A multivariate detection method, system and device for soil mineral nutrient elements

Abstract

The present application discloses a multivariate detection method, system, and apparatus for soil mineral nutrient element detection, relating to the field of soil mineral nutrient element detection. The method comprises: obtaining spectral information of a soil sample to be tested; extracting spectral intensity data corresponding to predetermined spectral variables based on the obtained spectral information; and inputting the data into a trained multivariate weighted network model to obtain the contents of multiple elements in the soil sample to be tested; wherein the parallel fully connected layers of the multivariate weighted network model are used to output preliminary predicted contents of a first group of multiple elements, and the weighted multivariate output layer performs weighted calculations on each element to obtain predicted contents of the multiple elements in the soil sample to be tested. The present application detects soil mineral nutrient elements by combining laser-induced breakdown spectroscopy technology with a trained multivariate weighted network model, thereby achieving rapid detection of soil mineral nutrient elements while ensuring the accuracy of the test results.

Description

Technical Field

[0001] The present application relates to the field of soil mineral nutrient element detection, and in particular to a multivariate detection method, system and equipment for soil mineral nutrient elements. Background Art

[0002] Soil mineral nutrients are essential components of soil and are crucial for plant growth, development, soil biological processes, and agricultural production. Therefore, accurate, rapid, and pollution-free testing of soil mineral nutrients plays a crucial role in guiding optimal fertilization, improving crop yield and quality, and protecting the ecological environment.

[0003] Currently, the main methods for detecting mineral nutrients in soil include atomic absorption spectrometry, inductively coupled plasma atomic emission spectrometry, and inductively coupled plasma mass spectrometry. Although these methods offer high accuracy, their cumbersome processing and long detection cycles make them inadequate for rapid detection and often exhibit lags. Summary of the Invention

[0004] The purpose of this application is to provide a multivariate detection method, system and equipment for soil mineral nutrient elements, which can quickly and accurately detect the content of multiple elements in the soil.

[0005] To achieve the above objectives, this application provides the following solutions:

[0006] In a first aspect, the present application provides a multivariate detection method for soil mineral nutrients, comprising:

[0007] Acquire spectral information of a soil sample to be tested; the spectral information includes a number of spectral variables and spectral intensity data corresponding to each spectral variable; the spectral information is obtained by processing the soil sample to be tested using laser induced breakdown spectroscopy technology.

[0008] According to the spectral information of the soil sample to be tested, the spectral intensity data corresponding to the predetermined spectral variable is extracted.

[0009] The spectral intensity data corresponding to the predetermined spectral variables are input into the trained multivariate weighted network model to obtain the predicted contents of multiple elements in the soil sample to be tested.

[0010] The multivariate weighted network model includes an input layer, a weighted conversion layer, a parallel fully connected layer and a weighted multivariate output layer.

[0011] The input layer is used to receive spectral intensity data corresponding to the predetermined spectral variable; the number of neurons in the input layer is equal to the number of the predetermined spectral variables.

[0012] The weighted conversion layer is used to weight the numerical values ​​output by each neuron in the input layer to obtain corresponding weighted values.

[0013] The parallel fully connected layer is used to calculate the preliminary content of multiple elements based on all weighted values ​​output by the weighted transformation layer; the parallel fully connected layer includes a fully connected network, and the output of the fully connected network includes a first number of neurons; wherein the first number of neurons is used to output the preliminary predicted content of a first group of multiple elements.

[0014] The weighted multivariate output layer is used to perform weighted calculations on each element based on the preliminary contents of the first group of multiple elements output by the parallel fully connected layer, so as to obtain the predicted contents of the multiple elements in the soil sample to be tested.

[0015] In a second aspect, the present application provides a multivariate detection system for soil mineral nutrients, including a detection system and a LIBS device; wherein,

[0016] The LIBS device is a device that obtains spectral information of a soil sample to be tested based on laser-induced breakdown spectroscopy technology; the LIBS device includes a pulsed laser and a spectrometer.

[0017] The detection system includes a spectrum information acquisition module, a spectrum intensity data extraction module and a soil mineral nutrient element prediction module; wherein,

[0018] The spectral information acquisition module is connected to the LIBS device and is used to obtain spectral information of the soil sample to be tested; the spectral information includes a number of spectral variables and spectral intensity data corresponding to each spectral variable.

[0019] The spectral intensity data extraction module is used to extract spectral intensity data corresponding to predetermined spectral variables based on the spectral information of the soil sample to be tested.

[0020] The soil mineral nutrient element prediction module is used to input the spectral intensity data corresponding to the predetermined spectral variables into the trained multivariate weighted network model to obtain the content of multiple elements in the soil sample to be tested.

