An artificial intelligence-based soil pollution assessment method and system

Abstract

The present invention relates to the technical field of soil pollution detection, and discloses an artificial-intelligence-based soil pollution assessment method and system. The artificial-intelligence-based soil pollution assessment method of the present invention includes the following steps: Step S101, determining the sensitive bands and contribution degree values of pollutants; Step S102, judging whether the pollutants are included in the pollutant list according to the second eigenvalue; Step S103, constructing an association matrix; Step S104, correcting the hyperspectral data of the pollutants in the pollutant list; Step S105, obtaining the concentrations of the pollutants through a pollutant concentration prediction model; Step S106, judging whether the soil is polluted according to the comprehensive pollution index. The present invention utilizes the characteristics of the sensitive bands of pollutants in hyperspectral data to convert the enhanced absorption effect between pollutants into an association matrix representation, and utilizes the nonlinear learning ability of the pollutant concentration prediction model to improve the accuracy of pollutant concentration measurement.

Description

Technical Field

[0001] The present invention relates to the technical field of soil pollution detection, and more specifically, it relates to a method and system for soil pollution assessment based on artificial intelligence. Background Art

[0002] With the acceleration of the industrialization process and the increase in the use of chemical fertilizers and pesticides, the concentration of soil pollutants has been continuously rising, endangering the ecological environment and human health. At present, most soil pollutant detections rely on traditional manual field sampling and experimental analysis methods. Although these methods can provide accurate pollutant concentrations for local areas, they have problems such as poor timeliness, large workload, and cumbersome measurement processes. In addition, it is difficult to achieve large-scale and real-time detection of soil pollutant concentrations, resulting in the inability to comprehensively evaluate the spread of soil pollution.

[0003] With the development of remote sensing technology, especially the emergence of hyperspectral data, currently, the spectral reflectance related to soil pollutants in hyperspectral data is extracted through the correlation coefficient, and then an inversion model between it and the measured value of the pollutant concentration is established through a polynomial to achieve large-scale and real-time detection of soil pollutant concentrations. However, the above scheme does not consider the enhanced absorption effect between pollutants, that is, when multiple pollutants coexist in the soil, the spectral absorption characteristics of these pollutants will interact with each other, resulting in the enhancement or suppression of the spectral reflectance in some bands. For example, heavy metal elements in the soil will form complexes with organic matter, clay minerals, oxides, etc., thus changing the hyperspectral data and affecting the accuracy of pollutant concentration measurement. Summary of the Invention

[0004] The present invention provides a method and system for soil pollution assessment based on artificial intelligence, which solves the technical problem in the related art that the prior art does not consider the influence of the enhanced absorption effect between pollutants on hyperspectral data, resulting in low accuracy of pollutant concentration measurement.

[0005] The present invention provides a method for soil pollution assessment based on artificial intelligence, including the following steps:

[0006] Step S101, determine the sensitive bands and corresponding contribution values of different pollutants in the soil to be detected in the hyperspectral data according to the hyperspectral fingerprint library;

[0007] The sum of the contribution values corresponding to different sensitive bands of the same pollutant in the hyperspectral data is 1;

[0008] The total number M of sensitive bands of each pollutant in the hyperspectral fingerprint library in the hyperspectral data is 3;

[0009] Step S102: Calculate the first eigenvalue based on the sensitive bands of pollutants in the hyperspectral data, and calculate the second eigenvalue by combining the contribution degree values corresponding to the sensitive bands. If the second eigenvalue is greater than or equal to the preset eigenvalue threshold, include the pollutant in the pollutant list.

[0010] Step S103: Construct a correlation matrix based on the peak spectral reflectances of the sensitive bands of all pollutants in the pollutant list in the hyperspectral data.

[0011] The correlation matrix includes A rows and A columns, where A represents the total number of pollutants in the pollutant list. The element value in the i-th row and j-th column is represented by three groups of weight coefficients and bias coefficients of the influence of the i-th pollutant on the j-th pollutant. When i = j, all the weight coefficients and bias coefficients in the i-th row and j-th column are assigned 1 and 0 respectively, where 1 ≤ i ≤ A and 1 ≤ j ≤ A.

[0012] Step S104: Correct the sensitive bands of all pollutants in the pollutant list in the hyperspectral data according to the correlation matrix, and extract the characteristic parameters corresponding to each pollutant.

[0013] Step S105: Input the characteristic parameters corresponding to each pollutant into the pollutant concentration prediction model respectively, and output the concentration of each pollutant.

[0014] Step S106: Calculate the comprehensive pollution index based on the concentration of each pollutant, and if it is greater than or equal to the preset pollution index threshold, it indicates that the soil to be detected is polluted.

