Moisture prediction method and system based on airborne multispectral sensitive band combination

Abstract

The invention discloses a moisture prediction method and system based on airborne multispectral sensitive wave band combination, and relates to the technical field of agricultural remote sensing and image processing, and the method comprises the steps: on the basis of the response characteristics of soil and plant canopy reflection spectrums to moisture, combining with the physical mechanism of O-H bond vibration in water molecules, and predicting the moisture content of the soil and plant canopy reflection spectrums; determining the bandwidths of all central wavelengths by considering the influence of atmospheric water vapor on the near infrared spectrum; generating reflection spectrums of the soil and the vegetation canopies, generating a spectral response function by using the relationship between the central wavelength and the bandwidth, converting the simulated spectrums into a moisture sensitive wave band, and calculating the reflectivity and the spectral index; and collecting reflection spectrum images of the soil and the vegetation canopies at the selected positions, and inputting the reflection spectrum images into the moisture prediction model to predict moisture. By optimizing wave band selection and bandwidth allocation, the problems that when an existing airborne multispectral sensor executes a moisture monitoring task, the pertinence of wave band selection is insufficient, and the monitoring precision is not high can be effectively solved.

Description

Technical Field

[0001] The present invention relates to the technical field of agricultural remote sensing and image processing, and in particular to a moisture prediction method and system based on airborne multi-spectral sensitive band combination. Background Art

[0002] With the intensification of global climate change and the increasing shortage of water resources, the sustainable management of agricultural production and ecological environment faces unprecedented challenges. As a key factor affecting plant growth and soil health, accurate monitoring of water is of great significance for improving agricultural output, optimizing water resource utilization efficiency, coping with extreme weather such as drought, and protecting the ecological environment. However, traditional water monitoring methods mainly rely on ground sampling and laboratory analysis, which are not only time-consuming and labor-intensive, but also have limitations such as limited spatial coverage and lack of real-time performance. This limitation significantly affects the efficiency and accuracy of agricultural management and environmental monitoring, especially in large areas of farmland or ecosystems.

[0003] At present, moisture monitoring methods based on remote sensing technology have been widely used. With the rapid development of remote sensing technology, moisture monitoring methods based on airborne sensors have gradually become an efficient alternative. Compared with ground monitoring, airborne sensing systems have the advantages of wide coverage, fast acquisition speed, and strong flexibility. In particular, multispectral sensors mounted on UAV platforms can obtain reflectance data of large areas in a short time, and realize rapid monitoring of soil and vegetation moisture information. In addition, the advancement of modern multispectral imaging technology, especially the precision of spectral band selection and the improvement of sensor resolution, makes it possible to monitor the moisture status of soil and vegetation in detail.

[0004] Under drought stress, the structure and water content of plant leaves will change significantly, resulting in changes in the response characteristics of the plant canopy reflectance spectrum. For example, when plants are short of water, the synthesis of chlorophyll decreases or the decomposition accelerates, resulting in an increase in the reflectance in the visible light region, accompanied by a red-edge blue shift phenomenon. Water shortage can also lead to a decrease in leaf thickness, changes in cell structure, and closure of stomata, which will affect the reflection and absorption characteristics of leaves in the near-infrared region. In addition, the light passing through the canopy will be scattered and reflected multiple times on the leaves and soil surfaces, which makes the canopy reflectance spectrum not only contain information about the plant itself, but also can predict the moisture content of the root zone soil by analyzing the canopy spectrum. For the exposed soil surface, the spectral reflectance characteristics of the bare soil are mainly affected by factors such as soil particle structure, moisture content, and organic matter content. Changes in moisture content will especially significantly affect the reflectance of the soil in the mid-infrared and short-wave infrared regions. Therefore, by combining the canopy spectrum and bare soil spectrum response characteristics, a comprehensive prediction of the moisture content of the root zone and bare soil can be achieved.

[0005] However, most of the existing airborne multispectral sensors are concentrated in the visible light and near-infrared bands. Although these bands can provide certain moisture information, their pertinence and accuracy are limited, and they lack sufficient spectral resolution and sensitivity to accurately capture the absorption characteristics of moisture, which limits their effectiveness in moisture monitoring applications. Summary of the invention

[0006] The purpose of the present invention is to address the deficiencies of the above-mentioned prior art and to provide a moisture prediction method and system based on an airborne multi-spectral sensitive band combination, so as to solve the problem that the prior art lacks sufficient spectral resolution and sensitivity and cannot accurately capture the absorption characteristics of moisture, resulting in limited effects in moisture monitoring applications.

[0007] The present invention specifically provides the following technical solutions:

[0008] A moisture prediction method based on airborne multi-spectral sensitive band combination includes the following steps:

[0009] Based on the response characteristics of soil and plant canopy reflectance spectra to moisture, the central wavelengths of multiple moisture-sensitive bands are selected. According to the physical mechanism of O-H bond vibration in water molecules, the central wavelengths of multiple different moisture-sensitive bands are added and selected, and the bandwidths of all central wavelengths are determined.

[0010] Simulate and generate the reflectance spectra of soil and vegetation canopy, generate a spectral response function using the relationship between the central wavelengths and bandwidths of the multiple different moisture-sensitive bands, convert the simulated reflectance spectra to multiple moisture-sensitive bands through the spectral response function, obtain the reflectance and spectral index of the multiple moisture-sensitive bands, and construct a moisture prediction model using the reflectance and spectral index;

[0011] All moisture-sensitive bands with a determined bandwidth are combined and integrated into the airborne multispectral sensor to collect reflectance spectral images of the soil and vegetation canopy at the selected location. The reflectance spectral images are then input into the moisture prediction model to predict the moisture content of the soil and plant canopy at the selected location.