[0021] The multivariate weighted network model includes an input layer, a weighted conversion layer, a parallel fully connected layer and a weighted multivariate output layer.

[0022] The input layer is used to receive spectral intensity data corresponding to the predetermined spectral variable; the number of neurons in the input layer is equal to the number of the predetermined spectral variables.

[0023] The weighted conversion layer is used to weight the numerical values ​​output by each neuron in the input layer to obtain corresponding weighted values.

[0024] The parallel fully connected layer is used to calculate the preliminary content of multiple elements based on all weighted values ​​output by the weighted transformation layer; the parallel fully connected layer includes a fully connected network, and the output of the fully connected network includes a first number of neurons; wherein the first number of neurons is used to output the preliminary predicted content of a first group of multiple elements.

[0025] The weighted multivariate output layer is used to perform weighted calculations on each element based on the preliminary predicted contents of the first group of multiple elements output by the parallel fully connected layer, so as to obtain the predicted contents of the multiple elements in the soil sample to be tested.

[0026] In a third aspect, the present application provides a computer device 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 the multivariate detection method for soil mineral nutrient elements.

[0027] According to the specific embodiments provided in this application, this application discloses the following technical effects:

[0028] The present application provides a multivariate detection method, apparatus, medium, and product for soil mineral nutrient elements. The method comprises: obtaining a spectrum of a soil sample to be tested using laser-induced breakdown spectroscopy technology; extracting spectral intensity data corresponding to predetermined spectral variables based on the spectrum of the soil sample to be tested; inputting the spectral intensity data corresponding to the predetermined spectral variables into a trained multivariate weighted network model to obtain predicted contents of multiple elements in the soil sample to be tested; the multivariate weighted network model comprises an input layer, a weighted transformation layer, a parallel fully connected layer, and a weighted multivariate output layer; wherein the number of neurons in the input layer is equal to the number of predetermined spectral variables; the weighted transformation layer weights the numerical values ​​output by each neuron in the input layer to obtain corresponding weighted values; the parallel fully connected layer calculates preliminary predicted contents of the multiple elements based on all weighted values; the parallel fully connected layer comprises a fully connected network, the output of the fully connected network comprises a first number of neurons, and correspondingly outputs preliminary predicted contents of a first group of multiple elements; and the weighted multivariate output layer performs weighted calculations on each element based on the preliminary predicted contents of the first group of multiple elements output by the parallel fully connected layer to obtain predicted contents of the multiple elements in the soil sample to be tested.

[0029] This application combines laser-induced breakdown spectroscopy (LIBS) technology with a trained multivariate weighted network model to detect soil mineral nutrients. Compared with traditional methods for detecting mineral nutrients in soil, LIBS technology has the characteristics of simple processing, rapidity and low damage, eliminating the cumbersome processing process of traditional methods. The trained multivariate weighted network model first calculates the preliminary predicted contents of multiple elements in a first group, and then performs weighted calculations on the preliminary predicted contents of multiple elements in the first group to obtain the final predicted content of each element, thereby ensuring the accuracy of the detection. Therefore, this application realizes the rapid detection of soil mineral nutrients while ensuring the accuracy of the detection results. BRIEF DESCRIPTION OF THE DRAWINGS

[0030] In order to more clearly illustrate the embodiments of the present application or the technical solutions in the prior art, the following briefly introduces the drawings required for use in the embodiments. Obviously, the drawings described below are only some embodiments of the present application. For ordinary technicians in this field, other drawings can be obtained based on these drawings without creative work.

[0031] Figure 1 A schematic flow chart of a multivariate detection method for soil mineral nutrients provided in one embodiment of the present application;

[0032] FIG2( a ) is a LIBS spectrum of an original soil sample provided in an embodiment of the present application;

[0033] FIG2( b ) is a LIBS spectrum of a soil sample after denoising provided in an embodiment of the present application;

[0034] Figure 3 A schematic diagram of a multivariable output weighted network structure provided in one embodiment of the present application;

[0035] Figure 4 A schematic diagram of a multivariate detection system for soil mineral nutrients provided in one embodiment of the present application;

[0036] FIG5( a ) is a schematic diagram of nitrogen LIBS detection results based on MOW-Net according to an embodiment of the present application;

[0037] FIG5( b ) is a schematic diagram of the potassium LIBS detection results based on MOW-Net provided in one embodiment of the present application;

[0038] FIG5( c ) is a schematic diagram of the calcium LIBS detection results based on MOW-Net provided in one embodiment of the present application;

[0039] Figure 6 A schematic diagram of the structure of a computer device provided in one embodiment of the present application. DETAILED DESCRIPTION

[0040] The following will be combined with the drawings in the embodiments of this application to clearly and completely describe the technical solutions in the embodiments of this application. Obviously, the embodiments described are only part of the embodiments of this application, not all of the embodiments. Based on the embodiments in this application, all other embodiments obtained by ordinary technicians in this field without making creative efforts are within the scope of protection of this application.

