CN111257240B - Near-earth hyperspectral data and integrated model-based rape nitrogen-phosphorus-potassium nutrient diagnosis method - Google Patents
Near-earth hyperspectral data and integrated model-based rape nitrogen-phosphorus-potassium nutrient diagnosis method Download PDFInfo
- Publication number
- CN111257240B CN111257240B CN202010051787.XA CN202010051787A CN111257240B CN 111257240 B CN111257240 B CN 111257240B CN 202010051787 A CN202010051787 A CN 202010051787A CN 111257240 B CN111257240 B CN 111257240B
- Authority
- CN
- China
- Prior art keywords
- deficiency
- nutrient
- nitrogen
- potassium
- treatment
- 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
Landscapes
- Physics & Mathematics (AREA)
- Spectroscopy & Molecular Physics (AREA)
- Health & Medical Sciences (AREA)
- Life Sciences & Earth Sciences (AREA)
- Chemical & Material Sciences (AREA)
- Analytical Chemistry (AREA)
- Biochemistry (AREA)
- General Health & Medical Sciences (AREA)
- General Physics & Mathematics (AREA)
- Immunology (AREA)
- Pathology (AREA)
- Investigating Or Analysing Materials By Optical Means (AREA)
Abstract
The invention discloses a rape nitrogen-phosphorus-potassium nutrient diagnosis method based on near-earth hyperspectral data and an integration model, which is a method for converting spectral reflectance characteristics into new probability characteristics by integrating a plurality of random forest models based on canopy spectral characteristic wave bands sensitive to different nutrient deficiency conditions of rape, constructing a diagnosis model by using the new probability characteristics, carrying out nondestructive diagnosis on the nitrogen, phosphorus and potassium nutrient conditions of rape and verifying the performance of the model by specific example application. The invention provides an efficient, rapid and lossless rape nutrient state diagnosis method, which effectively improves the precision of a nutrient diagnosis model and solves the problem of higher error diagnosis rate caused by similar spectral characteristics under different nutrient stresses in the existing method.
Description
Technical Field
The invention relates to the field of crop nutrient deficiency rapid detection and artificial intelligence application, in particular to a rape nitrogen-phosphorus-potassium nutrient diagnosis method based on near-earth hyperspectral data and an integrated model.
Background
The rape is the first major oil crop in China, and the rape seed oil accounts for 57.2 percent of the oil yield of the domestic oil crop and plays an important role in edible oil supply in China. Since nearly half a century, the rape production in China is rapidly developed, and a large number of researches show that the reasonable application of nitrogenous fertilizer, phosphate fertilizer and potash fertilizer has very obvious yield-increasing effect on rape. The method has the advantages that the nutrient condition of the crops, particularly the condition of main nutrients such as nitrogen, phosphorus, potassium and the like under the macroscopic scale can be accurately monitored in real time, and the method has important significance for assisting the decision of government functional departments, guiding the production of farmers and improving the yield and quality of the crops.
The hyperspectral crop nutrient diagnosis method can realize comprehensive and rapid nutrient diagnosis of a large-scale farmland by extracting spectral response of crops under nutrient shortage on the basis of not damaging the crops. However, the spectral characteristic responses of crops with different nutrient abundance have many similarities, for example, the severe deficiency of nitrogen and potassium can cause the chlorophyll content to be reduced and further cause the reflectivity of red wave band to be improved, the deficiency of nitrogen and phosphorus can cause the leaf area to be slowly increased and further cause the reflectivity of near infrared wave band to be reduced, and the like.
Therefore, the current spectral nutrient diagnosis research mainly aims at identifying the deficiency or absence of one nutrient element, and a method for enhancing the spectral response discrimination of crops under different nutrient conditions to realize the discrimination of different nutrient stresses and improve the nutrient diagnosis precision is not provided.
Disclosure of Invention
The invention aims to overcome the defects of the prior art and provides a method for diagnosing nitrogen, phosphorus and potassium nutrients of rape based on near-earth hyperspectral data and an integrated model; the method is based on near-earth hyperspectral data, canopy characteristic spectrum bands sensitive to different nutrient deficiency conditions of rape are mined, the selected canopy characteristic spectrum bands are transformed in a mode of integrating a plurality of models, the characteristic discrimination of the canopy characteristic spectrum bands under different nutrient deficiency conditions is enhanced, and the nondestructive diagnosis of the nitrogen, phosphorus and potassium nutrient conditions of the rape is realized.