[0012] Preferably, the simulation of generating the reflectance spectra of soil and vegetation canopy includes:

[0013] The MARMIT2 radiation transfer model is used to simulate the soil reflectance spectrum in the solar domain, and the soil moisture content SMC of the simulated soil reflectance spectrum is normalized as the dimensionless soil factor psoil;

[0014] The soil reflectance spectra with the lowest and highest water content are used as inputs to the PROSAIL radiative transfer model to simulate the reflectance spectra of the vegetation canopy.

[0015] Among them, after the MARMIT2 radiation transfer model simulates the reflectance spectrum of the soil in the solar domain, the sigmoid function is used to fit the soil moisture content SMC representing the soil reflectance spectrum according to the product of the input water layer thickness and the surface water coverage; the specific expression is:

[0016] BRF Soil (λ) = f MARMIT2 (λ,L,ε,δ);

[0017] Among them, BRF Soil (λ) is soil reflectance, soil moisture content SMC characterizing soil reflectance spectrum, f MARMIT2 (·) is the MARMIT2 radiation transfer model operation function, λ is the wavelength, L is the water layer thickness, ε is the surface water coverage, and δ is the soil particle volume fraction;

[0018] Among them, in the PROSAIL radiation transfer model, the equivalent water thickness is obtained, and the equivalent water thickness of the vegetation canopy and the normalized soil moisture content are represented by equivalent water thickness and psoil respectively; the specific expression is:

[0019] CEWT=LAI×C w ;

[0020] DHR Leaf (λ) = f PROSPECT (λ,θ DHR );

[0021] BRF Canopy (λ) = f SAIL (λ,DHR(λ),θ BRF );

[0022] Where CEWT is the canopy equivalent water thickness, LAI is the leaf area index, C w is the equivalent water thickness, DHR Leaf (λ) is the PROSPECT model f PROSPECT When the simulated wavelength is λ, the blade directional hemispherical reflectivity, θ DHR is the input parameter of PROSPECT model; BRF Canopy (λ) is the value obtained when the 4SAIL model is used. SAIL Coupling DHR Leaf (λ), the bidirectional reflectance of the canopy at the simulated wavelength λ, θ BRF are the input parameters of the 4SAIL model.

[0023] Preferably, a spectral response function is generated by using the relationship between the central wavelengths and bandwidths of the multiple different moisture-sensitive bands, and the simulated reflection spectrum is converted to multiple moisture-sensitive bands through the spectral response function to obtain the reflectance and spectral index of the multiple moisture-sensitive bands, including:

[0024] According to the central wavelength and bandwidth of the selected moisture sensitive band, the spectral response function is approximated by a Gaussian function; the specific expression is:

[0025]

[0026] Among them, SRF(λ) wavelength is the spectral index at λ, λ0 is the central wavelength, σ is the standard deviation, and the width of the spectral response function curve is controlled by the bandwidth FWMH calculation;

[0027] The soil reflectance spectrum and the canopy reflectance spectrum are converted to obtain the selected multiple moisture-sensitive band combinations and the reflectance of the multiple moisture-sensitive band combinations; the specific expression is:

[0028]

[0029] Where R band is the reflectivity after resampling, λ1 and λ2 are the wavelength ranges of the band, and BRF(λ) is the reflectivity when the simulated wavelength is λ;

[0030] Multiple moisture-sensitive bands are combined and input into the spectral response function to obtain the spectral index.

[0031] Preferably, before inputting the reflectance spectrum image into the moisture prediction model, the method further includes:

[0032] Collect real soil and vegetation canopy reflectance spectral data, and use spectral response functions to convert the real soil and vegetation canopy reflectance spectral data into multiple moisture-sensitive bands to obtain the reflectance and spectral index of multiple moisture-sensitive bands under real conditions;

[0033] The reflectance and spectral index of moisture-sensitive bands under multiple real conditions are used to perform nested cross-validation on the prediction results of the moisture prediction model. The moisture prediction model is optimized through the nested cross-validation results to obtain the optimal moisture prediction model.

[0034] Preferably, the method of using the reflectance and spectral index of multiple moisture-sensitive bands under real conditions to perform nested cross-validation on the prediction results of the moisture prediction model includes:

[0035] The outer cross-validation of the nested cross-validation randomly divides the data into 10 folds, retains 1 fold as the test set each time, and the remaining 9 folds as the training set to evaluate the optimal model selected in the inner cross-validation; the inner cross-validation divides the training set of the outer cross-validation into 10 folds to optimize and select the hyperparameters of the moisture prediction model.

[0036] Preferably, the central wavelengths of multiple moisture-sensitive bands are selected based on the response characteristics of the soil and plant canopy reflectance spectra to moisture, and the central wavelengths of multiple different moisture-sensitive bands are added and selected according to the physical mechanism of the O—H bond vibration in water molecules, including:

[0037] Based on the response characteristics of soil and plant canopy reflectance spectra to moisture, 560nm, 705nm, 750nm, 800nm, 900nm, 970nm, 1100nm, 1200nm, 1650nm and 2200nm were selected as the central wavelengths of the moisture-sensitive bands;

[0038] Based on the physical mechanism of O—H bond vibration in water molecules, 450 nm and 660 nm were selected as the central wavelengths of the moisture-sensitive band.

[0039] Preferably, determining the bandwidth of all central wavelengths includes:

[0040] The spectral information and spatiotemporal resolution are balanced by selecting the bandwidth in the moisture-sensitive band, where the bandwidths in the selected moisture-sensitive band are 30nm, 30nm, 30nm, 10nm, 15nm, 40nm, 40nm, 10nm, 50nm, 50nm, 50nm and 100nm respectively.

