Hyperspectral prediction method and system for soil moisture

Abstract

The invention belongs to the technical field of soil moisture prediction, and discloses a soil moisture hyperspectral prediction method and system.The influence of other factors on an experimental result is eliminated through a variable control method, the same soil sample is selected (the influence of soil parent material, texture and organic matter content is eliminated), salt solutions of different gradients are added, and the soil moisture hyperspectral prediction result is obtained. Analyzing water spectral characteristics of the coastal saline soil through a simulation experiment; on the basis of characteristic analysis and moisture characteristic analysis of measured soil moisture hyperspectral data, moisture and spectral characteristics are determined, a soil moisture spectrum prediction model covering all salinization degrees and moisture states of a coastal region is established, and variables selected by the model are clear, simple, convenient, easy to operate and high in universality. Realization results show that the method contributes to deepening understanding of spectral characteristics of specific soil moisture such as coastal saline soil, provides a new thought and method for rapid and accurate measurement of soil moisture, and has important significance in promoting agricultural sustainable development and efficient utilization of water resources.

Description

Technical Field

[0001] The present invention belongs to the technical field of soil moisture prediction, and particularly relates to a hyperspectral prediction method and system for soil moisture. Background Art

[0002] Soil moisture is an important factor for plant growth and development, and plays a very important role in soil productivity. The quality of soil moisture conditions plays a decisive role in soil formation and soil development, and also greatly affects soil properties. The amount of soil water content cannot be underestimated in its influence on factors such as soil salinity, soil temperature, and soil fertility. Especially for coastal saline-alkali areas, the water-salt movement plays a crucial role in aspects such as agricultural production and hydrological changes. Therefore, timely understanding of the content and distribution of soil moisture is of great significance for the efficient utilization of soil water resources and the sustainable development of agriculture.

[0003] Traditional methods for measuring soil water content, such as the drying method, tensiometer method, neutron method, and time domain reflectometry (TDR) and frequency domain reflectometry (FDR) methods, etc., although they can provide high-precision measurement results, are cumbersome to operate and time-consuming, and it is difficult to meet the need for large-area and rapid acquisition of soil moisture information. With the rapid development of remote sensing technology, it has become possible to monitor soil moisture by remote sensing means, providing a new way for rapid large-scale measurement of soil moisture.

[0004] Since the 1970s, foreign scholars have begun to combine ground test results with airborne microwave remote sensing data to study the relationship between soil moisture and parameters such as brightness temperature, initiating the research on microwave remote sensing monitoring of soil water content. Subsequently, with the launch and wide application of meteorological satellites, researchers have further deepened the research on using remote sensing technology to monitor soil water content, including the application of methods such as thermal inertia method and crop water stress index, as well as the exploration of data such as modis, and certain research results have been achieved. However, due to the few bands and low spectral resolution of conventional multispectral remote sensing images, there are large errors in regional or plot-level soil moisture monitoring. In contrast, hyperspectral remote sensing has a higher spectral resolution and band range, can form a continuous spectral curve, and can achieve dynamic, rapid, and real-time monitoring of soil moisture by detecting surface soil components. Domestic and foreign scholars have conducted a large number of studies on hyperspectral characteristics of soil moisture and selection of sensitive bands, and found that soil moisture has obvious absorption bands in specific bands (such as 1400nm, 1900nm, 2200nm, etc.), and these bands are regarded as sensitive bands of soil moisture.

[0005] In the aspect of hyperspectral remote sensing inversion models for soil moisture, researchers mostly use methods such as establishing multiple linear regression equations based on sensitive bands to predict soil moisture. However, spectral prediction methods for soil moisture under coastal saline-alkali conditions are rare. The basis of hyperspectral remote sensing inversion of soil moisture is to obtain the spectral characteristics of soil moisture. Soil spectra are comprehensively affected by various soil properties (soil parent material, soil type, soil organic matter content, soil moisture content). There are differences in moisture spectral characteristics under different soil properties. How the soil spectra of coastal saline soils are affected by moisture remains unknown. Even though some people such as Li Chen have carried out a small amount of research on the moisture spectral characteristics of coastal saline soils, because a large number of soil samples of different types or properties are used, the influence of factors such as organic matter content and soil texture cannot be eliminated, and the external conditions cannot be unified. Therefore, the moisture spectral characteristics of coastal saline soils have not been systematically elucidated. Especially, the samples used in the currently established soil moisture spectral inversion models cannot fully cover all moisture conditions and cannot fully represent coastal saline soil conditions. Therefore, it is difficult to have universality and promotion.

[0006] Through the above analysis, the problems and defects existing in the prior art are as follows:

[0007] (1) Traditional methods for measuring soil water content are cumbersome to operate and time-consuming, and it is difficult to meet the need for obtaining soil moisture information over a large area and quickly.

[0008] (2) When using multispectral remote sensing images to predict soil moisture, due to the few bands and low spectral resolution of multispectral remote sensing data, there are large errors in soil moisture monitoring at the regional or plot level.

[0009] (3) In the aspect of hyperspectral remote sensing inversion models for soil moisture, there is a lack of spectral prediction methods for soil moisture under coastal saline-alkali conditions, and the soil spectral characteristics of coastal saline soils affected by moisture are not clear.

[0010] (4) The samples used in the currently established soil moisture spectral inversion models cannot fully cover all moisture conditions and cannot fully represent coastal saline soil conditions. Therefore, it is difficult to have universality and promotion. Summary of the Invention

[0011] To overcome the problems existing in the related technologies, the disclosed embodiments of the present invention provide a hyperspectral prediction method and system for soil moisture, especially a hyperspectral prediction method and system for soil moisture based on data simulating water-salt gradient changes.

[0012] The technical solutions are as follows:

[0013] The present invention is implemented as follows. The hyperspectral prediction method for soil moisture specifically includes the following steps:

[0014] S1, Soil sample collection and preparation: Samples of the surface soil under different land use types in the area to be analyzed are collected, and soil samples with different gradient salt contents are simulated by adding chlorides at different concentrations.

[0015] S2, Soil moisture measurement: By measuring the total weights of the petri dishes containing soil samples and the empty aluminum boxes respectively, and the total weights of the petri dishes and aluminum boxes with soil samples and salt solutions added respectively, the moisture content of the soil samples is determined.

[0016] S3, Spectral data collection: Using a portable spectrometer, during the process of the soil sample changing from completely wet to air-dried, the spectral reflectance of soil moisture is measured every day at the same time period.

[0017] S4, Spectral data preprocessing: The 9-point weighted moving average method is used to smooth the noise of the measured soil moisture spectral curve to determine the reflectance value of the sample after smoothing.

[0018] S5, Spectral data resampling: In the ENVI software, the hyperspectral data collected by the portable spectrometer and preprocessed is resampled by 10 nm by loading the central wavelength and setting the full width at half maximum data parameters. The average value of all spectral reflectances within every 10 nm range is taken as the reflectance of the central wavelength within this range.

[0019] S6, Soil moisture characteristic analysis: Analyze the change characteristics of soil moisture content, the spectral reflectance characteristics of soil moisture under different conditions, and the spectral absorption characteristics of soil moisture respectively to clarify the moisture and spectral characteristics.

