Multispectral data prediction method and system based on deep learning

Abstract

The present invention relates to the field of data processing technology, and specifically to a multispectral data prediction method and system based on deep learning; a peak point is obtained according to the data change characteristics of a multispectral characteristic curve; a shear point and a non-shear point are obtained according to the data distribution characteristics of the peak point and the data fluctuation characteristics of the neighborhood; a target shearing degree is obtained according to the data discrete characteristics of the non-shear point and the change characteristics of the adjacent data and the curve smoothness characteristics of the shearing point neighborhood; and a preset light intensity is adjusted according to the target shearing degree to obtain an adaptive light intensity. The present invention obtains an iterative multispectral characteristic curve of sample detection according to the adaptive light intensity, and identifies the sample to be detected through a preset sample prediction model according to the finally obtained iterative multispectral characteristic curve, thereby improving the prediction accuracy of the sample to be detected.

Description

Technical Field

[0001] The present invention relates to the field of data processing technology, and in particular to a multispectral data prediction method and system based on deep learning. Background Art

[0002] The multispectral data prediction method has great potential in the customs detection of dangerous chemicals. Multispectral technology can provide high-dimensional information for accurate identification and classification of different types of chemicals. The prediction model can be constructed by combining the multispectral curve of the determined sample with the neural network algorithm, and the multispectral curve of the sample to be detected can be put into the prediction model for matching to predict the sample results. When collecting the multispectral characteristic curve of the sample to be detected, when the illumination intensity is too high, the spectral signal intensity will be too high, exceeding the detection capability of the sensor, and the output of the sensor will reach the maximum limit, so that the peak of the strong signal in the multispectral characteristic curve is clipped and presents a flat top. Therefore, the clipped multispectral characteristic curve is difficult to reflect the accurate composition of the sample to be detected, making the prediction result inaccurate. If a lower illumination intensity is used directly, the spectral signal intensity received by the sensor will be reduced, resulting in a low signal-to-noise ratio, making it difficult to extract effective spectral features, which will also make the prediction result inaccurate. Summary of the invention

[0003] In order to solve the technical problem that the high light intensity will lead to incomplete multi-spectral characteristic curve and affect the prediction accuracy of the sample to be detected, the purpose of the present invention is to provide a multi-spectral data prediction method and system based on deep learning. The technical solution adopted is as follows:

[0004] Acquire a multi-spectral characteristic curve of the sample to be detected according to a preset light intensity; obtain a peak point according to the data change characteristics of the multi-spectral characteristic curve;

[0005] Obtaining a suspected shearing degree according to the data distribution characteristics of the peak point and the data fluctuation characteristics of the neighborhood; obtaining a shearing point and a non-shearing point according to the suspected shearing degree of the peak point;

[0006] Obtain a first shearing coefficient according to the data discrete characteristics of the non-shearing point and the variation characteristics of the adjacent data; obtain a second shearing coefficient according to the curve smoothness characteristics of the neighborhood of the shearing point; obtain the shearing degree of the shearing point according to the first shearing coefficient and the second shearing coefficient; obtain a target shearing degree according to the shearing degree;

[0007] The preset light intensity is adjusted according to the target shearing degree to obtain an adaptive light intensity; an iterative multi-spectral characteristic curve of sample detection is obtained according to the adaptive light intensity, and the process of obtaining the iterative multi-spectral characteristic curve is repeated until there is no shearing point in the iterative multi-spectral characteristic curve; and the sample to be detected is identified through a preset sample prediction model according to the finally obtained iterative multi-spectral characteristic curve.

[0008] Furthermore, the step of obtaining a peak point according to the data variation characteristics of the multi-spectral characteristic curve includes:

[0009] For any data point in the multi-spectral characteristic curve, if the slope of the line connecting the arbitrary data point and the previous data point is a positive number, and the slope of the line connecting the arbitrary data point and the next data point is a negative number, the arbitrary data point is a peak point; if the values ​​of consecutive data points are equal and greater than the values ​​of the adjacent data points on both sides, the data point in the middle of the consecutive data points is taken as the peak point.

