[0132] cD1=[-0.21,-0.21,-0.21,-0.21,-0.21,-0.21,-0.21,-0.21];

[0133] c. Based on the minimum mean square error algorithm, the wavelet coefficients after soft threshold processing are filtered to obtain the wavelet coefficients after filtering.

[0134] The principle of the Least Mean Squares Algorithm (LMS) is to update the weight of the filter by minimizing the mean square error between the signal and the expected value. By filtering the wavelet coefficients after soft threshold processing, the effective information of the signal can be retained and the noise can be suppressed.

[0135] d. Reconstruct the wavelet coefficients after filtering to obtain the signal after noise reduction.

[0136] Specifically, in the wavelet transform method, the wavelet reconstruction formula can be used to reconstruct the processed wavelet coefficients to obtain the denoised electrical signal. For example, the following formula can be used for reconstruction: y = IDWT (cA, cD, wname) where IDWT represents inverse wavelet transform, cA and cD represent the approximate coefficients and detail coefficients after wavelet decomposition, respectively, and wname represents the selected wavelet function. In this example, by using the db4 wavelet to reconstruct the processed wavelet coefficients, the denoised electrical signal can be obtained as y_denoised = [9.79, 11.79, 14.79, 17.79, 19.79, 21.79, 24.79, 27.79].

[0137] (4) Data reconstruction module 144,

[0138] The data reconstruction module 144 is used to perform noise repair on the spectral data obtained after noise reduction processing based on the noise estimation curve, and perform data correction on the noise-repaired signal through a correction coefficient to obtain corrected spectral data, wherein the correction coefficient is obtained by linear regression of the standard sample data.

[0139] The data reconstruction module includes a noise repair unit and a data correction unit, wherein the noise repair unit is configured to: subtract the spectral data of each sampling point after noise reduction processing from the noise estimation value of the corresponding sampling point in the noise estimation curve to obtain a noise-repaired signal, which can be specifically achieved by the following formula:

[0140] y_repaired=y_denoised-n

[0141] Where n is the noise estimation curve, y_denoised represents the spectral data of each sampling point after denoising, and y_repaired represents the signal after noise repair.

[0142] For example, y_denoised = [9.79, 11.79, 14.79, 17.79, 19.79, 21.79, 24.79, 27.79], n = [0.01, -0.01, -0.03, -0.02, -0.01, -0.03, -0.02, -0.02], we get:

[0143] y_repaired=([9.79, 11.79, 14.79, 17.79, 19.79, 21.79, 24.79, 27.79])-([0.01,-0.01,-0.03, -0.02, -0.01, -0.03, -0.02, -0.02]) = ([9.78, 11.80, 14.82, 17.81, 19.80, 21.82, 24.81, 27.81])

[0144] Therefore, the spectral data after noise repair is:

[0145] y_repaired=[9.78, 11.80, 14.82, 17.81, 19.80, 21.82, 24.81, 27.81]

[0146] The data correction unit is configured to perform data correction on the noise-repaired signal through a correction coefficient to obtain corrected spectral data, which can be specifically achieved through the following correction formula:

[0147] y_corrected = a × y_repaired + b

[0148] Wherein, y_corrected represents the corrected spectral data, y_repaired represents the spectral data after noise repair, a and b are correction coefficients, which can be obtained by linear regression of the standard sample data, wherein the standard sample data includes the spectral intensity value x and the spectral intensity value y after noise suppression and data correction.

[0149] Table 3 Spectral intensity information of standard sample data

[0150] x y 10 20 20 40

[0151] For example, by performing linear regression on the standard sample data in the table below, the correction coefficient can be obtained as follows:

[0152] a=1,b=0

[0153] Based on the above correction formula, the corrected spectral data can be obtained as follows:

[0154] y_corrected=[9.78,11.80,14.82,17.81,19.80,21.82,24.81,27.81].

[0155] In this embodiment, after the noise is repaired on the original electrical signal, data correction is performed in combination with known standard sample data, which can improve the accuracy of the measurement result.