[0020] S7, Establishment of soil moisture hyperspectral prediction model: Use the full band to perform stepwise regression to extract 5 reliable bands and establish a least squares regression model; and use the full band data to extract the spectral principal components to establish a principal component-based partial least squares regression model, thereby realizing the hyperspectral prediction of soil moisture.

[0021] In step S1, the soil sample collection and preparation include:

[0022] Samples of the surface soil under different land use types in the area to be analyzed are all collected; the collected surface soil is placed in a plastic bag and left to dry in the laboratory; after air-drying, the soil samples are picked out of debris and plant residues and ground finely so that the soil samples pass through a 10-mesh / 2-mm aperture sieve for standby.

[0023] The ground soil samples are grouped and then put into aluminum boxes respectively. The surface is gently leveled with a ruler and placed in a petri dish; each group of soil is moistened with NaCl solutions with different concentrations respectively to simulate soils with different salt concentrations in coastal areas; the experiment is set up with 2 replicates.

[0024] In step S2, the soil moisture measurement includes:

[0025] Label the petri dishes, weigh the total weight of the petri dish containing the soil sample and the empty aluminum box and record it as Z; place the soil samples in the aluminum boxes respectively and weigh the total weight of the petri dish, the aluminum box and the soil samples again, and record it as Z1; add the salt solution with the corresponding concentration to the petri dish, start weighing after the soil samples are completely wetted, and weigh again at the same time every day hereafter until the weight of the soil samples does not change. After the water is completely evaporated, the weight after the salt solution wets the soil is recorded as Z2; the moisture content calculation formula for the soil sample corresponding to each aluminum box every day is:

[0026] Soil moisture content (%) = (Z2 - Z1) / (Z2 - Z) * 100%.

[0027] In step S3, the spectral data acquisition includes:

[0028] Use an Avantes portable spectrometer to collect hyperspectral data. The test range of the spectrometer is 350 - 1100 nm; before the test, whiteboard and blackboard corrections are performed to remove the influence of dark current.

[0029] The sensor probe of the spectrometer is vertically downward, and the distance between the probe and the soil sample is 12 cm; during the process of the soil sample changing from completely wet to air-dried, the soil moisture spectral reflectance is measured at the same time period every day when weighing, until the soil sample is air-dried, and spectral and water content data are obtained.

[0030] In step S4, the spectral data preprocessing includes:

[0031] Use the 9-point weighted moving average method to smooth the noise of the spectral curve; the spectral curve gives the reflectance values of N samples. The value range of the i-th point is 4 points before and after this point, including the i-th point. The weighted average calculation is performed on this point, and the new value is the smoothed value of the i-th point. The calculation process is as follows:

[0032] R i ' = 0.04R i-4 + 0.08R i-3 + 0.12R i-2 + 0.16R i-1 + 0.20R i + 0.16R i+1 + 0.12R i+2 + 0.08R i+3

[0033] + 0.04R i+4

[0034] Wherein, R i' is the smoothed sample reflectance value, and i-4, i-3, i-2, i-1, i, i+1, i+2, i+3, i+4 respectively represent 9 measurement points centered on i.

[0035] In step S5, the spectral data resampling includes:

[0036] In ENVI software, click Spectral - Spectral Libraries - Spectral LibrariesBuiler, select First Input Spectum - Option - From PlotWindows to open the hyperspectral data collected by Avantes portable spectrometer and pre - processed, and create a spectral library to be resampled. Subsequently, open Spectral - Spectral Libraries - Spectral Libraries Resampling in the main menu, select the created spectral library to be resampled, load the central wavelength and set the full - width at half - maximum data in Spectral Resampling Parameters to determine the resampling spectral range and resolution. The wavelength unit is selected as nanometers, and finally the resampled spectral data is obtained and saved. After such operations, the average value of all spectral reflectances within every 10 nm range is taken as the reflectance of the central wavelength within this range, that is, the 10 - nm resampling of the spectrum is completed.

[0037] In step S6, during the analysis of the spectral reflectance characteristics of soil moisture under different conditions, the spectral bands of 350 - 380 nm and 1050 - 1100 nm are excluded, and the continuum removal processing is performed on the resampled spectral data using the envelope method.

[0038] In step S7, the variables used in the least - squares regression model are the spectral reflectances of the measured bands of 895 nm, 675 nm, 740 nm, 860 nm, and 680 nm;

[0039] Y = 0.569 - 0.031*V 1 +0.203*V 2 -0.135*V 3 +0.069*V 4 -0.118*V 5 , (R 2 is 0.891)

[0040] The use of all - band data to extract spectral principal components and establish a principal - component - based partial least - squares regression model includes:

[0041] X = TP T +E

[0042] Y = UQ T + F

[0043] Wherein, Y is an n×p response matrix; X is an n×m modeling matrix; T is the projection of X and is the factor matrix; U is the projection of Y; P and Q are orthogonal loading matrices; E and F represent errors; based on the estimated factors T and U and the loading matrices P and Q, a linear model between Y and X is established through the PLSR model;

[0044] Y = Xb + e

[0045] Wherein, b is the coefficient of the PLSR model and e is the error vector.

[0046] Another object of the present invention is to provide a soil moisture hyperspectral prediction system, which is used to regulate the above-mentioned soil moisture hyperspectral prediction method, and the system includes:

[0047] A soil sample collection and preparation module, which is used to sample the surface soil under different land use types in the area to be analyzed, and simulate soil samples with different gradient salt contents by adding chlorides with different concentrations;

[0048] A soil moisture measurement module, which is used to determine the moisture content of the soil sample by measuring the total weight of the culture dish containing the soil sample and the empty aluminum box, and the total weight of the culture dish and the aluminum box after adding the soil sample and the salt solution respectively;

[0049] A spectral data collection module, which is used to use a portable spectrometer to measure the spectral reflectance of soil moisture every day at the same time period during the process of the soil sample changing from completely wet to air-dried;

[0050] A spectral data preprocessing module, which is used to perform smoothing noise processing on the measured soil moisture spectral curve by using a 9-point weighted moving average method to determine the smoothed sample reflectance value;

[0051] A spectral data resampling module, in the ENVI software, resamples the hyperspectral data collected by the portable spectrometer and preprocessed by data, and resamples the spectral data by loading the central wavelength and setting the full width at half maximum data parameter by 10 nm, and takes the average value of all spectral reflectances within every 10 nm range as the reflectance of the central wavelength within this range;

[0052] A soil moisture characteristic analysis module, which is used to analyze the change characteristics of soil moisture content, the spectral reflection characteristics of soil moisture under different conditions, and the spectral absorption characteristics of soil moisture to clarify the moisture and spectral characteristics;

[0053] The soil moisture hyperspectral prediction model establishment module is used to perform stepwise regression using the full band to extract 5 reliable bands and establish a least squares regression model; and is used to extract the spectral principal components using the full band data to establish a principal component-based partial least squares regression model, thereby realizing the hyperspectral prediction of soil moisture.

[0054] Combining all the above technical solutions, the advantages and positive effects of the present invention are as follows:

[0055] The present invention takes the saline soil in a certain area as the analysis object, excludes the influence of other factors on the experimental results by the method of controlling variables, selects the same soil sample, and analyzes the moisture spectral characteristics of coastal saline soil by adding salt solutions with different gradients; the results show that by analyzing the spectral differences of soils with different water contents, a spectral prediction model of soil water content suitable for coastal salinized areas is established, which not only helps to deepen the understanding of the moisture spectral characteristics of specific soils such as coastal saline soil, but also provides new ideas and methods for the rapid and accurate measurement of soil moisture, and is of great significance for promoting the sustainable development of agriculture and the efficient utilization of water resources.