In order to achieve the aim, the invention designs a near-earth hyperspectral data and integrated model-based rape nitrogen-phosphorus-potassium nutrient diagnosis method, which comprises the following steps:
1) establishing a canopy high spectral reflectivity data subset under different rape nutrient conditions
a. Collecting near-ground hyperspectral data of a rape canopy in a region to be detected to form a set, mapping the collected canopy hyperspectral reflectance data set with nutrient conditions of the rape in the region to be detected,
b. then dividing the high spectral reflectance data of the canopy into three groups of fertilization nutrient elements of nitrogen, phosphorus and potassium;
c. according to the physical and chemical parameters of local soil, the fertilization treatment of each group of nutrient elements is divided into four nutrient conditions of severe deficiency treatment, moderate deficiency treatment, normal treatment and excessive fertilization treatment according to the fertilization gradient;
d. matching the canopy high spectral reflectance data in each group with the nutrient condition, and using the canopy high spectral reflectance data subset under each nutrient condition as a data subset;
2) optimizing canopy hyperspectral reflectance data
A. Resampling the canopy hyperspectral reflectivity data at intervals, and deleting noise wave bands;
B. calculating the reflectivity mean value and standard deviation of each wave band in all canopy spectra of each data subgroup, checking each spectrum one by one, and rejecting the spectrum containing abnormal wave bands;
3) six random forest classification models are constructed for each nutrient element, and classification of nutrient abundance conditions is carried out for six times by utilizing the constructed models
a. Combining a plurality of data subgroups in each nutrient element (extracting four subgroups, three subgroups and two subgroups in the plurality of data subgroups in each nutrient element data to be combined) to generate six different data sets of each nutrient element, wherein the six data sets are respectively as follows:
(1) data set 1 is a combination of the severe deficient treatment subgroup, the moderate deficient treatment subgroup, the excess treatment subgroup, the normal treatment subgroup,
(2) data set 2 is a combination of the severe deficiency treatment subgroup, the moderate deficiency treatment subgroup, the normal treatment subgroup,
(3) data set 3 is a combination of a severe deficiency treatment subgroup, an excess treatment subgroup, a normal treatment subgroup,
(4) data set 4 is a combination of the severe deficiency treatment subgroup and the normal treatment subgroup,
(5) data set 5 is a combination of medium treatment subgroups and normal treatment subgroups,
(6) data set 6 is a combination of the excess treatment sub-group, the normal treatment sub-group;
b. respectively constructing six random forest classification models by using six different data sets of each nutrient element, wherein the classification category is the same as the nutrient condition in the utilized data set, and adjusting two parameters ntree and mtry when constructing each random forest classification model so as to achieve the optimal overall classification precision;
4) calculating the importance score of each wave band on the nutrient element abundance diagnosis
For each nutrient element, classifying nutrient abundance conditions by using each constructed random forest classification model to obtain importance (VI) and total classification precision of each wave band in the hyperspectral data of the canopy, calculating the mean (mu) and variance (sigma) of the importance of all wave bands of the hyperspectral of the canopy in each classification result, and normalizing the importance of each wave band to obtain the normalized wave band importance (VI'):
in the formula, VI' is the importance of the normalized wave band, VI is the importance of each wave band in the canopy hyperspectral data,
the mean value (mu) and the variance (sigma) of the importance of all wave bands of the hyperspectral of the canopy in each classification result;
5) screening of characteristic bands sensitive to abundance of each nutrient element
Carrying out six times of classification of nutrient abundance conditions on six data sets in each nutrient element to obtain normalized importance of each characteristic waveband, carrying out weighted summation, and weighting to obtain corresponding total precision (w) of each classificationi) And calculating the total contribution (score) of each characteristic waveband:
where score is the total contribution of each eigenband, wiVI' is the normalized wave band importance for the corresponding total classification precision of each time;
selecting 10 spectral bands with the highest total contribution degree score as diagnostic characteristic bands sensitive to nutrient element deficiency aiming at each nutrient element, and removing the rest spectral bands;
6) reestablishing a defect condition classification model of each nutrient element by utilizing the screened characteristic wave bands
Respectively utilizing 10 important characteristic wave bands sensitive to nutrient element abundance and deficiency to re-establish a random forest classification model for nutrient diagnosis, wherein the ntree and mtry parameters need to be corrected when the random forest classification model is established every time, and the correction basis is preferably the best classification total precision;
7) generating new probability characteristic factors by utilizing three nutrient element abundance condition classification models
a. Aiming at each nutrient element of nitrogen, phosphorus and potassium, selecting the wave band which is screened out in the step 5) and sensitive to the deficiency of each nutrient element from the hyperspectral data of each canopy, classifying and distinguishing the wave band by utilizing the abundance condition classification model of each nutrient element established in the step 6) to obtain the probability value of each spectrum data classified into each category,
b. then, all probability values form new characteristics which are used as characteristic factors for finally distinguishing the rich and deficient conditions of nitrogen, phosphorus and potassium nutrients;
8) constructing a random forest classification model by using the newly generated probability characteristic factors to obtain a diagnosis result of the nutrient deficiency condition of nitrogen, phosphorus and potassium;
and (3) constructing a new random forest classification model by using the newly generated probability characteristic factors (the ntree and mtry parameters need to be corrected when the random forest classification model is established, and the correction basis is optimal in classification total precision), and determining the newly generated probability characteristic factors to belong to which one of severe nitrogen deficiency, mild nitrogen deficiency, severe phosphorus deficiency, mild phosphorus deficiency, severe potassium deficiency, mild potassium deficiency and normal fertilization is selected, so that the diagnosis of the nitrogen, phosphorus and potassium nutrient deficiency condition is completed.
Further, in the c-th substep of step 1), four treatment subgroups in each group of nutrient elements are as follows:
in the nitrogen group treatment process, the application amount of a phosphate fertilizer is 90 kg/hectare, the application amount of a potassium fertilizer is 120 kg/hectare, and the normal application amounts are adopted, and meanwhile, the pest and disease control management work is carried out; wherein the content of the first and second substances,
the nitrogen fertilizer dosage for the severe nitrogen deficiency treatment is 0 kg/hectare,
the nitrogen fertilizer dosage for moderate nitrogen deficiency treatment is 75-90 kg/ha,
the nitrogen fertilizer dosage for normal treatment is 150-180 kg/ha,
the nitrogen fertilizer dosage for excessive nitrogen application treatment is 225 plus 270 kg/ha;
in the phosphorus group treatment process, the application amount of the nitrogen fertilizer is 180 kg/hectare, the application amount of the potassium fertilizer is 120 kg/hectare, and the nitrogen fertilizer and the potassium fertilizer are normal application amounts and are used for pest and disease control management; wherein
The dosage of the phosphate fertilizer for severe phosphorus deficiency treatment is 0 kg/ha,
the dosage of the phosphate fertilizer for moderate phosphorus deficiency treatment is 30-45 kg/ha,
the amount of phosphate fertilizer normally treated is 90 kg/ha,
the excessive phosphorus application treatment is that the usage amount of phosphate fertilizer is 135-;
in the potassium group treatment process, the application amount of the nitrogen fertilizer is 180 kg/hectare, the application amount of the phosphate fertilizer is 90 kg/hectare, both normal application amounts are adopted, and meanwhile, the pest and disease control management work is carried out; wherein the content of the first and second substances,
the dosage of the potassium fertilizer for severe potassium deficiency treatment is 0 kg/hectare,
the consumption of the potassium fertilizer for moderate potassium deficiency treatment is 60-75 kg/ha,
the normal potassium fertilizer dosage is 120 kg/ha,
the dosage of the potassium fertilizer subjected to excessive potassium application treatment is 135-180 kg/ha.
