CN112630161A - Remote sensing inversion model and method for total organic carbon content of overground part of rice unit area based on K nearest neighbor regression algorithm - Google Patents
Remote sensing inversion model and method for total organic carbon content of overground part of rice unit area based on K nearest neighbor regression algorithm Download PDFInfo
- Publication number
- CN112630161A CN112630161A CN202011382700.3A CN202011382700A CN112630161A CN 112630161 A CN112630161 A CN 112630161A CN 202011382700 A CN202011382700 A CN 202011382700A CN 112630161 A CN112630161 A CN 112630161A
- Authority
- CN
- China
- Prior art keywords
- rice
- carbon content
- organic carbon
- total organic
- unit area
- 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.)
- Pending
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
- G01N21/55—Specular reflectivity
-
- 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
-
- G—PHYSICS
- G01—MEASURING; TESTING
- G01S—RADIO DIRECTION-FINDING; RADIO NAVIGATION; DETERMINING DISTANCE OR VELOCITY BY USE OF RADIO WAVES; LOCATING OR PRESENCE-DETECTING BY USE OF THE REFLECTION OR RERADIATION OF RADIO WAVES; ANALOGOUS ARRANGEMENTS USING OTHER WAVES
- G01S17/00—Systems using the reflection or reradiation of electromagnetic waves other than radio waves, e.g. lidar systems
- G01S17/88—Lidar systems specially adapted for specific applications
-
- G—PHYSICS
- G01—MEASURING; TESTING
- G01S—RADIO DIRECTION-FINDING; RADIO NAVIGATION; DETERMINING DISTANCE OR VELOCITY BY USE OF RADIO WAVES; LOCATING OR PRESENCE-DETECTING BY USE OF THE REFLECTION OR RERADIATION OF RADIO WAVES; ANALOGOUS ARRANGEMENTS USING OTHER WAVES
- G01S7/00—Details of systems according to groups G01S13/00, G01S15/00, G01S17/00
- G01S7/48—Details of systems according to groups G01S13/00, G01S15/00, G01S17/00 of systems according to group G01S17/00
- G01S7/4802—Details of systems according to groups G01S13/00, G01S15/00, G01S17/00 of systems according to group G01S17/00 using analysis of echo signal for target characterisation; Target signature; Target cross-section
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/21—Design or setup of recognition systems or techniques; Extraction of features in feature space; Blind source separation
- G06F18/214—Generating training patterns; Bootstrap methods, e.g. bagging or boosting
-
- 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
- G01N2021/1793—Remote sensing
Landscapes
- Engineering & Computer Science (AREA)
- Physics & Mathematics (AREA)
- General Physics & Mathematics (AREA)
- Life Sciences & Earth Sciences (AREA)
- Immunology (AREA)
- Computer Networks & Wireless Communication (AREA)
- Remote Sensing (AREA)
- Radar, Positioning & Navigation (AREA)
- Pathology (AREA)
- Data Mining & Analysis (AREA)
- General Health & Medical Sciences (AREA)
- Biochemistry (AREA)
- Theoretical Computer Science (AREA)
- Health & Medical Sciences (AREA)
- Chemical & Material Sciences (AREA)
- Analytical Chemistry (AREA)
- Artificial Intelligence (AREA)
- General Engineering & Computer Science (AREA)
- Bioinformatics & Cheminformatics (AREA)
- Evolutionary Computation (AREA)
- Bioinformatics & Computational Biology (AREA)
- Evolutionary Biology (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Electromagnetism (AREA)
- Investigating Or Analysing Materials By Optical Means (AREA)
Abstract
The invention provides a remote sensing inversion model of total organic carbon content of the overground part of a unit area of rice based on a K-nearest neighbor regression algorithm, which is a K-nearest neighbor regression model of Python language and further provides model parameters of the K-nearest neighbor regression model. The remote sensing inversion method of the total organic carbon content of the overground part of the rice unit area based on the K nearest neighbor regression algorithm is also provided. The rice unit area overground part total organic carbon content remote sensing inversion model based on the K nearest neighbor regression algorithm can quickly and accurately acquire the rice unit area overground part total organic carbon content information, overcomes the difficulty that the characteristic wave band of the rice unit area overground part total organic carbon content is difficult to determine due to the spectrum superposition effect caused by complex rice components, greatly improves the accuracy of the rice unit area overground part total organic carbon content inversion model, and has the advantages of ingenious design, simple and convenient calculation, easy realization and low cost.
Description
Technical Field
The invention relates to the technical field of agricultural remote sensing, in particular to the technical field of measurement of total organic carbon content of the overground part of a unit area of rice, and specifically relates to a remote sensing inversion model and method for the total organic carbon content of the overground part of the unit area of rice based on a K nearest neighbor regression algorithm.
Background
The total organic carbon content of the overground part of the rice refers to the content of carbon element in organic matters in the overground part of the rice, is an important index reflecting the carbon-nitrogen metabolism of the rice, the growth vigor of the rice and the physiological state of the rice, and reflects the influence of the physiological and growth vigor of the rice and the external management measures of the fertility and water on the rice (Schelier red, Yanglin, Fanhui. the rice nitrogen content and carbon-nitrogen ratio spectrum estimation based on the carbon-nitrogen metabolism [ J ]. the crop science and newspaper, 2006,32(3): 430-.
The method has the advantages that the total organic carbon content of the overground part of the rice is monitored, the yield and the quality of rice production can be guaranteed, the physiological state and the growth activity of the rice can be dynamically managed, management measures such as fertilizer water and the like are effectively implemented, the carbon-nitrogen metabolism of the rice is adjusted, the use amount of nitrogen fertilizers and irrigation water in the rice production is reduced, the environmental problems caused by the application of a large amount of nitrogen fertilizers and the water for rice production are relieved, and remarkable economic and social benefits (Chengqing, Tianyongtao, Yaxia and the like) are generated. The traditional method for monitoring the total organic carbon content of the overground part of the rice mainly adopts a destructive sampling method, needs to be measured indoors, wastes time and labor, has poor timeliness, cannot timely acquire the total organic carbon content of the overground part of the rice, and is not beneficial to popularization and application.