[0041] The present invention provides a moisture prediction system based on an airborne multi-spectral sensitive band combination, comprising:

[0042] A selection module is used to select the central wavelengths of multiple moisture-sensitive bands based on the response characteristics of soil and plant canopy reflectance spectra to moisture, add and select the central wavelengths of multiple different moisture-sensitive bands based on the physical mechanism of O-H bond vibration in water molecules, and determine the bandwidth of all central wavelengths;

[0043] A model building module, used to simulate and generate the reflectance spectra of soil and vegetation canopy, generate a spectral response function using the relationship between the central wavelengths and bandwidths of the multiple different moisture-sensitive bands, convert the simulated reflectance spectra to multiple moisture-sensitive bands through the spectral response function, obtain the reflectance and spectral index of the multiple moisture-sensitive bands, and build a moisture prediction model using the reflectance and spectral index;

[0044] The prediction module is used to combine all moisture-sensitive bands of a certain bandwidth and integrate them into the airborne multispectral sensor, collect the reflectance spectral images of the soil and vegetation canopy at the selected location, and input the reflectance spectral images into the moisture prediction model to predict the moisture of the soil and plant canopy at the selected location.

[0045] The present invention provides a computer device, including a memory and a processor, wherein a program is stored in the memory, and when the program is executed by the processor, the processor executes the steps of the above-mentioned moisture prediction method based on airborne multi-spectral sensitive band combination.

[0046] The present invention provides a storage medium on which a computer program is stored. When the computer program is executed by a processor, the steps of the above-mentioned moisture prediction method based on airborne multi-spectral sensitive band combination are realized.

[0047] Compared with the prior art, the present invention has the following significant advantages:

[0048] The present invention selects the central wavelengths of multiple moisture-sensitive bands through the response characteristics of soil and plant canopy reflectance spectra to moisture, and on this basis, according to the physical mechanism of O-H bond vibration in water molecules, increases and selects the central wavelengths of multiple different moisture-sensitive bands, and performs corresponding bandwidth allocation, which can effectively solve the problems of insufficient band selection targeting and low monitoring accuracy of existing multi-spectral sensors when performing moisture monitoring tasks, and converts the simulated reflectance spectrum into moisture-sensitive bands using the spectral response function generated by the central wavelength and bandwidth, thereby constructing a moisture prediction model through the converted reflectance and spectral index, because the converted reflectance has sufficient spectral resolution and sensitivity, so that the moisture prediction model can be used to collect reflectance spectrum images at any location to achieve moisture prediction, thereby improving the prediction effect. BRIEF DESCRIPTION OF THE DRAWINGS

[0049] Figure 1 Schematic diagram of typical vegetation and soil reflectivity at the band positions selected for the present invention; wherein Figure 1 (a) is the reflectance variation diagram of different chlorophyll a+b contents in different bands. Figure 1 (b) is the reflectivity variation diagram of different canopy equivalent water thickness in the band between 1000-2400. Figure 1 (c) is the reflectivity variation diagram of different canopy equivalent water thickness in the band between 400-1000. Figure 1 (d) is the reflectance variation diagram of different soil moisture contents in different bands;

[0050] Figure 2 Schematic diagram of the stretching motion of the O—H bond of water molecules in the present invention; Figure 2 (a) is a water molecule diagram. Figure 2(b) is the symmetric stretching vibration diagram of water molecules. Figure 2 (c) is the asymmetric stretching vibration diagram of water molecules. Figure 2 (d) is the bending vibration diagram of water molecules;

[0051] Figure 3 A schematic diagram of the spectral response function of the moisture sensitive band provided by the present invention;

[0052] Figure 4 The invention is a schematic diagram of the reflectivity of soil and vegetation under different moisture conditions in the moisture sensitive band; Figure 4 (a) is a schematic diagram of the reflectance of vegetation under different moisture conditions in the moisture-sensitive band. Figure 4 (b) is a schematic diagram of the reflectivity of soil under different moisture conditions in the moisture sensitive band;

[0053] Figure 5 It is a flow chart of a moisture prediction method based on airborne multi-spectral sensitive band combination of the present invention;

[0054] Figure 6 This is a cross-validation diagram provided by the present invention. DETAILED DESCRIPTION

[0055] The following is a clear and complete description of the technical solutions of the embodiments of the present invention in conjunction with the drawings in the present invention. Obviously, the described embodiments are part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work should fall within the scope of protection of the present invention.

[0056] In traditional technology, although NDVI (Normalized Difference Vegetation Index) is often used to assess vegetation coverage and health, it is difficult to accurately reflect the difference in moisture in different types of vegetation or soil because it only relies on red and near-infrared light bands. At the same time, when water molecules absorb radiation energy, the O-H bonds vibrate and induce a combination of symmetrical stretching, asymmetrical stretching and bending motions, forming significant water absorption bands in multiple bands above 1000nm. These bandwidth positions usually include 970nm, 1200nm, 1650nm and 2200nm.

[0057] like Figure 1 and Figure 5 As shown, the present invention provides a moisture prediction method based on airborne multi-spectral sensitive band combination, which specifically includes the following steps:

[0058] Step S1: Based on the response characteristics of soil and plant canopy reflectance spectra to moisture, the central wavelengths of multiple moisture-sensitive bands are selected. According to the physical mechanism of O-H bond vibration in water molecules, the central wavelengths of multiple different moisture-sensitive bands are added and selected, and the bandwidth of all central wavelengths is determined by considering the influence of atmospheric water vapor on the near-infrared spectrum.