Still further, in the sub-step B of step 2), a specific step of rejecting spectra containing abnormal bands is:
when the reflectivity of a certain wave band of a certain spectrum is larger than the value of +/-two times of standard deviation of the mean value, marking the wave band of the certain spectrum as an abnormal wave band;
and if the number of the abnormal wave bands of the piece of spectral data is more than or equal to 20, removing the piece of spectral data from the data subgroup.
Still further, in the sub-step b of step 3), the value range of ntree is 100-1000, the step size is 100, the value range of mtry is 3-10, and the step size is 1.
Further, in the step 6), establishing a random forest classification model for nutrient diagnosis specifically comprises the following steps:
i, establishing a random forest classification model capable of distinguishing severe nitrogen deficiency, mild nitrogen deficiency and normal fertilization by using 10 screened important characteristic wave bands sensitive to nitrogen nutrient abundance;
ii, establishing a random forest classification model capable of distinguishing severe phosphorus deficiency, mild phosphorus deficiency and normal fertilization by using the screened 10 important characteristic wave bands sensitive to phosphorus nutrient abundance;
and iii, establishing a random forest classification model capable of distinguishing severe potassium deficiency, mild potassium deficiency and normal fertilization by using the screened 10 important characteristic wave bands sensitive to the abundance and the deficiency of potassium nutrients.
Still further, in the step 7), the a-th sub-step, nine probability values of each spectral data classified into each category are obtained by the following method:
i, selecting spectral data of 10 important characteristic wave bands sensitive to nitrogen nutrient deficiency, inputting a random forest classification model obtained by training aiming at the nitrogen nutrient, and obtaining probability values of each spectral data divided into severe nitrogen deficiency, mild nitrogen deficiency and normal fertilization;
ii, selecting spectral data of 10 important characteristic wave bands sensitive to phosphorus nutrient deficiency, inputting a random forest classification model obtained by training aiming at phosphorus nutrients, and obtaining probability values of each spectrum divided into severe phosphorus deficiency, mild phosphorus deficiency and normal fertilization;
and iii, selecting spectral data of 10 important characteristic wave bands sensitive to potassium nutrient deficiency, inputting a random forest classification model obtained by training aiming at potassium nutrients, and obtaining probability values of each spectral data divided into severe potassium deficiency, mild potassium deficiency and normal fertilization.
The invention has the beneficial effects that:
(1) based on the near-earth hyperspectral data, a characteristic wave band screening method is designed, the information dimension of the hyperspectral data is reduced, the processing process is simplified, and the processing speed is increased.
(2) The original characteristic spectrum is converted into a new probability characteristic by constructing an integrated model, the new probability characteristic obviously enhances the discrimination of the spectrum characteristics of the rape under different nutrient states, the accuracy of the estimation model is effectively improved, and a practical and reliable scientific method is provided for the remote sensing diagnosis of the nitrogen, phosphorus and potassium nutrients of the rape.
Drawings
FIG. 1 is a flow chart of a method for diagnosing nitrogen, phosphorus and potassium nutrients of rape based on near-earth hyperspectral data and an integration model;
FIG. 2 is a comparison graph of a new probability characteristic (b) and an original spectral reflectance characteristic (a) obtained by a near-earth hyperspectral data and integrated model based rape nitrogen phosphorus potassium nutrient diagnosis method;
Detailed Description
The present invention is described in further detail below with reference to specific examples so as to be understood by those skilled in the art.