[0056] The advantages of the present invention compared with the prior art further include:

[0057] The existing technologies rarely use hyperspectral data to invert and predict when establishing soil moisture models. The present invention uses hyperspectral technology to predict soil moisture, which has the advantages of rapidity and non-destruction.

[0058] Even though some people have carried out a small amount of analysis on establishing hyperspectral models to predict moisture at present, the range of soil moisture data they used is not wide enough to cover all field moisture conditions, so the established models do not have large-scale popularization.

[0059] At the same time, the currently established hyperspectral models do not fully consider the spectroscopic physical basis of soil moisture, while the hyperspectral model established by the present invention fully analyzes and considers the hyperspectral characteristics under different moisture and salt conditions. The selected spectral variables have clear physical meanings, and the spectral bands used are more comprehensive than the spectral bands published in the prior art. The established model has a physical basis and reliable results.

[0060] The present invention determines the best model by comparing three different types of models respectively, and the prediction accuracy of the model is higher than that of the current moisture spectral model.

[0061] The present invention has established a soil moisture hyperspectral monitoring model with high accuracy and simple method. The variables used in the model are measured hyperspectral data, the input variables are relatively few, the accuracy is high, and the operation is convenient.

[0062] The data used in the present invention is obtained by simulating all salinization degrees and moisture states in coastal areas. The established model covers a wide range of soil moisture and salt gradients, so it has good universality and accuracy. Compared with traditional soil water content measurement methods, this model can achieve rapid and non-contact soil moisture monitoring, greatly improving the monitoring efficiency and accuracy. BRIEF DESCRIPTION OF THE DRAWINGS

[0063] The accompanying drawings herein are incorporated into and constitute a part of this specification, showing embodiments consistent with the present disclosure, and are used together with the specification to explain the principles of the present disclosure;

[0064] Figure 1 is a flowchart of the soil moisture hyperspectral prediction method provided by an embodiment of the present invention;

[0065] Figure 2 is a spectral curve graph before and after smoothing provided by an embodiment of the present invention;

[0066] Figure 3 is a spectral curve graph under different moisture conditions in a 1 g / kg solution provided by an embodiment of the present invention;

[0067] Figure 4 is a spectral curve graph under different moisture conditions in a 6 g / kg solution provided by an embodiment of the present invention;

[0068] Figure 5 is a spectral curve graph under different moisture conditions in a 20 g / kg solution provided by an embodiment of the present invention;

[0069] Figure 6 is a spectral curve graph after continuum removal processing provided by an embodiment of the present invention. DETAILED DESCRIPTION OF THE EMBODIMENTS

[0070] To make the above objects, features, and advantages of the present invention more apparent and understandable, the following detailed description of the specific embodiments of the present invention is provided in conjunction with the accompanying drawings. Many specific details are set forth in the following description to facilitate a thorough understanding of the present invention. However, the present invention can be implemented in many other ways different from those described herein, and those skilled in the art can make similar improvements without departing from the spirit of the present invention. Therefore, the present invention is not limited by the specific embodiments disclosed below.

[0071] The innovation points of the soil moisture hyperspectral prediction method and system provided by the embodiments of the present invention are as follows:

[0072] The present invention excludes the influence of other factors on the experimental results by the method of controlling variables, selects the same soil sample (excluding the influence of soil parent material, texture, and organic matter content), and simulates and analyzes the moisture spectral characteristics of coastal saline soil by adding salt solutions with different gradients.

[0073] Based on the characteristic analysis of hyperspectral data for measuring soil moisture and the moisture characteristic analysis, the present invention has clarified the moisture and spectral characteristics, and established a soil moisture spectral prediction model covering all salinization degrees and moisture states in coastal areas: Y = 0.569 - 0.031*V 1 + 0.203*V 2 - 0.135*V 3 + 0.069*V 4 - 0.118*V 5 . The variables used in this model are the spectral reflectances of the measured bands at 895 nm, 675 nm, 740 nm, 860 nm, and 680 nm. The selected variables are clear, simple, easy to operate, and have high universality.

[0074] The present invention uses full-band data, extracts the spectral principal components, and establishes a partial least squares regression model based on the spectral principal components. The accuracy R of this model 2 is as high as 0.96, and it can achieve high-precision prediction of soil moisture.

[0075] Example 1, as Figure 1 shown, the soil moisture hyperspectral prediction method provided by the embodiment of the present invention specifically includes the following steps:

[0076] S1. Soil sample collection and preparation: Sample the surface soil under different land use types in the area to be analyzed, and simulate soil samples with different gradient salt contents by adding different concentrations of chlorides;

[0077] S2. Soil moisture measurement: Determine the moisture content of the soil sample by measuring the total weights of the culture dish containing the soil sample and the empty aluminum box, and the total weights of the culture dish and the aluminum box after adding the soil sample and the salt solution respectively;

[0078] S3. Spectral data collection: Use a portable spectrometer to measure the spectral reflectance of soil moisture at the same time period every day during the process of the soil sample from being completely wet to air-dried;

[0079] S4. Spectral data preprocessing: Use the 9-point weighted moving average method to smooth the noise of the measured soil moisture spectral curve, and determine the reflectance value of the sample after smoothing;

[0080] S5. Spectral data resampling: In the ENVI software, resample the hyperspectral data collected by the portable spectrometer and preprocessed by data, and resample the spectral data by 10 nm by loading the central wavelength and setting the full width at half maximum data parameter. Take the average value of all spectral reflectances within every 10 nm range as the reflectance of the central wavelength within this range;

[0081] S6. Soil moisture characteristic analysis: Analyze the variation characteristics of soil moisture content, the spectral reflectance characteristics of soil moisture under different conditions, and the spectral absorption characteristics of soil moisture respectively;

[0082] S7. Establishment of hyperspectral prediction model for soil moisture: Use the full band to perform stepwise regression to extract 5 reliable bands and establish a least squares regression model; and use the full band data to extract the spectral principal components to establish a principal component-based partial least squares regression model, thereby realizing the hyperspectral prediction of soil moisture.

[0083] As can be seen from the above embodiments, the present invention can improve the prediction accuracy of soil moisture and provide more accurate data support for fields such as agricultural production, water resource management, and environmental monitoring. Traditional soil moisture monitoring methods require field sampling, indoor drying, weighing and other links, with a long cycle and high costs. While using hyperspectral technology for soil moisture monitoring does not require sampling, and accurate soil moisture data can be obtained quickly in a few seconds. If this model is extended to hyperspectral satellite sensors, the distribution characteristics data of regional soil moisture can be obtained quickly and over a large area. For traditional soil moisture surveys, calculated at the county scale, generally about 400 soil sample points need to be surveyed to obtain soil moisture distribution data with higher accuracy, the survey cycle is approximately half a month, and the cost is approximately 50,000 yuan; while using the model method established by the present invention, after being extended to hyperspectral satellite sensors, the soil moisture distribution data of the entire county can be obtained in one day, saving 40,000 yuan in direct costs; and the accuracy is not lower than the data obtained by the traditional drying method, and the acquisition is very timely. If soil moisture data of a city is obtained, calculated according to 9 counties in a city, 360,000 yuan can be saved; if soil moisture data of a province is obtained, calculated according to 136 counties in Shandong Province, 5.44 million yuan can be saved at one time. For agricultural production, it is necessary to pay attention to the soil moisture status at all times, which requires obtaining soil moisture data at a relatively high frequency. If data is obtained at least twice a month, the present invention can save 130.56 million yuan in a year. If it can be applied to business, it can bring very promising prospects.

