CN110687053B - Regional organic matter content estimation method and device based on hyperspectral image - Google Patents
Regional organic matter content estimation method and device based on hyperspectral image Download PDFInfo
- Publication number
- CN110687053B CN110687053B CN201910950279.2A CN201910950279A CN110687053B CN 110687053 B CN110687053 B CN 110687053B CN 201910950279 A CN201910950279 A CN 201910950279A CN 110687053 B CN110687053 B CN 110687053B
- Authority
- CN
- China
- Prior art keywords
- dimensional
- organic matter
- spectrum
- model
- data
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Active
Links
Images
Classifications
-
- G—PHYSICS
- G01—MEASURING; TESTING
- G01N—INVESTIGATING OR ANALYSING MATERIALS BY DETERMINING THEIR CHEMICAL OR PHYSICAL PROPERTIES
- G01N21/00—Investigating or analysing materials by the use of optical means, i.e. using sub-millimetre waves, infrared, visible or ultraviolet light
- G01N21/17—Systems in which incident light is modified in accordance with the properties of the material investigated
- G01N21/25—Colour; Spectral properties, i.e. comparison of effect of material on the light at two or more different wavelengths or wavelength bands
-
- G—PHYSICS
- G01—MEASURING; TESTING
- G01N—INVESTIGATING OR ANALYSING MATERIALS BY DETERMINING THEIR CHEMICAL OR PHYSICAL PROPERTIES
- G01N21/00—Investigating or analysing materials by the use of optical means, i.e. using sub-millimetre waves, infrared, visible or ultraviolet light
- G01N21/17—Systems in which incident light is modified in accordance with the properties of the material investigated
- G01N21/47—Scattering, i.e. diffuse reflection
- G01N21/4738—Diffuse reflection, e.g. also for testing fluids, fibrous materials
Abstract
The application relates to a regional organic matter content estimation method and device based on hyperspectral images, soil organic matter data of certain places in a certain region and hyperspectral image data of the places are collected firstly, 9 one-dimensional spectrums and two-dimensional spectrums corresponding to the 9 one-dimensional spectrums are analyzed through data, a plurality of different models are respectively established according to the one-dimensional spectrums and/or the two-dimensional spectrums, fitting accuracy under four models of all the one-dimensional spectrums and/or the two-dimensional spectrums is compared, the spectrum with the highest accuracy and the used model are selected, and soil organic matter data of the remaining regions in the region are evaluated according to the selected spectrum and the model. Through the steps, the one-dimensional spectrum or two-dimensional spectrum type with the highest precision and the used model used for evaluation are screened out, and finally the estimation method with the highest precision for the organic matter content in the region can be obtained.
Description
Technical Field
The application belongs to the technical field of spectrum image application, and particularly relates to a regional organic matter content estimation method and device based on a hyperspectral image.
Background
Soil organic matter is an important index for evaluating soil fertility. The efficient monitoring of the content is a necessary condition and a necessary trend for effective management and utilization of land resources and fine operation of the planting industry. The traditional soil organic matter monitoring is mainly based on indoor measurement, and is complex to operate, time-consuming and labor-consuming. The characteristics of high efficiency, environmental protection and the like of the aerial and space remote sensing are increasingly widely applied to soil organic matter monitoring, but the defects of long image acquisition period and incomplete coverage restrict the application of the remote sensing in the field scale. Compared with the prior art, the advantages of rapidness, real-time performance, accuracy and no destruction of the airborne hyperspectral information acquisition are gradually highlighted [2], and the application of the method to monitoring of soil organic matters in field scale can certainly promote the progress of modern agriculture and promote the rapid development of precise agriculture.
In the 21 st century, Unmanned Aerial Vehicle (UAV) remote sensing technology developed rapidly and was applied to high-resolution soil survey and mapping on a field scale. Scholars at home and abroad develop a lot of researches on soil organic matters by using the remote sensing hyperspectral data of the unmanned aerial vehicle. The Peon J establishes a soil organic matter multivariate regression prediction model by using A Hyperspectral Scanner (AHS) as a data source of a vegetation coverage area in a mountainous area and obtains a good effect. The Xu SX and the like monitor the organic matters of the rice soil by using Vis-NIR spectra and SVMR in combination, the SVMR has great potential for detecting the organic matters of the rice soil, and the accuracy R2 of the established soil organic matter prediction model reaches 0.88. Qin Kai [6] researches an inversion model of soil organic matters based on reconstructed airborne hyperspectral data, and develops information extraction algorithm research of rock minerals and soil organic matters. Winsiana and the like acquire soil spectrum data of the karst region by utilizing an airborne hyperspectral imaging system, and more technical means are provided for rapidly, widely and real-timely monitoring the content of organic matters in the soil in the karst region. Zhangdonghui and the like acquire organic matter hyperspectral data of three river areas in Heilongjiang by a CASI-1500 aviation hyperspectral imaging system (ITRES in Canada), and the informatization level of soil information inversion is obviously improved.
The spectrum of organic matters in soil is influenced by soil self factors such as soil particle size, soil type, soil pH and the like to show different forms, and the soil organic matter monitoring model has larger difference and poor universality for different soil types. In the prior art, all the methods for estimating the organic matter content of a specific area exist, and no method for estimating the organic matter content of the land which can adapt to different areas exists.
Disclosure of Invention
The technical problem to be solved by the invention is as follows: in order to overcome the defects in the prior art, the regional organic matter content estimation method based on the hyperspectral image is provided.