A PSR +3500 portable terrestrial object spectrometer (Spectral Evolution, Haverhill, MA, USA) is used for collecting nutrient processing data sets and near-field hyperspectral data sets of winter rape overwintering period test cells, the Spectral range is 400nm to 2500nm, and the Spectral resolution is 1 nm. In sunny, windless weather, in the morning of 11: 00 to 1 in the afternoon: and collecting spectral data between 00. When data are collected, the optical fiber probe is placed 1 meter above the rape canopy. 3-5 spectra were collected for each test cell. 13 winter rape nutrient scarce field tests are carried out in Wuhan city, Wu-Hu-Hi city and Sanyo city of Hubei province between 2013 and 2019, the tested varieties are Hua-YOU No. 9 and Hua-YOU No. 62, the nutrient gradient setting of each test is different, the aim is to widely obtain hyperspectral data sets of different nutrient scarce conditions, 910 rape canopy spectral data are collected in total, and 30% of the data are used as independent test data sets for testing and comparing the diagnosis accuracy of the integration method and the common method;
method for diagnosing deficiency of nitrogen, phosphorus and potassium nutrients of rape by constructing integration model by utilizing near-earth hyperspectral data according to method flow shown in figure 1
Step 1: establishing a canopy high spectral reflectivity data subset under different rape nutrient conditions
Step 1.1: mapping the collected high spectral reflectance data of the rape canopy with the nutrient status of the rape in the area to be detected;
step 1.2: thirdly, dividing the high spectral reflectance data of the canopy into three groups of fertilization nutrient elements of nitrogen, phosphorus and potassium;
step 1.3: according to the physical and chemical parameters of local soil, the fertilization treatment of each group of nutrient elements is divided into four nutrient conditions of severe deficiency treatment, moderate deficiency treatment, normal treatment and excessive fertilization treatment according to the fertilization gradient; the severe nitrogen deficiency treatment is 0 kg/hectare of nitrogen fertilizer, the moderate nitrogen deficiency treatment is 75-90 kg/hectare of nitrogen fertilizer, the excessive nitrogen application treatment is 225 plus-nitrogen fertilizer plus 270 kg/hectare of nitrogen fertilizer, the severe phosphorus deficiency treatment is 0 kg/hectare of phosphate fertilizer, the moderate phosphorus deficiency treatment is 30-45 kg/hectare of phosphate fertilizer, the excessive phosphorus application treatment is 180 plus-phosphorus fertilizer plus-potassium fertilizer, the severe potassium deficiency treatment is 0 kg/hectare of potassium fertilizer, the moderate potassium deficiency treatment is 60-75 kg/hectare of potassium fertilizer, the excessive potassium application treatment is 180 plus-potassium fertilizer plus 180 kg/hectare of nitrogen fertilizer, the normal fertilizer application treatment is 150 plus-potassium fertilizer plus-180 kg/hectare of phosphate fertilizer, 90 kg/potassium fertilizer plus 120 kg/hectare of potassium fertilizer;
step 1.4: matching the high spectral reflectance data of the canopy in each group with the nutrient condition, wherein the high spectral reflectance data subset of the canopy under each nutrient condition is used as a data subset, and the seven subsets are respectively a severe nitrogen deficiency subset, a moderate nitrogen deficiency subset, a severe phosphorus deficiency subset, a moderate phosphorus deficiency subset, a severe potassium deficiency subset, a moderate potassium deficiency subset and a normal fertilization subset;
step 2: optimizing canopy hyperspectral reflectance data
Step 2.1: averaging 1nm resolution spectrum data into 10nm resolution spectrum data, and deleting noise bands (less than 400nm,1800-2000nm and more than 2300 nm);
step 2.2: calculating the mean value and standard deviation of each waveband spectrum in all canopy spectra of each subgroup data, checking each spectrum one by one, if the reflectivity of a certain waveband of a certain spectrum is greater than the value of +/-two times of the standard deviation of the mean value, marking the waveband of the certain spectrum as an abnormal waveband, and if the abnormal waveband of the certain spectrum data is greater than or equal to 20, rejecting the certain spectrum data;
and step 3: combining a plurality of subgroup data in each nutrient element to generate six different data sets, respectively training a nutrient diagnosis model by using the six different data sets, calculating the value of each waveband feature according to the feature importance of an input waveband obtained by the model, selecting ten important feature wavebands with the highest values as feature wavebands sensitive to the abundance and the deficiency of the nutrient element for each nutrient element, selecting a random forest classification algorithm by using the modeling method, realizing the random forest classification model by using Python3.5, and mainly using a sklern 0.19.2 machine learning framework;
step 3.1: combining a plurality of data subgroups in each nutrient element (extracting four subgroups, three subgroups and two subgroups in the plurality of data subgroups in each nutrient element data to be combined) to generate six different data sets of each nutrient element, wherein the six data sets are respectively as follows:
(1) data set 1 is a combination of the severe deficient treatment subgroup, the moderate deficient treatment subgroup, the excess treatment subgroup, the normal treatment subgroup,
(2) data set 2 is a combination of the severe deficiency treatment subgroup, the moderate deficiency treatment subgroup, the normal treatment subgroup,
(3) data set 3 is a combination of a severe deficiency treatment subgroup, an excess treatment subgroup, a normal treatment subgroup,
(4) data set 4 is a combination of the severe deficiency treatment subgroup and the normal treatment subgroup,
(5) data set 5 is a combination of medium treatment subgroups and normal treatment subgroups,
(6) data set 6 is a combination of the excess treatment sub-group, the normal treatment sub-group;
step 3.2: respectively constructing six random forest classification models by using six different data sets of each nutrient element, wherein the classification category is the same as the nutrient condition in the utilized data set, and adjusting two parameters ntree and mtry when each random forest classification model is established, wherein the value range of ntree is 100-1000, the step length is 100, the value range of mtry is 3-10, the step length is 1, and the correction basis is that the optimal total precision of the classification of the training set is taken as the best basis;
and 4, step 4: calculating the importance score of each wave band on the nutrient element abundance diagnosis
For each nutrient element, classifying nutrient abundance conditions by using each constructed random forest classification model to obtain importance (VI) and total classification precision of each wave band in the hyperspectral data of the canopy, calculating the mean (mu) and variance (sigma) of the importance of all wave bands of the hyperspectral of the canopy in each classification result, and normalizing the importance of each wave band to obtain the normalized wave band importance (VI'):
in the formula, VI' is the importance of the normalized wave band, VI is the importance of each wave band in the canopy hyperspectral data,
the mean value (mu) and the variance (sigma) of the importance of all wave bands of the hyperspectral of the canopy in each classification result;
and 5: screening of characteristic bands sensitive to abundance of each nutrient element