[0084] Furthermore, the present invention rarely uses hyperspectral data for soil moisture prediction models. Secondly, the data used in the present invention comes from simulated evaporation experiments, with multiple moisture and salt gradient levels set, and the model data represents various water-salt conditions in the field; therefore, compared with other research and inventions, the model established by the present invention has high universality. By obtaining hyperspectral data at different levels for moisture model establishment, the prediction accuracy and efficiency are greatly improved.

[0085] Furthermore, the present invention first addresses the issue of accuracy. Through the technical solution of the present invention, the accuracy of moisture prediction can be efficiently improved. Secondly, it has high universality. All along, the soil improvement and application in coastal saline-alkali areas require accurate knowledge of the water-salt status. This technical solution has established a prediction model covering all water-salt conditions existing in the wild by simulating all degrees of salinization and moisture states in coastal areas, showing good universality and having great application value for the effective monitoring of saline-alkali lands.

[0086] In previous studies, most used all spectral data for inversion. However, in this solution, through spectral data processing, a few spectral variables with high stability and high sensitivity were selected to participate in the modeling, which can well overcome the disadvantages of too many variables, too low efficiency, and poor stability in traditional models, and significantly improve the prediction accuracy of soil moisture.

[0087] Example 2. Preferably, the hyperspectral prediction method for soil moisture provided by the embodiments of the present invention specifically includes:

[0088] (1) Collection and preparation of soil samples.

[0089] All the samples used in the present invention were collected from the Yellow River Delta region, which is specifically located between 37°24′-38°10′N and 118°15′-119°19′E. When collecting soil, different land use types were fully considered, and surface soils under different land use types in Kenli County were sampled. The collected surface soils were put into plastic bags and left to dry in the laboratory. After air-drying, the soil samples were picked out of debris such as gravel and plant residues, and ground finely to pass through a 10-mesh / 2-mm aperture sieve for experimental use.

[0090] Since the distribution of soil salt content in the collected soil samples was not uniform enough, in order to control the influence of variable factors on the analysis results during the analysis process, samples with less salt content were selected from all soil samples for experiments to reduce the influence of the soil samples themselves on the experiments. Because the salts in coastal areas are mainly chlorides, the present invention simulated soil samples with different gradient salt contents by adding chlorides with different concentrations. The ground soil samples were divided into 2 groups, and each group was further divided into 8 parts, which were respectively put into 16 aluminum boxes with a diameter of 7 cm and a depth of 1.8 cm. The surface layer was gently leveled with a ruler and placed in a culture dish. Each group of soils was moistened with NaCl solutions with concentrations of 0 g / kg, 1 g / kg, 2 g / kg, 4 g / kg, 6 g / kg, 10 g / kg, 15 g / kg, and 20 g / kg respectively to simulate soils with different salt concentrations in coastal areas. The experiment was set up with 2 repeated trials.

[0091] (2) Measurement of soil moisture.

[0092] First, label the petri dishes with serial numbers, weigh the total weight of the petri dishes containing soil samples and empty aluminum boxes in the experiment and record it as Z; then, place the soil samples in the aluminum boxes respectively, weigh the total weight of the petri dishes, aluminum boxes and soil samples again, and record it as Z1; add the salt solution with the corresponding concentration to the petri dishes, start weighing after the soil samples are completely wetted, and weigh again at the same time every day thereafter until the weight of the soil samples hardly changes, at which time the moisture in them has completely evaporated. The weighing after the salt solution wets the soil is recorded as Z2, and the moisture content of the soil samples in the aluminum boxes every day can be calculated by the following formula:

[0093] Soil moisture content (%) = (Z2 - Z1) / (Z2 - Z) * 100%

[0094] (3) Spectral data acquisition.

[0095] For the acquisition of hyperspectral data, the Avantes portable spectrometer used in the present invention has a spectral range of 350 - 1100 nm. Before testing, whiteboard and blackboard corrections are carried out to remove the influence of dark current. The sensor probe of the spectrometer is 12 cm away from the soil sample and is perpendicular downward. During the process of the soil sample changing from completely wet to air-dried, the spectral reflectance of soil moisture is measured every day at the same time period (when weighing). Until the soil sample is air-dried, a total of 8 days of spectral and moisture content data are obtained, totaling 132.

[0096] (4) Spectral data preprocessing.

[0097] There are certain differences in the energy response among the bands of the spectrometer, and these small differences make the spectral curve always have some noises, small burrs, and non-smooth phenomena; in addition, due to the strong change in the reflectivity of the spectrometer itself and the low signal-to-noise ratio, etc., it is necessary to smooth the spectral curve. Therefore, the measured spectra are smoothed. The present invention uses the 9-point weighted moving average method to smooth the noise of the spectral curve.

[0098] The reflectance values of N samples are given in the spectral curve. At this time, the value range of the i-th point is 4 points before and after this point, including the i-th point. The weighted average calculation is carried out for this point, and the new value is the smoothed value of the i-th point. The specific calculation process is as follows:

[0099] R i ' = 0.04R i-4 + 0.08R i-3 + 0.12R i-2 + 0.16R i-1 + 0.20R i + 0.16R i+1 + 0.12R i+2 + 0.08R i+3

[0100] +0.04R i+4

[0101] Wherein, Ri' is the smoothed sample reflectivity value, and i - 4, i - 3, i - 2, i - 1, i, i + 1, i + 2, i + 3, i + 4 respectively represent 9 measurement points centered on i.

[0102] (5) Hyperspectral data resampling.

[0103] Hyperspectral data has many bands, a small sampling interval, and a large amount of information. At the same time, it also shows a serious autocorrelation phenomenon between bands, high information redundancy, which brings difficulties to the construction of spectral estimation models for soil components and data analysis. Existing technologies show that resampling spectral data can reduce the autocorrelation between spectral bands and greatly reduce the redundancy of spectral information. Therefore, the present invention resamples the spectral data at 10 nm. In the ENVI software, click Spectral - Spectral Libraries - Spectral Libraries Builer, select First Input Spectum - Option - From PlotWindows to open the hyperspectral data collected by the Avantes portable spectrometer and pre - processed, so as to create the spectral library to be resampled. Subsequently, open Spectral - Spectral Libraries - Spectral Libraries Resampling in the main menu, select the created spectral library to be resampled, load the central wavelength and set the full width at half maximum data in the Spectral Resampling Parameters to determine the resampling spectral range and resolution, and select the wavelength unit as nanometers. Finally, the resampled spectral data is obtained and saved. After such an operation, the average value of all spectral reflectivities within every 10 nm range is taken as the reflectivity of the central wavelength within this range, that is, the 10 - nm resampling of the spectrum is completed. A total of 130 bands are obtained after resampling.