In the physiological and biochemical processes of rice, the change of certain specific substances and cell structures in rice plants results in the change of rice reflectance spectra. Therefore, the change of the spectrum can be used for acquiring the rice growth information such as the total organic carbon content of the overground part of the rice (Zhoudouqin. monitoring of the nitrogen nutrition and the quality of grains of the rice based on the canopy reflection spectrum [ D ]. Nanjing agriculture university, 2007). Currently, hyperspectrum is used for monitoring the growth state of rice in crop growth monitoring. With the development and popularization of the spectrum technology, the hyperspectral data can quickly and rapidly acquire the information of the total organic carbon content of the overground part of the rice, and the hyperspectral data becomes a consensus of more and more rice production practitioners and researchers. The most common mode is to use a portable full-wave-band spectrometer to obtain rice growth information and select a characteristic wave band capable of reflecting the total organic carbon content of the overground part to construct an inversion model. In the process of constructing the rice overground part total organic carbon content inversion model, the spectral range measured by the full-waveband spectrometer covers 350-2500 nm, but because the components of the rice are complex, the characteristic wave bands of the component spectra are partially overlapped, the determination of the characteristic spectrum of the rice overground part total organic carbon content is difficult, and meanwhile, the rapid processing of hyperspectral data becomes an urgent technical problem to be solved for estimating the rice overground part total organic carbon content based on the hyperspectral data.
Therefore, it is desirable to provide a remote sensing inversion model of the total organic carbon content of the overground part of the unit area of the rice, which can quickly and accurately acquire the information of the total organic carbon content of the overground part of the unit area of the rice, overcome the difficulty that the characteristic waveband of the total organic carbon content of the overground part of the unit area of the rice is difficult to determine due to the spectrum superposition effect caused by complex rice components, and greatly improve the accuracy of the inversion model of the total organic carbon content of the overground part of the unit area of the rice.
Disclosure of Invention
In order to overcome the defects in the prior art, one object of the present invention is to provide a K-nearest neighbor regression algorithm-based remote sensing inversion model of total organic carbon content in the upper part of a rice unit area, which can quickly and accurately obtain the information of the total organic carbon content in the upper part of the rice unit area, overcome the difficulty that the characteristic waveband of the total organic carbon content in the upper part of the rice unit area is difficult to determine due to the spectrum superposition effect caused by complex rice components, greatly improve the accuracy of the inversion model of the total organic carbon content in the upper part of the rice unit area, and is suitable for large-scale popularization and application.
The invention also aims to provide a remote sensing inversion model of the total organic carbon content of the overground part of the unit area of the rice based on the K nearest neighbor regression algorithm, which has the advantages of ingenious design, simple and convenient calculation, easy realization and low cost, and is suitable for large-scale popularization and application.
The invention also aims to provide a K nearest neighbor regression algorithm-based remote sensing inversion method for the total organic carbon content of the overground part of the rice in unit area, which can quickly and accurately acquire the information of the total organic carbon content of the overground part of the rice in unit area, overcome the difficulty that the characteristic waveband of the total organic carbon content of the overground part of the rice in unit area is difficult to determine due to the spectrum superposition effect caused by complex rice components, greatly improve the inversion precision of the total organic carbon content of the overground part of the rice in unit area, and is suitable for large-scale popularization and application.
The invention also aims to provide a remote sensing inversion method of the total organic carbon content of the overground part of the unit area of the rice based on the K nearest neighbor regression algorithm, which has the advantages of ingenious design, simple and convenient operation and low cost, and is suitable for large-scale popularization and application.
In order to achieve the above object, in a first aspect of the present invention, there is provided a K-nearest neighbor regression algorithm-based remote sensing inversion model of total organic carbon content in the upper part of a rice unit area, which is characterized in that the K-nearest neighbor regression algorithm-based remote sensing inversion model of total organic carbon content in the upper part of a rice unit area is a K-nearest neighbor regression model in Python language, and model parameters of the K-nearest neighbor regression model are as follows: 'leaf _ size' 56, 'metric' 9, 'n _ neighbors' 8, 'algorithm' auto 'and' weights 'uniform'.
Preferably, the K-nearest neighbor regression model is trained by using a rice data set, the data set includes canopy reflectances of m sampling points of the rice and a total organic carbon content of an overground part in a unit area, the m sampling points are uniformly distributed in a rice planting area, and the canopy reflectivity is a canopy reflectivity of n characteristic bands.
More preferably, m is 36, n characteristic bands are 2151 characteristic bands, and the 2151 characteristic bands are from 350nm band to 2500nm band.
In a second aspect of the invention, the invention provides a remote sensing inversion method of total organic carbon content of overground parts of a unit area of rice based on a K nearest neighbor regression algorithm, which is characterized by comprising the following steps:
(1) measuring the canopy reflectance of the rice;
(2) measuring the total organic carbon content of the overground part of the unit area of the rice;
(3) calculating by using the canopy reflectivity as input data and adopting a K nearest neighbor regression model of Python language to obtain an inversion value, and calculating a decision coefficient R according to the inversion value and the total organic carbon content of the overground part of the unit area2Changing the value of the model parameter, R, of the K nearest neighbor regression model2The larger the change of the model parameter is, the greater the importance of the model parameter is, the model parameter is arranged from large to small according to the importance to construct a model parameter tuning rank matrix;
(4) training the K nearest neighbor regression model by taking the canopy reflectivity as the input data and the total organic carbon content of the overground part in unit area as an output result, and sequentially tuning the model parameters according to the model parameter tuning order matrix to obtain tuning values of the model parameters;
(5) the canopy reflectivity is used as the input data, the total organic carbon content of the overground part in unit area is used as the output result, the model parameter tuning value is adopted to train the K neighbor regression model, after the training of the K neighbor regression model is finished, the remote sensing inversion model of the total organic carbon content of the overground part in unit area of rice based on the K neighbor regression algorithm is obtained, the remote sensing inversion model of the total organic carbon content of the overground part in unit area of rice based on the K neighbor regression algorithm is stored by using a save method, if the remote sensing inversion model of the total organic carbon content of the overground part in unit area of rice based on the K neighbor regression algorithm is needed, the remote sensing inversion model of the total organic carbon content of the overground part in unit area of rice based on the K neighbor regression algorithm is loaded by using a load method for use.