[0015] Further, obtain the starting wavelength spectral reflectance and peak spectral reflectance of the sensitive bands of pollutants in the hyperspectral data, equally divide the bands between the starting wavelength spectral reflectance and the peak spectral reflectance into N bands, and calculate the first eigenvalue according to the spectral reflectances of the N bands, where N is a user-defined parameter.

[0016] The first eigenvalue is calculated as follows:

[0017] ;

[0018] where and represent the starting wavelength and ending wavelength of the n-th band respectively, and represent the starting wavelength spectral reflectance and ending wavelength spectral reflectance of the n-th band respectively, and represent the peak spectral reflectance and standard spectral reflectance of the sensitive band respectively, where is a user-defined parameter.

[0019] Further, the second eigenvalue is calculated as follows:

[0020] ;

[0021] where represents the first eigenvalue corresponding to the m-th sensitive band, represents the contribution value corresponding to the m-th sensitive band.

[0022] Further, constructing the correlation matrix includes the following steps:

[0023] Step S201, randomly select H pollutants from the hyperspectral fingerprint library, and randomly generate concentration values of each pollutant within the upper and lower limits to form a soil complex;

[0024] where both H and the upper and lower limits of each pollutant are user-defined parameters;

[0025] Step S202, obtain the peak spectral reflectance of the H pollutants in the soil complex in the sensitive bands of the hyperspectral data as the sample data of the training samples;

[0026] Step S203, respectively obtain the peak spectral reflectance of the H individual pollutants in the sensitive bands of the hyperspectral data as the sample labels of the training samples;

[0027] Step S204, repeat steps S201 to S203 until a training data set consisting of G training samples is obtained, and train the spectral correction model through the training data set;

[0028] where G is a user-defined parameter;

[0029] Step S205, respectively input the peak spectral reflectance of each pollutant in the pollutant list in the sensitive bands of the hyperspectral data into the trained spectral correction model, and output the weight coefficient and bias coefficient of the influence of other pollutants on this pollutant to form the correlation matrix.

[0030] Further, the calculation formula of the spectral correction model is as follows:

[0031] ;

[0032] where and respectively represent the peak spectral reflectance after and before correction of the h-th pollutant in the m-th sensitive band of the hyperspectral data, and respectively represent the weight coefficient and bias coefficient of the h-th pollutant in the m-th sensitive band of the hyperspectral data;

[0033] The calculation formula of the loss function Loss of the spectral correction model is as follows:

[0034] ;

[0035] where represents the peak spectral reflectance after the spectral correction model corrects the m-th sensitive band of the h-th pollutant corresponding to the sample data of the g-th training sample in the hyperspectral data, represents the peak spectral reflectance of the h-th pollutant corresponding to the sample label of the g-th training sample in the m-th sensitive band of the hyperspectral data.

[0036] Furthermore, the starting wavelength spectral reflectance and the corrected peak spectral reflectance of the sensitive band of the pollutant in the hyperspectral data are divided into K bands at equal intervals, and the normalized reflection index of each band is calculated as a characteristic parameter, where K is a user-defined parameter;

[0037] The normalized reflection index of the k-th band of the m-th corrected sensitive band is calculated as follows:

[0038] , where and represent the starting wavelength spectral reflectance and the ending wavelength spectral reflectance of the k-th band of the m-th corrected sensitive band respectively, and 1 ≤ m ≤ M = 3.

[0039] Furthermore, the pollutant concentration prediction model consists of M hidden layers, each hidden layer includes K hidden units, the k-th hidden unit of the m-th hidden layer inputs the normalized reflection index of the k-th band of the m-th corrected sensitive band in the input feature parameters, and outputs an update vector, where the dimension number of the update vector is a user-defined parameter;

[0040] The update vector output by the K-th hidden unit of the M-th hidden layer is input into the classifier, and the category space of the classifier represents the concentration of the pollutant.

[0041] Furthermore, the calculation formula of the pollutant concentration prediction model includes:

[0042] ;

[0043] ;

[0044] ;

[0045] ;

[0046] ;

[0047] Among them represents the updated vector output by the k-th hidden unit of the m-th hidden layer, represents the updated vector output by the (k - 1)-th hidden unit of the m-th hidden layer, represents the updated vector output by the k-th hidden unit of the (m - 1)-th hidden layer, represents the normalized reflection index of the k-th band of the m-th sensitive band after correction among the feature parameters input to the k-th hidden unit of the m-th hidden layer, 、 、 and represent the reset vector, update gate value, intermediate conversion vector, and hidden vector of the k-th hidden unit of the m-th hidden layer respectively, 、 、 and represent the first, second, third, and fourth weight parameters of the k-th hidden unit of the m-th hidden layer respectively, 、 、 and represent the first, second, third, and fourth bias parameters of the k-th hidden unit of the m-th hidden layer respectively, represents element-wise multiplication, sigmoid represents the sigmoid activation function, Swish represents the Swish activation function, and Mish represents the Mish activation function.