The technical scheme adopted by the invention for solving the technical problems is as follows:
the invention provides a regional organic matter content estimation method based on hyperspectral images, which comprises the following steps of:
s1: the method comprises the steps of obtaining hyperspectral image data of a certain area, collecting soil organic matter data of certain places in the area, and screening out the hyperspectral image data of the places;
s2: extracting original reflectivity from the hyperspectral image data of the screened place, and performing 8 spectrum transformations of envelope curve, reciprocal, logarithm, first-order differential, second-order differential, reciprocal first-order differential, logarithmic first-order differential and reciprocal logarithm removal on the original reflectivity to obtain 9 one-dimensional spectra; extracting two wave bands from all the wave bands of the 9 kinds of one-dimensional spectrums, combining the two wave bands pairwise, and calculating a soil ratio index and a soil normalization index for the combined spectrums to obtain a two-dimensional spectrum of soil organic matters; the calculation formula is as follows:
RSI(λ1,λ2)＝Rλ1/Rλ2(1)
NDSI(λ1,λ2)＝(Rλ1-Rλ2)/(Rλ1+Rλ2)(2)
λ1、λ2is any two wave bands, wherein1-λ2≠0，Rλ1Reflectivity, R λ, corresponding to the respective band2The reflectivity corresponding to the corresponding wave band;
s3: respectively establishing PLSR, BPNN and SVM models according to the extraction result of the two-dimensional spectrum in the step S2;
s4: comparing the fitting data of the soil organic matter data of different two-dimensional spectrums under different models with the actual soil organic matter data collected in the step S1, calculating the fitting precision of the fitting data, and selecting the spectrum with the highest precision and the used model;
s5: and evaluating soil organic matter data of the rest area of the area by using the selected spectrum with the highest precision and the used model.
Preferably, in the method for estimating regional organic matter content based on hyperspectral image, in the step S3, models are respectively built for extraction results of 9 kinds of one-dimensional spectra;
and in the step S3, the fitting data of the soil organic matter data of all the different one-dimensional spectrums and the two-dimensional spectrums under different models are compared with the accuracy of the actual soil organic matter data collected in the step S1.
Preferably, in the hyperspectral image-based regional organic matter content estimation method, the sampling points are distributed by adopting a fishing net point distribution method in the selection of the place in the step S1.
Preferably, the method for estimating the organic matter content of the area based on the hyperspectral image, provided by the invention, comprises the steps of establishing PLSR, BPNN and SVM models respectively by taking a one-dimensional spectrum and/or a two-dimensional spectrum as input variables, and adjusting modeling parameters to achieve a required modeling effect;
the modeling precision and the inspection precision are determined by a coefficient R2And the root mean square error RMSE.
Preferably, the regional organic matter content estimation method based on hyperspectral image of the invention,
step S3, an MLR model is also established, the establishment of the MLR model is completed in SPSS software, 95% of the MLR model is set as a variable error characterization level to carry out variable selection and elimination, and models based on one-dimensional spectrums and/or two-dimensional spectrums are sequentially established;
when an MLR model is established, a plurality of organic matter response sensitive wave bands are screened out and used for establishing the MLR model.
The invention provides a regional organic matter content estimation device based on hyperspectral images, which comprises:
the data acquisition screening module: the hyperspectral image data acquisition system is used for acquiring hyperspectral image data of a certain area, acquiring soil organic matter data of certain places in the area and screening out the hyperspectral image data of the places;
a spectrum calculation module: the system is used for extracting an original reflectivity from the screened hyperspectral image data of the place, and performing 8 spectrum transformations of envelope curve, reciprocal, logarithm, first-order differential, second-order differential, reciprocal first-order differential, logarithm first-order differential and reciprocal logarithm on the original reflectivity to obtain 9 one-dimensional spectra; extracting two wave bands from all the wave bands of the 9 kinds of one-dimensional spectrums, combining the two wave bands pairwise, and calculating a soil ratio index and a soil normalization index for the combined spectrums to obtain a two-dimensional spectrum of soil organic matters; the calculation formula is as follows:
RSI(λ1,λ2)＝Rλ1/Rλ2(1)
NDSI(λ1,λ2)＝(Rλ1-Rλ2)/(Rλ1+Rλ2)(2)
λ1、λ2is any two wave bands, wherein1-λ2≠0，Rλ1Reflectivity, R λ, corresponding to the respective band2The reflectivity corresponding to the corresponding wave band;
a model building module: respectively establishing PLSR, BPNN and SVM models according to the extraction result of the two-dimensional spectrum in the spectrum calculation module;
a model screening module: comparing fitting data of soil organic matter data of different two-dimensional spectra under different models with actual soil organic matter data acquired in a data acquisition screening module, calculating fitting precision of the fitting data, and selecting a spectrum with highest precision and a used model;
an evaluation module: and evaluating soil organic matter data of the rest area of the area by using the selected spectrum with the highest precision and the used model.
Preferably, in the device for estimating the content of organic matters in the region based on the hyperspectral image, the model building module further builds models for the extraction results of the 9 kinds of one-dimensional spectra respectively;
and the model establishing module compares the fitting data of the soil organic matter data of all different one-dimensional spectrums and two-dimensional spectrums under different models with the precision of the actual soil organic matter data acquired by the data acquisition and screening module.
Preferably, according to the regional organic matter content estimation method based on the hyperspectral image, the sampling points are distributed by adopting a fishing net point distribution method for selecting the places in the data acquisition and screening module.
Preferably, the device for estimating the organic matter content of the hyperspectral image-based region of the invention,
the PLSR, the BPNN and the SVM models are respectively established by taking the one-dimensional spectrum and/or the two-dimensional spectrum as input variables and adjusting modeling parameters to achieve the required modeling effect;
the modeling precision and the inspection precision are determined by a coefficient R2And the root mean square error RMSE.