[0010] Furthermore, the step of obtaining the suspected shear degree according to the data distribution characteristics of the peak point and the data fluctuation characteristics of the neighborhood includes:

[0011] Calculate the average of the absolute values ​​of the slope of the line connecting each data point from the nearest trough point on the left side of the peak point to the nearest trough point on the right side and the previous data point to obtain the variation amplitude characteristic value; calculate the variance of the absolute values ​​of the slope of the line connecting each data point from the nearest trough point on the left side of the peak point to the nearest trough point on the right side and the previous data point to obtain the fluctuation characteristic value; calculate the product of the variation amplitude characteristic value and the fluctuation characteristic value and negatively correlate them to obtain the variation coefficient; normalize the value of the peak point to obtain the peak coefficient; calculate the average of the variation coefficient and the peak coefficient to obtain the suspected shear degree of the peak point.

[0012] Furthermore, the step of obtaining the shearing point and the non-shearing point according to the suspected shearing degree of the crest point includes:

[0013] When the suspected shearing degree of the crest point exceeds a preset shearing threshold, the crest point is a shearing point; otherwise, the crest point is a non-shearing point.

[0014] Furthermore, the step of obtaining the first shear coefficient according to the discrete characteristics of the data of the non-shear point and the variation characteristics of the adjacent data includes:

[0015] Calculate the numerical average of all non-shear points and negatively map them to obtain the low peak degree of the non-shear points; calculate the average of the numerical difference between the non-shear point and the nearest trough points on the left and right to obtain the amplitude difference; calculate the continuous slope of the non-shear point and the previous data point to obtain the first change value, and the slope of the line connecting the non-shear point and the next data point to obtain the second change value; calculate the absolute value of the difference between the first change value and the second change value to obtain the change difference; calculate the product of the amplitude difference of the non-shear point and the change difference to obtain the fluctuation degree; calculate the average of the fluctuation degree of all non-shear points and negatively map them to obtain the fluctuation characteristic value; calculate the average of the low peak degree and the fluctuation characteristic value to obtain the first shear coefficient.

[0016] Furthermore, the step of obtaining the second shear coefficient according to the curve smoothness feature of the shear point neighborhood includes:

[0017] The accumulated value of the difference between the rate of change of each data point and the subsequent data point from the nearest trough point on the left side of the shearing point to the nearest trough point on the right side is calculated to obtain a smooth eigenvalue; the smooth eigenvalue is normalized to obtain a second shear coefficient of the shearing point.

[0018] Further, the step of obtaining the shear degree of the shear point according to the first shear coefficient and the second shear coefficient comprises:

[0019] An average value of the first shear coefficient and the second shear coefficient is calculated to obtain the shear degree of the shear point.

[0020] Further, the step of obtaining a target shearing degree according to the shearing degree comprises:

[0021] The maximum value of the shearing degree of all shearing points in the multi-spectral characteristic curve is taken as the target shearing degree.

[0022] Furthermore, the step of adjusting the preset illumination intensity according to the target shearing degree to obtain the adaptive illumination intensity includes:

[0023] The product of the target shearing degree, the preset light intensity and the preset adjustment coefficient is calculated to obtain the reduction degree; and the difference between the preset light intensity and the reduction degree is calculated to obtain the adaptive light intensity.

[0024] The present invention also proposes a multispectral data prediction system based on deep learning, comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the computer program to implement any step of a multispectral data prediction method based on deep learning.

[0025] The present invention has the following beneficial effects:

[0026] In the present invention, obtaining the peak point can determine the position in the multi-spectral characteristic curve that is prone to shearing, and obtaining the suspected shearing degree can judge the degree to which different peak points may be sheared according to the data characteristics at the peak point, thereby distinguishing the peak points into shearing points and non-shearing points. Obtaining the first shearing coefficient can determine the degree of shearing according to the curve data characteristics at the non-shearing point in the multi-spectral characteristic curve, and obtaining the second shearing coefficient can determine the degree of shearing according to the smoothness characteristics of the curve at the shearing point in the multi-spectral characteristic curve; obtaining the shearing degree can accurately reflect the degree to which different shearing points may be sheared. Obtaining the target shearing degree can adaptively determine the degree of adjustment required according to the shearing characteristics in the multi-spectral characteristic curve, and obtaining the adaptive illumination intensity can enable the collected multi-spectral characteristic curve to more accurately reflect the composition of the sample to be detected, ultimately improving the prediction accuracy of the sample to be detected. BRIEF DESCRIPTION OF THE DRAWINGS

[0027] In order to more clearly illustrate the technical solutions and advantages in the embodiments of the present invention or the prior art, the drawings required for use in the embodiments or the prior art descriptions are 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.