[0104] (6) Analysis of the variation characteristics of soil moisture content.

[0105] Table 1 shows the moisture changes of different soil samples during the experiment. It can be seen from this that during the whole experiment, the soil moisture content changes significantly with time. For soil samples infiltrated with the same concentration of salt, the soil moisture content changes greatly in the first three days. Especially during the process from the 1st day to the 2nd day, the soil moisture content decreases by about 23%, and the maximum can reach 27.76%. As time goes by, the change of soil moisture content gradually tends to be gentle.

[0106] The soil moisture content varies in soils with different salt concentrations. On the first day, the initial moisture content of the soil without salt solution infiltration reached 63.66%, while the moisture content of other soils infiltrated with salt solution was generally lower than 50%. Since the natural air-drying method was used in the experiment, affected by factors such as water vapor in the air, the soil samples could not reach a state with a moisture content of 0. This invention generally monitors until the soil weight no longer changes. As the experiment progresses, when the weight no longer changes, the moisture content of the soil with a higher salt concentration still exceeds 1%, while the moisture content of the soil with a lower concentration is generally lower than 0.5%. This shows that in the air-dried state, soils containing more salt generally absorb a certain amount of water.

[0107] Table 1 Changes in soil moisture content

[0108]

[0109] (7) Analysis of the spectral reflectance characteristics of soil moisture under different conditions.

[0110] The original spectra collected by the instrument have a large number of spikes, and the entire spectral range has strong fluctuations, which is not conducive to spectral analysis. Therefore, this invention performs smoothing processing on it. The smoothing of the spectral curve is mainly through the convolution smoothing algorithm. The Savitzky-Golay (S-G) filter is a weighted average algorithm with a moving window, but its weighting coefficients are not simple constants, but are obtained by the least squares fitting of a given high-order polynomial within the sliding window. The biggest feature of this filter is that it can ensure that the shape and width of the signal remain unchanged while filtering out noise (the specific smoothing process and formula are clearly shown in step (4) of the spectral data preprocessing). The smoothed spectral curve (see Figure 2 ) is much smoother compared to before smoothing, and the noise is eliminated. However, due to the influence of water vapor at both ends of the spectral range, spikes still exist, and individual values appear negative. Therefore, in actual applications, this invention eliminates the ranges of 350 - 380 nm and 1050 - 1100 nm and does not analyze them.

[0111] Soil spectra are affected by various physical and chemical properties of the soil. Under different moisture conditions, the spectral reflectance of the soil is different. To better observe the soil spectral curve, this invention selects several spectral curves under different infiltration solution concentrations. Figure 3 The figure shows the spectral curves under different moisture conditions in a 1 g / kg solution. When the water content is relatively large (30%), the greater the moisture, the higher the soil spectral reflectance; when the water content is relatively small (30% - 0.55%), as the moisture increases, the soil spectral reflectance will decrease, which is consistent with the existing results.

[0112] Different from the prior art, in this group of experiments, when the water content is very small (less than 0.55%), the soil spectra generally no longer show a regular relationship with the water content in the soil. However, there are still differences between different soil spectra, indicating that when the soil water content is less than 0.55%, the water no longer has a significant impact on the soil spectra. Since the basic physical and chemical properties of the soil samples used in this invention are the same, except for the different salt contents added, it is considered that the factor affecting the spectral differences at this time is the salt. Through Figure 3 It can be seen that when the water content is very small, the soil spectra generally no longer show a regular relationship with the water content in the soil. However, there are still differences between different soil spectra. After removing the spectral interference of water, and with the same basic physical and chemical properties of the soil samples, only the different salt gradients set in this invention affect the spectral differences.

[0113] To verify the generality of the above conclusion, this invention also extracted the spectral curves at concentrations of 6 g / kg (see Figure 4 ) and 20 g / kg (see Figure 5 ). The results also show that when the water content is relatively large (6 g / kg concentration / 25.75%, 20 g / kg concentration / 24.07%), the greater the water content, the higher the soil spectral reflectance; when the water content is relatively small (6 g / kg concentration / 25.75% - 1.15%, 20 g / kg concentration / 24.07% - 2.50%), as the water content increases, the soil spectral reflectance will decrease; when the water content is even smaller, the curve is controlled by the salt.

[0114] The prior art shows that there is a critical value between the soil water content and the soil spectrum (usually greater than the field capacity). When the water content is lower than the critical value, the soil spectral reflectance decreases as the soil humidity increases, and when it exceeds the critical value, it decreases as the soil humidity increases. This critical value at which the curve reflectance is affected by water is not much larger than the field capacity and is often equated with the field water holding rate. In fact, in agricultural applications, the water content on the second day after irrigation is usually considered to be equivalent to the field water holding rate. This group of experiments in this invention shows that the approximate field water holding rate is 24%. The above experiments show that the higher the salt content, the lower the general field water holding capacity, and at the same time, the higher the soil water content node at which the influence is dominated by water changes to being dominated by salt.

[0115] (8) Analysis of the spectral absorption characteristics of soil water.

[0116] The continuum removal processing is carried out on the spectral data by using the envelope method, and the reflectance is normalized between 0 and 1 to highlight the absorption characteristics of the spectral curve. Through the comparative analysis of the spectral absorption valleys, the bands with obvious absorption peak characteristics are screened out. The envelope is usually defined as connecting the protruding peak points on the spectral curve with straight lines point by point, and the external angle of the broken line at the peak point is greater than 180°. Dividing the value on the original spectral curve by the corresponding value on the envelope is the spectral de-envelope, and this processing can be achieved by setting certain parameters in the ENVI software. The continuum removal processing is carried out on the resampled spectrum. After processing, the absorption valleys and reflection peaks of the spectral curve are obvious, so that the spectral absorption characteristic analysis and spectral characteristic band selection can be better carried out. The result of the continuum removal processing (see Figure 6 ) shows that the absorption valleys are mainly distributed near the five bands of 410nm, 430nm, 500nm, 760nm, 830nm, and 940nm, and the soil with higher water content shows more absorption valleys. For the sake of uniformity, the present invention can use the above five absorption bands for moisture prediction.

[0117] Another exemplary one is that in the process of analyzing the spectral absorption characteristics of soil moisture, the continuum removal processing is carried out on the resampled spectral data by using the envelope method. The envelope is usually defined as connecting the protruding peak points on the spectral curve with straight lines point by point, and the external angle of the broken line at the peak point is greater than 180°. Dividing the value on the original spectral curve by the corresponding value on the envelope is the spectral de-envelope. The specific implementation is through the ENVI 5.3 software. Click on Spectral, select Mapping Mathods, and use the Continuum Removal function to complete the continuum removal processing of the spectral data. After such processing, the reflectance is normalized between 0 and 1 to highlight the absorption characteristics of the spectral curve; the five bands of 410nm, 430nm, 500nm, 760nm, 830nm, and 940nm are used for moisture prediction.

[0118] (9) Establishment of the hyperspectral prediction model for soil moisture. Use all bands to perform stepwise regression to extract reliable five bands and establish a least squares regression model; and use all-band data to extract the spectral principal components to establish a principal component-based partial least squares regression model, so as to realize the hyperspectral prediction of soil moisture.