[0059] The central wavelength of the moisture-sensitive band is selected based on the response characteristics of the soil and vegetation reflectance spectra to moisture. Figure 1 As shown in the figure, the reflectance spectrum of soil generally decreases with the increase of soil moisture content (SMC), while the spectrum of vegetation canopy is relatively complex. In the visible light region, the influence of vegetation moisture on the canopy reflectance spectrum is small. The band selection mainly needs to consider the effect of reduced synthesis and accelerated decomposition of chlorophyll under water stress on the canopy reflectance spectrum. Affected by the absorption characteristics of chlorophyll, the reflectance spectrum of vegetation canopy forms a reflection peak in the green light region near 560nm, and there are two significant absorption valleys in the blue light region near 450nm and the red light region near 660nm. At around 705-750nm, the reflectance increases rapidly, forming a "red edge". When the chlorophyll content decreases, the reflectance in the visible light region increases, accompanied by the phenomenon of red edge blue shift. In the near-infrared region, water shortage in plants will lead to the contraction of plant cells and the shrinkage of leaves, resulting in thinning of leaves and closure of stomata, affecting the absorption and reflection characteristics of light in the near-infrared region. Affected by plant moisture and cell structure, the canopy reflectance spectrum has several moisture absorption bands near 970nm, 1200nm, 1450nm, 1900nm and 2500nm, and produces several reflection peaks around 1100nm, 1300nm, 1650nm and 2200nm. Among them, the absorption valley near 1450nm is only affected by leaf moisture, while the three absorption valleys at 970, 1200 and 1900nm are also affected by leaf starch and protein. At the same time, considering the strong influence of water vapor near wavelengths of 1300-1500nm, 1750-2000nm and 2300-2500nm, the 1550-1750nm range and near 2200nm are more suitable bands for monitoring plant canopy moisture. In addition, 800nm ​​and 900nm can be used to construct the normalized vegetation index NDVI and the moisture index WI. Therefore, 560nm, 705nm, 750nm, 800nm, 900nm, 970nm, 1100nm, 1200nm, 1650nm and 2200nm are selected as the central wavelengths of the moisture sensitive band.

[0060] The central wavelength of the water-sensitive band is selected based on the physical mechanism of the O-H bond vibration in water molecules. Water molecules absorb radiation to obtain energy, producing rotational transitions, intermolecular vibrational transitions, intramolecular vibrational transitions, and electronic transitions. Rotational transitions and intermolecular vibrational transitions determine the absorption in the microwave and far-infrared ranges, electronic transitions determine the absorption in the ultraviolet range, and intramolecular vibrational transitions determine the absorption in the mid-infrared range. In the visible and near-infrared regions, water absorption is mainly due to the stretching overtones and vibrational absorption of the O-H bond vibrations in water molecules. In liquid water, each water molecule can form up to four hydrogen bonds with other surrounding water molecules. Hydrogen bonds are constantly breaking and recombining, which restricts the rotation and vibration of water molecules. This restriction leads to the vibration modes of water molecules, including symmetric stretching vibrations, asymmetric stretching vibrations, and bending vibrations, such as Figure 2 The O—H bond stretching vibration mainly occurs in the high frequency region (about 3400cm -1 ), the bending vibration mainly occurs in the mid-frequency region (about 1600cm -1 ). Since the energy required for symmetric stretching and asymmetric stretching is very similar, the absorption peaks merge into a large absorption peak at around 3000nm. Bending vibrations occur at lower energy levels, resulting in an absorption peak at around 6000nm. The harmonics of these vibrations occur at higher energy levels (i.e. shorter wavelengths), increasing the required energy and decreasing the absorption intensity. The harmonics of bending and stretching vibrations cause the absorption spectrum of water to produce shoulders in the visible and near-infrared ranges. Since the absorption intensity of continuous overtone transitions of water decreases by 10 to 20 times as the number of stretching quanta increases, overtone transitions with more than 5 stretching quanta have no significant effect on reflectivity. The location of the vibration absorption of liquid water is shown in Table 1:

[0061] Table 1 Location of vibration absorption of liquid water

[0062] Wavelength (nm) <![CDATA[Wavenumber (cm -1 )]]> explain 1940 5260 av1+v2+bv3(a+b=1) 1450 6800 av1+bv3(a+b=2) 1190 8330 av1+v2+bv3(a+b=2) 970 10310 av1+bv3(a+b=3) 836 11960 av1+v2+bv3(a+b=3) 750 13330 av1+bv3(a+b=4) 660 15150 av1+v2+bv3(a+b=4) 605 16500 av1+bv3(a+b=5) 514 19460 av1+bv3(a+b=6) 449 22270 av1+bv3(a+b=7) 401 24940 av1+bv3(a+b=8)

[0063] Therefore, 450 nm and 660 nm are selected as the central wavelengths of the moisture sensitive band.

[0064] The bandwidth of the central wavelength of the moisture-sensitive band is determined from the aspects of spectral absorption characteristics, atmospheric transmittance, and practical feasibility. For example, 970nm is located in the water absorption band and 705nm is located in the red edge region, so a narrow bandwidth is required to accurately capture the change in moisture content; in areas with low atmospheric transmittance, the bandwidth should be relatively wide to ensure sufficient signal acquisition, such as in the near-infrared region of 800nm ​​and 900nm; considering the flight speed and image resolution of the airborne platform, the bandwidth setting needs to strike a balance between capturing sufficient spectral information and ensuring spatiotemporal resolution. Based on these principles, the bandwidth in the moisture-sensitive band is selected to balance spectral information and spatiotemporal resolution, and the band bandwidths in the selected moisture-sensitive bands are 30nm, 30nm, 30nm, 10nm, 15nm, 40nm, 40nm, 10nm, 50nm, 50nm, 50nm and 100nm respectively.

[0065] Step S2: simulate and generate the reflectance spectra of soil and vegetation canopy, add Gaussian noise to the generated reflectance spectra to simulate real data, generate a spectral response function using the relationship between the central wavelength and bandwidth of multiple different moisture-sensitive bands, convert the simulated reflectance spectra to multiple moisture-sensitive bands through the spectral response function, obtain the reflectance and spectral index of multiple moisture-sensitive bands, and use the reflectance and spectral index to build a moisture prediction model.

[0066] The spectral response function describes the sensitivity of the sensor to light of different wavelengths. Simulating the spectral response function is a key step in designing a multi-spectral sensor. According to the present invention, the central wavelength and bandwidth of the moisture-sensitive band are selected, and the spectral response function is approximated using a Gaussian function. Specifically:

[0067]

[0068] Among them, SRF(λ) wavelength is the spectral index at λ, λ0 is the central wavelength, σ is the standard deviation, and the width of the spectral response function curve is controlled by the bandwidth FWMH calculation. The spectral response function is as follows Figure 3 shown.