Preferably, in the step (1), the measurement is performed by using a hyperspectral radiometer, the measurement time is 10: 00-14: 00, the hyperspectral radiometer adopts a lens with a 25-degree field angle, a sensor probe of the portable field hyperspectral radiometer vertically points to the canopy of the rice and has a vertical height of 1 m from the top layer of the canopy, the ground field range diameter of the sensor probe is 0.44 m, the sensor probe faces the sunlight, the measurement is corrected by using a standard board, and the standard board is a standard white board with a reflectivity of 95% -99%.
Preferably, in the step (2), the step of measuring the total organic carbon content per unit area of the rice specifically includes:
collecting the overground part plants of the rice, deactivating enzyme, drying to constant weight to obtain dry plants, crushing the dry plants, determining the total organic carbon content of the overground part, and converting the total organic carbon content of the overground part into the total organic carbon content of the overground part in unit area according to a sampling area.
More preferably, in the step (2), the water-removing temperature is 105 ℃, the water-removing time is 20-30 minutes, the drying temperature is 80-90 ℃, and the determination of the total organic carbon content of the overground part adopts a potassium dichromate volumetric method.
Preferably, in the step (3), the model parameter tuning rank matrix is:
Params＝{'leaf_size','metric','p','n_neighbors','algorithm','weights'}。
preferably, in the step (4), the optimized values of the model parameters are:
'leaf_size'＝56,'metric'＝'euclidean','p'＝9,'n_neighbors'＝8,'algorithm'＝'auto','weights'＝'uniform'。
preferably, in the step (1), the step of measuring the canopy reflectance of the rice is specifically to measure the canopy reflectance of m sampling points of a rice planting area, the m sampling points are uniformly distributed in the rice planting area, and the canopy reflectance is the canopy reflectance of n characteristic bands; in the step (2), the step of measuring the total organic carbon content per unit area of the above-ground parts of the rice is to measure the total organic carbon content per unit area of the m sampling points.
More preferably, in the step (1), the m is 36, the n characteristic bands are 2151 characteristic bands, and the 2151 characteristic bands are from 350nm to 2500 nm.
The invention has the following beneficial effects:
1. the remote sensing inversion model of the total organic carbon content of the overground part of the rice unit area based on the K neighbor regression algorithm is a K neighbor regression model of Python language, and the model parameters of the K neighbor regression model are as follows: 'leaf _ size' 56, 'metric' euclidean ', p' 9, 'n _ neighbors' 8, 'algorithm' auto 'and' weights 'uniform' are tested, and R is tested2Above 0.85, the method can quickly and accurately acquire the information of the total organic carbon content of the overground part of the unit area of the rice, overcome the difficulty that the characteristic waveband of the total organic carbon content of the overground part of the unit area of the rice is difficult to determine due to the spectrum superposition effect caused by complex rice components, greatly improve the accuracy of an inversion model of the total organic carbon content of the overground part of the unit area of the rice, and is suitable for large-scale popularization and application.
2. The remote sensing inversion model of the total organic carbon content of the overground part of the rice unit area based on the K neighbor regression algorithm is a K neighbor regression model of Python language, and the model parameters of the K neighbor regression model are as follows: 'leaf _ size' 56, 'metric' euclidean ', p' 9, 'n _ neighbors' 8, 'algorithm' auto 'and' weights 'uniform' are tested, and R is tested2Above 0.85, therefore, the method has the advantages of ingenious design, simple and convenient calculation, easy realization and low cost, and is suitable for large-scale popularization and application.
3. K nearest neighbor regression-based methodThe remote sensing inversion method of the total organic carbon content of the overground part of the unit area of the rice by the algorithm comprises the following steps: measuring the canopy reflectance of the rice; measuring the total organic carbon content of the overground part of the unit area of the rice: taking the reflectivity of the canopy as input data, calculating by adopting a K nearest neighbor regression model of Python language to determine a coefficient R2Constructing a model parameter tuning order matrix; training a K nearest neighbor regression model by taking the reflectivity of the canopy as input data and the total organic carbon content of the overground part in unit area as an output result, and sequentially tuning model parameters according to a model parameter tuning order matrix to obtain tuning values of the model parameters; training a K-nearest neighbor regression model by using the canopy reflectivity as input data and the total organic carbon content of the overground part in unit area as an output result and adopting the adjusted values of model parameters to obtain a remote sensing inversion model of the total organic carbon content of the overground part in unit area of rice based on the K-nearest neighbor regression algorithm, inspecting the model, and performing R2Above 0.85, therefore, the method can quickly and accurately acquire the information of the total organic carbon content of the overground part of the unit area of the rice, overcome the difficulty that the characteristic waveband of the total organic carbon content of the overground part of the unit area of the rice is difficult to determine due to the spectrum superposition effect caused by complex rice components, greatly improve the inversion precision of the total organic carbon content of the overground part of the unit area of the rice, and is suitable for large-scale popularization and application.