[0156] The technical solution provided by this embodiment can remove background noise and high-frequency noise in the spectral data to be processed by the data preprocessing module in the signal processor when using a spectrometer for near-infrared spectrum detection. The preprocessed spectral data can be subjected to noise reduction and filtering through the noise suppression module, and appropriate thresholds and parameters can be determined according to the noise estimation curve for noise removal through the data reconstruction module. The noise estimation curve can be used to repair the noise of the spectral data after noise removal, and the data can be corrected in combination with known standard sample data to improve the accuracy of the measurement results. By integrating the signal processor provided by this embodiment into the spectrometer, the problem of poor measurement accuracy caused by the high noise level during the measurement of the chip-level spectrometer is avoided, and the accuracy and stability of the chip-level spectrometer in near-infrared spectrum detection are improved.

[0157] Embodiment 2

[0158] Figure 2 This is a flow chart of a noise processing method for spectral data provided in the second embodiment of the present invention. The method can be performed by a noise processing device for spectral data. The device can be implemented by software and / or hardware, and the device can be integrated in a spectrometer. Figure 2 , the method of this embodiment specifically includes:

[0159] S210: Receive spectral data to be processed, and pre-process the spectral data to be processed.

[0160] The processed spectral data is the electrical signal data obtained by the near-infrared spectrum formed by the sample to be tested after being irradiated by the light source in the spectrometer, the near-infrared spectrum is decomposed by the grating in the spectrometer in turn, and converted by the sensor array.

[0161] In this embodiment, the preprocessing includes removing background noise and removing noise higher than a preset frequency. Specifically, the preprocessing of the spectral data can be achieved by using methods such as baseline correction and smoothing filtering.

[0162] S220 , estimating the noise of each sampling point in the spectral data to be processed based on a preset noise model, obtaining a noise estimation value of each sampling point, and forming a noise estimation curve.

[0163] There are multiple preset noise models, for example, a Gaussian white noise model can be used, and this embodiment does not limit the specific noise model. The parameters of the preset noise model are obtained by analyzing the background spectrum data. Specifically, the noise of each sampling point in the spectrum data to be processed is estimated based on the preset noise model to obtain the noise estimation value of each sampling point, including:

[0164] Calculate the noise mean and noise variance of each sampling point in the background spectrum data, and use the obtained noise mean and noise variance as parameter values ​​of the preset noise model;

[0165] According to the parameter value, the noise of each sampling point in the spectral data to be processed is estimated based on the maximum likelihood estimation method to obtain the noise estimation value of each sampling point.

[0166] This embodiment does not specifically limit the execution order of step S210 and step S220.

[0167] S230: performing noise reduction processing on the preprocessed spectral data.

[0168] The noise reduction process can be achieved through the following steps:

[0169] Based on the wavelet transform algorithm, the preprocessed spectral data is decomposed into wavelet coefficients of different scales and frequencies;

[0170] Perform soft threshold processing on the wavelet coefficients to obtain the wavelet coefficients after soft threshold processing;

[0171] Based on the least mean square error (LMS) algorithm, the wavelet coefficients after soft threshold processing are filtered to obtain the wavelet coefficients after filtering;

[0172] The wavelet coefficients after filtering are reconstructed to obtain the signal after noise reduction.

[0173] S240 , based on the noise estimation curve, performing noise repair on the spectral data obtained after the noise reduction process, and correcting the spectral data after the noise repair by using a correction coefficient to obtain corrected spectral data.

[0174] The correction coefficient is obtained by performing linear regression on the standard sample data.

[0175] In this embodiment, step S240 can be implemented in the following manner:

[0176] Subtract the spectral data of each sampling point after noise reduction processing from the noise estimation value of the corresponding sampling point in the noise estimation curve to obtain the spectral data after noise repair, and correct the spectral data after noise repair by the correction coefficient according to the following formula:

[0177] y_corrected = a × y_repaired + b

[0178] Among them, y_corrected represents the corrected spectral data, y_repaired represents the spectral data after noise repair, and a and b are correction coefficients.