[0041] Traditional soil mineral nutrient element detection methods are time-consuming and labor-intensive, consume a lot of chemical reagents, and are prone to environmental pollution. Laser induced breakdown spectroscopy (LIBS) technology is an atomic spectroscopy technology with the characteristics of simple processing, rapid and minimal damage, and remote detection. The principle of soil element measurement is to laser ablate soil element samples to generate plasma, and collect spectral signals in the plasma through a spectrometer. The intensity of element-related spectral lines often increases with the increase of the concentration of elements in the sample. Therefore, a relationship model between the LIBS emission line intensity of the element and the element concentration in the soil sample can be established to achieve rapid quantitative detection of soil elements. However, LIBS spectrum noise interference is serious, and the prediction model is complex and diverse. In response to the problem that soil LIBS spectra have many variables and serious spectral noise interference, the present application provides a soil LIBS spectrum noise removal method based on nearly 0 standard deviation, which can effectively remove spectral noise and retain effective information. In addition, to address the problem of the large variety of soil mineral nutrients and the complex and diverse prediction models, the present application provides a LIBS multivariate detection network for soil mineral nutrients, namely a multivariable output weighting-network (MOW-Net), which can realize the simultaneous detection of multiple soil mineral nutrients using one model, avoiding the complexity of multiple detection methods and multiple prediction models for different elements, and improving detection efficiency.

[0042] This application adopts LIBS spectroscopy rapid detection technology, which can detect soil mineral nutrients without chemical treatment, improves the detection speed, and is environmentally friendly.

[0043] In order to make the above-mentioned purposes, features and advantages of the present application more obvious and easy to understand, the present application is further described in detail below with reference to the accompanying drawings and specific implementation methods.

[0044] In an exemplary embodiment, Figure 1 As shown, the present application provides a multivariate detection method for soil mineral nutrients, comprising the following steps:

[0045] S100: Acquire spectral information of a soil sample to be tested; the spectral information includes a plurality of spectral variables and spectral intensity data corresponding to each spectral variable; the spectral information is obtained by processing the soil sample to be tested using laser-induced breakdown spectroscopy. Different spectral variables correspond to different wavelengths.

[0046] S101: Extracting spectral intensity data corresponding to predetermined spectral variables based on spectral information of the soil sample to be tested.

[0047] S102: Inputting the spectral intensity data corresponding to the predetermined spectral variables into a trained multivariate weighted network model to obtain predicted contents of multiple elements in the soil sample to be tested.

[0048] The multivariate weighted network model includes an input layer, a weighted conversion layer, a parallel fully connected layer and a weighted multivariate output layer.

[0049] The input layer is used to receive spectral intensity data corresponding to the predetermined spectral variable; the number of neurons in the input layer is equal to the number of the predetermined spectral variables.

[0050] The weighted conversion layer is used to weight the numerical values ​​output by each neuron in the input layer to obtain corresponding weighted values.

[0051] The parallel fully connected layer is configured to calculate preliminary contents of the multiple elements based on all weighted values ​​output by the weighted transformation layer. The parallel fully connected layer includes a fully connected network, and the output of the fully connected network includes a first number of neurons. The first number of neurons is configured to output preliminary predicted contents of a first group of multiple elements. The value of the first number is equal to the first group multiplied by the number of types of the multiple elements.

[0052] The weighted multivariate output layer is used to perform weighted calculations on each element based on the preliminary contents of the first group of multiple elements output by the parallel fully connected layer, so as to obtain the predicted contents of the multiple elements in the soil sample to be tested.

[0053] Furthermore, the process of determining the predetermined spectral variables includes:

[0054] S200: Calculating the standard deviation of the spectral intensity corresponding to each spectral variable based on spectral information of a plurality of experimental soil samples, wherein each soil sample contains each of the spectral variables. The spectral information of the experimental soil samples is also obtained by processing the soil samples using laser-induced breakdown spectroscopy.

[0055] S201: Determine a spectral variable whose standard deviation is greater than a preset standard deviation threshold as a predetermined spectral variable.

[0056] The preset standard deviation threshold is a value close to 0.