4. The invention discloses a remote sensing inversion method of total organic carbon content of overground parts in unit area of rice based on a K nearest neighbor regression algorithm, which comprises the following steps: measuring the canopy reflectance of the rice; measuring the total organic carbon content of the overground part of the unit area of the rice: taking the reflectivity of the canopy as input data, calculating by adopting a K nearest neighbor regression model of Python language to determine a coefficient R2Constructing a model parameter tuning order matrix; training a K nearest neighbor regression model by taking the reflectivity of the canopy as input data and the total organic carbon content of the overground part in unit area as an output result, and sequentially tuning model parameters according to a model parameter tuning order matrix to obtain tuning values of the model parameters; training a K nearest neighbor regression model by using the canopy reflectivity as input data and the total organic carbon content of the overground part in unit area as an output result and adopting the adjusted values of model parameters to obtain a rice unit based on a K nearest neighbor regression algorithmRemote sensing inversion model of total organic carbon content in the overground part of the area, and inspecting the model, R2Above 0.85, therefore, the design is ingenious, the operation is simple and convenient, the cost is low, and the method is suitable for large-scale popularization and application.
These and other objects, features and advantages of the present invention will become more fully apparent from the following detailed description, the accompanying drawings and the claims, and may be realized by means of the instrumentalities, devices and combinations particularly pointed out in the appended claims.
Drawings
FIG. 1 is a schematic flow chart of a specific embodiment of the remote sensing inversion method of the total organic carbon content of the overground part of a rice unit area based on the K-nearest neighbor regression algorithm.
FIG. 2 is a schematic diagram of a model building process of the embodiment shown in FIG. 1.
FIG. 3 is a graph showing the results of the model test of the embodiment shown in FIG. 1, wherein the measured values and the predicted values are in% by weight.
Detailed Description
The invention provides a remote sensing inversion model of total organic carbon content of overground parts of rice unit area based on a K neighbor regression algorithm, aiming at the requirement of estimating the total organic carbon content of the overground parts of rice unit area based on hyperspectrum, overcoming the difficulties that the characteristic wave band of the total organic carbon content of the overground parts of rice unit area is difficult to determine and the characteristic wave band of hyperspectral data is time-consuming and labor-consuming in screening because of complex components of rice, the remote sensing inversion model of the total organic carbon content of the overground parts of rice unit area based on the K neighbor regression algorithm is a K neighbor regression model of Python language, and the model parameters of the K neighbor regression model are as follows: 'leaf _ size' 56, 'metric' 9, 'n _ neighbors' 8, 'algorithm' auto 'and' weights 'uniform'.
The K-nearest neighbor regression model may be trained by using any suitable data set, and preferably, the K-nearest neighbor regression model is trained by using a data set of rice, the data set includes canopy reflectivities of m sample points of the rice and total organic carbon content of overground parts per unit area, the m sample points are uniformly distributed in a rice planting area, and the canopy reflectivity is a canopy reflectivity of n characteristic bands. The rice planting area can be a plurality of ecological points and a plurality of varieties of rice planting areas.
M and n are positive integers, which can be determined according to needs, and more preferably, m is 36, n characteristic bands are 2151 characteristic bands, and the 2151 characteristic bands are from 350nm to 2500 nm.
The invention also provides a remote sensing inversion method of the total organic carbon content of the overground part of the rice unit area based on the K nearest neighbor regression algorithm, which comprises the following steps:
(1) measuring the canopy reflectance of the rice;
(2) measuring the total organic carbon content of the overground part of the unit area of the rice;
(3) calculating by using the canopy reflectivity as input data and adopting a K nearest neighbor regression model of Python language to obtain an inversion value, and calculating a decision coefficient R according to the inversion value and the total organic carbon content of the overground part of the unit area2Changing the value of the model parameter, R, of the K nearest neighbor regression model2The larger the change of the model parameter is, the greater the importance of the model parameter is, the model parameter is arranged from large to small according to the importance to construct a model parameter tuning rank matrix;
(4) training the K nearest neighbor regression model by taking the canopy reflectivity as the input data and the total organic carbon content of the overground part in unit area as an output result, and sequentially tuning the model parameters according to the model parameter tuning order matrix to obtain tuning values of the model parameters;
(5) the canopy reflectivity is used as the input data, the total organic carbon content of the overground part in unit area is used as the output result, the model parameter tuning value is adopted to train the K neighbor regression model, after the training of the K neighbor regression model is finished, the remote sensing inversion model of the total organic carbon content of the overground part in unit area of rice based on the K neighbor regression algorithm is obtained, the remote sensing inversion model of the total organic carbon content of the overground part in unit area of rice based on the K neighbor regression algorithm is stored by using a save method, if the remote sensing inversion model of the total organic carbon content of the overground part in unit area of rice based on the K neighbor regression algorithm is needed, the remote sensing inversion model of the total organic carbon content of the overground part in unit area of rice based on the K neighbor regression algorithm is loaded by using a load method for use.
In the step (1), the measurement may be performed by any suitable spectrometer and method, preferably, in the step (1), the measurement is performed by using a hyperspectral radiometer, the measurement time is 10:00 to 14:00, the hyperspectral radiometer uses a lens with a field angle of 25 degrees, a sensor probe of the portable field hyperspectral radiometer vertically points to the canopy of the rice and has a vertical height of 1 meter from the top layer of the canopy, the ground field range diameter of the sensor probe is 0.44 meter, the sensor probe faces the sun, the measurement is corrected by using a standard board, and the standard board is a standard white board with a reflectivity of 95% to 99%.
In the step (2), the step of measuring the total organic carbon content per unit area of the rice may specifically include any suitable method, and preferably, in the step (2), the step of measuring the total organic carbon content per unit area of the rice specifically includes:
collecting the overground part plants of the rice, deactivating enzyme, drying to constant weight to obtain dry plants, crushing the dry plants, determining the total organic carbon content of the overground part, and converting the total organic carbon content of the overground part into the total organic carbon content of the overground part in unit area according to a sampling area.
In the step (2), the water-removing and the drying can adopt any suitable conditions, and the total organic carbon content in the overground part can be measured by any suitable method, and preferably, in the step (2), the water-removing temperature is 105 ℃, the water-removing time is 20-30 minutes, the drying temperature is 80-90 ℃, and the total organic carbon content in the overground part is measured by a potassium dichromate volumetric method.