[0028] Figure 1 A flow chart of a multispectral data prediction method based on deep learning provided by an embodiment of the present invention. DETAILED DESCRIPTION

[0029] In order to further explain the technical means and effects adopted by the present invention to achieve the predetermined invention purpose, the following is a detailed description of the multispectral data prediction method and system based on deep learning proposed by the present invention, its specific implementation, structure, features and effects, in combination with the accompanying drawings and preferred embodiments. In the following description, different "one embodiment" or "another embodiment" does not necessarily refer to the same embodiment. In addition, specific features, structures or characteristics in one or more embodiments may be combined in any suitable form.

[0030] Unless defined otherwise, all technical and scientific terms used herein have the same meaning as commonly understood by one of ordinary skill in the art to which this invention belongs.

[0031] The following is a detailed description of a multispectral data prediction method and system based on deep learning provided by the present invention in conjunction with the accompanying drawings.

[0032] See also Figure 1, which shows a flow chart of a multispectral data prediction method based on deep learning provided by an embodiment of the present invention, the method comprising the following steps:

[0033] Step S1, obtaining a multi-spectral characteristic curve of a sample to be detected according to a preset light intensity; and obtaining a peak point according to a data variation characteristic of the multi-spectral characteristic curve.

[0034] In an embodiment of the present invention, the implementation scenario is to obtain a more accurate multi-spectral characteristic curve of the sample to be detected and improve the accuracy of sample prediction. First, the multi-spectral characteristic curve of the sample to be detected is obtained according to the preset light intensity. Since a lower preset light intensity will reduce the signal intensity received by the sensor, resulting in a lower signal-to-noise ratio and difficulty in extracting effective spectral features, a higher preset light intensity is used. The implementer can determine the preset light intensity according to the implementation scenario; it should be noted that obtaining a multi-spectral characteristic curve through a multi-spectral imaging system belongs to the prior art, and the specific process will not be repeated. The horizontal axis of the multi-spectral characteristic curve is different wavelengths, and the vertical axis is the reflection intensity of the sample to be detected at different wavelengths.

[0035] Because the preset light intensity may be too high, the multi-spectral characteristic curve is clipped. In order to ensure the integrity of the multi-spectral characteristic curve, it is necessary to analyze whether there is clipping based on the curve characteristics of the multi-spectral characteristic curve. The peak in the multi-spectral characteristic curve will show the characteristics of being sheared, so the peak point is first obtained according to the data change characteristics of the multi-spectral characteristic curve; preferably, in an embodiment of the present invention, the step of obtaining the peak point includes: for any data point in the multi-spectral characteristic curve, if the slope of the line connecting the arbitrary data point and the previous data point is a positive number, and the slope of the line connecting the arbitrary data point and the next data point is a negative number, the arbitrary data point is a peak point, the previous data point is the data point adjacent to the left side of the arbitrary data point in the multi-spectral characteristic curve, and the next data point is the data point adjacent to the right side of the arbitrary data point in the multi-spectral characteristic curve; if the values ​​of consecutive data points are equal and greater than the values ​​of the adjacent two-side data points, the adjacent two-side data points refer to the two data points adjacent to the continuous data point on the left and right. At this time, it may be that the top of the peak shows the characteristics of being sheared, or a horizontal line segment may appear in the multi-spectral characteristic curve; in order to accurately analyze whether it is sheared, it is necessary to take the data point in the middle of the continuous data points as the peak point.

[0036] Step S2, obtaining a suspected shearing degree according to the data distribution characteristics of the crest point and the data fluctuation characteristics of the neighborhood; obtaining a shearing point and a non-shearing point according to the suspected shearing degree of the crest point.

[0037] After obtaining all the peak points in the multispectral characteristic curve, it can be determined whether there is shearing at the peak point; if there is shearing at the peak point, the curve at the peak point is relatively horizontal, the curve changes slowly, and the peak feature is not obvious; at the same time, in order to eliminate the interference of slow curve changes at the peak points of lower intensity in the multispectral characteristic curve, the larger the value of the peak point, the higher the peak, and the easier it is to be sheared; therefore, the suspected shearing degree is obtained based on the data distribution characteristics of the peak point and the data fluctuation characteristics of the neighborhood.