[0179] In this embodiment, the specific implementation process of steps S210 to S240 can refer to the description of the above embodiment, which will not be repeated here.

[0180] By adopting the above technical solution, noise suppression of spectral data can be achieved, and the accuracy and stability of spectral data detection can be improved.

[0181] Embodiment 3

[0182] Figure 3 This is a flow chart of a liquid concentration prediction method provided in the third embodiment of the present invention. The method can be executed by a liquid concentration prediction device. The device can be implemented by software and / or hardware. The device can be integrated in a spectrometer. Figure 3 , the method of this embodiment specifically includes:

[0183] S310, after the liquid to be tested is subjected to near-infrared spectrum detection by a spectrometer, a signal to be tested output by the spectrometer after noise processing is obtained.

[0184] The spectrometer is the spectrometer provided in the above embodiment. The specific processing process of the spectrometer on the liquid to be tested can refer to the description of the above embodiment, which will not be repeated here.

[0185] S320, inputting the signal to be measured into the trained preset neural network model to obtain the liquid concentration information corresponding to the signal to be measured.

[0186] The preset neural network model may be a multilayer perceptron (MLP), and the structure of the network may be configured according to actual scene requirements, for example, 8 input nodes, 10 hidden nodes, and 1 output node may be configured. The training method for the neural network model may adopt a back propagation (BP) algorithm.

[0187] Specifically, the preset neural network model is trained in the following way:

[0188] Each group of sample liquids is placed in a spectrometer in turn for near-infrared spectrum detection, and the sample spectrum data after noise suppression corresponding to each group of sample liquids is obtained; each group of sample spectrum data and its corresponding concentration information are used as training samples and input into a preset neural network model for training. When the objective function of the preset neural network model reaches convergence, the training of the preset neural network model is completed, wherein the preset neural network model is represented by the following formula:

[0189]

[0190] Wherein, y is the signal to be tested after noise processing, W1 and W2 are the weight matrices of the neural network, b1 and b2 are the bias vectors of the neural network, and σ is the activation function of the preset neural network model, for example, it can be a sigmoid function or a tanh function.

[0191] Specifically, four sample data with known concentrations can be used to train the neural network:

[0192] Table 4 Sample data information used to train neural networks

[0193]

[0194]

[0195] The parameters of the neural network model after training are:

[0196] W_1=[[0.01,-0.02,-0.01,0.02,0.01,-0.02,-0.01,0.02],[0.02,-0.01,

[0197] -0.02,0.01,0.02,-0.01,-0.02,0.01],[-0.01,0.02,0.01,-0.02,-0.01,0.02,0.01,-0.02],[0.02,0.01,-0.02,-0.01,0.02,0.01,-0.02,-0.01],[-0.02,-0.01,0 .02,0.01,-0.02,-0.01,0.02,0.01],[0.01,0.02,-0.01,-0.02,0.01,0.02,-0.01,-0.02],[-0.02,0.01,0.02,-0.01,-0.02,0.01,0.02,-0.01],[0.01,-0.02,0.01,

[0198] -0.02,0.01,-0.02,0.01,-0.02],[-0.01,0.02,-0.01,0.02,-0.01,0.02,-0.01,0.02],[0.02,-0.01,0.02,-0.01,0.02,-0.01,0.02,-0.01]];

[0199] W_2=[[0.01],[-0.02],[0.01],[-0.02],[0.01],[-0.02],[0.01],[-0.02],[0.01],[-0.02]];

[0200] b_1=[0.01,-0.02,0.01,-0.02,0.01,-0.02,0.01,-0..02,0.01,-0.02];

[0201] b_2=[0.01].