In the step (3), the model parameter tuning rank matrix is based onDetermining the coefficient R2Determining, preferably, in the step (3), that the model parameter tuning rank matrix is:
Params＝{'leaf_size','metric','p','n_neighbors','algorithm','weights'}。
in the step (4), the tuning values of the model parameters are sequentially determined according to the model parameter tuning rank matrix, and more preferably, in the step (4), the tuning values of the model parameters are:
'leaf_size'＝56,'metric'＝'euclidean','p'＝9,'n_neighbors'＝8,'algorithm'＝'auto','weights'＝'uniform'。
in order to improve the accuracy of the remote sensing inversion model of the total organic carbon content of the overground part of the rice in unit area based on the K-nearest neighbor regression algorithm, a plurality of sampling points of a rice planting area can be selected, and the canopy reflectivity of a plurality of characteristic wave bands of the sampling points and the total organic carbon content of the overground part of the unit area of the sampling points are measured, preferably, in the step (1), the step of measuring the canopy reflectivity of the rice is specifically to measure the canopy reflectivity of m sampling points of the rice planting area, the m sampling points are uniformly distributed in the rice planting area, and the canopy reflectivity is the canopy reflectivity of n characteristic wave bands; in the step (2), the step of measuring the total organic carbon content per unit area of the above-ground parts of the rice is to measure the total organic carbon content per unit area of the m sampling points.
In the step (1), m and n are positive integers, which can be determined as required, and more preferably, in the step (1), m is 36, the n characteristic bands are 2151 characteristic bands, and the 2151 characteristic bands are from 350nm to 2500 nm.
The invention will be further illustrated with reference to the following specific examples. It should be understood that these examples are for illustrative purposes only and are not intended to limit the scope of the present invention.
Examples
The K-nearest neighbor regression algorithm-based remote sensing inversion method for the total organic carbon content of the overground part of the rice unit area is based on actually measured hyperspectral data, and rice canopy reflectance spectrum data and total organic carbon content data of the overground part of the rice unit area which are collected by a rice planting area (a rice and wheat planting base in Huaian area of agricultural science research institute of Huaian city, Jiangsu province, the rice variety is No. 5, and the sampling period is the rice jointing period) are adopted, so that 48 sampling points are totally distributed, and are uniformly distributed and completely cover the whole area of the rice planting area. The data of 48 sampling points are divided into two parts by a random method, wherein the data of 36 sampling points is used for model construction, and the data of 12 sampling points is used for model inspection. The flow of the remote sensing inversion method of the total organic carbon content of the overground part of the rice unit area based on the K nearest neighbor regression algorithm is shown in figure 1, and the method comprises the following steps:
1. and (4) performing spectral measurement.
The rice canopy spectrum measurement is carried out by using a field Spec Pro portable field hyperspectral radiometer produced by American ASD in clear weather, no wind or small wind speed within the time range of 10: 00-14: 00, and the sampling testers wear dark clothes to reduce the influence or interference on the spectrometer. During sampling, a lens with a 25-degree field angle is selected, a sensor probe vertically points to a measurement target, namely a canopy, the vertical height of the sensor probe is about 1 meter from the top layer of the canopy, the diameter of the ground field range is 0.44 meter, the average value of reflection spectra measured for 10 times is taken as the spectral data of the sampling point. And in the measurement process, the standard white board is corrected before and after the measurement of each sampling point. If the distribution of the environmental light field changes in the test process, the standard white board is also corrected, and the reflectivity of the standard white board used in the embodiment is 99%. The measured spectrum data is checked by using field Spec Pro portable field hyperspectral radiometer random software RS3 or ViewSpec Pro software, abnormal spectrum files are removed, interpolation calculation is carried out on the spectrum data to obtain spectrum data with the range of 350 nm-2500 nm and the resolution of 1nm, the average value of parallel sampling spectra of the spectrum is calculated, and finally, the spectrum data is exported and stored as ASCII files.
2. Determination of total organic carbon content of overground part of unit area of rice
The overground part of rice plants of the spectral measurement field uniformly distributed at each sampling point was collected, the number of the plants was 6, the plants were wrapped with absorbent paper and brought back to the laboratory, the water was removed at 105 ℃ for 20 minutes, then the plants were dried at 85 ℃ until the weight was constant, the dried plants were pulverized, and the total organic Carbon Content (CC) of the overground part was determined by the potassium dichromate volumetric method (Liyukai. general analytical method for soil agricultural chemistry [ M ]. Beijing: Beijing scientific Press, 1983:79,272), the unit being (% by weight).
3. Model construction
The model construction is implemented by adopting a K nearest neighbor regression model of Python language, please refer to FIG. 2, and the model construction mainly comprises the following steps:
3.1 data verification
And checking the acquired rice canopy reflectivity data, and rejecting abnormal whole spectral curve data. The abnormal spectrum in the invention means that adjacent spectrum changes by more than 100%, and spectrum values including null values and negative values are included.
3.2 preprocessing of data
And preprocessing the verified rice canopy reflectivity data and the rice unit area overground part total organic carbon content data, including removing the paired rice canopy reflectivity data and rice unit area overground part total organic carbon content data containing the missing value and the null value.
3.3 partitioning of data sets
To ensure a reasonable evaluation of the model training and inversion results, a stochastic approach was used to divide the entire data set (48 groups) into two parts, with 75% (36 groups) of data used for model training and 25% (12 groups) for post-training effectiveness evaluation.
3.4 partitioning of training data sets
In order to ensure the effect of model training, a random method is used, and a training data set is divided into 5 parts to train the model when the model is trained and iterated every time.
3.5 construction of model parameter tuning rank matrix
In the invention, the tuning of the model parameters in the model training process is very important, and in order to ensure that the best model tuning is obtained as much as possible, a trial-and-error method is used for tuning the model parameters. The present invention uses the coefficient of determination R2(R2The closer to 1, the better) as the test parameter, a parameter rank matrix for evaluating the weight of the model parameter is constructed. According to a training data set, firstly, a default value of a model parameter is used for calculation to obtain an inversion value, and a decision coefficient R is calculated according to the inversion value and the total organic carbon content of the overground part in unit area2Then changing the value of the model parameter, R2The larger the change of the model parameter is, the greater the importance of the model parameter is, the model parameter is arranged from large to small according to the importance to construct a model parameter tuning rank matrix for subsequent calculation.