[0057] The original soil sample LIBS spectrum contains a large amount of spectral noise, which brings information redundancy and reduces the accuracy of modeling. Therefore, it is necessary to remove the spectral noise. The noise spectrum has less heterogeneity among all soil sample spectra. In comparison, the spectrum containing information has greater heterogeneity because it can reflect different soil sample characteristics. Therefore, according to the difference in heterogeneity between the noise spectrum and the information spectrum, this application provides a spectral noise removal method based on near-zero standard deviation. The standard deviation can reflect the size of heterogeneity to a certain extent. The smaller the standard deviation, the smaller the heterogeneity, and the larger the standard deviation, the greater the heterogeneity. Therefore, removing a portion of the spectral variables with near-zero standard deviation is conducive to removing spectral noise.

[0058] As an optional implementation, the process of removing spectral noise (also the process of determining the predetermined spectral variables) may also be as follows:

[0059] First, based on the spectral information of all experimental soil samples, the standard deviation of the spectral intensity corresponding to each spectral variable (also known as spectral wavelength) was calculated; second, the spectral variables were sorted in ascending order according to the size of the standard deviation; finally, the spectral variables with a standard deviation close to 0 were removed.

[0060] The experimental results show that in the spectral information of the experimental soil samples, the standard deviation of about 12,000 spectral variables is close to 0, so 8,000-12,000 spectral variables can be removed. This embodiment removes 10,000 spectral variables. Figure 2 (a) and Figure 2 (b) show the original spectrum and the denoised spectrum, respectively. It can be seen that the number of variables in a soil LIBS spectrum has been reduced from 26,864 to 16,864, and more than 1 / 3 of the spectral variables have been removed, but the overall spectral profile has hardly changed, and the spectral variables with higher intensity are retained. This method not only removes the noise spectrum, but also retains the information spectrum, improves the information concentration, and is conducive to improving the modeling accuracy.

[0061] A total of 202 soil samples were collected from Xiangshan County, Ningbo City, Zhejiang Province, and Jiangxi Province. The collected soil samples were air-dried, ground, and sieved (2 mm) to a fine, uniform powder. Each soil sample was divided equally into two portions: one portion was used to determine the true values ​​of the soil mineral nutrients total nitrogen (N), total potassium (K), and exchangeable calcium (Ca), using the corresponding national standard methods. The other portion was used for laser-induced breakdown spectroscopy (LIBS) analysis.

[0062] Furthermore, the process of processing the soil sample using laser induced breakdown spectroscopy technology includes:

[0063] S300: Pressing each soil sample into a sheet to obtain a sheet soil sample.

[0064] S301: Using laser induced breakdown spectroscopy technology, collect spectral information at preset positions of each flaky soil sample to obtain a predetermined amount of spectral information; wherein the predetermined amount is equal to the number of preset positions in each flaky soil sample multiplied by the number of times spectral information is collected at each preset position.

[0065] S302: A predetermined number of spectral information are averaged to obtain spectral information of each soil sample.

[0066] As an optional implementation, the process of processing the soil sample using laser-induced breakdown spectroscopy technology can be specifically as follows:

[0067] A 0.5g sample of experimental soil powder was pressed into a tablet. Notably, no chemical treatment was performed in this step. LIBS was used to ablate 16 different locations on the tablet (15-20 locations are acceptable, depending on the size of the soil tablet). The same location was ablated five times (4-6 times are acceptable). Each ablation acquired a spectrum, taking 1 second, for a total of 80 spectra, taking 80 seconds. The average of these 80 spectra represented the LIBS spectrum of the experimental soil sample.

[0068] In another exemplary embodiment of the present application, the training process of the multivariate weighted network model includes:

[0069] S400: Testing a plurality of experimental soil samples using a standard testing method to obtain test results of the plurality of experimental soil samples; wherein the test results include the actual content of the plurality of elements in each soil sample.

[0070] S401: Divide the spectral intensity data corresponding to the predetermined spectral variables of several experimental soil samples and the detection results into a modeling set, a verification set and a prediction set.

[0071] S402: Training the multivariable weighted network model according to the modeling set to obtain a preliminarily trained multivariable weighted network model.

[0072] S403: Validating the preliminarily trained multivariate weighted network model according to the validation set to determine whether the preliminarily trained multivariate weighted network model is over-optimized, and obtaining a first result.

[0073] S404: If the first result is yes, return to step S402.

[0074] S405: If the first result is no, the preliminarily trained multivariate weighted network model is tested according to the prediction set to determine whether the prediction ability of the preliminarily trained multivariate weighted network model meets the requirements, and obtain a second result.

[0075] S406: If the second result is no, return to step S402.

[0076] S407: If the second result is yes, the preliminarily trained multivariate weighted network model is determined as the trained multivariate weighted network model.

[0077] Furthermore, the standard detection methods include atomic absorption spectroscopy, inductively coupled plasma atomic emission spectroscopy and inductively coupled plasma mass spectrometry.