[0038] Preferably, in an embodiment of the present invention, the step of obtaining the suspected shearing degree includes: calculating the average value of the absolute value of the slope of the line connecting each data point from the nearest trough point on the left side of the peak point to the nearest trough point on the right side and the previous data point, and obtaining the variation amplitude characteristic value; when shearing occurs, the curve near the peak point is relatively smooth, and the slope of the line connecting the data point and the previous data point is close to 0, and then the smaller the variation amplitude characteristic value, the greater the possibility of shearing at the peak point. Calculate the variance of the absolute value of the slope of the line connecting each data point from the nearest trough point on the left side of the peak point to the nearest trough point on the right side and the previous data point, and obtain the fluctuation characteristic value; when the variance is smaller, the fluctuation characteristic value is smaller, which means that the curve near the peak point is smoother, there is no obvious peak feature, and the peak point is more likely to be sheared. Calculate the product of the variation amplitude characteristic value and the fluctuation characteristic value and negatively correlate them to obtain the variation coefficient; when the variation coefficient is larger, it means that the peak point is more likely to be sheared. Normalize the value of the peak point to obtain the peak coefficient; the larger the value of the peak point in the multi-spectral characteristic curve, the higher the vertical axis direction, the easier it is to be sheared, and the larger the peak coefficient. Calculate the average value of the coefficient of variation and the peak coefficient to obtain the suspected shear degree of the peak point. The larger the suspected shear degree, the more likely it is that the peak point is sheared. The formula for obtaining the suspected shear degree includes:

[0039]

[0040] Where R represents the suspected shear degree at the peak point, represents an exponential function with a natural constant as the base, S represents the fluctuation characteristic value, and N represents the number of data points from the nearest trough point on the left side of the peak point to the nearest trough point on the right side. It represents the slope of the line connecting the nth data point and the previous data point. represents the characteristic value of the variation amplitude, represents the coefficient of variation, represents the normalized function, and A represents the value of the peak point.

[0041] Furthermore, after obtaining the suspected shearing degree of different crest points, the shearing point and the non-shearing point can be obtained according to the suspected shearing degree of the crest point; preferably, in the embodiment of the present invention, the step of obtaining the shearing point and the non-shearing point includes: when the suspected shearing degree of the crest point exceeds the preset shearing threshold, the crest point is a shearing point, otherwise the crest point is a non-shearing point; in the embodiment of the present invention, the preset shearing threshold is 0.7, which can be determined by the implementer according to the implementation scenario. The non-shearing point is a crest point that is not sheared; the shearing point is a crest point that may be sheared. Even if it is not sheared, it is because the preset light intensity is too high, resulting in the reflection intensity obtained at the crest point being close to the detection proximity value of the sensor, and the curve at the crest point changes too slowly, and the multi-spectral characteristic curve can no longer reflect its true characteristics, so it is necessary to adjust the preset light intensity.

[0042] Step S3, obtaining a first shearing coefficient according to the discrete characteristics of the data of the non-shearing point and the change characteristics of the adjacent data; obtaining a second shearing coefficient according to the smooth characteristics of the curve in the neighborhood of the shearing point; obtaining the shearing degree of the shearing point according to the first shearing coefficient and the second shearing coefficient; and obtaining a target shearing degree according to the shearing degree.

[0043] When there is no shear point in the multi-spectral characteristic curve, it means that the multi-spectral characteristic curve can fully reflect the composition of the sample to be detected, and there is no need to adjust the preset light intensity. When there is a shear point, it is necessary to reduce the preset light intensity to make the collected multi-spectral characteristic curve complete. If the preset light intensity is reduced too much, it will cause the amplitude of the multi-spectral characteristic curve to decrease, the noise to be obvious, and the signal-to-noise ratio to decrease, making it difficult to extract effective information, which will also lead to inaccurate prediction results. Therefore, it is necessary to adaptively adjust the preset light intensity according to the degree of shearing at the shear point. In the multi-spectral characteristic curve of a chemical substance, different peak shapes, positions, and extension trends of the peaks reflect key information about the characteristics of the substance; an obvious peak can represent the absorption of a chemical bond or functional group, which is very abundant in the sample to be detected and interacts with the incident light. Therefore, in the multi-spectral characteristic curve, when the peak at the shear point presents a single peak feature and the other peaks are lower in height and change more slowly, it is more likely that there is a main component in the sample to be tested, and the concentration of the main component is higher, and thus the peak is more likely to be sheared; therefore, the first shear coefficient can be obtained based on the discrete characteristics of the data at the non-shear point and the changing characteristics of the adjacent data.