Preferably, the device for estimating the regional organic matter content based on the hyperspectral image also comprises an MLR model, wherein the MLR model is established in SPSS software, 95% of the MLR model is set as a variable error representation level to perform variable selection and elimination, and models based on one-dimensional spectrums and/or two-dimensional spectrums are sequentially established;
when an MLR model is established, a plurality of organic matter response sensitive wave bands are screened out and used for establishing the MLR model.
The invention has the beneficial effects that:
the invention provides a regional organic matter content estimation method and device based on a hyperspectral image, which are characterized by firstly acquiring soil organic matter data of certain places in a certain region and hyperspectral image data of the places, analyzing 9 one-dimensional spectrums and two-dimensional spectrums corresponding to the 9 one-dimensional spectrums through the data, respectively establishing four models of MLR, PLSR, BPNN and SVM according to the one-dimensional spectrums and/or the two-dimensional spectrums, comparing the fitting accuracy of the four models of all the one-dimensional spectrums and/or the two-dimensional spectrums, selecting the spectrum with the highest accuracy and the used model, and evaluating the soil organic matter data of the rest regions in the region according to the selected spectrum and the model. Through the steps, the one-dimensional spectrum or two-dimensional spectrum type with the highest precision and the used model used for evaluation are screened out, and finally the estimation method with the highest precision for the organic matter content in the region can be obtained.
Drawings
The technical solution of the present application is further explained below with reference to the drawings and the embodiments.
FIG. 1 is a diagram showing the location of an investigation region in an embodiment of the present application;
FIG. 2 is a distribution diagram of modeling points and sampling points in an embodiment of the present application;
FIG. 3 is an MLR, PLSR, BPNN, SVM modeling effect based on one-dimensional spectra in an effect embodiment of the present application;
FIG. 4 is an MLR, PLSR, BPNN, SVM modeling effect based on RSI two-dimensional spectrum in the effect embodiment of the present application;
FIG. 5 is an MLR, PLSR, BPNN, SVM modeling effect based on NDSI two-dimensional spectra in an effect embodiment of the present application;
FIG. 6a is a diagram of a test area in an embodiment of the present application;
FIG. 6b is a graph showing the test results in the working example of the present application;
fig. 7 is a diagram of a result of verifying an estimated value by a measured value in an embodiment of the present application;
fig. 8 is a flowchart of the regional organic matter content estimation method based on hyperspectral images according to the present application.
Detailed Description
It should be noted that the embodiments and features of the embodiments in the present application may be combined with each other without conflict.
The technical solutions of the present application will be described in detail below with reference to the accompanying drawings in conjunction with embodiments.
Example 1
The embodiment provides a regional organic matter content estimation method based on hyperspectral images, as shown in fig. 1, comprising the following steps:
s1: acquiring Hyperspectral Image data of a certain area (a spectral Image with spectral resolution within the range of 10l magnitude order is called a Hyperspectral Image), acquiring soil organic matter data of certain places (sampling points) in the area (the sampling points can be distributed by adopting a fishing net point distribution method), and screening the Hyperspectral Image data of the places;
s2: extracting original Reflectivity (RAW) from the screened hyperspectral image data of the place, and performing 8 spectral transformations of envelope Curve (CR), reciprocal (IR), Logarithm (LR), first order differential (FDR), second order differential (SDR), reciprocal first order differential (IFDR), logarithmic first order differential (LFDR) and reciprocal logarithm (ILR) on the original reflectivity to obtain 9 one-dimensional spectra; extracting two wave bands from all the wave bands of the 9 kinds of one-dimensional spectrums, combining the two wave bands pairwise, and calculating a soil ratio index (RSI) and a soil normalization index (NDSI) for the combined spectrums to obtain a two-dimensional spectrum of soil organic matters; the calculation formula is as follows:
RSI(λ1,λ2)＝Rλ1/Rλ2 (1)
NDSI(λ1,λ2)＝(Rλ1-Rλ2)/(Rλ1+Rλ2) (2)
λ1、λ2is any two wave bands, wherein1-λ2≠0，Rλ1Reflectivity (original reflectivity or spectral conversion reflectivity) corresponding to the corresponding band, R lambda2A reflectance (original reflectance or spectrally transformed reflectance) corresponding to the respective band;
s3: respectively establishing four models of MLR, PLSR, BPNN and SVM (if the calculation is saved, the models can be directly selected from the three models of PLSR, BPNN and SVM, the total number of the 4 models is 108 according to different spectra) according to the extraction results of the 9 one-dimensional spectra and the two-dimensional spectra in the step S2; generally, the accuracy of the model of the one-dimensional spectrum is necessarily lower than that of the two-dimensional spectrum, and therefore, the four models of MLR, PLSR, BPNN and SVM can also be respectively established directly according to the extraction result of the two-dimensional spectrum (the establishment of the models belongs to the prior art, and is not described herein again);
s4: comparing the fitting data of the soil organic matter data of different one-dimensional spectrums and two-dimensional spectrums under different models with the actual soil organic matter data collected in the step S1, calculating the fitting precision of the fitting data, and selecting the spectrum with the highest precision and the used model;
s5: and evaluating soil organic matter data of the rest area of the area by using the selected spectrum with the highest precision and the used model.
Generally speaking, the one-dimensional spectrum transformation is difficult to screen out the organic matter response sensitive band influenced by the spectrum information of other substances, the two-dimensional spectra (RSI, NDSI) can remove redundant information, effectively amplify the weak correlation between the organic matter and the spectra, strengthen the interaction relation between the organic matter and the spectra, and quickly extract the organic matter response sensitive band (the sensitive band of the one-dimensional spectra is between 410 and 440nm, and the sensitive band of the two-dimensional spectra (RSI, NDSI) is 500 and 550nm, 900 nm).