[0202] Specifically, the trained neural network model can be used to predict the glucose concentration of an unknown sample. If the spectral data collected for an unknown sample is:

[0203] y=[12,14,17,20,22,24,27,30]

[0204] By using the spectral data processing method provided in the above embodiment, the electrical signal after noise suppression can be obtained as follows:

[0205] y_corrected=[11.82,13.84,16.84,19.9,21.9,23.95,26.95,29.98]

[0206] Then, the neural network model can be used to predict the electrical signal after noise suppression, and the glucose concentration of the unknown sample can be obtained as:

[0207] c = f(y_corrected)

[0208] Among them, f is the function expression of the neural network model.

[0209] In this embodiment, by using a spectrometer to perform near-infrared spectrum detection on a liquid of unknown concentration, a noise-processed spectrum signal to be measured can be obtained. The spectrum information to be measured is input into a trained neural network model to obtain the liquid concentration, thereby achieving a fast and accurate prediction of the liquid concentration.

[0210] Embodiment 4

[0211] Figure 4 A structural block diagram of a noise processing device for spectral data provided by the fourth embodiment of the present invention, such as Figure 4 As shown, the device includes: a data preprocessing module 410, a noise modeling module 420, a noise suppression module 430 and a data reconstruction module 440, wherein:

[0212] The data preprocessing module 410 is used to receive the spectral data to be processed and preprocess the spectral data to be processed, wherein the preprocessing includes removing background noise and removing noise higher than a preset frequency, wherein the spectral data to be processed is the electrical signal data obtained after the sample to be tested is irradiated by the light source in the spectrometer to form a near-infrared spectrum, and the near-infrared spectrum passes through the grating and sensor array in the spectrometer in sequence;

[0213] A noise modeling module 420 is used to estimate the noise of each sampling point in the spectral data to be processed based on a preset noise model, obtain a noise estimation value of each sampling point, and form a noise estimation curve;

[0214] A noise suppression module 430 is used to perform noise reduction on the pre-processed spectral data;

[0215] The data reconstruction module 440 is used to perform noise repair on the spectral data obtained after noise reduction processing based on the noise estimation curve, and correct the noise-repaired spectral data through a correction coefficient to obtain corrected spectral data, wherein the correction coefficient is obtained by linear regression of the standard sample data.

[0216] Optionally, the preset noise model is a Gaussian white noise model, and the noise modeling module is specifically used for:

[0217] Calculate the noise mean and noise variance of each sampling point in the background spectrum data, and use the obtained noise mean and noise variance as parameter values ​​of the preset noise model;

[0218] According to the parameter values, the noise of each sampling point in the spectral data to be processed is estimated based on the maximum likelihood estimation method to obtain the noise estimation value of each sampling point.

[0219] Optional, noise suppression module, specifically used for:

[0220] Based on the wavelet transform algorithm, the preprocessed spectral data is decomposed into wavelet coefficients of different scales and frequencies;

[0221] Perform soft threshold processing on the wavelet coefficients to obtain the wavelet coefficients after soft threshold processing;

[0222] Based on the least mean square error LMS algorithm, the wavelet coefficients after soft threshold processing are filtered to obtain the wavelet coefficients after filtering;

[0223] The wavelet coefficients after filtering are reconstructed to obtain the signal after noise reduction.

[0224] Optional, data reconstruction module, including:

[0225] A noise restoration unit, used for subtracting the spectral data of each sampling point after noise reduction processing from the spectral data of each sampling point in the noise estimation curve to obtain the spectral data after noise restoration;

[0226] The data correction unit is used to correct the noise-repaired spectral data using a correction coefficient according to the following formula:

[0227] y_corrected = a × y_repaired + b

[0228] Among them, y_corrected represents the corrected spectral data, y_repaired represents the spectral data after noise repair, and a and b are correction coefficients.