[0048] Furthermore, the calculation formula of the comprehensive pollution index comp is as follows:

[0049] ;

[0050] where D represents the total number of pollutants in the pollutant list, 、 and represent the concentration, standard concentration, and weight coefficient of the d-th pollutant in the pollutant list respectively, E represents the custom default concentration, where the standard concentration and weight coefficient of the pollutant are both custom parameters, and max represents taking the maximum value.

[0051] The present invention provides an artificial intelligence-based soil pollution assessment system, including:

[0052] The first module, which is used to determine the sensitive bands and corresponding contribution values of different pollutants in the soil to be detected in the hyperspectral data according to the hyperspectral fingerprint library;

[0053] The second module is used to calculate a first eigenvalue based on the sensitive bands of pollutants in the hyperspectral data, calculate a second eigenvalue by combining the contribution degree values corresponding to the sensitive bands, and if it is determined that the second eigenvalue is greater than or equal to a preset eigenvalue threshold, then include the pollutant in the pollutant list;

[0054] The third module is used to construct a correlation matrix based on the peak spectral reflectances of the sensitive bands of all pollutants in the pollutant list in the hyperspectral data;

[0055] The fourth module is used to correct the sensitive bands of all pollutants in the pollutant list in the hyperspectral data according to the correlation matrix, and extract the characteristic parameters corresponding to each pollutant;

[0056] The fifth module is used to input the characteristic parameters corresponding to each pollutant into the pollutant concentration prediction model respectively, and output the concentration of each pollutant;

[0057] The sixth module is used to calculate a comprehensive pollution index according to the concentration of each pollutant, and if it is determined that the comprehensive pollution index is greater than or equal to a preset pollution index threshold, it indicates that the soil to be detected is polluted.

[0058] The beneficial effects of the present invention are as follows: The present invention utilizes the characteristics of the sensitive bands of pollutants in the hyperspectral data to convert the enhanced absorption effect between pollutants into a correlation matrix representation, corrects the hyperspectral data of pollutants through the correlation matrix, and uses the non-linear learning ability of the pollutant concentration prediction model to realize the mapping with the pollutant concentration, thereby improving the accuracy of pollutant concentration determination. Description of the Drawings

[0059] Figure 1 is a flowchart of an artificial intelligence-based soil pollution assessment method of the present invention;

[0060] Figure 2 is a flowchart of constructing a correlation matrix of the present invention;

[0061] Figure 3 is a schematic diagram of an artificial intelligence-based soil pollution assessment system of the present invention;

[0062] Figure 4 is a comparison chart of lead concentration detection of the present invention;

[0063] Figure 5 is a comparison chart of cadmium concentration detection of the present invention;

[0064] Figure 6 is a comparison chart of perfluorooctane sulfonic acid concentration detection of the present invention.

[0065] In the figure: the first module 301, the second module 302, the third module 303, the fourth module 304, the fifth module 305, the sixth module 306. Detailed implementation manners

[0066] Now, the subject matter described herein will be discussed with reference to example embodiments. It should be understood that discussing these embodiments is only to enable those skilled in the art to better understand and thus implement the subject matter described herein. Without departing from the scope of protection of the content of this specification, changes can be made to the functions and arrangements of the elements discussed. Each example can omit, substitute, or add various processes or components as needed. Additionally, the features described in relation to some examples can also be combined in other examples.

[0067] It should be noted that, unless otherwise defined, the technical terms or scientific terms used in one or more embodiments of the present invention should have the ordinary meaning understood by those of ordinary skill in the art to which the present invention pertains. The "first", "second", and similar terms used in one or more embodiments of the present invention do not denote any order, quantity, or importance, but are only used to distinguish different components. Words such as "including" or "comprising" mean that the elements or objects appearing before this word cover the elements or objects listed after this word and their equivalents, without excluding other elements or objects. Words such as "connected" or "coupled" are not limited to physical or mechanical connections, but can include electrical connections, whether direct or indirect. "Upper", "lower", "left", "right", etc. are only used to represent relative positional relationships, and when the absolute position of the object being described changes, the relative positional relationship may also change accordingly.