[0078] As an optional implementation, step S401 includes:

[0079] 202 soil samples were compressed and randomly divided into a calibration set, a validation set, and a prediction set in a ratio of approximately 3:1:1. The three sets correspond to 124, 39, and 39 soil samples, respectively. Because each soil sample corresponds to one LIBS spectrum, the three sets also correspond to 124, 39, and 39 spectra, respectively. The calibration and validation sets are used to train the model, and the performance of the calibration set reflects the model's training capability. The prediction set is used to test the model, and its performance reflects the model's predictive ability.

[0080] In another exemplary embodiment of the present application, the plurality of elements include nitrogen, potassium and calcium.

[0081] Further, if Figure 3 As shown, the output of the fully connected network includes 15 neurons; among them, the first to third neurons are used to output the preliminary contents of the first group of nitrogen, potassium and calcium elements; the fourth to sixth neurons are used to output the preliminary contents of the second group of nitrogen, potassium and calcium elements; the seventh to ninth neurons are used to output the preliminary contents of the third group of nitrogen, potassium and calcium elements; the tenth to twelfth neurons are used to output the preliminary contents of the fourth group of nitrogen, potassium and calcium elements; the thirteenth to fifteenth neurons are used to output the preliminary contents of the fifth group of nitrogen, potassium and calcium elements.

[0082] The weighted multivariate output layer performs weighted calculations on the five groups of preliminary contents of nitrogen, potassium, and calcium output by the parallel fully connected layer, respectively, to obtain the contents of nitrogen, potassium, and calcium in the soil sample to be tested.

[0083] That is to say, in order to realize a kind of model and predict multiple soil mineral nutrients at the same time, the present embodiment constructs a kind of LIBS multivariate detection network, i.e. multivariable output weighting-network (MOW-Net). Wherein, MOW-Net has four layers, the first layer is the soil sample LIBS spectral information input layer, which is used to input soil LIBS spectral information (i.e. spectral intensity). The soil sample LIBS spectrum after removing the noise spectrum has 16864 spectral variables, and neurons correspond to spectral variables one by one, so this layer is provided with 16864 neurons. The second layer is the weighted conversion layer, and the input data is the neuron value output by the first layer. The neuron value of the first layer is multiplied by the weight coefficient α to obtain the weighted neuron value. The weight coefficient α is obtained by network self-learning. The network learning method adopts the gradient descent algorithm, and the output data is the weighted neuron value, so the second layer also has 16864 neurons. The third layer is a parallel fully connected layer. Its input data is the neuron values ​​output by the second layer. The 16,864 neuron values ​​are outputted by the fully connected network into 15 neurons. These 15 neurons are divided into five groups, each of which outputs preliminary predicted N, K, and Ca levels. This is equivalent to performing five parallel calculations, outputting a total of 15 neuron values. The fourth layer is a weighted multivariate output layer. Its input data is the 15 neuron values ​​from the third layer. For example, the five parallel N element predictions are weighted and summed to obtain the final predicted N content. The weight W is obtained in the same way as the weight coefficient α. The output of this layer is three neurons, representing the predicted N, K, and Ca levels, respectively.

[0084] In an exemplary embodiment, Figure 4 As shown, a multivariate detection system for soil mineral nutrient elements is provided, which includes a detection system and a LIBS device; wherein,

[0085] The LIBS device is a device that obtains spectral information of a soil sample to be tested based on laser-induced breakdown spectroscopy technology; the LIBS device includes a pulsed laser and a spectrometer.

[0086] The detection system includes a spectrum information acquisition module M1, a spectrum intensity data extraction module M2 and a soil mineral nutrient element prediction module M3; wherein,

[0087] The spectral information acquisition module M1 is connected to the LIBS device and is used to obtain spectral information of the soil sample to be tested; the spectral information includes several spectral variables and spectral intensity data corresponding to each spectral variable.

[0088] The spectral intensity data extraction module M2 is used to extract spectral intensity data corresponding to predetermined spectral variables based on the spectral information of the soil sample to be tested.

[0089] The soil mineral nutrient element prediction module M3 is used to input the spectral intensity data corresponding to the predetermined spectral variables into the trained multivariate weighted network model to obtain the content of multiple elements in the soil sample to be tested.

[0090] The multivariate weighted network model is the same as that in the previous embodiment.

[0091] Furthermore, the spectrometer has a delay time of 3.5-4.5 μs and a gate width time of 15-17 μs for sampling the spectral information of the soil sample to be tested. Delay time and gate width are important parameters for LIBS detection.