According to the canopy reflectivity data in the training data set and the corresponding data of the total organic carbon content of the overground part in the unit area, the model parameter tuning order matrix obtained by calculation is as follows:
Params＝{'leaf_size','metric','p','n_neighbors','algorithm','weights'}。
3.6 model construction
Adjusting the order matrix according to the obtained model parameters, modeling data including actually measured canopy reflectivity data and corresponding actually measured total organic carbon content data of the overground part of the unit area, taking the actually measured canopy reflectivity data as input data and the actually measured total organic carbon content data of the overground part of the unit area as output results, training a K nearest neighbor regression model, sequentially adjusting the model parameters according to the model parameter adjustment order matrix to obtain complete parameters and values of the model, wherein the data comprises the following data:
'leaf_size'＝56,'metric'＝'euclidean','p'＝9,'n_neighbors'＝8,'algorithm'＝'auto','weights'＝'uniform'。
after the model training is finished, the save method is used for saving the model, and if the model is required to be used, the load method is operated for loading and using.
3.7 model test
Using 12 sampling points except for the constructed model to input the hyperspectral data into the model, using the adjusted model parameters to calculate to obtain a predicted value, analyzing the relationship between the predicted value and an actual measured value (the total organic carbon content of the overground part in unit area), and obtaining a result shown in figure 3, wherein R of the model is2Is 0.8707. Model R using default parameters2Is 0.6332.
In the embodiment, Matlab software (version: R2020a 9.8.0.1380330) and Python (version:3.7.0) developed by MathWorks company in the United states are used for random division of training data and test data and construction, training and test of models, and the K-nearest neighbor regression model of Python is called through the Matlab software.
Therefore, the invention provides a new remote sensing inversion model of the total organic carbon content of the overground part of the rice in unit area based on K nearest neighbor regression algorithm based on actual measurement hyperspectral remote sensing data, the information of the total organic carbon content of the overground part of the rice can be rapidly and accurately obtained based on the actual measurement rice canopy reflectivity data and the data of the total organic carbon content of the overground part of the rice collected on the spot, the difficulty that the characteristic waveband of the total organic carbon content of the overground part of the rice is difficult to determine caused by the spectrum superposition effect caused by complex rice components is overcome, the model parameter is optimized by constructing a model parameter optimization order matrix, the model parameter is optimized by using a trial-and-error method, the phenomenon of overfitting of a linear model is effectively reduced, the inversion precision of the total organic carbon content of the overground part of the rice is greatly improved, and the model is suitable for quantitative inversion of the total organic carbon content of the overground, therefore, the carbon and nitrogen metabolism, the physiological state and the water and fertilizer supply state of the rice are obtained, the growth information acquisition efficiency in the rice cultivation and planting process is improved, and basic scientific data are provided for the water and fertilizer operation in rice production.
Compared with the prior art, the invention has the following advantages:
(1) the K-Nearest neighbor (KNN, K-near Neighbors) regression model used by the invention is suitable for the inversion of the total organic carbon content of the overground part of the rice based on the hyperspectrum, on the basis of comprehensively considering the information of the waveband range of 350-2500 nm of the hyperspectrum, the influence and superposition effect of various substance compositions and cell structures in the rice body, especially the influence and superposition effect of complex components on the characteristic waveband of the total organic carbon content of the overground part of the rice are considered, and the inversion of the total organic carbon content of the overground part of the rice is carried out by fully utilizing the total organic carbon content information of the overground part of the rice contained in different wavebands in remote sensing data.
(2) A machine learning algorithm of an extreme random tree is used, a model of 350-2500 nm reflectivity and total organic carbon content of the overground part of the rice is constructed, linear effect and nonlinear effect of high-resolution spectral data and the total organic carbon content of the overground part of the rice are comprehensively considered, the phenomenon of low precision caused by the fact that a linear regression model is used alone can be effectively reduced, meanwhile, the adverse effect of outliers of the high-resolution spectral data and the total organic carbon content of the overground part of the rice on modeling can be reduced, and therefore the precision and the efficiency of the rice overground part total organic carbon content inversion model based on hyperspectral information are improved.
(3) The independence of model training and model inspection is fully considered, the training data set and the inspection data set are divided by a random segmentation method, the training data set is only used for model training, and the inspection data set is only used for model inspection, so that the reasonability of model effect inspection is ensured.
(4) Because the parameter tuning of the model is very important to the calculation accuracy of the model, the model parameter rank matrix is constructed in the invention, the decision coefficient R2 is used as an evaluation parameter, and the model parameter tuning is carried out by using a trial-and-error method, so that the model training and parameter tuning speed is greatly improved on the basis of ensuring the parameter tuning effect.
(5) The inversion method for the total organic carbon content of the overground part of the rice is simple and convenient to calculate, is suitable for remote sensing quantitative inversion of the total organic carbon content of the overground part of the rice in different ecological regions, different varieties and different growth periods, can accurately invert the total organic carbon content of the overground part of the rice, can quickly acquire information such as carbon-nitrogen metabolism, physiological conditions and growth vigor of the rice, and meanwhile provides scientific data for water and fertilizer operational management of rice planting and cultivation.
In conclusion, the remote sensing inversion model of the total organic carbon content of the overground part of the rice unit area based on the K nearest neighbor regression algorithm can quickly and accurately acquire the information of the total organic carbon content of the overground part of the rice unit area, overcomes the difficulty that the characteristic wave band of the total organic carbon content of the overground part of the rice unit area is difficult to determine due to the spectrum superposition effect caused by complex rice components, greatly improves the precision of the inversion model of the total organic carbon content of the overground part of the rice unit area, and is ingenious in design, simple and convenient to calculate, easy to implement, low in cost and suitable for large-scale popularization and application.