[0229] Embodiment 5

[0230] Figure 5 A structural block diagram of a liquid concentration measuring device provided in Embodiment 5 of the present invention, such as Figure 5 As shown, the device includes: a signal acquisition module 510 and a liquid concentration determination module 520, wherein:

[0231] The signal acquisition module 510 is used to acquire the signal to be tested after the noise is processed and output by the spectrometer after the liquid to be tested is subjected to near-infrared spectrum detection by the spectrometer provided by any embodiment of the present invention;

[0232] The liquid concentration determination module 520 is used to input the signal to be measured into the trained preset neural network model to obtain the liquid concentration information corresponding to the signal to be measured, wherein the preset neural network model is trained in the following manner:

[0233] Placing each group of sample liquids in turn in a spectrometer for near-infrared spectrum detection to obtain sample spectrum data corresponding to each group of sample liquids after noise processing;

[0234] The spectral data of each group of samples and their corresponding concentration information are used as training samples and input into the preset neural network model for training. When the objective function of the preset neural network model reaches convergence, the training of the preset neural network model is completed, wherein the preset neural network model is represented by the following formula:

[0235]

[0236] Among them, y is the signal to be tested after noise processing, W1 and W2 are the weight matrices of the neural network, b1 and b2 are the bias vectors of the neural network, and σ is the activation function of the preset neural network model.

[0237] Embodiment 6

[0238] Figure 6 This is a schematic diagram of the structure of a computing device provided in Embodiment 6 of the present invention. Figure 6 As shown, the computing device may include:

[0239] A memory 701 storing executable program codes;

[0240] a processor 702 coupled to the memory 701;

[0241] The processor 702 calls the executable program code stored in the memory 701 to execute the liquid concentration prediction method provided by any embodiment of the present invention.

[0242] An embodiment of the present invention discloses a computer-readable storage medium storing a computer program, wherein the computer program enables a computer to execute the method for predicting liquid concentration provided by any embodiment of the present invention.

[0243] An embodiment of the present invention further provides a computer-readable storage medium on which a computer program is stored. When the program is executed by a processor, the noise processing method for spectral data provided by any embodiment of the present invention is implemented.

[0244] In various embodiments of the present invention, it should be understood that the size of the serial numbers of the above-mentioned processes does not mean the necessary order of execution. The execution order of each process should be determined by its function and internal logic, and should not constitute any limitation on the implementation process of the embodiments of the present invention.

[0245] In the embodiments provided by the present invention, it should be understood that "B corresponding to A" means that B is associated with A, and B can be determined according to A. However, it should also be understood that determining B according to A does not mean determining B only according to A, and B can also be determined according to A and / or other information.

[0246] In addition, each functional unit in each embodiment of the present invention may be integrated into one processing unit, or each unit may exist physically separately, or two or more units may be integrated into one unit. The above-mentioned integrated unit may be implemented in the form of hardware or in the form of software functional units.

[0247] If the above integrated unit is implemented in the form of a software functional unit and sold or used as an independent product, it can be stored in a computer-accessible memory. Based on this understanding, the technical solution of the present invention, in essence, or the part that contributes to the prior art, or all or part of the technical solution, can be embodied in the form of a software product, which is stored in a memory and includes several requests for a computer device (which can be a personal computer, a server or a network device, etc., specifically a processor in a computer device) to perform some or all of the steps of the above methods of various embodiments of the present invention.

[0248] Those skilled in the art can understand that all or part of the steps in the various methods of the above embodiments can be completed by instructing related hardware through a program, and the program can be stored in a computer-readable storage medium, and the storage medium includes a read-only memory (ROM), a random access memory (RAM), a programmable read-only memory (PROM), an erasable programmable read-only memory (EPROM), a one-time programmable read-only memory (OTPROM), an electronically erasable rewritable read-only memory (EEPROM), a compact disc (CD-ROM) or other optical disc storage, magnetic disk storage, magnetic tape storage, or any other computer-readable medium that can be used to carry or store data.

[0249] Those skilled in the art can understand that the accompanying drawings are only schematic diagrams of an embodiment, and the modules or processes in the accompanying drawings are not necessarily required to implement the present invention.