Carrying out six times of classification of nutrient abundance conditions on six data sets in each nutrient element to obtain normalized importance of each characteristic waveband, carrying out weighted summation, and weighting to obtain corresponding total precision (w) of each classificationi) And calculating the total contribution (score) of each characteristic waveband:
where score is the total contribution of each eigenband, wiThe corresponding total precision of each classification is the normalized importance of the wave band;
selecting 10 spectral bands with the highest total contribution degree score as diagnosis characteristic bands sensitive to nutrient element deficiency aiming at each nutrient element, and removing the rest spectral bands; the 10 screened characteristic wave bands sensitive to the nitrogen nutrient shortage are as follows: 640, 2070, 650, 2000, 680, 690, 630, 660, 670, 2020 nm; the 10 screened characteristic wave bands sensitive to the shortage of phosphorus nutrients are as follows: 1120, 910, 2000, 2040, 810, 1090, 760, 690, 680, 1420 nm; the 10 screened characteristic wave bands sensitive to the shortage of potassium nutrients are as follows: 2100, 2260, 2030, 2040, 530, 2290, 680, 2070, 650, 2080;
step 6: reestablishing a defect condition classification model of each nutrient element by utilizing the screened characteristic wave bands
Respectively utilizing 10 important characteristic wave bands sensitive to nutrient element abundance and deficiency to reestablish a random forest classification model for nutrient diagnosis, which comprises the following specific steps:
i, establishing a random forest classification model capable of distinguishing severe nitrogen deficiency, mild nitrogen deficiency and normal fertilization by using 10 screened important characteristic wave bands sensitive to nitrogen nutrient abundance;
ii, establishing a random forest classification model capable of distinguishing severe phosphorus deficiency, mild phosphorus deficiency and normal fertilization by using the screened 10 important characteristic wave bands sensitive to phosphorus nutrient abundance;
and iii, establishing a random forest classification model capable of distinguishing severe potassium deficiency, mild potassium deficiency and normal fertilization by using the screened 10 important characteristic wave bands sensitive to the abundance and the deficiency of potassium nutrients.
And correcting ntree and mtry parameters every time a random forest classification model is established, wherein the value range of ntree is 100-1000, the step length is 100, the value range of mtry is 3-10, the step length is 1, the correction basis is that the total precision of the classification of the training set is optimal, and the corrected optimal nutrient diagnosis model is stored for next nutrient deficiency diagnosis;
and 7: generating new probability characteristic factors by utilizing three nutrient element abundance condition classification models
Step 7.1: aiming at each nutrient element of nitrogen, phosphorus and potassium, selecting the wave band which is screened out in the step 5 and sensitive to deficiency of each nutrient element from the hyperspectral data of each canopy, classifying and distinguishing the wave band by utilizing the abundance condition classification model of each nutrient element established in the step 6, obtaining the probability value of each spectrum data classified into each category, and counting nine probability values to obtain the method as follows:
i, selecting spectral data of 10 important characteristic wave bands sensitive to nitrogen nutrient deficiency, inputting a random forest classification model obtained by training aiming at the nitrogen nutrient, and obtaining probability values of each spectral data divided into severe nitrogen deficiency, mild nitrogen deficiency and normal fertilization;
ii, selecting spectral data of 10 important characteristic wave bands sensitive to phosphorus nutrient deficiency, inputting a random forest classification model obtained by training aiming at phosphorus nutrients, and obtaining probability values of each spectrum divided into severe phosphorus deficiency, mild phosphorus deficiency and normal fertilization;
and iii, selecting spectral data of 10 important characteristic wave bands sensitive to potassium nutrient deficiency, inputting a random forest classification model obtained by training aiming at potassium nutrients, and obtaining probability values of each spectral data divided into severe potassium deficiency, mild potassium deficiency and normal fertilization.
Step 7.2: then, all probability values form new characteristics which are used as characteristic factors for finally distinguishing the rich and deficient conditions of nitrogen, phosphorus and potassium nutrients;
and 8: constructing a random forest classification model by using the newly generated probability characteristic factors to obtain a diagnosis result of the nutrient deficiency condition of nitrogen, phosphorus and potassium;
establishing a new random forest classification model by using the newly generated probability characteristic factors (the establishment of the random forest classification model needs to correct ntree and mtry parameters, wherein the value range of ntree is 100-1000, the step length is 100, the value range of mtry is 3-10, the step length is 1, and the correction is preferably based on the best classification total precision), and determining which category of severe nitrogen deficiency, mild nitrogen deficiency, severe phosphorus deficiency, mild phosphorus deficiency, severe potassium deficiency, mild potassium deficiency and normal fertilization the newly generated probability characteristic factors belong to, namely completing the diagnosis of the nitrogen, phosphorus and potassium nutrient deficiency condition;
the results of the method in the examples were compared with three other commonly used diagnostic methods, including a single random forest classification model, a support vector machine classification model, and an artificial neural network classification model. All models carry out rape nutrient diagnosis by utilizing the reflectivity of the characteristic wave band screened by the method, and the models are subjected to precision evaluation by utilizing the same independent test data set, and a classification confusion matrix, overall precision and kappa coefficient are calculated; the results of the comparison of the different methods are shown in table 1; it can be seen that the embodiment shows better performance on various evaluation indexes, wherein the total precision is improved by 16.74% and the kappa coefficient is improved by 0.20 relative to a single random forest classification model; the total precision is improved by 18.91 percent relative to the classification model of the support vector machine, and the kappa coefficient is improved by 0.20 percent; the total precision is improved by 36.20 percent relative to the artificial neural network classification model, and the kappa coefficient is improved by 0.40 percent; particularly, the diagnosis accuracy of moderate nitrogen deficiency, moderate potassium deficiency and severe potassium deficiency is greatly improved, for example, the user precision of moderate potassium deficiency is improved from 0.00% to 87.50%, the production precision and the user precision of severe potassium deficiency and moderate potassium deficiency are both 0.00% in a support vector machine and an artificial neural network classification model, and the methods in the embodiments are obviously improved, wherein the scientificity and the rationality of the method are proved.