[0069] According to the approximate spectral response function, the soil reflectance spectrum and the canopy reflectance spectrum are converted to obtain a plurality of selected moisture-sensitive band combinations (12 moisture-sensitive band combinations selected in the present invention) and the reflectance of the plurality of moisture-sensitive band combinations, specifically:

[0070]

[0071] Where R bandis the reflectivity after resampling, λ1 and λ2 are the wavelength ranges of the band, and BRF(λ) is the reflectivity when the simulated wavelength is λ. Add 10% Gaussian noise to the resampled reflectivity data to simulate the real data. The resampling results are as follows: Figure 4 .

[0072] Multiple moisture-sensitive bands are combined and input into the spectral response function to obtain the spectral index.

[0073] The MARMIT2 model is used to simulate the soil reflectance spectrum in the solar domain. The SMC of the simulated soil reflectance spectrum is normalized as the dimensionless soil factor psoil. The soil reflectance spectra with the lowest and highest moisture contents are used as inputs to the PROSAIL radiation transfer model to simulate the canopy reflectance spectrum.

[0074] The MARMIT2 model is used to simulate about 100,000 soil reflectance spectra. That is, after the MARMIT2 model simulates the soil reflectance spectrum in the solar domain, the sigmoid function is used to fit the soil moisture content SMC representing the soil reflectance spectrum according to the product of the input water layer thickness and the surface water coverage. Specifically:

[0075] BRF Soil (λ) = f MARMIT2 (λ,L,ε,δ);

[0076] Among them, BRF Soil (λ) is soil reflectance, soil moisture content SMC characterizing soil reflectance spectrum, f MARMIT2 (·) is the MARMIT2 radiation transfer model operation function, λ is the wavelength, L is the water layer thickness, ε is the surface water coverage, and δ is the soil particle volume fraction.

[0077] The PROSAIL radiation transfer model was used to simulate 50,000 canopy reflectance spectra and obtain the equivalent water thickness. The equivalent water thickness of the vegetation canopy and the normalized soil moisture content were expressed as canopy equivalent water thickness (CEWT) and psoil, respectively. Specifically:

[0078] CEWT=LAI×C w ;

[0079] DHR Leaf (λ) = f PROSPECT (λ,θ DHR );

[0080] BRF Canopy (λ) = f SAIL (λ,DHR(λ),θ BRF );

[0081] Where CEWT is the canopy equivalent water thickness, LAI is the leaf area index, C w is the equivalent water thickness, DHR Leaf (λ) is the value obtained by using the PROSPECT model PROSPECT The directional hemispherical reflectance (DHR) of the blade when the simulated wavelength is λ, θ DHR are the input parameters of the PROSPEC T model; BRF Canopy (λ) is the value obtained when the 4SAIL model is used. SAIL Coupling DHR Leaf (λ), Bidirectional Reflectance Factor (BRF) of the canopy when the simulated wavelength is λ, θ BRF are the input parameters of the 4SAIL model. The input parameter range is shown in Table 2:

[0082] Table 2 Input parameter ranges of MARMIT2 and PROSAIL radiation transfer models

[0083] Input Parameters Physical meaning unit scope L Water layer thickness cm 0-0.05 ε Surface water coverage Dimensionless 0-1 δ Soil particle volume fraction Dimensionless 0-0.25 N Blade structure parameters Dimensionless 1.5 Cab Chlorophyll a+b content <![CDATA[μg / cm 2 ]]> 30-70 Car Carotenoid content <![CDATA[μg / cm 2 ]]> Cab / 4 C Equivalent water thickness cm 0-0.05 Cm Dry matter content <![CDATA[g / cm 2 ]]> 0.012 LAI Leaf Area Index <![CDATA[m 2 / m 2 ]]> 1-5 ALA Average leaf inclination angle ° 20-75 Hotspot Hot spot effect parameters Dimensionless 0.5 / LAI tts Solar zenith angle ° 25-50 tto Observation zenith angle ° 0 psi Relative azimuth ° 0-360 psoil Soil Factors Dimensionless 0-1

[0084] The vegetation index is calculated using the constructed data set, where the spectral index formula is shown in Table 3-1 and Table 3-2:

[0085] Table 3-1 Vegetation index formula

[0086] Vegetation Index formula RVI R800 / R660 NDVI (R800-R660) / (R800+R660) WI R900 / R970 SR2 R560 / R660 mNDVI (R750-R705) / (R750+R705) WI / mNDVI WI / mNDVI WI / NDVI WI / NDVI RDVI (R800-R660) / sqrt(R800+R660) GNDVI (R800-R560) / (R800+R560) RGRI R660 / R560 OSAVI 1.16*(R800-R660) / (R800+R660+0.16) TCARI 3*((R705-R660)-0.2*(R705-R560)*R705 / R660) OSAVI / TCARI OSAVI / TCARI MCARI TCARI / 3 MCARI1 1.2*(2.5*(R800-R660)-1.3*(R800-R560)) MSR (R800 / R660-1) / sqrt(R800 / R660+1) MTVI3 1.2*(1.2*(R800-R560)-2.5*(R660-R560)) <![CDATA[mSR 705 ]]> (R750-R450) / (R705-R450) <![CDATA[mNDVI 705 ]]> (R750-R705) / (R750+R705-2*R450) SIPI (R800-R450) / (R800+R660) TVI 0.5*(120*(R750-R560)-200*(R660-R560)) NDRE (R800-R750) / (R800+R750) VARI (R560-R660) / (R560+R660-R450)

[0087] Wherein, R450, R560, R660, R705, R750, R800, R900, R970, R1100, R1200, R1650, and R2200 represent the reflectivity at 450nm, 560nm, 660nm, 705nm, 750nm, 800nm, 900nm, 970nm, 1100nm, 1200nm, 1650nm, and 2200nm, respectively.