[0044] Preferably, in an embodiment of the present invention, the step of obtaining the first shear coefficient includes: calculating the numerical average of all non-shear points and negatively mapping them to obtain the low peak degree of the non-shear point; the height of the peak where the non-shear point is located is relatively low and is not sheared. The smaller the data of the non-shear point, the greater the low peak degree, and the easier it is for the multi-spectral characteristic curve to present a single obvious high peak. Calculate the average value of the numerical difference between the non-shear point and the nearest trough point on the left and right to obtain the amplitude difference; when the amplitude difference is smaller, it means that the height of the peak where the non-shear point is located is lower. Calculate the continuous slope of the non-shear point and the previous data point to obtain the first change value, and the slope of the line connecting the non-shear point and the next data point to obtain the second change value; calculate the absolute value of the difference between the first change value and the second change value to obtain the change difference; when the change difference is smaller, it means that the peak where the non-shear point is located is flatter. Calculate the product of the amplitude difference and the change difference of the non-shear point to obtain the degree of fluctuation; when the amplitude difference and the change difference of the non-shear point are smaller, the degree of fluctuation is smaller, and the peak of the non-shear point is more gentle and less obvious, and then the multi-spectral characteristic curve is more likely to present a single obvious high peak, and the high peak is more likely to be sheared. Calculate the average value of the degree of fluctuation of all non-shear points and negatively map them to obtain the fluctuation characteristic value; calculate the average value of the low peak degree and the fluctuation characteristic value to obtain the first shear coefficient; when the first shear coefficient is larger, it means that more peaks in the multi-spectral characteristic curve show the characteristics of low height and gentle change, and then it is more likely that there is an obvious high peak to be sheared. The formula for obtaining the first shear coefficient includes:

[0045]

[0046] Where W represents the first shear coefficient, D represents the low peak degree, represents an exponential function with a natural constant as the base, I represents the number of non-shear points, represents the amplitude difference of the i-th non-shearing point, represents the change difference of the i-th non-shear point, Indicates the degree of fluctuation, Represents the fluctuation characteristic value.

[0047] Furthermore, when the curves on both sides of the higher peak are smoother, it indicates that the purity of the material components of the sample to be tested is higher, there is no interference from impurities or other components, and it is easier to present the characteristics of a single high peak; therefore, the smoother the curves on both sides of the peak, the higher the height of the peak, and the greater the degree of shearing. Then, the second shear coefficient is obtained according to the smooth characteristics of the curve in the neighborhood of the shear point. Preferably, in an embodiment of the present invention, the step of obtaining the second shear coefficient includes: calculating the cumulative value of the difference between the rate of change of each data point from the nearest trough point on the left side of the shear point to the nearest trough point on the right side and the next data point, and obtaining a smooth characteristic value; normalizing the smooth characteristic value to obtain the second shear coefficient of the shear point. The rate of change is obtained through the point-slope formula based on the data point and the previous data point. If the two sides of the peak where the shear point is located are relatively smooth with fewer subtle fluctuations, the slope from the nearest trough point on the left to the shear point will gradually decrease from a positive number to 0, and the slope from the shear point to the nearest trough point on the right will gradually decrease from 0 to a negative number, and the slope will not fluctuate significantly. Therefore, the larger the smooth eigenvalue and the larger the second shear coefficient, the smoother the curves on both sides of the peak where the shear point is located, the more likely the peak is to be a single obvious high peak, and the greater the degree of shearing.

[0048] After obtaining the first shear coefficient and the second shear coefficient of different shear points, the shearing degree of the peaks at different shear points can be analyzed. Therefore, the shearing degree of the shear point is obtained according to the first shear coefficient and the second shear coefficient, which specifically includes: calculating the average value of the first shear coefficient and the second shear coefficient to obtain the shearing degree of the shear point. The greater the shearing degree, the higher the degree of shearing at the shear point, and then it is necessary to adjust according to the shear point with the highest shearing degree, so that the multi-spectral characteristic curve can display the complete components of the sample to be detected; therefore, the target shearing degree is obtained according to the shearing degree, which specifically includes: taking the maximum value of the shearing degree of all shearing points in the multi-spectral characteristic curve as the target shearing degree.

[0049] Step S4, adjusting the preset light intensity according to the target shearing degree to obtain an adaptive light intensity; obtaining an iterative multispectral characteristic curve for sample detection according to the adaptive light intensity, repeating the process of obtaining the iterative multispectral characteristic curve until there is no shearing point in the iterative multispectral characteristic curve; identifying the sample to be detected through a preset sample prediction model according to the finally obtained iterative multispectral characteristic curve.