[0092] This application uses the following cases to illustrate the soil mineral nutrient element detection effect of the method proposed in this application:

[0093] Determination coefficient (R 2 ) represents the degree of explanation of the independent variable (LIBS spectrum) on the dependent variable (element concentration), R 2 It is usually between 0 and 1. The closer to 1, the better the effect. The root mean square error (RMSE) represents the numerical deviation between the true value of the element concentration and the model predicted value. The smaller the RMSE, the better the effect. 2 and RMSE to evaluate the model performance. 2 Use R 2 C and R 2 P , and RMSE is represented by RMSEC and RMSEP respectively.

[0094] The partial least squares regression (PLSR) model is a classic LIBS quantitative detection model and can be used as a comparison model. Table 1 shows the results of simultaneous LIBS detection of soil mineral nutrients based on the PLSR and MOW-Net models. From the modeling results, for N, K, and Ca, the MOW-Net-based R 2 CThe RMSEC based on MOW-Net is lower than that of PLSR, indicating that the learning ability of MOW-Net model is better than that of PLSR. From the prediction set effect, the RMSEP of N, K and Ca prediction based on PLSR are 3.1%, 2.966g / kg and 951.522mg / kg respectively, and the corresponding values ​​based on MOW-Net are 2.0%, 2.333g / kg and 873.780mg / kg respectively. The RMSEP based on MOW-Net is lower than that of PLSR, indicating that the prediction accuracy of MOW-Net is better than that of PLSR, which illustrates the rationality and effectiveness of the MOW-Net model design in this patent.

[0095] Table 1: LIBS simultaneous detection results of soil mineral nutrients based on PLSR and MOW-Net models

[0096]

[0097] Figure 5(a)-Figure 5(c) The MOW-Net-based LIBS synchronous detection results of soil mineral nutrients are intuitively displayed, using the predicted set data. The black straight line is y = x, and the black dotted line is the linear fit line between the actual value and the predicted value of the element. The formula and R of the dotted line are marked in the lower right corner of the figure. 2 Overall, the predicted points of N, K and Ca elements are all near the black straight line, indicating a good prediction trend of the MOW-Net model. Separately, the linear fitting lines of N, K and Ca elements are getting closer and closer to the black straight line, with slopes of 0.76, 0.80 and 0.96 respectively, and R 2 They are 0.75, 0.83 and 0.94 respectively, indicating that the MOW-Net model can accurately predict the content of soil mineral nutrients, and the prediction effect of the three elements is ranked as Ca>K>N.

[0098] Compared with the prior art, the present invention has the following advantages:

[0099] (1) It can realize rapid and pollution-free detection of soil mineral nutrients.

[0100] (2) It can achieve effective denoising of soil LIBS spectra.

[0101] (3) A LIBS multivariate detection network MOW-Net is provided. Compared with the classic PLSR method, it can effectively improve the detection accuracy and realize the simultaneous and accurate LIBS detection of soil mineral nutrients. The detection R 2 P They were 0.751, 0.826 and 0.935 respectively, and the RMSEP were 2.0%, 2.333 g / kg and 873.780 mg / kg respectively.

[0102] The present application also provides an application scenario, which applies the above-mentioned multivariate detection method for soil mineral nutrients. Specifically, the multivariate detection method for soil mineral nutrients provided in this embodiment can be applied in the soil mineral nutrient detection scenario.

[0103] In an exemplary embodiment, a computer device is provided. The computer device may be a server or a terminal. The internal structure diagram thereof may be as follows: Figure 6 As shown. The computer device includes a processor, a memory, an input / output interface (Input / Output, abbreviated as I / O) and a communication interface. The processor, memory and input / output interface are connected through a system bus, and the communication interface is connected to the system bus through the input / output interface. The processor of the computer device is used to provide computing and control capabilities. The memory of the computer device includes a non-volatile storage medium and an internal memory. The non-volatile storage medium stores an operating system, a computer program and a database. The internal memory provides an environment for the operation of the operating system and the computer program in the non-volatile storage medium. The database of the computer device is used for spectral information of the soil sample to be tested. The input / output interface of the computer device is used to exchange information between the processor and an external device. The communication interface of the computer device is used to communicate with an external terminal through a network connection. When the computer program is executed by the processor, a multivariate detection method of soil mineral nutrients is implemented.

[0104] Those skilled in the art will understand that Figure 6 The structure shown in the figure is only a block diagram of a part of the structure related to the solution of the present application, and does not constitute a limitation on the computer device to which the solution of the present application is applied. The specific computer device may include more or fewer components than shown in the figure, or combine certain components, or have a different component arrangement.

[0105] In an exemplary embodiment, a computer device is further provided, including a memory and a processor. The memory stores a computer program, and the processor implements the steps in the above method embodiments when executing the computer program.