Example 2
The embodiment provides a regional organic matter content estimation device based on hyperspectral image, including:
the data acquisition screening module: the hyperspectral image data acquisition system is used for acquiring hyperspectral image data of a certain area, acquiring soil organic matter data of certain places in the area and screening out the hyperspectral image data of the places;
a spectrum calculation module: the system is used for extracting an original reflectivity from the screened hyperspectral image data of the place, and performing 8 spectrum transformations of envelope curve, reciprocal, logarithm, first-order differential, second-order differential, reciprocal first-order differential, logarithm first-order differential and reciprocal logarithm on the original reflectivity to obtain 9 one-dimensional spectra; extracting two wave bands from all the wave bands of the 9 kinds of one-dimensional spectrums, combining the two wave bands pairwise, and calculating a soil ratio index and a soil normalization index for the combined spectrums to obtain a two-dimensional spectrum of soil organic matters; the calculation formula is as follows:
RSI(λ1,λ2)＝Rλ1/Rλ2 (1)
NDSI(λ1,λ2)＝(Rλ1-Rλ2)/(Rλ1+Rλ2) (2)
λ1、λ2is any two wave bands, wherein1-λ2≠0，Rλ1Reflectivity, R λ, corresponding to the respective band2The reflectivity corresponding to the corresponding wave band;
a model building module: respectively establishing four models of MLR, PLSR, BPNN and SVM according to the extraction result of the two-dimensional spectrum in the spectrum calculation module;
a model screening module: comparing fitting data of soil organic matter data of different two-dimensional spectra under different models with actual soil organic matter data acquired in a data acquisition screening module, calculating fitting precision of the fitting data, and selecting a spectrum with highest precision and a used model;
an evaluation module: and evaluating soil organic matter data of the rest area of the area by using the selected spectrum with the highest precision and the used model.
The model establishing module is also used for respectively establishing an MLR model, a PLSR model, a BPNN model and an SVM model according to the extraction results of the 9 kinds of one-dimensional spectra (if the model is used for saving calculation power, the model can be directly selected from the PLSR model, the BPNN model and the SVM model);
and the model establishing module compares the fitting data of the soil organic matter data of all different one-dimensional spectrums and two-dimensional spectrums under different models with the precision of the actual soil organic matter data acquired by the data acquisition and screening module.
And selecting the place in the data acquisition and screening module and laying sampling points by adopting a fishing net point laying method.
The establishment of the MLR model is completed in SPSS software, 95% of the MLR model is set as a variable error characterization level to carry out variable selection and elimination, and models based on one-dimensional spectrum and/or two-dimensional spectrum are sequentially established;
when an MLR model is established, a plurality of organic matter response sensitive wave bands are screened out and used for establishing the MLR model.
The PLSR, the BPNN and the SVM models are respectively established by taking the one-dimensional spectrum and/or the two-dimensional spectrum as input variables and adjusting modeling parameters to achieve the required modeling effect;
the modeling precision and the inspection precision are determined by a coefficient R2And the root mean square error RMSE.
Effects of the embodiment
Effect example a shaze town of suzhou city was selected as a research area (see fig. 1) and hyperspectral data and soil organic matter data were collected. The research area is located in a subtropical humid monsoon climate area, the annual average precipitation is 800 mm-1500 mm, the climate is warm and humid, and the research area is suitable for rice growth. The soil is mostly silt clay soil, the particles are fine, the porosity is high, precipitation infiltration is facilitated, the bottom soil layer clay soil and the silt clay soil have an effect of intercepting infiltration precipitation, the water and fertilizer retention performance is good, and the soil is ideal soil for rice planting.
Airborne hyperspectral data are acquired by a DJ Mpro600 unmanned aerial vehicle carrying a GaiaSky-Mini 2-VN airborne hyperspectral imaging system (parameters of the hyperspectral imaging system are shown in Table 1). In order to reduce the interference of the external environment on the soil spectrum, the method comprises the following steps of selecting a soil sample with clear weather and small wind level, namely 12 days 1 month and 25 days: 30-13: data acquisition is performed on the study area 30. 60 waypoints are set on the collection day, the flight is carried out for 2 times, the lateral coverage rate is 70 percent, and the flight height is 300 meters. The spatial resolution is 0.073 m, and the maximum spectral resolution can reach 3.2 nm. The obtained hyperspectral data are subjected to lens correction, reflectivity calibration and atmospheric correction by using SpecView in sequence, and the splicing of 60-scene images is completed by using HiSpectraLStitcher.
TABLE 1 GaiaSky-Mini 2-VN airborne imaging sensor parameters
Soil organic matter data acquisition
59 sampling points are distributed in a research area by adopting a fishing net distribution method, and after sampling points of non-bare soil are removed in actual sampling, the total number of the sampling points is 45 (see figure 2). Removing straws and soil surface layers, collecting surface soil with the depth of 0-10cm from 5 points by a plum blossom sampling method, mixing to obtain a soil sample, collecting 1Kg of each soil sample, and storing in a dark place.
The collected soil sample is subjected to pretreatment such as freeze-drying, grinding, sieving and the like, inorganic carbon is digested by hydrochloric acid, organic carbon is measured by a German Jena Multi NC3100TOC analyzer, and the measurement result is multiplied by 1.724 soil organic matter conversion coefficient to obtain the soil organic matter content. Compared with the traditional potassium dichromate volumetric method, the method greatly reduces human errors, shortens the determination time and improves the determination efficiency.