[0050] After determining the target shearing degree of the multi-spectral characteristic curve, the preset illumination intensity can be adjusted according to the target shearing degree to obtain an adaptive illumination intensity; preferably, in an embodiment of the present invention, obtaining the adaptive illumination intensity includes: calculating the product of the target shearing degree, the preset illumination intensity and the preset adjustment coefficient to obtain the degree of reduction; in an embodiment of the present invention, the preset adjustment coefficient is 0.1 to avoid excessive reduction resulting in a low signal-to-noise ratio, and the implementer can determine it by himself according to the implementation scenario; when the target shearing degree is greater, the greater the degree of shearing, the greater the degree of reduction. Calculate the difference between the preset illumination intensity and the degree of reduction to obtain the adaptive illumination intensity. After obtaining the adaptive illumination intensity, the newly collected multi-spectral characteristic curve needs to continue to determine whether there is a shearing point. Therefore, the iterative multi-spectral characteristic curve of the sample detection is obtained according to the adaptive illumination intensity to determine whether there is a shearing point in the iterative multi-spectral characteristic curve. If it still exists, the acquisition process of the iterative multi-spectral characteristic curve is repeated until there is no shearing point in the iterative multi-spectral characteristic curve. When there is no shear point in the iterative multispectral characteristic curve finally obtained, the iterative multispectral characteristic curve can fully reflect the reflection intensity of the sample to be detected at different wavelengths, and then identify the sample to be detected through the preset sample prediction model based on the iterative multispectral characteristic curve finally obtained. The preset sample prediction model is constructed by deep learning through the multispectral characteristic curve of the determined sample through the neural network. The iterative multispectral characteristic curve of the sample to be detected is put into the preset sample prediction model. The neural network can match and predict according to the curve change characteristics to predict the name of the sample to be detected. At this point, it is judged whether there is a shear point according to the curve change characteristics of the multispectral characteristic curve, and the shear degree of the shear point is calculated, so as to adjust the preset illumination degree of the multispectral characteristic curve, so that the final iterative multispectral characteristic curve can fully reflect the reflection intensity of the sample to be detected at different wavelengths, and improve the accuracy of the prediction result.

[0051] In summary, the embodiment of the present invention provides a multispectral data prediction method based on deep learning; the peak point is obtained according to the data change characteristics of the multispectral characteristic curve; the shear point and the non-shear point are obtained according to the data distribution characteristics of the peak point and the data fluctuation characteristics of the neighborhood; the target shear degree is obtained according to the data discrete characteristics of the non-shear point and the change characteristics of the adjacent data, and the curve smoothness characteristics of the shear point neighborhood; the preset light intensity is adjusted according to the target shear degree to obtain an adaptive light intensity. The present invention obtains an iterative multispectral characteristic curve of sample detection according to the adaptive light intensity, and identifies the sample to be detected through a preset sample prediction model according to the finally obtained iterative multispectral characteristic curve, thereby improving the prediction accuracy of the sample to be detected.

[0052] The present invention also proposes a multispectral data prediction system based on deep learning, including a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the computer program to implement any step of a multispectral data prediction method based on deep learning.

[0053] It should be noted that the sequence of the above embodiments of the present invention is for description only and does not represent the advantages and disadvantages of the embodiments. The processes depicted in the accompanying drawings do not necessarily require the specific order or continuous order shown to achieve the desired results. In some embodiments, multitasking and parallel processing are also possible or may be advantageous.

[0054] The various embodiments in this specification are described in a progressive manner, and the same or similar parts between the various embodiments can be referenced to each other, and each embodiment focuses on the differences from other embodiments.