It will thus be seen that the objects of the invention have been fully and effectively accomplished. The functional and structural principles of the present invention have been shown and described in the embodiments, and the embodiments may be modified without departing from the principles. Therefore, this invention includes all modifications encompassed within the spirit and scope of the claims.
Claims (11)
1. The remote sensing inversion model of the total organic carbon content of the overground part of the rice unit area based on the K neighbor regression algorithm is characterized in that the remote sensing inversion model of the total organic carbon content of the overground part of the rice unit area based on the K neighbor regression algorithm is a K neighbor regression model of Python language, and model parameters of the K neighbor regression model are as follows: 'leaf _ size' 56, 'metric' 9, 'n _ neighbors' 8, 'algorithm' auto 'and' weights 'uniform'.
2. The K-nearest neighbor regression algorithm-based remote sensing inversion model for total organic carbon content of underground parts of rice in unit area according to claim 1, wherein the K-nearest neighbor regression model is trained by using a rice data set, the data set comprises canopy reflectivity of m sampling points of the rice and total organic carbon content of underground parts of unit area, the m sampling points are uniformly distributed in a rice planting area, and the canopy reflectivity is the canopy reflectivity of n characteristic bands.
3. The K-nearest neighbor regression algorithm-based remote sensing inversion model of total organic carbon content in rice unit area above ground according to claim 2, wherein m is 36, the n characteristic bands are 2151 characteristic bands, and the 2151 characteristic bands are from 350nm to 2500 nm.
4. A remote sensing inversion method of total organic carbon content of overground parts of a unit area of rice based on a K nearest neighbor regression algorithm is characterized by comprising the following steps:
(1) measuring the canopy reflectance of the rice;
(2) measuring the total organic carbon content of the overground part of the unit area of the rice;
(3) calculating by using the canopy reflectivity as input data and adopting a K nearest neighbor regression model of Python language to obtain an inversion value, and calculating a decision coefficient R according to the inversion value and the total organic carbon content of the overground part of the unit area2Changing the value of the model parameter, R, of the K nearest neighbor regression model2The larger the change of the model parameter is, the greater the importance of the model parameter is, the model parameter is arranged from large to small according to the importance to construct a model parameter tuning rank matrix;
(4) training the K nearest neighbor regression model by taking the canopy reflectivity as the input data and the total organic carbon content of the overground part in unit area as an output result, and sequentially tuning the model parameters according to the model parameter tuning order matrix to obtain tuning values of the model parameters;
(5) the canopy reflectivity is used as the input data, the total organic carbon content of the overground part in unit area is used as the output result, the model parameter tuning value is adopted to train the K neighbor regression model, after the training of the K neighbor regression model is finished, the remote sensing inversion model of the total organic carbon content of the overground part in unit area of rice based on the K neighbor regression algorithm is obtained, the remote sensing inversion model of the total organic carbon content of the overground part in unit area of rice based on the K neighbor regression algorithm is stored by using a save method, if the remote sensing inversion model of the total organic carbon content of the overground part in unit area of rice based on the K neighbor regression algorithm is needed, the remote sensing inversion model of the total organic carbon content of the overground part in unit area of rice based on the K neighbor regression algorithm is loaded by using a load method for use.
5. The remote sensing inversion method of total organic carbon content of rice in unit area and above ground based on K nearest neighbor regression algorithm as claimed in claim 4, wherein in said step (1), said measurement is performed with a hyperspectral radiometer, said measurement time is 10:00 ~ 14:00, said hyperspectral radiometer uses a lens with a field angle of 25 °, a sensor probe of said portable hyperspectral radiometer is directed vertically to the canopy of said rice and has a vertical height of 1 meter from the top layer of said canopy, a ground field range diameter of said sensor probe is 0.44 meters, said sensor probe faces the sun, said measurement is corrected with a standard board, said standard board is a standard white board with a reflectivity of 95% -99%.
6. The remote sensing inversion method of total organic carbon content of rice unit area above ground based on K-nearest neighbor regression algorithm as claimed in claim 4, wherein in said step (2), said step of measuring total organic carbon content of rice unit area above ground specifically comprises:
collecting the overground part plants of the rice, deactivating enzyme, drying to constant weight to obtain dry plants, crushing the dry plants, determining the total organic carbon content of the overground part, and converting the total organic carbon content of the overground part into the total organic carbon content of the overground part in unit area according to a sampling area.
7. The remote sensing inversion method of total organic carbon content in the overground part of rice unit area based on K-nearest neighbor regression algorithm as claimed in claim 6, wherein in said step (2), said temperature of de-enzyming is 105 ℃, said time of de-enzyming is 20-30 minutes, said temperature of drying is 80-90 ℃, said method of determining total organic carbon content in the overground part adopts potassium dichromate volumetric method.
8. The remote sensing inversion method of total organic carbon content of rice unit area overground part based on K-nearest neighbor regression algorithm as claimed in claim 4, characterized in that in said step (3), said model parameter tuning rank matrix is:
Params＝{'leaf_size','metric','p','n_neighbors','algorithm','weights'}。
9. the remote sensing inversion method of total organic carbon content of rice unit area overground part based on K-nearest neighbor regression algorithm as claimed in claim 8, characterized in that in said step (4), said model parameters are adjusted to values:
'leaf_size'＝56,'metric'＝'euclidean','p'＝9,'n_neighbors'＝8,'algorithm'＝'auto','weights'＝'uniform'。
10. the K-nearest neighbor regression algorithm-based remote sensing inversion method for total organic carbon content of overground parts of rice per unit area according to claim 4, wherein in the step (1), the step of measuring the canopy reflectance of rice is specifically to measure the canopy reflectance of m sampling points of a rice planting area, the m sampling points are uniformly distributed in the rice planting area, and the canopy reflectance is the canopy reflectance of n characteristic bands; in the step (2), the step of measuring the total organic carbon content per unit area of the above-ground parts of the rice is to measure the total organic carbon content per unit area of the m sampling points.