Modeling set inspection set partitioning
The 45 sample total set of the study area was randomly divided into 32 modeling sets, 13 testing sets using spss.22 (see fig. 2). The total set, the modeling set and the inspection set mean values are more than 2 percent (see table 2), all reach the national third-level soil fertility standard, the soil organic matter content in the research area is high, and the response of iron or manganese in the soil to the soil spectrum can be covered. The variation coefficients of the total set, the modeling set and the inspection set are all larger than 2.4% (see table 2), so that the phenomenon that the precision of estimating the content of the organic matters in the soil by the spectral reflectivity is low due to the fact that the variation coefficient of the content of the organic matters in the soil sample is too small is avoided. The modeling set and the inspection set are reasonably divided, and the method can be used for estimating and researching the content of the organic matters in the soil.
TABLE 2 organic matter characterization statistics for sampling points in research area
One-dimensional spectrum and two-dimensional spectrum extraction
The original reflectivity of 176 wave bands of 45 sampling points is subjected to 8 kinds of transformations such as envelope Curve (CR), reciprocal (IR), Logarithm (LR), first order differential (FDR), second order differential (SDR), reciprocal first order differential (IFDR), logarithmic first order differential (LFDR), reciprocal logarithm (ILR) and the like by utilizing Excel and origin.2018, and a one-dimensional spectrum is obtained. The correlation analysis of the one-dimensional spectrum with organic matter was performed in spss.22.
The extraction of two-dimensional characteristic spectra was done in matlab.7 and Excel. And combining the one-dimensional spectrums of all the wave bands pairwise by utilizing Matlab.7, and calculating a soil ratio index (RSI) and a soil normalization index (NDSI) of the combined spectrums to obtain a two-dimensional spectrum of the soil organic matter.
The calculation formula is as follows:
RSI(λ1,λ2)＝Rλ1/Rλ (1)
NDSI(λ1,λ2)＝(Rλ1-Rλ2)/(Rλ1+Rλ2) (2)
λ1、λ2is any two wave bands, wherein1-λ2≠0，Rλ1Reflectivity (original reflectivity or spectral conversion reflectivity) corresponding to the corresponding band, R lambda2A reflectance (original reflectance or spectrally transformed reflectance) corresponding to the respective band;
and (3) calculating a correlation coefficient between the two-dimensional spectrum and the organic matter, wherein the correlation coefficient is realized in Matlab.7 through a Corrcoef function.
The screening of one-dimensional and two-dimensional characteristic spectra is completed in Excel. In order to reduce unnecessary difference factors between one-dimensional modeling and two-dimensional modeling, increase comparability of modeling effect and persuasiveness of result, the number of input variables for one-dimensional modeling and two-dimensional modeling is unified (see table 3). 10 organic matter response sensitive wave bands are sequentially screened out from the original and 8 transformed one-dimensional and two-dimensional spectrums and used for establishing a multiple linear regression Model (MLR). And simultaneously screening out two-dimensional characteristic spectra with the same number as the full-waveband one-dimensional spectra as input variables of a Partial Least Squares (PLSR), a Back Propagation Neural Network (BPNN) and a Support Vector Machine (SVM) modeling method.
TABLE 3 number of model input variables
Model building and inspection
Establishing a multiple regression (MLR) and Partial Least Squares (PLSR) linear model and a Back Propagation Neural Network (BPNN) nonlinear model of soil organic matter hyperspectral prediction and a Support Vector Machine (SVM) nonlinear model.
The establishment of the MLR model is completed in SPSS.22, 95% of the MLR model is set as a variable error characterization level to carry out variable selection and elimination, and a one-dimensional characteristic spectrum model and a two-dimensional characteristic spectrum model based on the original reflectivity and 8 mathematical transformations of the original reflectivity are sequentially established. The PLSR, the BPNN and the SVM models are respectively established by taking the full-waveband one-dimensional spectrum and the two-dimensional characteristic spectrum as input variables and achieving the required modeling effect through adjustment of modeling parameters. The modeling precision and the inspection precision are determined by a coefficient R2And root mean square error RMSEEvaluation was made of R2RMSE formula is as follows:
(1) and (2) in the formula
All represent the predicted value of organic matter of the ith sampling point, and the formula is shown in (1)
The average value of measured values of organic matter is shown in the formulae (1) and (2)iAnd (4) showing the measured value of the organic matter of the ith sampling point. R2The larger the RMSE, the higher the modeling accuracy and the more stable the model.
Based on the original reflectivity and the extraction results of the one-dimensional spectrum and the two-dimensional spectrum (RSI and NDSI) feature spectra of 8 kinds of transformation, four models (shown in fig. 4-5) of MLR, PLSR, BPNN and SVM are respectively established, the number of the models is 108 in total, wherein BPNN obtains higher modeling precision and inspection precision in the one-dimensional and two-dimensional (RSI and NDSI) spectrum modeling, and the overall modeling effect is better. The NDSI two-dimensional spectrum modeling test precision corresponding to the SDR spectrum transformation is low, the modeling effect is poor, the NDSI two-dimensional spectrum modeling effect corresponding to the SDR spectrum transformation is far inferior to that corresponding to the SDR spectrum transformation, the result is consistent with the two-dimensional characteristic spectrum extraction result, and the correlation between the NDSI two-dimensional spectrum corresponding to the SDR spectrum transformation and organic matters is smaller than that of the RSI two-dimensional spectrum. Compared with a one-dimensional spectrum, the modeling precision and the inspection precision of a linear model (MLR, PLSR) are improved to a greater extent in the two-dimensional spectrum, wherein the linear model (MLR, PLSR) established based on the RSI two-dimensional spectrum has the best effect, directly because the correlation between the RSI two-dimensional characteristic spectrum and organic matters is larger than the correlation between the NDSI two-dimensional characteristic spectrum and the organic matters. The degree of correlation between the visible characteristic spectrum and the organic matter is closely related to the modeling effect, and proper spectrum transformation is the basis for establishing a high-quality model. Besides LFDR transformation, the SVM model modeling precision and the detection precision of different spectrum transformations have larger difference in one-dimensional and two-dimensional spectra and poorer stability. The method shows that the original reflectivity can accurately position the characteristic wave band of organic matter response after LR and FDR combined transformation, the fitting degree between the organic matter and the spectrum is improved, and LFDR is the optimal spectrum variable of the soil organic matter prediction model in the research area. The modeling accuracy and inspection accuracy of two-dimensional (RSI, NDSI) spectra are overall higher than for one-dimensional spectra. The Ratio (RSI) two-dimensional spectrum modeling effect is superior to the Normalization (NDSI) two-dimensional spectrum modeling effect, and the established model has better robustness.