Table 1 the integration method of the present invention (a), the single random forest classification model (b), the support vector machine classification model (c), and the artificial neural network classification model (d) compare the diagnosis accuracy with the confusion matrix for the diagnosis of nitrogen, phosphorus, and potassium nutrients in rape.
(a) The total precision of the integration method is 80.09 percent; kappa coefficient 0.75
(b) The total precision of the single random forest classification model is 63.35 percent; kappa coefficient 0.55
(c) 61.99% of the total precision of the classification model of the support vector machine; kappa coefficient 0.55
(d) The total precision of the artificial neural network classification model is 43.89 percent; kappa coefficient 0.35
Other parts not described in detail are prior art. Although the present invention has been described in detail with reference to the above embodiments, it is only a part of the embodiments of the present invention, not all of the embodiments, and other embodiments can be obtained without inventive step according to the embodiments, and the embodiments are within the scope of the present invention.
Claims (5)
1. A rape nitrogen-phosphorus-potassium nutrient diagnosis method based on near-earth hyperspectral data and an integrated model is characterized by comprising the following steps of: the method comprises the following steps:
1) establishing a canopy high spectral reflectivity data subset under different rape nutrient conditions
a. Collecting near-ground hyperspectral data of a rape canopy in a region to be detected to form a set, mapping the collected canopy hyperspectral reflectance data set with nutrient conditions of the rape in the region to be detected,
b. then dividing the high spectral reflectance data of the canopy into three groups of fertilization nutrient elements of nitrogen, phosphorus and potassium;
c. according to the physical and chemical parameters of local soil, the fertilization treatment of each group of nutrient elements is divided into four nutrient conditions of severe deficiency treatment, moderate deficiency treatment, normal treatment and excessive fertilization treatment according to the fertilization gradient;
d. matching the canopy high spectral reflectance data in each group with the nutrient condition, and using the canopy high spectral reflectance data subset under each nutrient condition as a data subset;
2) optimizing canopy hyperspectral reflectance data
A. Resampling the canopy hyperspectral reflectivity data at intervals, and deleting noise wave bands;
B. calculating the reflectivity mean value and standard deviation of each wave band in all canopy spectra of each data subgroup, checking each spectrum one by one, and rejecting the spectrum containing abnormal wave bands;
3) six random forest classification models are constructed for each nutrient element, and classification of nutrient abundance conditions is carried out for six times by utilizing the constructed models
a. Combining a plurality of data subgroups in each nutrient element to generate six different data sets of each nutrient element, wherein the six data sets are respectively as follows:
(1) data set 1 is a combination of the severe deficient treatment subgroup, the moderate deficient treatment subgroup, the excess treatment subgroup, the normal treatment subgroup,
(2) data set 2 is a combination of the severe deficiency treatment subgroup, the moderate deficiency treatment subgroup, the normal treatment subgroup,
(3) data set 3 is a combination of a severe deficiency treatment subgroup, an excess treatment subgroup, a normal treatment subgroup,
(4) data set 4 is a combination of the severe deficiency treatment subgroup and the normal treatment subgroup,
(5) data set 5 is a combination of medium treatment subgroups and normal treatment subgroups,
(6) data set 6 is a combination of the excess treatment sub-group, the normal treatment sub-group;
b. respectively constructing six random forest classification models by using six different data sets of each nutrient element, wherein the classification category is the same as the nutrient condition in the utilized data set, and adjusting two parameters ntree and mtry when constructing each random forest classification model so as to achieve the optimal overall classification precision; wherein, the value range of ntree is 100-1000, the step length is 100, the value range of mtry is 3-10, and the step length is 1;
4) calculating the importance score of each wave band on the nutrient element abundance diagnosis
For each nutrient element, classifying nutrient abundance conditions by using each constructed random forest classification model to obtain importance (VI) and total classification precision of each wave band in the hyperspectral data of the canopy, calculating the mean (mu) and variance (sigma) of the importance of all wave bands of the hyperspectral of the canopy in each classification result, and normalizing the importance of each wave band to obtain the normalized wave band importance (VI'):
in the formula, VI' is the importance of the normalized wave band, VI is the importance of each wave band in the canopy hyperspectral data,
the mean value (mu) and the variance (sigma) of the importance of all wave bands of the hyperspectral of the canopy in each classification result;
5) screening of characteristic bands sensitive to abundance of each nutrient element
Carrying out six times of classification of nutrient abundance conditions on six data sets in each nutrient element to obtain normalized importance of each characteristic waveband, carrying out weighted summation, and weighting to obtain corresponding total precision (w) of each classificationi) And calculating the total contribution (score) of each characteristic waveband:
where score is the total contribution of each eigenband, wiVI' is the normalized wave band importance for the corresponding total classification precision of each time;
selecting 10 spectral bands with the highest total contribution degree score as diagnostic characteristic bands sensitive to nutrient element deficiency aiming at each nutrient element, and removing the rest spectral bands;
6) reestablishing a defect condition classification model of each nutrient element by utilizing the screened characteristic wave bands
Respectively utilizing 10 important characteristic wave bands sensitive to nutrient element abundance and deficiency to re-establish a random forest classification model for nutrient diagnosis, wherein the ntree and mtry parameters need to be corrected when the random forest classification model is established every time, and the correction basis is preferably the best classification total precision;
7) generating new probability characteristic factors by utilizing three nutrient element abundance condition classification models
a. Aiming at each nutrient element of nitrogen, phosphorus and potassium, selecting the wave band which is screened out in the step 5) and sensitive to the deficiency of each nutrient element from the hyperspectral data of each canopy, classifying and distinguishing the wave band by utilizing the abundance condition classification model of each nutrient element established in the step 6) to obtain the probability value of each spectrum data classified into each category,
b. then, all probability values form new characteristics which are used as characteristic factors for finally distinguishing the rich and deficient conditions of nitrogen, phosphorus and potassium nutrients;
8) constructing a random forest classification model by using the newly generated probability characteristic factors to obtain a diagnosis result of the nutrient deficiency condition of nitrogen, phosphorus and potassium;
and constructing a new random forest classification model by using the newly generated probability characteristic factors, and determining which category the newly generated probability characteristic factors belong to nitrogen severe deficiency, nitrogen mild deficiency, phosphorus severe deficiency, phosphorus mild deficiency, potassium severe deficiency, potassium mild deficiency and normal fertilization, namely completing the diagnosis of the nitrogen, phosphorus and potassium nutrient deficiency condition.