[0068] As Figures 1 to 6 shown, an artificial intelligence-based soil pollution assessment method includes the following steps:

[0069] Step S101: Determine the sensitive bands and corresponding contribution values of different pollutants in the hyperspectral data of the soil to be detected according to the hyperspectral fingerprint library;

[0070] The sum of the contribution values corresponding to different sensitive bands of the same pollutant in the hyperspectral data is 1;

[0071] The total number M of sensitive bands of each pollutant in the hyperspectral fingerprint library in the hyperspectral data is 3;

[0072] Step S102: Calculate a first eigenvalue according to the sensitive bands of the pollutant in the hyperspectral data, and calculate a second eigenvalue in combination with the contribution value corresponding to the sensitive band, and if it is determined that it is greater than or equal to a preset eigenvalue threshold, then include the pollutant in the pollutant list;

[0073] Step S103: Construct an association matrix based on the peak spectral reflectance of all pollutants in the pollutant list in the hyperspectral data;

[0074] The association matrix includes A rows and A columns, where A represents the total number of pollutants in the pollutant list. The element value in the i-th row and j-th column is represented by the sum of three weight coefficients and a bias coefficient of the influence of the i-th pollutant on the j-th pollutant. When i = j, all the weight coefficients and bias coefficients in the i-th row and j-th column are assigned 1 and 0 respectively, where 1 ≤ i ≤ A and 1 ≤ j ≤ A;

[0075] Step S104: Correct the sensitive bands of all pollutants in the pollutant list in the hyperspectral data according to the association matrix, and extract the characteristic parameters corresponding to each pollutant;

[0076] Step S105: Input the characteristic parameters corresponding to each pollutant into the pollutant concentration prediction model respectively, and output the concentration of each pollutant;

[0077] Step S106: Calculate the comprehensive pollution index based on the concentration of each pollutant, and judge that it is greater than or equal to the preset pollution index threshold, which indicates that the soil to be detected is polluted.

[0078] It should be noted that the hyperspectral data is represented by the spectral reflectance corresponding to the bands, and the sensitive bands of different pollutants in the hyperspectral fingerprint library and the contribution degree values corresponding to the sensitive bands are all user-defined parameters. The sensitive band represents the band with the most significant spectral reflectance of the pollutant in the hyperspectral data, and the contribution degree value represents the relative importance of the spectral reflectance corresponding to different sensitive bands for identifying pollutants. For example, lead (Pb) includes three sensitive bands, namely 530 - 550 nm, 830 - 850 nm, and 1660 - 1700 nm, and the corresponding contribution degree values are set to 0.25, 0.45, and 0.3 respectively. Another example is that cadmium (Cd) includes three sensitive bands, namely 520 - 540 nm, 910 - 930 nm, and 2250 - 2350 nm, and the corresponding contribution degree values are set to 0.2, 0.5, and 0.3 respectively.

[0079] It should be noted that the present invention first uses the hyperspectral fingerprint library to determine the types of pollutants in the soil to be detected, then corrects the sensitive bands of the pollutants in the hyperspectral data (correcting the spectral reflectance), and finally realizes the determination of the pollutant concentration through the pollutant concentration prediction model. In addition, in order to improve the measurement accuracy, one pollutant type can correspond to one pollutant concentration prediction model, but the structures and calculation formulas of all pollutant concentration prediction models are the same.

[0080] In one embodiment of the present invention, the starting wavelength spectral reflectance and the peak wavelength spectral reflectance of the sensitive band of the pollutant in the hyperspectral data are obtained, the band between the starting wavelength spectral reflectance and the peak wavelength spectral reflectance is equally divided into N bands, and a first eigenvalue is calculated based on the spectral reflectance of the N bands, where N is a user-defined parameter, for example, N is set to 20;

[0081] The first eigenvalue is calculated as follows:

[0082] ;

[0083] where and respectively represent the starting wavelength and the ending wavelength of the nth band, and respectively represent the starting wavelength spectral reflectance and the ending wavelength spectral reflectance of the nth band, and respectively represent the peak wavelength spectral reflectance and the standard spectral reflectance of the sensitive band, where is a user-defined parameter, for example is set to 100.

[0084] In one embodiment of the present invention, the second eigenvalue is calculated as follows:

[0085] ;

[0086] where represents the first eigenvalue corresponding to the mth sensitive band, represents the contribution value corresponding to the mth sensitive band.

[0087] It should be noted that one sensitive band of the pollutant in the hyperspectral data corresponds to one first eigenvalue, and the calculation formulas of the first eigenvalue and the second eigenvalue do not involve dimensional calculations, and the preset eigenvalue threshold is also a user-defined parameter, and the second eigenvalue can be calculated based on the sensitive band of the pollutant at a low concentration (for example, 10% of the pollution standard) as the preset eigenvalue threshold.

[0088] In one embodiment of the present invention, as Figure 2 shown, an association matrix is constructed, including the following steps:

[0089] Step S201, randomly select H pollutants from the hyperspectral fingerprint library, and randomly generate concentration values of each pollutant within the upper and lower limits to form a soil complex;

[0090] where both H and the upper and lower limits of each pollutant are user-defined parameters;

[0091] Step S202: Obtain the peak spectral reflectance of the sensitive bands of H pollutants in the hyperspectral data in the soil complex as the sample data of the training samples.