In order to quantitatively evaluate the modeling quality of different methods, an inspection precision error (an absolute value of a difference between model inspection precision and modeling precision) is defined as a measurement index (see table 4). Compared with a one-dimensional spectrum, the detection precision error of the MLR, PLSR and BPNN model established based on the two-dimensional spectrum is reduced to some extent, wherein the detection precision error of the MLR model is obviously reduced and reaches 0.25. The MLR modeling method is characterized in that MLR modeling input variables are few, one-dimensional characteristic spectrum information is single, part of hidden characteristic spectra are omitted, and internal relations among the characteristic spectra are omitted. The model established by the two-dimensional spectrum is superior to the model established by the one-dimensional spectrum in the aspects of applicability and stability.
The correlation analysis between the raw reflectance and the 8 spectral transforms and the organic matter is shown in table 4. The correlation between the spectrum and the organic matter after LFDR and IFDR conversion is better, the correlation at the most sensitive wave band is more than 0.5, wherein the correlation between the IFDR and the organic matter is up to 0.561 at most, and is improved by 0.16 compared with the optimal correlation between the RAW characteristic spectrum and the organic matter; the reflectivity after FDR, SDR, IFDR and LFDR transformation and the correlation variation coefficient between organic matters are greatly improved, so that the visible differential transformation can effectively decompose overlapped peaks between wave bands, the spectral characteristic difference between samples is enlarged, and the spectral sensitivity is improved.
TABLE 4 correlation between one-dimensional spectra and organic matter
Selecting a fallow paddy field (shown in figure 6a) in Wujiang district, Suzhou city as a test area, collecting airborne hyperspectral image data in 3 months and 4 days in 2019 in sunny weather, and simultaneously collecting 8 soil samples to determine the organic matter content of the soil samples for verification of research results.
The application of the model in the field is realized through the Bandmath function of Envi5.3, and the estimation result of the soil organic matter is shown in figure 6 b. The organic matter content of the field is estimated to be 18.521-29.514 g/Kg from fig. 7 without the non-field area, wherein the organic matter content of most of the area exceeds 20g/Kg, and the organic matter content is relatively rich. The organic matter content and the estimation result of the actual measurement points are shown in figure 7, the verification precision (R2) is 0.813, the model estimation effect is good, the verification Root Mean Square Error (RMSE) is 0.764, the model stability is good, the relative analysis error (RPD) is 1.983, and the model is reliable.
In light of the foregoing description of the preferred embodiments according to the present application, it is to be understood that various changes and modifications may be made without departing from the spirit and scope of the invention. The technical scope of the present application is not limited to the contents of the specification, and must be determined according to the scope of the claims.
As will be appreciated by one skilled in the art, embodiments of the present application may be provided as a method, system, or computer program product. Accordingly, the present application may take the form of an entirely hardware embodiment, an entirely software embodiment or an embodiment combining software and hardware aspects. Furthermore, the present application may take the form of a computer program product embodied on one or more computer-usable storage media (including, but not limited to, disk storage, CD-ROM, optical storage, and the like) having computer-usable program code embodied therein.
The present application is described with reference to flowchart illustrations and/or block diagrams of methods, apparatus (systems), and computer program products according to embodiments of the application. It will be understood that each flow and/or block of the flow diagrams and/or block diagrams, and combinations of flows and/or blocks in the flow diagrams and/or block diagrams, can be implemented by computer program instructions. These computer program instructions may be provided to a processor of a general purpose computer, special purpose computer, embedded processor, or other programmable data processing apparatus to produce a machine, such that the instructions, which execute via the processor of the computer or other programmable data processing apparatus, create means for implementing the functions specified in the flowchart flow or flows and/or block diagram block or blocks.
These computer program instructions may also be stored in a computer-readable memory that can direct a computer or other programmable data processing apparatus to function in a particular manner, such that the instructions stored in the computer-readable memory produce an article of manufacture including instruction means which implement the function specified in the flowchart flow or flows and/or block diagram block or blocks.
These computer program instructions may also be loaded onto a computer or other programmable data processing apparatus to cause a series of operational steps to be performed on the computer or other programmable apparatus to produce a computer implemented process such that the instructions which execute on the computer or other programmable apparatus provide steps for implementing the functions specified in the flowchart flow or flows and/or block diagram block or blocks.