2. The method for diagnosing the nitrogen, phosphorus and potassium nutrients of the rapes based on the near-earth hyperspectral data and the integrated model as claimed in claim 1, wherein the method comprises the following steps: in the step c of the step 1), four fertilization treatments in each group of nutrient elements are as follows:
in the nitrogen group treatment process, the application amount of a phosphate fertilizer is 90 kg/hectare, the application amount of a potassium fertilizer is 120 kg/hectare, and the normal application amounts are adopted, and meanwhile, the pest and disease control management work is carried out; wherein the content of the first and second substances,
the nitrogen fertilizer dosage for the severe nitrogen deficiency treatment is 0 kg/hectare,
the nitrogen fertilizer dosage for moderate nitrogen deficiency treatment is 75-90 kg/ha,
the nitrogen fertilizer dosage for normal treatment is 150-180 kg/ha,
the nitrogen fertilizer dosage for excessive nitrogen application treatment is 225 plus 270 kg/ha;
in the phosphorus group treatment process, the application amount of the nitrogen fertilizer is 180 kg/hectare, the application amount of the potassium fertilizer is 120 kg/hectare, and the nitrogen fertilizer and the potassium fertilizer are normal application amounts and are used for pest and disease control management; wherein
The dosage of the phosphate fertilizer for severe phosphorus deficiency treatment is 0 kg/ha,
the dosage of the phosphate fertilizer for moderate phosphorus deficiency treatment is 30-45 kg/ha,
the amount of phosphate fertilizer normally treated is 90 kg/ha,
the excessive phosphorus application treatment is that the usage amount of phosphate fertilizer is 135-;
in the potassium group treatment process, the application amount of the nitrogen fertilizer is 180 kg/hectare, the application amount of the phosphate fertilizer is 90 kg/hectare, both normal application amounts are adopted, and meanwhile, the pest and disease control management work is carried out; wherein the content of the first and second substances,
the dosage of the potassium fertilizer for severe potassium deficiency treatment is 0 kg/hectare,
the consumption of the potassium fertilizer for moderate potassium deficiency treatment is 60-75 kg/ha,
the normal potassium fertilizer dosage is 120 kg/ha,
the dosage of the potassium fertilizer subjected to excessive potassium application treatment is 135-180 kg/ha.
3. The method for diagnosing the nitrogen, phosphorus and potassium nutrients of the rapes based on the near-earth hyperspectral data and the integrated model as claimed in claim 1, wherein the method comprises the following steps: in the step B of the step 2), the specific step of rejecting the spectrum containing the abnormal waveband is as follows:
when the reflectivity of a certain wave band of a certain spectrum is larger than the value of +/-two times of standard deviation of the mean value, marking the wave band of the certain spectrum as an abnormal wave band;
and if the number of the abnormal wave bands of the piece of spectral data is more than or equal to 20, removing the piece of spectral data from the data subgroup.
4. The method for diagnosing the nitrogen, phosphorus and potassium nutrients of the rapes based on the near-earth hyperspectral data and the integrated model as claimed in claim 1, wherein the method comprises the following steps: in the step 6), establishing a random forest classification model for nutrient diagnosis specifically comprises the following steps:
i, establishing a random forest classification model capable of distinguishing severe nitrogen deficiency, mild nitrogen deficiency and normal fertilization by using 10 screened important characteristic wave bands sensitive to nitrogen nutrient abundance;
ii, establishing a random forest classification model capable of distinguishing severe phosphorus deficiency, mild phosphorus deficiency and normal fertilization by using the screened 10 important characteristic wave bands sensitive to phosphorus nutrient abundance;
and iii, establishing a random forest classification model capable of distinguishing severe potassium deficiency, mild potassium deficiency and normal fertilization by using the screened 10 important characteristic wave bands sensitive to the abundance and the deficiency of potassium nutrients.