[0106] Those skilled in the art will understand that all or part of the processes in the above-mentioned embodiment methods can be implemented by instructing the relevant hardware through a computer program, and the computer program can be stored in a non-volatile computer-readable storage medium. When the computer program is executed, it can include the processes of the embodiments of the above-mentioned methods. Among them, any reference to memory, database or other media used in the embodiments provided in this application may include at least one of non-volatile and volatile memory. Non-volatile memory may include read-only memory (ROM), magnetic tape, floppy disk, flash memory, optical memory, high-density embedded non-volatile memory, resistive random access memory (ReRAM), magnetic random access memory (MRAM), ferroelectric random access memory (FRAM), phase change memory (PCM), graphene memory, etc. Volatile memory may include random access memory (RAM) or external cache memory, etc. By way of illustration and not limitation, RAM may be in various forms, such as static random access memory (SRAM) or dynamic random access memory (DRAM).

[0107] The databases involved in the various embodiments provided herein may include at least one of a relational database and a non-relational database. Non-relational databases may include, but are not limited to, distributed databases based on blockchains. The processors involved in the various embodiments provided herein may include, but are not limited to, general-purpose processors, central processing units, graphics processing units, digital signal processors, programmable logic units, data processing logic units based on quantum computing, and the like.

[0108] The technical features of the above embodiments can be combined arbitrarily. To make the description concise, not all possible combinations of the technical features in the above embodiments are described. However, as long as there is no contradiction in the combination of these technical features, they should be considered to be within the scope of this specification.

[0109] This document uses specific examples to illustrate the principles and implementation methods of this application. The description of the above examples is only intended to help understand the method and core concept of this application. At the same time, for those skilled in the art, based on the concept of this application, there may be changes in the specific implementation methods and application scope. In summary, the content of this specification should not be understood as limiting this application.

Claims

1. A multivariate detection method for soil mineral nutrients, characterized in that: The multivariate detection method of soil mineral nutrient elements includes: Acquire spectral information of the soil sample to be tested; the spectral information includes a plurality of spectral variables and spectral intensity data corresponding to each spectral variable; the spectral information is obtained by processing the soil sample to be tested using laser induced breakdown spectroscopy technology; Extracting spectral intensity data corresponding to predetermined spectral variables based on spectral information of the soil sample to be tested; Inputting the spectral intensity data corresponding to the predetermined spectral variables into the trained multivariate weighted network model to obtain the contents of multiple elements in the soil sample to be tested; The multivariate weighted network model includes an input layer, a weighted conversion layer, a parallel fully connected layer and a weighted multivariate output layer; The input layer is used to receive spectral intensity data corresponding to the predetermined spectral variable; the number of neurons in the input layer is equal to the number of the predetermined spectral variables; The weighted conversion layer is used to weight the values ​​output by each neuron in the input layer to obtain corresponding weighted values; The parallel fully connected layer is used to calculate the preliminary content of multiple elements based on all weighted values ​​output by the weighted transformation layer; the parallel fully connected layer includes a fully connected network, and the output of the fully connected network includes a first number of neurons; wherein the first number of neurons is used to output the preliminary predicted content of the first group of multiple elements; The weighted multivariate output layer is used to perform weighted calculations on each element based on the preliminary predicted contents of the first group of multiple elements output by the parallel fully connected layer, so as to obtain the predicted contents of the multiple elements in the soil sample to be tested.

2. The multivariate detection method for soil mineral nutrients according to claim 1, characterized in that: The process of determining the predetermined spectral variables includes: Calculating the standard deviation of the spectral intensity corresponding to each spectral variable based on the spectral information of a plurality of experimental soil samples, wherein each soil sample contains each of the spectral variables; The spectral variable whose standard deviation is greater than a preset standard deviation threshold is determined as a predetermined spectral variable.

3. The multivariate detection method for soil mineral nutrients according to claim 1, characterized in that: The process of processing the soil sample using laser-induced breakdown spectroscopy technology includes: Pressing each soil sample to be tested into a sheet to obtain a sheet soil sample; Using laser-induced breakdown spectroscopy, spectral information is collected at a preset position of each flaky soil sample to obtain a predetermined amount of spectral information; wherein the predetermined amount is equal to the number of preset positions in each flaky soil sample multiplied by the number of times spectral information is collected at each preset position; The spectral information of a predetermined number of samples is averaged to obtain the spectral information of each soil sample.