[0119] Among them, the variables used in the least squares regression model are the spectral reflectances of the measured bands of 895nm, 675nm, 740nm, 860nm, and 680nm;

[0120] Y = 0.569 - 0.031*V 1 + 0.203*V 2 - 0.135*V 3 + 0.069*V 4-0.118*V 5 ,(R 2 is 0.891)

[0121] The use of full-band data to extract the spectral principal components and establish a partial least squares regression model based on the principal components includes:

[0122] X = TP T + E

[0123] Y = UQ T + F

[0124] where Y is an n×p response matrix; X is an n×m modeling matrix; T is the projection of X, which is the factor matrix; U is the projection of Y; P and Q are orthogonal loading matrices; E and F represent errors; based on the estimated factors T and U and the loading matrices P and Q, a linear model between Y and X is established through the PLSR model;

[0125] Y = Xb + e

[0126] where b is the coefficient of the PLSR model and e is the error vector.

[0127] Another exemplary aspect of the present invention is that three methods are used to predict soil moisture, and the best prediction model is selected by comparing the model accuracies. The first method performs regression prediction based on the reflection characteristic bands of the spectrum. The second method is to perform spectral correlation analysis on the spectral data within the full-band spectral data range using the stepwise regression analysis method, screen out the bands with significant correlation, and then perform regression analysis. The third method is to use the full-band data to establish a partial least squares regression model of the spectral principal components.

[0128] 1) Prediction model based on absorption characteristics: Bands with obvious spectral absorption peak characteristics are screened out (such as 410nm, 430nm, 500nm, 760nm, 830nm, 940nm, etc.). The results show that there is an extremely significant negative correlation between each absorption band and the moisture content (see Table 2). The above bands are subjected to stepwise regression analysis, and the model selects two bands, 830nm and 940nm. The determination coefficient R 2 is 0.676, and the adjusted determination coefficient Ra 2 is 0.665, RMSE = 0.10047. F = 63.557, sig = 0.000 < 0.01, indicating that the equation is significant and can withstand the test.

[0129] Table 2 Correlation between characteristic bands and soil water content

[0130] 410 430 500 760 830 940 Unsmoothed resampling -0.505** -0.556** -0.617** -0.701** -0.752** -0.729** Smoothed resampling -0.524** -0.578** -0.630** -0.733** -0.770** -0.175

[0131] When selecting the absorption band, since the present invention has smoothed the spectrum, the present invention further smooths the spectrum, then resamples it at 10 nm intervals, and selects the reflectance at the corresponding bands of 410 nm, 430 nm, 500 nm, 760 nm, 830 nm, and 940 nm as independent variables and the corresponding soil moisture content for correlation analysis first (see Table 2). The results show that except for 940 nm, there is an extremely significant correlation between the remaining bands and the soil moisture content (sig < 0.01). This is consistent with the prior art. The stepwise regression analysis is performed on the smoothed and resampled bands to obtain the regression model.

[0132] Soil moisture content = 0.513 - 0.082 * R830 + 0.072 * R760

[0133] At this time, the bands selected by the model are 760 nm and 830 nm, and the determination coefficient R 2 is 0.712, and the adjusted determination coefficient R 2 is 0.702, RMSE = 0.09477. F = 75.223, sig = 0.000 < 0.01, indicating that the significance of the equation can withstand the test. Compared with the previous equation, the accuracy has been improved, indicating that the 940 nm band before non-smoothed resampling has uncertainty in predicting soil moisture, and spectral processing should be performed during prediction.

[0134] The parameter test is performed on the coefficients of the established regression model. The results are shown in Table 3. The probabilities corresponding to the t-values of the constant term and the independent variables of the equation are extremely small and less than 0.01, indicating that the equation is effective and the selected spectral variables are also meaningful.

[0135] Table 3 Parameter Estimation Values and Test Results

[0136]

[0137] 2) Stepwise regression model based on all bands.

[0138] According to the second method, first, all spectra are smoothed and resampled, and then all spectra are used as the dependent variable. The sensitive bands are selected through correlation analysis, and the regression equation between the reflectance and the soil moisture is established. The determination coefficient R 2 of the equation is 0.900, and the adjusted determination coefficient R 2 is 0.891, and the coefficient of mean square error RMSE is 0.05733. Since there are more variables involved and the determination coefficient is large, the accuracy of the equation is relatively high. The equation is as follows:

[0139] Y = 0.569 - 0.031 * V 1 + 0.203 * V 2 - 0.135 * V 3+0.069*V 4 -0.118*V 5

[0140] F = 103.944, sig = 0.000 < 0.01, indicating that the significance of the equation can withstand the test. Parameter tests were conducted on the coefficients of the established regression model, and the results are shown in Table 4. The probabilities corresponding to the t-values of the constant term and the independent variables of the equation are extremely small and less than 0.01, indicating that the equation is effective and the selected spectral variables are also meaningful. The variables used in this model are the spectral reflectances measured in the 895nm, 675nm, 740nm, 860nm, and 680nm bands.

[0141] Table 4 Parameter Estimation Values and Test Results

[0142]

[0143]

[0144] 3) Partial Least Squares Regression Prediction Model Based on All Spectral Principal Components

[0145] Table 5 Model Results

[0146] RMSE <![CDATA[R 2 > Predicted RMSE <![CDATA[Predict R 2 > 0.0285 0.9726 0.0374 0.9528 Unsmoothed 0.0228 0.9825 0.0319 0.9657 Smoothed

[0147] Because there are a large number of hyperspectral bands, it is not convenient to establish a general multiple regression model and it is easy to generate information redundancy. The partial least squares regression method can extract the principal components from the spectra, and the principal components are uncorrelated. Therefore, in order to try to find the best model and to compare the influence of smoothing and non-smoothing on the model, the present invention uses all spectral bands to establish partial least squares models of spectral principal components respectively. As shown in Table 5, the independent verification accuracies of the partial least squares model established without smoothing and the partial least squares model of spectral principal components established with the smoothed spectra are both very high, but the accuracy of the smoothed model is higher than that of the unsmoothed model, indicating that smoothing is necessary for model establishment.

[0148] X = TP T + E

[0149] Y = UQ T + F

[0150] Where Y is an n×p response matrix; X is an n×m modeling matrix; T is the projection of X, also known as the factor matrix; U is the projection of Y; P and Q are orthogonal loading matrices; E and F represent errors. Based on the above estimated factors T and U and loading matrices P and Q, a linear model between Y and X can finally be established through the PLSR model:

[0151] Y = Xb + e

[0152] Among them, b is the coefficient of the PLSR model, and e is the error vector.

[0153] Objectively speaking, the accuracies of all three models are relatively high, and all can be used for predicting soil moisture. The multiple regression model established based on the spectral absorption band has the lowest accuracy. The multiple regression model (R 2 = 0.891) established based on the correlation of spectral reflectance and the partial least squares model established based on all spectral bands have higher accuracies. Among them, the partial least squares model based on the spectral principal components has the highest accuracy, and R 2 reaches 0.9657. However, from the perspective of practicality, the model established by least squares is more applicable because it has fewer variables and a simpler model.