[0250] Those skilled in the art can understand that the modules in the device in the embodiment can be distributed in the device in the embodiment according to the description of the embodiment, or can be changed accordingly and located in one or more devices different from the embodiment. The modules in the above embodiment can be combined into one module, or can be further divided into multiple sub-modules.

[0251] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, rather than to limit it. Although the present invention has been described in detail with reference to the aforementioned embodiments, those skilled in the art should understand that they can still modify the technical solutions described in the aforementioned embodiments, or make equivalent replacements for some of the technical features therein. However, these modifications or replacements do not deviate the essence of the corresponding technical solutions from the spirit and scope of the technical solutions of the embodiments of the present invention.

Claims

1. A spectrometer, characterized in that: include: light source, grating, sensor array and signal processor, wherein: The light source is used to emit a near-infrared light beam; The grating is used to receive the near-infrared spectrum formed after the near-infrared light beam passes through the sample to be tested, and decompose the near-infrared spectrum into spectral components of different wavelengths; The sensor array is used to receive the spectral components of different wavelengths, convert the spectral components of different wavelengths into electrical signals, and send the electrical signals as spectral data to be processed to the signal processor; The signal processor is connected to the sensor array and includes: a data preprocessing module, a noise modeling module, a noise suppression module and a data reconstruction module, wherein: The data preprocessing module is used to preprocess the spectral data to be processed, wherein the preprocessing includes removing background noise and removing noise higher than a preset frequency; The noise modeling module is used to estimate the noise of each sampling point in the spectral data to be processed based on a preset noise model, obtain the noise estimation value of each sampling point, and form a noise estimation curve; The noise suppression module is used to perform noise reduction processing on the preprocessed spectral data; The data reconstruction module is used to perform noise repair on the spectral data obtained after noise reduction processing based on the noise estimation curve, and to correct the spectral data after noise repair by a correction coefficient to obtain corrected spectral data, wherein the correction coefficient is obtained by performing linear regression on the standard sample data.

2. A noise processing method for spectral data, applied to the spectrometer as claimed in claim 1, characterized in that: include: Receive spectral data to be processed, and preprocess the spectral data to be processed, wherein the preprocessing includes removing background noise and removing noise higher than a preset frequency, wherein the spectral data to be processed is electrical signal data obtained after the sample to be tested is irradiated by a light source in a spectrometer to form a near-infrared spectrum, and the near-infrared spectrum passes through a grating and a sensor array in the spectrometer in sequence; The noise of each sampling point in the spectral data to be processed is estimated based on a preset noise model to obtain a noise estimation value of each sampling point to form a noise estimation curve; Performing noise reduction processing on the preprocessed spectral data; Based on the noise estimation curve, the spectral data obtained after the noise reduction process is subjected to noise repair, and the spectral data after the noise repair is corrected by a correction coefficient to obtain corrected spectral data, wherein the correction coefficient is obtained by performing linear regression on the standard sample data.

3. The method according to claim 2, characterized in that The preset noise model is a Gaussian white noise model. Accordingly, estimating the noise of each sampling point in the spectral data to be processed based on the preset noise model to obtain the noise estimation value of each sampling point includes: Calculate the noise mean and noise variance of each sampling point in the background spectrum data, and use the obtained noise mean and noise variance as parameter values ​​of the preset noise model; According to the parameter value, the noise of each sampling point in the spectral data to be processed is estimated based on the maximum likelihood estimation method to obtain the noise estimation value of each sampling point.

4. The method according to claim 2, characterized in that: The denoising process of the pre-processed spectral data comprises: Decomposing the preprocessed spectral data into wavelet coefficients of different scales and frequencies based on a wavelet transform algorithm; Perform soft threshold processing on the wavelet coefficients to obtain the wavelet coefficients after soft threshold processing; Based on the minimum mean square error LMS algorithm, filtering is performed on the wavelet coefficients after the soft threshold processing to obtain the wavelet coefficients after the filtering processing; The wavelet coefficients after the filtering process are reconstructed to obtain a signal after the noise reduction process.

