CN102435564B - Method for estimating plant nitrogen content based on three-band spectral index - Google Patents

Abstract

The invention discloses a method for estimating plant nitrogen content based on a three-band spectral index, belonging to the field of plant remote sensing monitoring. The invention overcomes the defect that the sensitivity is reduced when the existing two-waveband spectral index method tends to saturation, has higher precision and accuracy for monitoring the nitrogen content of rice and wheat leaves under different varieties, different water treatment and different nitrogen levels, realizes the real-time acquisition of the nitrogen information of crops, and promotes the wide application of the nondestructive crop monitoring technology based on the spectral technology.

Description

A kind of method based on triband spectrum index estimation plant nitrogen content

Technical field

The present invention relates to agriculture remote sensing of vegetation field, relate in particular to a kind of method based on triband spectrum index estimation plant nitrogen content.

Background technology

Using of nitrogenous fertilizer is the effective way that improves crop yield and quality, but unscientific nitrogenous fertilizer input can make crop absorb fully, causes economic loss and ecological pollution.Therefore, people's accurate nitrogenous fertilizer management that begins one's study in recent years, thus improve the effective way of utilization rate of nitrogen fertilizer, and obtain the basis that field crops nitrogen situation is accurate nitrogenous fertilizer management accurately and real-time.

Traditional method of obtaining plant nitrogen nutrition mainly depends on destructive sampling, in office analysis is measured, though the result is comparatively reliable, has certain hysteresis quality, and has destroyed the vegetation integrality.The remote sensing of vegetation spectral technique can in real time, fast, nondestructively be monitored the crop nitrogen nutrition situation, in time carries out growth of cereal crop seedlings diagnosis, for accurate nitrogen management provides reliable technical support.

Remote sensing of vegetation can be distinguished vegetation type and divide, and can deduce out the important parameter of vegetation, can also estimate some physical quantitys relevant with vegetation photosynthesis more exactly.The ultimate principle of remote sensing of vegetation is the spectral characteristic of plant.Different plants have different spectral signatures, particularly near-infrared band bigger difference are arranged because the institutional framework of blade is different with biochemical component.

Normalized differential vegetation index (NDVI, Normalized Difference Vegetation Index) is defined as the ratio of difference He these two wave band numerical value sums of near-infrared band and the red wave band numerical value of visible light, be the statistical parameter that two wave bands of near-infrared band and visible light red spectral band are represented, that is:

$\mathrm{NDVI}=\frac{R_{\mathrm{nir}}-R_{\mathrm{red}}}{R_{\mathrm{nir}}+R_{\mathrm{red}}}$

The application of NDVI is very extensive, and it is the best indicator of vegetation growth state and coverage, and is closely related with the vegetation distribution situation.Under the situation that has vegetation to cover, NDVI is on the occasion of (＞0), and increases with vegetation coverage.But the main defective of NDVI is: after vegetation coverage was greater than 80%, the NDVI value increased slowly, is state of saturation, has a strong impact on the sensitivity of monitoring.

Summary of the invention

Goal of the invention: the purpose of this invention is to provide a kind of method based on triband spectrum index estimation plant nitrogen content, be used for revising the oversaturated defective of two wave band normalized differential vegetation indexs, set up one based on efficient, the accurate plant nitrogen content monitoring model of triband index.

Technical scheme: for achieving the above object, a kind of method based on triband spectrum index estimation plant nitrogen content of the present invention comprises the steps:

(1) measures plant nitrogen content and carry out spectrum sample;

(2) make up triband spectrum index (mNDVI), described $\mathrm{mNDVI}=\frac{(R_{\mathrm{nir}}-(R_{\mathrm{red}}-k\×R_{\mathrm{green}/\mathrm{blue}}))}{(R_{\mathrm{nir}}+(R_{\mathrm{red}}-k\×R_{\mathrm{green}/\mathrm{blue}}))},$

R wherein _NirBe the reflectivity of near-infrared band, R _RedBe the reflectivity of visible light red spectral band, R _Green/blueBe visible light blue green light wave band reflectivity, k is correction factor;

(3) get mNDVI and the plant nitrogen content correlativity k value when the highest;

(4) respectively near-infrared band (760nm～1000nm), visible light red spectral band (620～760nm), visible light blue green light wave band (gets R arbitrarily in 400～600nm) _Nir, R _Red, R _Green/blueRandom groups is built the linear regression model (LRM) of upright mNDVI and plant nitrogen content jointly, gets the higher coefficient of determination (R ²) corresponding wave band is as optimum R _Nir, R _Red, R _Green/blue

(5) utilize mNDVI to set up the monitoring model of this plant nitrogen content;

(6) monitoring model that obtains of checking procedure (5).