11. The remote sensing inversion method of total organic carbon content in rice unit area based on K-nearest neighbor regression algorithm as claimed in claim 10, wherein in said step (1), said m is 36, said n characteristic bands are 2151 characteristic bands, and said 2151 characteristic bands are from 350nm band to 2500nm band.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202011382700.3A CN112630161A (en) | 2020-12-01 | 2020-12-01 | Remote sensing inversion model and method for total organic carbon content of overground part of rice unit area based on K nearest neighbor regression algorithm |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202011382700.3A CN112630161A (en) | 2020-12-01 | 2020-12-01 | Remote sensing inversion model and method for total organic carbon content of overground part of rice unit area based on K nearest neighbor regression algorithm |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN112630161A true CN112630161A (en) | 2021-04-09 |
Family
ID=75307156
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202011382700.3A Pending CN112630161A (en) | 2020-12-01 | 2020-12-01 | Remote sensing inversion model and method for total organic carbon content of overground part of rice unit area based on K nearest neighbor regression algorithm |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN112630161A (en) |
Citations (5)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN111829965A (en) * | 2020-08-04 | 2020-10-27 | 淮阴师范学院 | Remote sensing inversion model and method for starch accumulation amount of rice overground part |
| CN111855589A (en) * | 2020-08-04 | 2020-10-30 | 淮阴师范学院 | Remote sensing inversion model and method for rice leaf nitrogen accumulation |
| CN111855591A (en) * | 2020-08-04 | 2020-10-30 | 淮阴师范学院 | Rice overground part carbon-nitrogen ratio remote sensing inversion model and method |
| CN111855592A (en) * | 2020-08-04 | 2020-10-30 | 淮阴师范学院 | Remote sensing inversion model and method for upper dry matter weight in unit area of rice |
| CN111855590A (en) * | 2020-08-04 | 2020-10-30 | 淮阴师范学院 | Remote sensing inversion model and method for rice leaf starch accumulation |
-
2020
- 2020-12-01 CN CN202011382700.3A patent/CN112630161A/en active Pending
Patent Citations (5)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN111829965A (en) * | 2020-08-04 | 2020-10-27 | 淮阴师范学院 | Remote sensing inversion model and method for starch accumulation amount of rice overground part |
| CN111855589A (en) * | 2020-08-04 | 2020-10-30 | 淮阴师范学院 | Remote sensing inversion model and method for rice leaf nitrogen accumulation |
| CN111855591A (en) * | 2020-08-04 | 2020-10-30 | 淮阴师范学院 | Rice overground part carbon-nitrogen ratio remote sensing inversion model and method |
| CN111855592A (en) * | 2020-08-04 | 2020-10-30 | 淮阴师范学院 | Remote sensing inversion model and method for upper dry matter weight in unit area of rice |
| CN111855590A (en) * | 2020-08-04 | 2020-10-30 | 淮阴师范学院 | Remote sensing inversion model and method for rice leaf starch accumulation |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN111855589A (en) | Remote sensing inversion model and method for rice leaf nitrogen accumulation | |
| CN111855591A (en) | Rice overground part carbon-nitrogen ratio remote sensing inversion model and method | |
| CN109187441B (en) | Method for constructing summer corn nitrogen content monitoring model based on canopy spectral information | |
| CN111855590A (en) | Remote sensing inversion model and method for rice leaf starch accumulation | |
| CN110567892B (en) | Summer corn nitrogen hyperspectral prediction method based on critical nitrogen concentration | |
| CN107796764B (en) | Method for constructing wheat leaf area index estimation model based on three-band vegetation index | |
| CN111829965A (en) | Remote sensing inversion model and method for starch accumulation amount of rice overground part | |
| CN107505271B (en) | Plant nitrogen estimation method and system based on nitrogen component radiation transmission model | |
| CN105445229B (en) | Wheat spring nitrogen fertilizer applying method based on nitrogen balance spectroscopic methodology and its construction method for chasing after nitrogen quantity model | |
| CN113268923B (en) | Summer corn yield estimation method based on simulated multispectral | |
| CN110189793B (en) | Hyperspectrum-based wheat nitrogen fertilizer physiological utilization rate estimation model construction and wheat variety classification with different nitrogen efficiencies | |
| CN111855593A (en) | Remote sensing inversion model and method for starch content of rice leaf | |
| CN112270131A (en) | Remote sensing inversion model and method for rice leaf area index based on ARD regression algorithm | |
| CN111855592A (en) | Remote sensing inversion model and method for upper dry matter weight in unit area of rice | |
| CN104502283A (en) | Two-band hyperspectral index and prediction model for estimating yield and shoot dry weight of soybean | |
| CN112816618A (en) | Method for screening nitrogen-efficient wheat varieties | |
| CN103940748A (en) | Citrus canopy nitrogen content prediction and visualization method based on hyperspectrum technology | |
| CN113049499A (en) | Indirect remote sensing inversion method for water total nitrogen concentration, storage medium and terminal equipment | |
| CN112362812A (en) | Remote sensing inversion model and method for rice leaf chlorophyll carotenoid content ratio based on Lars algorithm | |
| CN113670913B (en) | Construction method for inversion hyperspectral vegetation index of nitrogen content of rice | |
| CN112525831A (en) | Remote sensing inversion model and method for protein nitrogen accumulation of rice overground part based on Catboost regression algorithm | |
| CN112630167A (en) | Rice leaf carbon-nitrogen ratio remote sensing inversion model and method based on support vector machine regression algorithm | |
| CN112632847A (en) | XGboost regression algorithm-based rice leaf starch content remote sensing inversion model and method | |
| CN112630161A (en) | Remote sensing inversion model and method for total organic carbon content of overground part of rice unit area based on K nearest neighbor regression algorithm | |
| CN112362810A (en) | Remote sensing inversion model and method for total organic carbon content of rice unit area leaf based on random forest regression algorithm |
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 |