5. The method according to claim 2, characterized in that: The performing noise repair on the spectral data obtained after the noise reduction process based on the noise estimation curve comprises: Subtract the spectral data of each sampling point after noise reduction processing from the noise estimation value of the corresponding sampling point in the noise estimation curve to obtain the spectral data after noise repair; Accordingly, the correction of the noise-repaired spectral data by the correction coefficient includes: The noise-repaired spectral data is corrected by the correction coefficient according to the following formula: in, represents the corrected spectral data, represents the spectral data after noise repair, and a and b are correction coefficients.

6. A method for predicting liquid concentration, characterized in that: include: After the liquid to be tested is subjected to near-infrared spectrum detection by the spectrometer as claimed in claim 1, a signal to be tested is obtained after noise processing output by the spectrometer; The signal to be measured is input into a trained preset neural network model to obtain liquid concentration information corresponding to the signal to be measured, wherein the preset neural network model is trained in the following manner: Placing each group of sample liquids in the spectrometer in turn for near-infrared spectrum detection, and obtaining sample spectrum data corresponding to each group of sample liquids after noise processing; Each group of sample spectral data and its corresponding concentration information are used as training samples and input into a preset neural network model for training. When the objective function of the preset neural network model reaches convergence, the training of the preset neural network model is completed, wherein the preset neural network model is represented by the following formula: Among them, y is the signal to be tested after noise processing, W1 and W2 are the weight matrices of the neural network, b1 and b2 are the bias vectors of the neural network, and σ is the activation function of the preset neural network model.

7. A noise processing device for spectral data, characterized in that: include: A data preprocessing module, used for receiving the spectral data to be processed and preprocessing the spectral data to be processed, wherein the preprocessing includes removing background noise and removing noise higher than a preset frequency, wherein the spectral data to be processed is electrical signal data obtained after the sample to be tested is irradiated by a light source in a spectrometer to form a near-infrared spectrum, and the near-infrared spectrum passes through a grating and a sensor array in the spectrometer in sequence; A noise modeling module, used for estimating the noise of each sampling point in the spectral data to be processed based on a preset noise model, obtaining a noise estimation value of each sampling point, so as to form a noise estimation curve; A noise suppression module, used for performing noise reduction processing on the preprocessed spectral data; The data reconstruction module is used to perform noise repair on the spectral data obtained after noise reduction processing based on the noise estimation curve, and to correct the spectral data after noise repair by a correction coefficient to obtain corrected spectral data, wherein the correction coefficient is obtained by linear regression of the standard sample data.

8. A device for predicting liquid concentration, characterized in that: include: A signal acquisition module for obtaining a signal to be tested, used for obtaining a signal to be tested output by the spectrometer after noise processing after the liquid to be tested is subjected to near-infrared spectrum detection by the spectrometer as claimed in claim 1; The liquid concentration determination module is used to input the signal to be measured into a preset neural network model that has been trained to obtain the liquid concentration information corresponding to the signal to be measured, wherein the preset neural network model is trained in the following manner: Placing each group of sample liquids in the spectrometer in turn for near-infrared spectrum detection, and obtaining sample spectrum data corresponding to each group of sample liquids after noise processing; Each group of sample spectral data and its corresponding concentration information are used as training samples and input into a preset neural network model for training. When the objective function of the preset neural network model reaches convergence, the training of the preset neural network model is completed, wherein the preset neural network model is represented by the following formula: Among them, y is the signal to be tested after noise processing, W1 and W2 are the weight matrices of the neural network, b1 and b2 are the bias vectors of the neural network, and σ is the activation function of the preset neural network model.

9. A computing device, characterized in that The computing device comprises: one or more processors; a storage device for storing one or more programs, When the one or more programs are executed by the one or more processors, the one or more processors implement the liquid concentration prediction method as described in claim 6.

10. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the program is executed by a processor, the noise processing method for spectral data as described in any one of claims 2 to 5 is implemented, or the liquid concentration prediction method as described in claim 6 is implemented.

Images