[0088] Table 3-2 Vegetation index formula

[0089] ARVI (R800-(2*R660-R450)) / (R800+(2*R660-R450)) <![CDATA[VARI rededge ]]> (R705-R660) / (R705+R660) DVI R800-R660 BNDVI (R800-R450) / (R800+R450) <![CDATA[RDVI green ]]> (R750-R560) / (R750+R560) <![CDATA[DVI green ]]> (R750-R560) CAR (R800-(1*R660+0.5*R450)) / (R800+(1*R660+0.5*R450)) GCI R800 / R450-1 EVI 2.5*(R800-R660) / (R800+6*R660-7.5*R450+1) LWVI1 (R1100-R900) / (R1100+R900) LWVI2 (R1100-R1200) / (R1100+R1200) DWSI (R800+R560) / (R1650+R660) DWSI1 R800 / R1650 DWSI2 R1650 / R560 DWSI3 R1650 / R660 DWSI4 (R800-R560) / (R1650+R660) SR R1650 / R2200

[0090] Step S3: All moisture-sensitive bands of the determined bandwidth are combined and integrated into the airborne multispectral sensor to collect the reflectance spectral images of the soil and vegetation canopy at the selected location, and the reflectance spectral images are input into the moisture prediction model to predict the moisture of the soil and plant canopy at the selected location.

[0091] Before inputting the reflectance spectral image into the moisture prediction model, the following are also included:

[0092] Real soil and vegetation canopy reflectance spectral data are collected, and the spectral response function is used to convert the real soil and vegetation canopy reflectance spectral data into multiple moisture-sensitive bands to obtain the reflectance and spectral index of multiple moisture-sensitive bands under real conditions.

[0093] The reflectance and spectral index of multiple moisture-sensitive bands under real conditions are used to perform nested cross-validation on the prediction results of the moisture prediction model. The moisture prediction model is optimized through the nested cross-validation results to obtain the optimal moisture prediction model. The details are as follows:

[0094] Based on the reflectance data and spectral index of moisture-sensitive bands, a random forest regression model was established using nested cross-validation to predict soil moisture content (SMC), canopy equivalent water thickness (CEWT) and soil factor (ps oil), and to evaluate the rationality of band selection for soil and vegetation moisture monitoring.

[0095] The outer cross-validation of the nested cross-validation randomly divides the data into 10 folds, retains 1 fold as the test set each time, and the remaining 9 folds are used as training sets to evaluate the optimal model selected in the inner cross-validation; the inner cross-validation divides the training set of the outer cross-validation into 10 folds to optimize and select the hyperparameters of the random forest regression model (water prediction model). Figure 6 shown.

[0096] Through independent inner and outer layer validation, information leakage in the model selection process is reduced, providing a more accurate estimate of the model's generalization ability.

[0097] The predicted performance for CEWT and psoil in the examples is as follows:

[0098] Table 4 Prediction performance of CEWT and psoil

[0099]

[0100] in, is the mean of the coefficient of determination of the 10 nested cross-validation prediction results, RMSE ncv and nRMSE ncv They represent the root mean square error and normalized root mean square error means of 10 nested cross validation results.

[0101] The prediction results show that the combination of simulated reflectance with noise and spectral index can well predict the three moisture parameters, which indicates that the band selection for soil and vegetation moisture monitoring in the present invention is theoretically reasonable. The rationality of the band selection is verified based on a real dataset of soil and vegetation canopy reflectance spectra.

[0102] Table 5 Dataset source

[0103] type Dataset name Sample size Prediction target Collection area soil Bablet_2016 106 SMC Tunisia, France soil Dupiau_2020 72 SMC Chile, Tunisia, United States, China soil Humper_2015 455 SMC Tunisia soil Lesaignoux_2008 190 SMC France soil Liu_2002 367 SMC France soil Lobell_2002 41 SMC USA soil Marcq_2012 258 SMC Germany, China, Tunisia soil Philpot_2014 405 SMC USA vegetation ISRAEL_2004 207 CEWT Israel vegetation ISRAEL_2005 124 CEWT Israel

[0104] The soil reflectance spectral dataset used to verify the rationality of band selection covers many countries and regions around the world and has good representativeness. The eight soil datasets were merged to predict SMC to verify the rationality of band selection; the vegetation canopy reflectance spectral dataset is the canopy reflectance spectrum of wheat during its entire growth period, providing dry weight (g / m 2 ), moisture content (%) and leaf area index (m 2 / m 2 ) and other parameters, as follows:

[0105]

[0106] Among them, WC is the leaf water content of the plant, DW is the dry weight of the leaf per unit area, and W f is the fresh weight of leaves, W d is the leaf dry weight, and A is the plant sampling area. The prediction target of the vegetation canopy reflectance spectral dataset is CEWT, whose value is equal to the product of LAI and EWT. EWT can be calculated by the following formula:

[0107]

[0108] Among them, ρ w is the density of water. Since some samples in the vegetation canopy reflectance spectrum dataset lack measured LAI data, the number of samples is only 150 and 96. The target variable is predicted by the modeling method to verify the rationality of the band selection. The results are shown in the table:

[0109] The prediction results show that the combination of reflectance and spectral index constructed based on the real data set can well predict CWET and SMC, which indicates that the band selection for soil and vegetation moisture monitoring in the present invention also has high performance in actual use.

[0110] Table 6 Verifies the rationality of band selection

[0111]

[0112] Table 7 Texture index formula

[0113]

[0114] For the actual collected image data, the texture index of each band and vegetation index grayscale image can be further calculated. The texture index can be extracted by the gray level co-occurrence matrix method, and the calculation formula is shown in Table 7.