[0154] Example 3. The hyperspectral prediction system for soil moisture provided by the embodiments of the present invention includes:

[0155] A soil sample collection and preparation module, which is used to sample the surface soil under different land use types in the area to be analyzed, and simulate soil samples with different gradient salt contents by adding chlorides with different concentrations;

[0156] A soil moisture measurement module, which is used to determine the moisture content of the soil sample by measuring the total weights of the culture dish containing the soil sample and the empty aluminum box, and the total weights of the culture dish and the aluminum box after adding the soil sample and the salt solution respectively;

[0157] A spectral data collection module, which is used to use a portable spectrometer to measure the spectral reflectance of soil moisture at the same time period every day during the process of the soil sample changing from completely wet to air-dried;

[0158] A spectral data preprocessing module, which is used to perform smoothing noise processing on the measured soil moisture spectral curve by using a 9-point weighted moving average method to determine the smoothed sample reflectance value;

[0159] A spectral data resampling module, which resamples the hyperspectral data collected by the portable spectrometer and preprocessed by data, and resamples the spectral data by 10 nm by loading the central wavelength and setting the full width at half maximum data parameter, and takes the average value of all spectral reflectances within every 10 nm range as the reflectance of the central wavelength within this range;

[0160] A soil moisture characteristic analysis module, which is used to analyze the change characteristics of soil moisture content, the spectral reflectance characteristics of soil moisture under different conditions, and the spectral absorption characteristics of soil moisture to clarify the moisture and spectral characteristics;

[0161] The soil moisture hyperspectral prediction model establishment module is used to perform stepwise regression using the full band to extract 5 reliable bands and establish a least squares regression model; and is used to extract the spectral principal components using the full band data to establish a principal component-based partial least squares regression model, thereby realizing the hyperspectral prediction of soil moisture.

[0162] In the above embodiments, the descriptions of the respective embodiments have their own focuses. For the parts not detailed or recorded in a certain embodiment, reference can be made to the relevant descriptions of other embodiments.

[0163] Regarding the information interaction, execution process, etc. between the above-mentioned devices / units, since they are based on the same concept as the method embodiments of the present invention, for their specific functions and the technical effects brought about, reference can be specifically made to the method embodiment part, and details are not elaborated here.

[0164] Those skilled in the art can clearly understand that, for the convenience and conciseness of description, only the above-mentioned division of each functional unit and module is used as an example. In actual applications, the above-mentioned functions can be allocated to different functional units and modules as needed, that is, the internal structure of the device is divided into different functional units or modules to complete all or part of the functions described above. Each functional unit and module in the embodiments can be integrated into one processing unit, or each unit can exist physically alone, or two or more units can be integrated into one unit. The above-mentioned integrated unit can be implemented in the form of hardware or in the form of a software functional unit. In addition, the specific names of each functional unit and module are only for the convenience of mutual distinction and do not limit the protection scope of the present invention. The specific working process of the units and modules in the above system can refer to the corresponding process in the foregoing method embodiments.

[0165] According to an embodiment of the present application, the present invention also provides a computer device, which includes: at least one processor, a memory, and a computer program stored in the memory and executable on the at least one processor. When the processor executes the computer program, the steps in any of the above method embodiments are implemented.

[0166] The embodiment of the present invention also provides a computer-readable storage medium, which stores a computer program. When the computer program is executed by a processor, the steps in each of the above method embodiments can be implemented.

[0167] The embodiment of the present invention also provides an information data processing terminal, which is used to provide a user input interface to implement the steps in each of the above method embodiments when executed on an electronic device. The information data processing terminal is not limited to mobile phones, computers, and switches.

[0168] An embodiment of the present invention further provides a server, which is used to provide a user input interface to implement the steps in the above method embodiments when executed on an electronic device.

[0169] An embodiment of the present invention further provides a computer program product. When the computer program product runs on an electronic device, the electronic device can be made to execute the steps in the above method embodiments.

[0170] If the integrated unit is implemented in the form of a software functional unit and sold or used as an independent product, it can be stored in a computer-readable storage medium. Based on this understanding, to implement all or part of the processes in the above method embodiments of the present application, a computer program can be used to instruct relevant hardware to complete. The computer program can be stored in a computer-readable storage medium. When the computer program is executed by a processor, the steps in the above method embodiments can be implemented. Among them, the computer program includes computer program code, and the computer program code can be in the form of source code, object code, executable file or some intermediate form, etc. The computer-readable medium can at least include: any entity or device that can carry the computer program code to the photographing device / terminal device, recording medium, computer memory, read-only memory (ROM), random access memory (RAM), electrical carrier signal, telecommunication signal, and software distribution medium. For example, a USB flash drive, a mobile hard disk, a magnetic disk or an optical disc, etc.

[0171] In the above embodiments, the descriptions of the various embodiments have their own emphases. For the parts not detailed or recorded in a certain embodiment, reference can be made to the relevant descriptions of other embodiments.

[0172] To further prove the positive effects of the above embodiments, the present invention conducts the following experiments based on the above technical solutions.

[0173] The technical solution proposed by the present invention has obvious advantages compared with the prior art. Specifically, at present, when establishing the vast majority of moisture or salinity prediction models, Landsat or Sentinel multispectral images are usually used. After preprocessing such as atmospheric correction and radiometric calibration, multiple spectral indices are calculated based on several basic bands. The calculation process not only needs to consider basic mathematical transformations, but even involves some relatively cumbersome differential or derivative forms. The obtained spectral indices are numerous and have a very low correlation with soil moisture or salinity content. Therefore, further screening of characteristic variables is required. During the screening process, it is very easy to eliminate variables that theoretically have a greater impact on moisture or salinity but have a low correlation, resulting in the loss of spectral information, or there may be characteristic variables with a high correlation but unable to theoretically explain moisture or salinity, resulting in a low overall prediction accuracy. Not only that, for the currently established prediction models, the model accuracy R 2 is hardly above 0.5, and it has little effect on quickly obtaining moisture or salinity content over a large area and guiding actual agricultural production. However, the accuracy of the few bands selected in the present invention reaches 0.89 after modeling, and the model accuracy reaches 0.96 after extracting the principal components from all bands, showing a very high prediction effect.

[0174] The above is only a preferred specific implementation manner of the present invention, but the protection scope of the present invention is not limited thereto. Any person skilled in the art within the technical scope disclosed by the present invention, any modifications, equivalent replacements, and improvements made within the spirit and principle of the present invention shall be covered by the protection scope of the present invention.

Claims

1. A soil moisture hyperspectral prediction method, characterized in that: The method includes: S1, soil sample collection and preparation: sampling the surface soil under different land use types in the analyzed area, and simulating soil samples with different gradient salt content by adding different concentrations of chloride; S2, soil moisture measurement: The moisture content of the soil sample was determined by measuring the total weight of the Petri dish containing the soil sample and the empty aluminum box, and the total weight of the Petri dish and the aluminum box with the soil sample and salt solution added respectively; S3, spectral data collection: using a portable spectrometer, the soil moisture spectral reflectance was measured at the same time period every day during the process of soil samples from being completely wet to being air-dried; S4, spectral data preprocessing: using a 9-point weighted moving average method to smooth the noise of the measured soil moisture spectrum curve and determine the reflectance value of the smoothed sample; S5, spectral data resampling: In the ENVI software, the hyperspectral data collected by the portable spectrometer and preprocessed are resampled by 10nm by loading the central wavelength and setting the half-width data parameters, and the average value of all spectral reflectances within every 10nm is taken as the reflectance of the central wavelength within the range; S6, soil moisture characteristics analysis: analyze the variation characteristics of soil moisture content, the spectral reflectance characteristics of soil moisture under different conditions, and the spectral absorption characteristics of soil moisture, and clarify the moisture and spectral characteristics; S7, establishment of hyperspectral prediction model for soil moisture: use full-band stepwise regression to extract 5 reliable bands and establish a least squares regression model; and use full-band data to extract spectral principal components and establish a partial least squares regression model based on principal components, thereby realizing hyperspectral prediction of soil moisture.