Claims

1. A multispectral data prediction method based on deep learning, characterized in that: The method comprises the following steps: Acquire a multi-spectral characteristic curve of the sample to be detected according to a preset light intensity; obtain a peak point according to the data change characteristics of the multi-spectral characteristic curve; Obtaining a suspected shearing degree according to the data distribution characteristics of the peak point and the data fluctuation characteristics of the neighborhood; obtaining a shearing point and a non-shearing point according to the suspected shearing degree of the peak point; Obtain a first shearing coefficient according to the data discrete characteristics of the non-shearing point and the variation characteristics of the adjacent data; obtain a second shearing coefficient according to the curve smoothness characteristics of the neighborhood of the shearing point; obtain the shearing degree of the shearing point according to the first shearing coefficient and the second shearing coefficient; obtain a target shearing degree according to the shearing degree; The preset illumination intensity is adjusted according to the target shearing degree to obtain an adaptive illumination intensity; an iterative multi-spectral characteristic curve of sample detection is obtained according to the adaptive illumination intensity, and the acquisition process of the iterative multi-spectral characteristic curve is repeated until there is no shearing point in the iterative multi-spectral characteristic curve; and the sample to be detected is identified by a preset sample prediction model according to the finally obtained iterative multi-spectral characteristic curve; The step of obtaining the suspected shear degree according to the data distribution characteristics of the peak point and the data fluctuation characteristics of the neighborhood includes: Calculate the average value of the absolute value of the slope of the line connecting each data point from the nearest trough point on the left side of the crest point to the nearest trough point on the right side and the previous data point to obtain the variation amplitude characteristic value; calculate the variance of the absolute value of the slope of the line connecting each data point from the nearest trough point on the left side of the crest point to the nearest trough point on the right side and the previous data point to obtain the fluctuation characteristic value; calculate the product of the variation amplitude characteristic value and the fluctuation characteristic value and negatively correlate them to obtain the variation coefficient; normalize the value of the crest point to obtain the crest coefficient; calculate the average value of the variation coefficient and the crest coefficient to obtain the suspected shear degree of the crest point; The step of obtaining the first shear coefficient according to the discrete characteristics of the data of the non-shear point and the variation characteristics of the adjacent data comprises: Calculate the numerical average of all non-shear points and negatively map them to obtain the low peak degree of the non-shear points; calculate the average of the numerical difference between the non-shear point and the nearest trough points on the left and right to obtain the amplitude difference; calculate the continuous slope between the non-shear point and the previous data point to obtain the first change value, and the slope of the line connecting the non-shear point and the next data point to obtain the second change value; calculate the absolute value of the difference between the first change value and the second change value to obtain the change difference; calculate the product of the amplitude difference of the non-shear point and the change difference to obtain the fluctuation degree; calculate the average of the fluctuation degree of all non-shear points and negatively map them to obtain the fluctuation characteristic value; calculate the average of the low peak degree and the fluctuation characteristic value to obtain the first shear coefficient; The step of obtaining the second shear coefficient according to the smooth curve feature of the shear point neighborhood comprises: Calculate the cumulative value of the difference between the change rate of each data point and the next data point from the nearest trough point on the left side of the shearing point to the nearest trough point on the right side to obtain a smooth eigenvalue; normalize the smooth eigenvalue to obtain a second shear coefficient of the shearing point; The step of obtaining the shear degree of the shear point according to the first shear coefficient and the second shear coefficient comprises: Calculating an average of the first shear coefficient and the second shear coefficient to obtain a shear degree at the shear point; The step of obtaining a target shearing degree according to the shearing degree comprises: The maximum value of the shearing degree of all shearing points in the multi-spectral characteristic curve is taken as the target shearing degree.

2. The multispectral data prediction method based on deep learning according to claim 1, characterized in that: The step of obtaining a peak point according to the data variation characteristics of the multi-spectral characteristic curve comprises: For any data point in the multi-spectral characteristic curve, if the slope of the line connecting the arbitrary data point and the previous data point is a positive number, and the slope of the line connecting the arbitrary data point and the next data point is a negative number, the arbitrary data point is a peak point; if the values ​​of consecutive data points are equal and greater than the values ​​of the adjacent data points on both sides, the data point in the middle of the consecutive data points is taken as the peak point.

3. The multispectral data prediction method based on deep learning according to claim 1, characterized in that: The step of obtaining shear points and non-shear points according to the suspected shear degree of the crest point comprises: When the suspected shearing degree of the crest point exceeds a preset shearing threshold, the crest point is a shearing point; otherwise, the crest point is a non-shearing point.

4. The multispectral data prediction method based on deep learning according to claim 1, characterized in that: The step of adjusting the preset light intensity according to the target shearing degree to obtain the adaptive light intensity comprises: The product of the target shearing degree, the preset light intensity and the preset adjustment coefficient is calculated to obtain the reduction degree; and the difference between the preset light intensity and the reduction degree is calculated to obtain the adaptive light intensity.

5. A multispectral data prediction system based on deep learning, comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, characterized in that: The processor executes the computer program to implement the steps of the method according to any one of claims 1 to 4.

Images