5. The method for diagnosing the nitrogen, phosphorus and potassium nutrients of the rapes based on the near-earth hyperspectral data and the integrated model as claimed in claim 1, wherein the method comprises the following steps: step 7) a, a small step, wherein nine probability values of each spectrum data classified into each category are obtained by the following method:
i, selecting spectral data of 10 important characteristic wave bands sensitive to nitrogen nutrient deficiency, inputting a random forest classification model obtained by training aiming at the nitrogen nutrient, and obtaining probability values of each spectral data divided into severe nitrogen deficiency, mild nitrogen deficiency and normal fertilization;
ii, selecting spectral data of 10 important characteristic wave bands sensitive to phosphorus nutrient deficiency, inputting a random forest classification model obtained by training aiming at phosphorus nutrients, and obtaining probability values of each spectrum divided into severe phosphorus deficiency, mild phosphorus deficiency and normal fertilization;
and iii, selecting spectral data of 10 important characteristic wave bands sensitive to potassium nutrient deficiency, inputting a random forest classification model obtained by training aiming at potassium nutrients, and obtaining probability values of each spectral data divided into severe potassium deficiency, mild potassium deficiency and normal fertilization.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202010051787.XA CN111257240B (en) | 2020-01-17 | 2020-01-17 | Near-earth hyperspectral data and integrated model-based rape nitrogen-phosphorus-potassium nutrient diagnosis method |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202010051787.XA CN111257240B (en) | 2020-01-17 | 2020-01-17 | Near-earth hyperspectral data and integrated model-based rape nitrogen-phosphorus-potassium nutrient diagnosis method |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN111257240A CN111257240A (en) | 2020-06-09 |
| CN111257240B true CN111257240B (en) | 2021-03-02 |
Family
ID=70954149
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202010051787.XA Active CN111257240B (en) | 2020-01-17 | 2020-01-17 | Near-earth hyperspectral data and integrated model-based rape nitrogen-phosphorus-potassium nutrient diagnosis method |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN111257240B (en) |
Families Citing this family (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN116108318B (en) * | 2023-04-06 | 2023-06-13 | 华中农业大学 | Rape nitrogen fertilizer recommended dressing amount calculation method based on unmanned aerial vehicle multispectral image |
Family Cites Families (6)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US20110125477A1 (en) * | 2009-05-14 | 2011-05-26 | Lightner Jonathan E | Inverse Modeling for Characteristic Prediction from Multi-Spectral and Hyper-Spectral Remote Sensed Datasets |
| CN103278467A (en) * | 2013-05-07 | 2013-09-04 | 江苏大学 | Rapid nondestructive high-accuracy method with for identifying abundance degree of nitrogen element in plant leaf |
| CN104198397B (en) * | 2014-07-30 | 2017-09-05 | 江苏大学 | The method that chamber crop nutrient content is detected under N P and K reciprocation |
| CN104198396B (en) * | 2014-07-30 | 2016-08-31 | 江苏大学 | The method that polarization-hyperspectral technique diagnosis crop N P and K wanes |
| CN106841051B (en) * | 2016-12-19 | 2019-05-10 | 中国科学院南京土壤研究所 | A kind of crop nitrogen nutrition detection method based on visual image fusion value |
| CN108898156A (en) * | 2018-05-28 | 2018-11-27 | 江苏大学 | A kind of green green pepper recognition methods based on high spectrum image |
-
2020
- 2020-01-17 CN CN202010051787.XA patent/CN111257240B/en active Active
Non-Patent Citations (1)
| Title |
|---|
| Using hyperspectral remote sensing techniques to monitor nitrogen, phosphorus, sulphur and potassium in wheat (Triticum aestivum L.);G. R. Mahajan et al.;《Precision Agric》;20141231;第15卷;第499-522页 * |
Also Published As
| Publication number | Publication date |
|---|---|
| CN111257240A (en) | 2020-06-09 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN109187441B (en) | Method for constructing summer corn nitrogen content monitoring model based on canopy spectral information | |
| CN102788752B (en) | Non-destructive detection device and method of internal information of crops based on spectrum technology | |
| CN110874617B (en) | Method for establishing winter wheat leaf nitrogen content estimation model | |
| CN111855589A (en) | Remote sensing inversion model and method for rice leaf nitrogen accumulation | |
| CN110189793B (en) | Hyperspectrum-based wheat nitrogen fertilizer physiological utilization rate estimation model construction and wheat variety classification with different nitrogen efficiencies | |
| CN111855591A (en) | Rice overground part carbon-nitrogen ratio remote sensing inversion model and method | |
| CN116258060A (en) | Soil testing formula fertilization method based on machine learning | |
| CN111855593A (en) | Remote sensing inversion model and method for starch content of rice leaf | |
| CN110569605A (en) | Non-glutinous rice leaf nitrogen content inversion model method based on NSGA2-ELM | |
| CN115479902A (en) | Chlorophyll content determination model construction method based on wavelength and system thereof | |
| CN111257240B (en) | Near-earth hyperspectral data and integrated model-based rape nitrogen-phosphorus-potassium nutrient diagnosis method | |
| CN111855592A (en) | Remote sensing inversion model and method for upper dry matter weight in unit area of rice | |
| CN114694023A (en) | Cotton aphid severity grading method based on spectral index data reconstruction | |
| CN113109272B (en) | Drip irrigation cotton nitrogen nutrition deficiency diagnosis method and system | |
| CN113139717B (en) | Crop seedling condition grading remote sensing monitoring method and device | |
| CN115684107B (en) | Early quantitative monitoring method for rice salt stress based on sunlight-induced chlorophyll fluorescence index | |
| CN114034672B (en) | Crop cold injury degree grading method based on chlorophyll fluorescence technology | |
| CN115479903A (en) | Chlorophyll content measuring method and system based on wavelength | |
| CN114062305A (en) | Single grain variety identification method and system based on near infrared spectrum and 1D-In-Resnet network | |
| CN114548220B (en) | Paddy rice glume blooming state detection method based on handheld spectroradiometer | |
| CN112990692B (en) | Winter wheat yield estimation method based on multiparameter coupling in sub-fertility stage | |
| CN115956499B (en) | Shadow-tolerance evaluation method for peanuts in seedling stage and shadow-tolerance peanut variety screening method | |
| CN111079672B (en) | Grape classification extraction method based on maximum entropy characteristics | |
| CN117391315B (en) | Agricultural meteorological data management method and device | |
| CN113504186B (en) | Method for estimating utilization rate of nitrogen fertilizer in wheat by remote sensing |
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 |