[0115] Where: i and j represent the grayscale values ​​at the i-th row and j-th column of the image respectively; P(i,j) represents the probability of the corresponding grayscale value at the i-th row and j-th column in the matrix; N is the grayscale of the image; Mean is the mean of the grayscale co-occurrence matrix, σ i and σ j is the standard deviation.

[0116] The embodiments in this description provide a selection basis and a moisture prediction method for using an airborne multispectral moisture-sensitive band combination, and achieve accurate and efficient inversion of soil and crop moisture information in agricultural areas by establishing a machine learning inversion model.

[0117] Based on the above method, the present invention provides a moisture prediction system based on airborne multi-spectral sensitive band combination, including: a selection module, a model building module and a prediction module.

[0118] Among them, the selection module is used to select the central wavelengths of multiple moisture-sensitive bands based on the response characteristics of the soil and plant canopy reflectance spectra to moisture, and according to the physical mechanism of the O-H bond vibration in water molecules, increase and select the central wavelengths of multiple different moisture-sensitive bands, and determine the bandwidth of all central wavelengths; the model construction module is used to simulate and generate the reflectance spectra of soil and vegetation canopy, generate a spectral response function using the relationship between the central wavelengths and bandwidths of multiple different moisture-sensitive bands, convert the simulated reflectance spectra into multiple moisture-sensitive bands through the spectral response function, obtain the reflectance and spectral index of multiple moisture-sensitive bands, and construct a moisture prediction model using the reflectance and spectral index; the prediction module is used to combine all moisture-sensitive bands with a determined bandwidth and integrate them into the airborne multispectral sensor, collect the reflectance spectral images of the soil and vegetation canopy at the selected location, and input the reflectance spectral images into the moisture prediction model to predict the moisture of the soil and plant canopy at the selected location.

[0119] The present invention also provides a computer device, including a memory and a processor, wherein a program is stored in the memory, and when the program is executed by the processor, the processor executes the steps of a moisture prediction method based on an airborne multi-spectral sensitive band combination.

[0120] In accordance with the disclosed embodiments, a computing device may communicate with one or more external devices (e.g., keyboards, pointing devices, Bluetooth communications, etc.), or with any device (e.g., routers, modems, etc.) that enables a computing device to communicate with one or more other computing devices.

[0121] The present invention also provides a storage medium on which a computer program is stored. When the computer program is executed by a processor, the steps of a moisture prediction method based on an airborne multi-spectral sensitive band combination are implemented.

[0122] According to the disclosed embodiments, the storage medium may be a non-volatile computer-readable storage medium, such as but not limited to: a portable computer disk, a hard disk, a random access memory (RAM), a read-only memory (ROM), an erasable programmable read-only memory (EPROM or flash memory), a portable compact disk read-only memory (CD-ROM), an optical storage device, a magnetic storage device, or any suitable combination thereof. In the present invention, the storage medium may be any tangible medium containing or storing a program that may be used by or in conjunction with an instruction execution system, apparatus, or device.

[0123] The above content is a further detailed description of the present invention in combination with a specific preferred embodiment. For technicians in the technical field to which the present invention belongs, several simple deductions or substitutions can be made without departing from the concept of the present invention, which should be regarded as belonging to the protection scope of the present invention.

Claims

1. A moisture prediction method based on airborne multi-spectral sensitive band combination, characterized in that: The steps include: Based on the response characteristics of soil and plant canopy reflectance spectra to moisture, the central wavelengths of multiple moisture-sensitive bands are selected. According to the physical mechanism of O-H bond vibration in water molecules, the central wavelengths of multiple different moisture-sensitive bands are added and selected, and the bandwidths of all central wavelengths are determined. Simulate and generate the reflectance spectra of soil and vegetation canopy, generate a spectral response function using the relationship between the central wavelengths and bandwidths of the multiple different moisture-sensitive bands, convert the simulated reflectance spectra to multiple moisture-sensitive bands through the spectral response function, obtain the reflectance and spectral index of the multiple moisture-sensitive bands, and construct a moisture prediction model using the reflectance and spectral index; All moisture-sensitive bands with a determined bandwidth are combined and integrated into the airborne multispectral sensor to collect reflectance spectral images of the soil and vegetation canopy at the selected location. The reflectance spectral images are then input into the moisture prediction model to predict the moisture content of the soil and plant canopy at the selected location.

2. A moisture prediction method based on airborne multi-spectral sensitive band combination as claimed in claim 1, characterized in that: The simulation generates reflectance spectra of soil and vegetation canopy, including: The MARMIT2 radiation transfer model is used to simulate the soil reflectance spectrum in the solar domain, and the soil moisture content SMC of the simulated soil reflectance spectrum is normalized as the dimensionless soil factor psoil; The soil reflectance spectra with the lowest and highest water content are used as inputs to the PROSAIL radiative transfer model to simulate the reflectance spectra of the vegetation canopy. Among them, after the MARMIT2 radiation transfer model simulates the reflectance spectrum of the soil in the solar domain, the sigmoid function is used to fit the soil moisture content SMC representing the soil reflectance spectrum according to the product of the input water layer thickness and the surface water coverage; the specific expression is: BRF Soil (λ)=f MARMIT2 (λ,L,e,d); Among them, BRF Soil (λ) is soil reflectance, soil moisture content SMC characterizing soil reflectance spectrum, f MARMIT2 (·) is the MARMIT2 radiation transfer model operation function, λ is the wavelength, L is the water layer thickness, ε is the surface water coverage, and δ is the soil particle volume fraction; Among them, in the PROSAIL radiation transfer model, the equivalent water thickness is obtained, and the equivalent water thickness of the vegetation canopy and the normalized soil moisture content are represented by equivalent water thickness and psoil respectively; the specific expression is: CEWT=LAI×C w ; DHR Leaf (λ)=f PROSPECT (λ,θ DHR ); BRF Canopy (λ)=f SAIL (λ,DHR(λ),θ BRF ); Where CEWT is the canopy equivalent water thickness, LAI is the leaf area index, C w is the equivalent water thickness, DHR Leaf (λ) is the PROSPECT model f PROSPECT When the simulated wavelength is λ, the blade directional hemispherical reflectivity, θ DHR is the input parameter of PROSPECT model; BRF Canopy (λ) is the value obtained when the 4SAIL model is used. SAIL Coupling DHR Leaf (λ), the bidirectional reflectance of the canopy at the simulated wavelength λ, θ BRF are the input parameters of the 4SAIL model.