4. The multivariate detection method for soil mineral nutrients according to claim 2, characterized in that: The training process of the multivariate weighted network model includes: Testing a plurality of experimental soil samples using a standard testing method to obtain test results for the plurality of experimental soil samples; wherein the test results include the actual content of the plurality of elements in each soil sample; Dividing the spectral intensity data corresponding to the predetermined spectral variables of a plurality of experimental soil samples and the test results into a modeling set, a validation set, and a prediction set; The multivariable weighted network model is trained according to the modeling set to obtain a preliminarily trained multivariable weighted network model; Verifying the preliminarily trained multivariate weighted network model based on the validation set to determine whether the preliminarily trained multivariate weighted network model is over-optimized, thereby obtaining a first result; If the first result is yes, return to step "training the multivariate weighted network model according to the modeling set to obtain a preliminarily trained multivariate weighted network model"; If the first result is no, the preliminarily trained multivariate weighted network model is tested according to the prediction set to determine whether the prediction ability of the preliminarily trained multivariate weighted network model meets the requirements, thereby obtaining a second result; If the second result is no, return to step "training the multivariate weighted network model according to the modeling set to obtain a preliminarily trained multivariate weighted network model"; If the second result is yes, the preliminarily trained multivariate weighted network model is determined as the trained multivariate weighted network model.

5. The multivariate detection method of soil mineral nutrients according to claim 4, characterized in that: The standard detection methods include atomic absorption spectroscopy, inductively coupled plasma atomic emission spectroscopy, and inductively coupled plasma mass spectrometry.

6. The multivariate detection method for soil mineral nutrients according to claim 1, characterized in that: The plurality of elements include nitrogen, potassium and calcium.

7. The multivariate detection method for soil mineral nutrients according to claim 6, characterized in that: The output of the fully connected network includes 15 neurons; among them, the first to third neurons are used to output the preliminary contents of the first group of nitrogen, potassium and calcium elements; the fourth to sixth neurons are used to output the preliminary contents of the second group of nitrogen, potassium and calcium elements; the seventh to ninth neurons are used to output the preliminary contents of the third group of nitrogen, potassium and calcium elements; the tenth to twelfth neurons are used to output the preliminary contents of the fourth group of nitrogen, potassium and calcium elements; and the thirteenth to fifteenth neurons are used to output the preliminary predicted contents of the fifth group of nitrogen, potassium and calcium elements. The weighted multivariate output layer performs weighted calculations on the five groups of preliminary predicted contents of nitrogen, potassium, and calcium output by the parallel fully connected layer, respectively, to obtain the predicted contents of nitrogen, potassium, and calcium in the soil sample to be tested.

8. A multivariate detection system for soil mineral nutrients, characterized in that: The multivariate detection system for soil mineral nutrient elements includes a detection system and a LIBS device; wherein, The LIBS device is a device that obtains spectral information of a soil sample to be tested based on laser-induced breakdown spectroscopy technology; the LIBS device includes a pulsed laser and a spectrometer; The detection system includes a spectrum information acquisition module, a spectrum intensity data extraction module and a soil mineral nutrient element prediction module; wherein, A spectral information acquisition module, connected to the LIBS device, for acquiring spectral information of the soil sample to be tested; the spectral information includes a plurality of spectral variables and spectral intensity data corresponding to each spectral variable; A spectral intensity data extraction module is used to extract spectral intensity data corresponding to predetermined spectral variables based on spectral information of the soil sample to be tested; A soil mineral nutrient element prediction module is used to input the spectral intensity data corresponding to the predetermined spectral variables into a trained multivariate weighted network model to obtain the content of multiple elements in the soil sample to be tested; The multivariate weighted network model includes an input layer, a weighted conversion layer, a parallel fully connected layer and a weighted multivariate output layer; The input layer is used to receive spectral intensity data corresponding to the predetermined spectral variable; the number of neurons in the input layer is equal to the number of the predetermined spectral variables; The weighted conversion layer is used to weight the values ​​output by each neuron in the input layer to obtain corresponding weighted values; The parallel fully connected layer is used to calculate the preliminary content of multiple elements based on all weighted values ​​output by the weighted transformation layer; the parallel fully connected layer includes a fully connected network, and the output of the fully connected network includes a first number of neurons; wherein the first number of neurons is used to output the preliminary predicted content of the first group of multiple elements; The weighted multivariate output layer is used to perform weighted calculations on each element based on the preliminary predicted contents of the first group of multiple elements output by the parallel fully connected layer, so as to obtain the predicted contents of the multiple elements in the soil sample to be tested.

9. The multivariate detection system for soil mineral nutrients according to claim 8, characterized in that: The delay time of the spectrometer in sampling the spectral information of the soil sample to be tested is 3.5-4.5 μs, and the gate width time is 15-17 μs.

10. A computer device 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 the multivariate detection method for soil mineral nutrients according to any one of claims 1 to 7.

Images