[0092] Step S203: Respectively obtain the peak spectral reflectance of the sensitive bands of H individual pollutants in the hyperspectral data as the sample labels of the training samples.

[0093] Step S204: Repeat Step S201 to Step S203 until a training data set composed of G training samples is obtained, and train the spectral correction model through the training data set.

[0094] where G is a user-defined parameter;

[0095] Step S205: Respectively input the peak spectral reflectance of the sensitive bands of each pollutant in the pollutant list in the hyperspectral data into the trained spectral correction model, and output the weight coefficient and bias coefficient of the influence of other pollutants on this pollutant to form an association matrix.

[0096] It should be noted that the number of pollutants selected from the hyperspectral fingerprint library each time may not be a fixed value. Setting upper and lower limits for the concentration value of each pollutant is to ensure the rationality of random generation. The purpose of the spectral correction model is to construct a mapping relationship between the peak spectral reflectance of the sensitive bands of multiple pollutants in the hyperspectral data and the peak spectral reflectance of a single pollutant in the hyperspectral data through learnable weight coefficients and bias coefficients, so as to achieve the effect of hyperspectral correction. In addition, the larger the number G of training samples, the better the training effect.

[0097] In an embodiment of the present invention, the calculation formula of the spectral correction model is as follows:

[0098] ;

[0099] where and respectively represent the peak spectral reflectance of the m-th sensitive band of the h-th pollutant in the hyperspectral data after and before correction, and respectively represent the weight coefficient and bias coefficient of the m-th sensitive band of the h-th pollutant in the hyperspectral data;

[0100] The calculation formula of the loss function Loss of the spectral correction model is as follows:

[0101] ;

[0102] where It represents the peak spectral reflectance after correcting the m-th sensitive band of the h-th pollutant corresponding to the sample data of the g-th training sample in the hyperspectral data through the spectral correction model. It represents the peak spectral reflectance of the h-th pollutant corresponding to the sample label of the g-th training sample in the m-th sensitive band of the hyperspectral data.

[0103] It should be noted that a gradient optimizer is used to update the weight coefficients and bias coefficients in the spectral correction model in the reverse direction, so as to minimize the loss value calculated forward using the loss function. Commonly used gradient optimizers include AMSGrad, AdaBelief, etc. In addition, the method of correcting the sensitive bands of all pollutants in the pollutant list in the hyperspectral data according to the correlation matrix is the same as the calculation formula of the spectral correction model.

[0104] In an embodiment of the present invention, the starting wavelength spectral reflectance and the corrected peak spectral reflectance of the sensitive band of the pollutant in the hyperspectral data are divided into K bands at equal intervals, and the normalized reflection index of each band is calculated as a characteristic parameter, where K is a user-defined parameter. For example, K is set to 10.

[0105] The normalized reflection index of the k-th band of the m-th corrected sensitive band The calculation formula is as follows:

[0106] , where and respectively represent the starting wavelength spectral reflectance and the ending wavelength spectral reflectance of the k-th band of the m-th corrected sensitive band, and 1 ≤ m ≤ M = 3.

[0107] It should be noted that the characteristic parameter is represented in the form of a 3-row and K-column matrix, that is, the rows of the matrix correspond to 3 sensitive bands, and the columns of the matrix correspond to K bands.

[0108] In an embodiment of the present invention, the pollutant concentration prediction model consists of M hidden layers, each hidden layer includes K hidden units, and the k-th hidden unit of the m-th hidden layer inputs the normalized reflection index of the k-th band of the m-th corrected sensitive band in the characteristic parameter and outputs an update vector, where the dimension number of the update vector is a user-defined parameter.

[0109] The update vector output by the K-th hidden unit of the M-th hidden layer is input into the classifier, and the category space of the classifier represents the concentration of the pollutant.

[0110] In an embodiment of the present invention, the calculation formula of the pollutant concentration prediction model includes:

[0111] ;

[0112] ;

[0113] ;

[0114] ;

[0115] ;

[0116] where represents the updated vector output by the k-th hidden unit of the m-th hidden layer, represents the updated vector output by the (k - 1)-th hidden unit of the m-th hidden layer, represents the updated vector output by the k-th hidden unit of the (m - 1)-th hidden layer, represents the normalized reflection index of the k-th band of the m-th sensitive band after correction among the feature parameters input to the k-th hidden unit of the m-th hidden layer, , , and represent the reset vector, update gate value, intermediate transformation vector, and hidden vector of the k-th hidden unit of the m-th hidden layer respectively, , , and represent the first, second, third, and fourth weight parameters of the k-th hidden unit of the m-th hidden layer respectively, , , and represent the first, second, third, and fourth bias parameters of the k-th hidden unit of the m-th hidden layer respectively, represents element-wise multiplication, sigmoid represents the sigmoid activation function, Swish represents the Swish activation function, and Mish represents the Mish activation function.