Claims (6)
1. A regional organic matter content estimation method based on hyperspectral images is characterized by comprising the following steps:
s1: the method comprises the steps of obtaining hyperspectral image data of a certain area, collecting soil organic matter data of certain places in the area, and screening out the hyperspectral image data of the places;
s2: extracting original reflectivity from the hyperspectral image data of the screened place, and performing 8 spectrum transformations of envelope curve, reciprocal, logarithm, first-order differential, second-order differential, reciprocal first-order differential, logarithmic first-order differential and reciprocal logarithm removal on the original reflectivity to obtain 9 one-dimensional spectra; extracting two wave bands from all the wave bands of the 9 kinds of one-dimensional spectrums, combining the two wave bands pairwise, and calculating a soil ratio index and a soil normalization index for the combined spectrums to obtain a two-dimensional spectrum of soil organic matters; the calculation formula is as follows:
RSI(λ1，λ2)＝Rλ1/Rλ2 (1)
NDSI(λ1，λ2)＝(Rλ1-Rλ2)/(Rλ1+Rλ2) (2)
wherein λ is1、λ2Is any two wave bands, wherein1-λ2≠0，Rλ1Reflectivity, R λ, corresponding to the respective band2The reflectivity corresponding to the corresponding wave band;
s3: respectively establishing MLR, PLSR, BPNN and SVM models according to the extraction result of the two-dimensional spectrum in the step S2, specifically:
sequentially screening 10 organic response sensitive wave bands for establishing an MLR model from the original and 8 transformed one-dimensional and two-dimensional spectrums; meanwhile, screening out two-dimensional characteristic spectra with the same number as the full-waveband one-dimensional spectra as input variables for modeling PLSR, BPNN and SVM;
the establishment of the MLR model is completed in SPSS.22, 95% of the MLR model is set as a variable error representation level to carry out variable selection and elimination, and a one-dimensional characteristic spectrum model and a two-dimensional characteristic spectrum model based on the original reflectivity and 8 mathematical transformations of the original reflectivity are sequentially established;
the PLSR, the BPNN and the SVM models are respectively established by taking the full-waveband one-dimensional spectrum and the two-dimensional characteristic spectrum as input variables and adjusting modeling parameters to achieve the required modeling effect; the modeling precision and the inspection precision are determined by a coefficient R2Evaluating two indexes of the error RMSE;
s4: comparing the fitting data of the soil organic matter data of different two-dimensional spectrums under different models with the actual soil organic matter data collected in the step S1, calculating the fitting precision of the fitting data, and selecting the spectrum with the highest precision and the used model;
s5: and evaluating soil organic matter data of the rest area of the area by using the selected spectrum with the highest precision and the used model.
2. The hyperspectral image-based regional organic matter content estimation method according to claim 1, wherein in the step S3, the extraction results of 9 one-dimensional spectra are respectively modeled;
and in the step S3, the fitting data of the soil organic matter data of all the different one-dimensional spectrums and the two-dimensional spectrums under different models are compared with the accuracy of the actual soil organic matter data collected in the step S1.
3. The regional organic matter content estimation method based on hyperspectral image according to claim 1 or 2, wherein the place is selected in the step S1 by arranging sampling points by using a fishing net point arrangement method.
4. A regional organic matter content estimation device based on hyperspectral image is characterized by comprising:
the data acquisition screening module: the hyperspectral image data acquisition system is used for acquiring hyperspectral image data of a certain area, acquiring soil organic matter data of certain places in the area and screening out the hyperspectral image data of the places;
a spectrum calculation module: the system is used for extracting an original reflectivity from the screened hyperspectral image data of the place, and performing 8 spectrum transformations of envelope curve, reciprocal, logarithm, first-order differential, second-order differential, reciprocal first-order differential, logarithm first-order differential and reciprocal logarithm on the original reflectivity to obtain 9 one-dimensional spectra;
extracting two wave bands from all the wave bands of the 9 kinds of one-dimensional spectrums, combining the two wave bands pairwise, and calculating a soil ratio index and a soil normalization index for the combined spectrums to obtain a two-dimensional spectrum of soil organic matters; the calculation formula is as follows:
RSI(λ1，λ2)＝Rλ1/Rλ2 (1)
NDSI(λ1，λ2)＝(Rλ1-Rλ2)/(Rλ1+Rλ2) (2)
wherein λ is1、λ2Is any two wave bands, wherein1-λ2≠0，Rλ1Reflectivity, R λ, corresponding to the respective band2The reflectivity corresponding to the corresponding wave band;
a model building module: respectively establishing PLSR, BPNN and SVM models according to the extraction results of the two-dimensional spectrum in the spectrum calculation module, and sequentially screening 10 organic response sensitive wave bands for establishing an MLR model according to the original and 8 transformed one-dimensional and two-dimensional spectrums; meanwhile, screening out two-dimensional characteristic spectra with the same number as the full-waveband one-dimensional spectra as input variables for modeling PLSR, BPNN and SVM;
the establishment of the MLR model is completed in SPSS.22, 95% of the MLR model is set as a variable error representation level to carry out variable selection and elimination, and a one-dimensional characteristic spectrum model and a two-dimensional characteristic spectrum model based on the original reflectivity and 8 mathematical transformations of the original reflectivity are sequentially established;
the PLSR, the BPNN and the SVM models are respectively established by taking the full-waveband one-dimensional spectrum and the two-dimensional characteristic spectrum as input variables and adjusting modeling parameters to achieve the required modeling effect; the modeling precision and the inspection precision are determined by a coefficient R2Evaluating two indexes of the error RMSE;
a model screening module: comparing fitting data of soil organic matter data of different two-dimensional spectra under different models with actual soil organic matter data acquired in a data acquisition screening module, calculating fitting precision of the fitting data, and selecting a spectrum with highest precision and a used model;
an evaluation module: and evaluating soil organic matter data of the rest area of the area by using the selected spectrum with the highest precision and the used model.