2. The soil moisture hyperspectral prediction method according to claim 1, characterized in that: In step S1, soil sample collection and preparation includes: The surface soil of different land use types in the analyzed area was sampled; the collected surface soil was placed in a plastic bag and placed in the laboratory to dry; the air-dried soil samples were cleaned of debris and plant residues and ground to make the soil samples pass through a 10-mesh / 2mm aperture sieve for later use; The ground soil samples were grouped and placed in aluminum boxes. The surface was gently scraped with a ruler and placed in a culture dish. Each group of soil was moistened with NaCl solutions of different concentrations to simulate soils with different salt concentrations in coastal areas. The experiment was set up with two repeated tests.

3. The soil moisture hyperspectral prediction method according to claim 1, characterized in that: In step S2, soil moisture measurement includes: Label the culture dishes, weigh the total weight of the culture dishes containing soil samples and the empty aluminum box and record it as Z; place the soil samples in the aluminum boxes and weigh the total weight of the culture dishes, aluminum boxes and soil samples again, record it as Z1; add the salt solution of the corresponding concentration to the culture dishes, start weighing after the soil samples are completely soaked, and weigh them again at the same time every day thereafter until the weight of the soil samples does not change. After the water evaporates completely, the weight of the salt solution soaking the soil is recorded as Z2; the moisture content calculation formula for the corresponding aluminum box soil sample on each day is: Soil moisture content (%) = (Z2-Z1) / (Z2-Z)*100%.

4. The soil moisture hyperspectral prediction method according to claim 1, characterized in that: In step S3, spectral data collection includes: Avantes portable spectrometer was used to collect hyperspectral data. The spectrometer test range was 350-1100nm. White and black board calibration were performed before the test to remove the influence of dark current. The sensor probe of the spectrometer is pointed vertically downward, with the probe 12 cm away from the soil sample. When the soil sample is completely wetted and then air-dried, the spectral reflectance of soil moisture is measured when it is weighed at the same time period every day until the soil sample is air-dried, and the spectrum and moisture content data are obtained.

5. The soil moisture hyperspectral prediction method according to claim 1, characterized in that: In step S4, spectral data preprocessing includes: The 9-point weighted moving average method is used to smooth the noise of the spectral curve. The spectral curve is given the reflectance value of N samples. The value range of the i-th point is the 4 points before and after the point, including the i-th point. The weighted average calculation is performed on the point, and the new value is the smoothed value of the i-th point. The calculation process is as follows: R i '=0.04R i-4 +0.08R i-3 +0.12R i-2 +0.16R i-1 +0.20R i +0.16R i+1 +0.12R i+2 +0.08R i+3 +0.04R i+4 Among them, R i ' is the smoothed sample reflectance value, and i-4, i-3, i-2, i-1, i, i+1, i+2, i+3, and i+4 represent the 9 measurement points centered on i.

6. The soil moisture hyperspectral prediction method according to claim 1, characterized in that: In step S5, the spectral data resampling includes: in the ENVI software, click Spectral-Spectral Libraries-Spectral LibrariesBuiler, select First Input Spectum-Option-From Plot Windows to open the hyperspectral data collected by the Avantes portable spectrometer and pre-processed, so as to create a spectral library to be resampled, then open Spectral-Spectral Libraries-Spectral Libraries Resampling in the main menu, select the created spectral library to be resampled, load the center wavelength and set the half-height width data in Spectral Resampling Parameters to determine the spectral range and resolution of the resampled spectrum, and select the wavelength unit as nanometers, and finally obtain the resampled spectral data and save it. After this operation, all spectral reflectances within the range of 10nm are averaged as the reflectance of the central wavelength within the range, that is, the resampling of the spectrum 10nm is completed.

7. The soil moisture hyperspectral prediction method according to claim 1, characterized in that: In step S6, during the analysis of soil moisture spectral reflectance characteristics under different conditions, the spectral bands of 350-380 nm and 1050-1100 nm are removed and not analyzed.

8. The soil moisture hyperspectral prediction method according to claim 1, characterized in that: In step S6, during the soil moisture spectral absorption characteristic analysis, the envelope method is used to perform continuum removal processing on the resampled spectral data.

9. The soil moisture hyperspectral prediction method according to claim 1, characterized in that: In step S7, the variables used in the least squares regression model are the measured spectral reflectances of the 895nm, 675nm, 740nm, 860nm, and 680nm bands; Y=0.569-0.031*V1+0.203*V2-0.135*V3+0.069*V4-0.118*V5,(R 2 0.891) The method of using full-band data to extract spectral principal components and establish a partial least squares regression model based on principal components includes: X=TP T +E Y=UQ T +F Among them, Y is the n×p response matrix; X is the n×m modeling matrix; T is the projection of X, which is a factor matrix; U is the projection of Y; P and Q are orthogonal loading matrices; E and F represent errors; Based on the estimated factors T and U and the loading matrices P and Q, a linear model between Y and X is established through the PLSR model; Y=Xb+e Where b is the coefficient of the PLSR model and e is the error vector.

10. A soil moisture hyperspectral prediction system, characterized in that: The system is used to control the soil moisture hyperspectral prediction method according to any one of claims 1 to 9, and the system includes: Soil sample collection and preparation module, used to sample the surface soil under different land use types in the area to be analyzed, and simulate soil samples with different gradient salt content by adding different concentrations of chloride; A soil moisture measurement module is used to determine the moisture content of the soil sample by measuring the total weight of the culture dish containing the soil sample and the empty aluminum box, and the total weight of the culture dish and the aluminum box to which the soil sample and the salt solution are added respectively; The spectral data acquisition module is used to measure the spectral reflectance of soil moisture at the same time period every day when the soil sample is from completely wet to air-dried using a portable spectrometer; The spectral data preprocessing module is used to perform smooth noise processing on the measured soil moisture spectrum curve using a 9-point weighted moving average method to determine the reflectance value of the sample after smooth processing; The spectral data resampling module uses the hyperspectral data collected by the portable spectrometer and preprocessed by the data to resample the spectral data by 10nm by loading the central wavelength and setting the half-width data parameters, and calculates the average value of all spectral reflectances within every 10nm as the reflectance of the central wavelength within the range; Soil moisture characteristic analysis module, used to analyze the variation characteristics of soil moisture content, the spectral reflection characteristics of soil moisture under different conditions, and the spectral absorption characteristics of soil moisture, and to clarify the moisture and spectral characteristics; The soil moisture hyperspectral prediction model establishment module is used to use the full band for stepwise regression to extract 5 reliable bands and establish a least squares regression model; and is used to use the full band data to extract the spectral principal components and establish a partial least squares regression model based on the principal components, thereby realizing soil moisture hyperspectral prediction.

Images