[0117] It should be noted that the dimension number of the updated vector can be set to 16. Then, the dimension numbers of the intermediate transformation vector and the hidden vector must also be 16, while the update gate value is a real number. The first weight parameter needs to be designed as a matrix of size (1 + 16) × 1, and the first bias parameter can be designed as a real number. The second weight parameter needs to be designed as a matrix of size (16 + 16) × 16, and the second bias parameter needs to be designed as a vector of size 1 × 16. The third weight parameter needs to be designed as a matrix of size 1 × 16, the dimension number of the reset vector is 16, the third bias parameter needs to be designed as a vector of size 1 × 16, the fourth weight parameter needs to be designed as a matrix of size (1 + 16) × 16, and the fourth bias parameter needs to be designed as a vector of size 1 × 16.

[0118] In one embodiment of the present invention, the calculation formula of the comprehensive pollution index comp is as follows:

[0119] ;

[0120] Where D represents the total number of pollutants in the pollutant list, , and respectively represent the concentration, standard concentration and weight coefficient of the d-th pollutant in the pollutant list, E represents a user-defined default concentration, for example, E is set to 0.01, where the standard concentration and weight coefficient of the pollutant are both user-defined parameters, and max represents taking the maximum value.

[0121] It should be noted that the calculation formula of the comprehensive pollution index does not involve dimension calculation, and the preset pollution index threshold is a user-defined parameter. In addition, the concentration of each pollutant can also be directly compared with the corresponding standard concentration, and an alarm message is generated if it is greater than the standard concentration, which will not be elaborated here.

[0122] In one embodiment of the present invention, as Figure 3 shown, the present invention provides an artificial intelligence-based soil pollution assessment system, including:

[0123] The first module 301 is used to determine the sensitive bands and corresponding contribution values of different pollutants in the soil to be detected in the hyperspectral data according to the hyperspectral fingerprint library;

[0124] The second module 302 is used to calculate the first eigenvalue according to the sensitive bands of the pollutants in the hyperspectral data, calculate the second eigenvalue in combination with the contribution values corresponding to the sensitive bands, and if it is judged that it is greater than or equal to the preset eigenvalue threshold, then include the pollutant in the pollutant list;

[0125] The third module 303 is used to construct an association matrix according to the peak spectral reflectance of the sensitive bands of all pollutants in the pollutant list in the hyperspectral data;

[0126] The fourth module 304 is used to correct the sensitive bands of all pollutants in the pollutant list according to the association matrix and extract the characteristic parameters corresponding to each pollutant;

[0127] The fifth module 305 is used to input the characteristic parameters corresponding to each pollutant into the pollutant concentration prediction model respectively and output the concentration of each pollutant;

[0128] The sixth module 306 is used to calculate the comprehensive pollution index according to the concentration of each pollutant, and if it is judged that it is greater than or equal to the preset pollution index threshold, it means that the soil to be detected is polluted.

[0129] In one embodiment of the present invention, as Figure 4 shown, the soil pollutant concentration detection methods provided by the present invention (spectral correction model and pollutant concentration prediction model) and the inversion model of the prior art (polynomial model mentioned in the background art) are respectively used to detect 10 groups of lead concentrations (unit: mg / kg) for comparison.

[0130] In one embodiment of the present invention, as Figure 5 shown, the soil pollutant concentration detection methods provided by the present invention (spectral correction model and pollutant concentration prediction model) and the inversion model of the prior art (polynomial model mentioned in the background art) are respectively used to detect 10 groups of cadmium concentrations (unit: mg / kg) for comparison.

[0131] It should be noted that the idea of the present invention can also be used to measure the concentrations of other compounds in soil. For example, per- and polyfluoroalkyl substances (PFAS for short), which are widely used in various consumer and industrial products due to their excellent thermal and chemical stability. Common PFAS include perfluorooctanoic acid (PFOA), perfluorooctane sulfonic acid (PFOS), etc.

[0132] In one embodiment of the present invention, as Figure 6 shown, the soil pollutant concentration detection methods provided by the present invention (spectral correction model and pollutant concentration prediction model) and the inversion model of the prior art (polynomial model mentioned in the background art) are respectively used to detect 10 groups of PFOS concentrations (unit: μg / kg) for comparison.