5. The device for estimating the content of organic matters in the hyperspectral image based area according to claim 4, wherein the model building module further builds models for the extraction results of the 9 one-dimensional spectra respectively;
and the model establishing module compares the fitting data of the soil organic matter data of all different one-dimensional spectrums and two-dimensional spectrums under different models with the precision of the actual soil organic matter data acquired by the data acquisition and screening module.
6. The device for estimating the organic matter content of the hyperspectral image-based area according to claim 4 or 5 is characterized in that the selection of the places in the data acquisition and screening module adopts a fishing net point arrangement method to arrange sampling points.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201910950279.2A CN110687053B (en) | 2019-10-08 | 2019-10-08 | Regional organic matter content estimation method and device based on hyperspectral image |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201910950279.2A CN110687053B (en) | 2019-10-08 | 2019-10-08 | Regional organic matter content estimation method and device based on hyperspectral image |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN110687053A CN110687053A (en) | 2020-01-14 |
| CN110687053B true CN110687053B (en) | 2021-03-02 |
Family
ID=69111565
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201910950279.2A Active CN110687053B (en) | 2019-10-08 | 2019-10-08 | Regional organic matter content estimation method and device based on hyperspectral image |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN110687053B (en) |
Families Citing this family (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN112268866A (en) * | 2020-11-06 | 2021-01-26 | 苏州科技大学 | Two-dimensional hyperspectral data-based paddy soil humic acid content calculation method and device |
| CN113420412B (en) * | 2021-05-26 | 2023-05-09 | 南京信息工程大学 | Soil organic carbon content continuous depth distribution extraction method based on imaging spectrum |
| CN113959973A (en) * | 2021-08-18 | 2022-01-21 | 北京工业大学 | Drug API prediction method based on DWI-BPLS |
| CN114740180A (en) * | 2022-04-07 | 2022-07-12 | 中山大学 | Soil organic carbon estimation method and device based on multi-source remote sensing data |
Family Cites Families (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN108801934A (en) * | 2018-04-10 | 2018-11-13 | 安徽师范大学 | A kind of modeling method of soil organic carbon EO-1 hyperion prediction model |
| CN109540811A (en) * | 2018-12-25 | 2019-03-29 | 核工业北京地质研究院 | A method of black earth nutrient content precision of prediction is improved by spectrum transform |
-
2019
- 2019-10-08 CN CN201910950279.2A patent/CN110687053B/en active Active
Also Published As
| Publication number | Publication date |
|---|---|
| CN110687053A (en) | 2020-01-14 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN110687053B (en) | Regional organic matter content estimation method and device based on hyperspectral image | |
| Kamal et al. | Assessment of multi-resolution image data for mangrove leaf area index mapping | |
| Malenovský et al. | Retrieval of spruce leaf chlorophyll content from airborne image data using continuum removal and radiative transfer | |
| Lu et al. | Prediction of soil properties using laboratory VIS–NIR spectroscopy and Hyperion imagery | |
| Schlerf et al. | Remote sensing of forest biophysical variables using HyMap imaging spectrometer data | |
| Schlerf et al. | Retrieval of chlorophyll and nitrogen in Norway spruce (Picea abies L. Karst.) using imaging spectroscopy | |
| Thapa et al. | Potential of high-resolution ALOS–PALSAR mosaic texture for aboveground forest carbon tracking in tropical region | |
| Muñoz et al. | Soil carbon mapping using on-the-go near infrared spectroscopy, topography and aerial photographs | |
| An et al. | Hyperspectral field estimation and remote-sensing inversion of salt content in coastal saline soils of the Yellow River Delta | |
| Heenkenda et al. | Quantifying mangrove chlorophyll from high spatial resolution imagery | |
| Abuelgasim et al. | Evaluation of national and global LAI products derived from optical remote sensing instruments over Canada | |
| CN114018833B (en) | Method for estimating heavy metal content of soil based on hyperspectral remote sensing technology | |
| Houborg et al. | Adapting a regularized canopy reflectance model (REGFLEC) for the retrieval challenges of dryland agricultural systems | |
| Yue et al. | Method for accurate multi-growth-stage estimation of fractional vegetation cover using unmanned aerial vehicle remote sensing | |
| CN112395808A (en) | Biomass remote sensing mapping method combining random forest and collaborative kriging | |
| CN109934109B (en) | Remote sensing-based method for extracting forest and grass vegetation information in loess plateau water and soil loss area | |
| Jiang et al. | BESS-STAIR: a framework to estimate daily, 30 m, and all-weather crop evapotranspiration using multi-source satellite data for the US Corn Belt | |
| CN115235997A (en) | Soil texture inversion method based on satellite hyperspectral image | |
| Qiao et al. | Estimating maize LAI by exploring deep features of vegetation index map from UAV multispectral images | |
| CN113436153A (en) | Method for predicting carbon components of undisturbed soil profile based on hyperspectral imaging and support vector machine technology | |
| CN114814167B (en) | Soil heavy metal content inversion method fusing multi-source environment variables and spectral information | |
| CN111783288A (en) | Inversion method for soil salinity of yellow river delta based on Landsat8 | |
| Hang et al. | Estimation of chlorophyll-a concentration in Lake Taihu from Gaofen-1 wide-field-of-view data through a machine learning trained algorithm | |
| Liu et al. | Improving artificial forest biomass estimates using afforestation age information from time series Landsat stacks | |
| CN113466143B (en) | Soil nutrient inversion method, device, equipment and medium |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| PB01 | Publication | ||
| PB01 | Publication | ||
| SE01 | Entry into force of request for substantive examination | ||
| SE01 | Entry into force of request for substantive examination | ||
| GR01 | Patent grant | ||
| GR01 | Patent grant |