3. A moisture prediction method based on airborne multi-spectral sensitive band combination as claimed in claim 2, characterized in that: The relationship between the central wavelengths and bandwidths of the multiple different moisture-sensitive bands is used to generate a spectral response function, and the simulated reflection spectrum is converted to multiple moisture-sensitive bands through the spectral response function to obtain the reflectance and spectral index of the multiple moisture-sensitive bands, including: According to the central wavelength and bandwidth of the selected moisture sensitive band, the spectral response function is approximated by a Gaussian function; the specific expression is: Among them, SRF(λ) wavelength is the spectral index at λ, λ0 is the central wavelength, σ is the standard deviation, and the width of the spectral response function curve is controlled by the bandwidth FWMH calculation; The soil reflectance spectrum and the canopy reflectance spectrum are converted to obtain the selected multiple moisture-sensitive band combinations and the reflectance of the multiple moisture-sensitive band combinations; the specific expression is: Where R band is the reflectivity after resampling, λ1 and λ2 are the wavelength ranges of the band, and BRF(λ) is the reflectivity when the simulated wavelength is λ; Multiple moisture-sensitive bands are combined and input into the spectral response function to obtain the spectral index.

4. The moisture prediction method based on airborne multi-spectral sensitive band combination according to claim 1, characterized in that: Before inputting the reflectance spectrum image into the moisture prediction model, the method further includes: Collect real soil and vegetation canopy reflectance spectral data, and use spectral response functions to convert the real soil and vegetation canopy reflectance spectral data into multiple moisture-sensitive bands to obtain the reflectance and spectral index of multiple moisture-sensitive bands under real conditions; The reflectance and spectral index of moisture-sensitive bands under multiple real conditions are used to perform nested cross-validation on the prediction results of the moisture prediction model. The moisture prediction model is optimized through the nested cross-validation results to obtain the optimal moisture prediction model.

5. The moisture prediction method based on airborne multi-spectral sensitive band combination according to claim 4, characterized in that: The method uses the reflectance and spectral index of multiple moisture-sensitive bands under real conditions to perform nested cross-validation on the prediction results of the moisture prediction model, including: The outer cross-validation of the nested cross-validation randomly divides the data into 10 folds, retains 1 fold as the test set each time, and the remaining 9 folds as the training set to evaluate the optimal model selected in the inner cross-validation; the inner cross-validation divides the training set of the outer cross-validation into 10 folds to optimize and select the hyperparameters of the moisture prediction model.

6. The moisture prediction method based on airborne multi-spectral sensitive band combination according to claim 1, characterized in that: The central wavelengths of multiple water-sensitive bands are selected based on the response characteristics of soil and plant canopy reflectance spectra to water. According to the physical mechanism of O-H bond vibration in water molecules, the central wavelengths of multiple different water-sensitive bands are added and selected, including: Based on the response characteristics of soil and plant canopy reflectance spectra to moisture, 560nm, 705nm, 750nm, 800nm, 900nm, 970nm, 1100nm, 1200nm, 1650nm and 2200nm were selected as the central wavelengths of the moisture-sensitive bands; Based on the physical mechanism of O—H bond vibration in water molecules, 450 nm and 660 nm were selected as the central wavelengths of the moisture-sensitive band.

7. The moisture prediction method based on airborne multi-spectral sensitive band combination according to claim 1, characterized in that: The step of determining the bandwidth of all central wavelengths comprises: The spectral information and spatiotemporal resolution are balanced by selecting the bandwidth in the moisture-sensitive band, where the bandwidths in the selected moisture-sensitive band are 30nm, 30nm, 30nm, 10nm, 15nm, 40nm, 40nm, 10nm, 50nm, 50nm, 50nm and 100nm respectively.

8. A moisture prediction system based on airborne multi-spectral sensitive band combination, characterized in that: include: A selection module is used to select the central wavelengths of multiple moisture-sensitive bands based on the response characteristics of soil and plant canopy reflectance spectra to moisture, add and select the central wavelengths of multiple different moisture-sensitive bands based on the physical mechanism of O-H bond vibration in water molecules, and determine the bandwidth of all central wavelengths; A model building module, used to simulate and generate the reflectance spectra of soil and vegetation canopy, generate a spectral response function using the relationship between the central wavelengths and bandwidths of the multiple different moisture-sensitive bands, convert the simulated reflectance spectra to multiple moisture-sensitive bands through the spectral response function, obtain the reflectance and spectral index of the multiple moisture-sensitive bands, and build a moisture prediction model using the reflectance and spectral index; The prediction module is used to combine all moisture-sensitive bands of a certain bandwidth and integrate them into the airborne multispectral sensor, collect the reflectance spectral images of the soil and vegetation canopy at the selected location, and input the reflectance spectral images into the moisture prediction model to predict the moisture of the soil and plant canopy at the selected location.

9. A computer device, characterized in that: It comprises a memory and a processor, wherein a program is stored in the memory, and when the program is executed by the processor, the processor executes the steps of a moisture prediction method based on an airborne multi-spectral sensitive band combination as described in any one of claims 1 to 7.

10. A storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by a processor, the steps of a moisture prediction method based on an airborne multi-spectral sensitive band combination according to any one of claims 1 to 7 are implemented.

Images