[0133] It should be noted that through observation, it is found that the detection accuracy of the soil pollutant concentration detection method provided by the present invention is higher.

[0134] It should be noted that the setting of the interval and threshold values is for the convenience of comparison. The size of the threshold depends on the amount of sample data and the base quantity set by those skilled in the art for each group of sample data, as long as it does not affect the proportional relationship between the parameters and the quantified values. And the above formulas are all calculations of taking the numerical values without dimensions. The formulas are all obtained by software simulation of collecting a large amount of data to get a formula closest to the actual situation. The preset parameters in the formulas are set by those skilled in the art according to the actual situation.

[0135] The above describes the embodiments of this embodiment, but this embodiment is not limited to the above specific implementation manners. The above specific implementation manners are only illustrative and not restrictive. Under the inspiration of this embodiment, those of ordinary skill in the art can also make many forms, all of which fall within the protection scope of this embodiment.

Claims

1. An artificial intelligence-based soil pollution assessment method, characterized in that, It includes the following steps: Step S101: Determine the sensitive bands of different pollutants in the soil to be detected in the hyperspectral data and the corresponding contribution values according to the hyperspectral fingerprint library; The sum of the contribution values corresponding to different sensitive bands of the same pollutant in the hyperspectral data is 1; The total number M of sensitive bands of each pollutant in the hyperspectral fingerprint library in the hyperspectral data is 3; Step S102: Calculate the first eigenvalue based on the sensitive bands of the pollutant in the hyperspectral data, and calculate the second eigenvalue in combination with the contribution value corresponding to the sensitive band. If it is greater than or equal to the preset eigenvalue threshold, include the pollutant in the pollutant list; Step S103: Construct an association matrix based on the peak spectral reflectances of the sensitive bands of all pollutants in the pollutant list in the hyperspectral data; The association matrix includes A rows and A columns. A represents the total number of pollutants in the pollutant list. The element value in the i-th row and j-th column is represented by the three groups of weight coefficients and bias coefficients of the influence of the i-th pollutant on the j-th pollutant. When i = j, all the weight coefficients and bias coefficients in the i-th row and j-th column are assigned 1 and 0 respectively, where 1 ≤ i ≤ A and 1 ≤ j ≤ A; Step S104: Correct the sensitive bands of all pollutants in the pollutant list in the hyperspectral data according to the association matrix, and extract the characteristic parameters corresponding to each pollutant; Step S105: Input the characteristic parameters corresponding to each pollutant into the pollutant concentration prediction model respectively, and output the concentration of each pollutant; Step S106: Calculate the comprehensive pollution index according to the concentration of each pollutant, and if it is greater than or equal to the preset pollution index threshold, it means that the soil to be detected is polluted; Obtain the starting wavelength spectral reflectance and peak spectral reflectance of the sensitive band of the pollutant in the hyperspectral data. Divide the band between the starting wavelength spectral reflectance and the peak spectral reflectance into N equal intervals. Calculate the first eigenvalue according to the spectral reflectances of the N bands, where N is a user-defined parameter; The first eigenvalue The calculation formula is as follows: ； wherein and respectively represent the starting wavelength and the ending wavelength of the n-th band, and respectively represent the spectral reflectance at the starting wavelength and the spectral reflectance at the ending wavelength of the n-th band, and respectively represent the spectral reflectance at the peak of the sensitive band and the standard spectral reflectance, wherein is a user-defined parameter; Second eigenvalue The calculation formula is as follows: ； wherein represents the first eigenvalue corresponding to the m-th sensitive band, represents the contribution value corresponding to the m-th sensitive band.

2. The soil pollution assessment method based on artificial intelligence according to claim 1, wherein, Construct an association matrix, including the following steps: Step S201: Randomly select H pollutants from the hyperspectral fingerprint library, and randomly generate the concentration values of each pollutant within the upper and lower limits to form a soil complex; Where H and the upper and lower limits of each pollutant are user-defined parameters; Step S202: Obtain the peak spectral reflectances of the sensitive bands of the H pollutants in the soil complex in the hyperspectral data as the sample data of the training samples; Step S203: Obtain the peak spectral reflectances of the sensitive bands of the H individual pollutants in the hyperspectral data as the sample labels of the training samples respectively; Step S204: Repeat Step S201 to Step S203 until a training data set composed of G training samples is obtained, and train the spectral correction model through the training data set; Where G is a user-defined parameter; Step S205: Input the peak spectral reflectances of the sensitive bands of each pollutant in the pollutant list in the hyperspectral data into the trained spectral correction model respectively, and output the weight coefficients and bias coefficients of the influence of other pollutants on the pollutant to form an association matrix.

3. The method for evaluating soil pollution based on artificial intelligence according to claim 2, wherein The calculation formula of the spectral correction model is as follows: ； where and respectively represent the peak spectral reflectance of the h-th pollutant in the m-th sensitive band of the hyperspectral data after and before correction, and respectively represent the weight coefficient and bias coefficient of the m-th sensitive band of the h-th pollutant in the hyperspectral data; The calculation formula of the loss function Loss of the spectral correction model is as follows: ； wherein represents the peak spectral reflectance of the h-th pollutant corresponding to the sample data of the g-th training sample after being corrected by the spectral correction model in the m-th sensitive band of the hyperspectral data, represents the peak spectral reflectance of the h-th pollutant corresponding to the sample label of the g-th training sample in the m-th sensitive band of the hyperspectral data.

4. The method for evaluating soil pollution based on artificial intelligence according to claim 1, wherein The starting wavelength spectral reflectance of the sensitive band of the pollutant in the hyperspectral data and the corrected peak spectral reflectance are divided into K bands at equal intervals, and the normalized reflection index of each band is calculated as a characteristic parameter, where K is a user-defined parameter; Normalized reflection index of the k-th band in the m-th corrected sensitive band The calculation formula is as follows: , where and respectively represent the starting wavelength spectral reflectance and the ending wavelength spectral reflectance of the k-th band of the m-th sensitive band after calibration, where 1 ≤ m ≤ M = 3.

5. The soil pollution assessment method based on artificial intelligence according to claim 4, wherein, The pollutant concentration prediction model consists of Q hidden layers, each hidden layer includes K hidden units. The k-th hidden unit of the q-th hidden layer inputs the normalized reflection index of the k-th band of the q-th sensitive band in the corrected characteristic parameters and outputs an update vector, where the dimension number of the update vector is a user-defined parameter, Q = M, 1 ≤ q ≤ Q; The update vector output by the K-th hidden unit of the Q-th hidden layer is input into the classifier, and the category space of the classifier represents the concentration of the pollutant.

6. The soil pollution assessment method based on artificial intelligence according to claim 5, characterized in that, The calculation formula of the pollutant concentration prediction model includes: ； ； ； ； ； Among them represents the updated vector output by the k-th hidden unit of the q-th hidden layer, represents the updated vector output by the (k - 1)-th hidden unit of the q-th hidden layer, represents the updated vector output by the k-th hidden unit of the (q - 1)-th hidden layer, represents the normalized reflection index of the k-th band of the q-th sensitive band after correction among the feature parameters input to the k-th hidden unit of the q-th hidden layer, 、 、 and represent the reset vector, update gate value, intermediate transformation vector, and hidden vector of the k-th hidden unit of the q-th hidden layer respectively, 、 、 and represent the first, second, third, and fourth weight parameters of the k-th hidden unit of the q-th hidden layer respectively, 、 、 and represent the first, second, third, and fourth bias parameters of the k-th hidden unit of the q-th hidden layer respectively, represents element-wise multiplication, sigmoid represents the sigmoid activation function, Swish represents the Swish activation function, and Mish represents the Mish activation function.

7. The method for evaluating soil pollution based on artificial intelligence according to claim 1, wherein The calculation formula of the comprehensive pollution index comp is as follows: ； where D represents the total number of pollutants in the pollutant list, , and respectively represent the concentration, standard concentration, and weight coefficient of the d-th pollutant in the pollutant list, E represents the custom default concentration, where the standard concentration and weight coefficient of the pollutant are both custom parameters, and max represents taking the maximum value.

8. An artificial intelligence-based soil pollution assessment system, characterized in that, Implement an artificial intelligence-based soil pollution assessment method according to any one of claims 1 to 7, including: The first module is used to determine the sensitive bands and the corresponding contribution values of different pollutants in the soil to be detected in the hyperspectral data according to the hyperspectral fingerprint library; The second module is used to calculate the first eigenvalue according to the sensitive bands of the pollutants in the hyperspectral data, and calculate the second eigenvalue in combination with the contribution values corresponding to the sensitive bands, and if it is judged that it is greater than or equal to the preset eigenvalue threshold, then include the pollutant in the pollutant list; The third module is used to construct an association matrix according to the peak spectral reflectances of the sensitive bands of all pollutants in the pollutant list in the hyperspectral data; The fourth module is used to correct the sensitive bands of all pollutants in the pollutant list in the hyperspectral data according to the association matrix and extract the characteristic parameters corresponding to each pollutant; The fifth module is used to input the characteristic parameters corresponding to each pollutant into the pollutant concentration prediction model respectively and output the concentration of each pollutant; The sixth module is used to calculate the comprehensive pollution index according to the concentration of each pollutant, and if it is judged that it is greater than or equal to the preset pollution index threshold, it means that the soil to be detected is polluted.

Images