The mensuration of plant nitrogen content adopts destructive sampling in the described step (1), and tested plant sample oven dry is weighed, and adopts Kjeldahl to measure and obtains surveying nitrogen content.Carry out spectrum sample simultaneously, adopt spectral radiometer observed samples in suitable weather, the spectra re-recorded reflectivity.

Simulation k is in (5,5) interval in the described step (3), at interval under 0.5 the step change, gets the highest k value of mNDVI and plant nitrogen content correlativity by simulation.The simulation step-length must be adjusted in conjunction with actual conditions, when interval steps is excessive, can omits best k value, and when interval steps is too small, can increase amount of calculation, and not significantly improve model accuracy.

Set up monitoring model in the described step (5) and set up monitoring model Y=a * mNDVI+b by a plurality of mNDVI with test statistics in the certain hour, and monitoring model is assessed.The monitoring coefficient of determination (R is adopted in the assessment of described monitoring model _Y ²) and standard error (SE) assessment models degree of fitting.

Described step (6) comprising: the blade nitrogen content (LNC) that utilizes this determination of plant of independent time to get is set up testing model y=c * LNC+d.Described testing model adopts testing accuracy (R _A ²), relatively root-mean-square error (RRMSE) and slope (slope) carry out comprehensive evaluation.

Wherein, $\mathrm{RRMSE}=\sqrt{\frac{1}{n}\×\underset{i=1}{\overset{n}{\mathrm{\Σ}}}{(P_i-O_i)}^2}\×\frac{100}{\stackrel{\‾}{O_i}},$ N is the model testing sample number, P _iBe model estimated value, O _iBe the experimental observation value.

This programme can carry out the nitrogen content monitoring to a certain standalone object, also can monitor the nitrogen content of different objects simultaneously.When monitoring the nitrogen content of different objects, the described definite optimum R of step (4) _Nir, R _Red, R _Green/blueThe time, get the higher coefficient of determination (R of each object ²) best result of preceding 10% common factor.

Beneficial effect: the defective that sensitivity descended when a kind of method based on triband spectrum index estimation plant nitrogen content of the present invention had overcome existing two band spectrum index methods and tends to saturated, especially the paddy rice under different cultivars, different in moisture processing, the different nitrogen level, the monitoring of wheat leaf blade nitrogen content had higher precision and accuracy, realize obtaining in real time of crop nitrogen information, promoted the crop non-destructive monitoring broad application based on spectral technique.

Description of drawings

Fig. 1 is the method flow diagram that the present invention estimates the plant nitrogen content;

Fig. 2 is the correlativity synoptic diagram of mNDVI and rice wheat blade nitrogen content under the different correction factor k of the present invention;

Fig. 3 is the modelling effect figure of mNDVI of the present invention and rice wheat blade nitrogen content monitoring model, and ordinate is represented blade nitrogen content (LNC (%)), and horizontal ordinate is corresponding triband vegetation index (mNDVI).Show among the figure through monitoring experiment result for many years and draw: the monitoring model Y=4.4366x-1.0648 of paddy rice from the jointing to the pustulation period, monitoring coefficient of determination R _Y ²=0.870, standard error SE=0.052, sample number N=183; The monitoring model Y=7.3912x-2.7266 of wheat from the jointing to the pustulation period, monitoring coefficient of determination R _Y ²=0.857, standard error SE=0.148, sample number N=228;

Fig. 4 is triband vegetation index mNDVI of the present invention and rice wheat blade nitrogen content model testing test design sketch, and sample is independent time data among the figure, and ordinate is the prediction nitrogen content, and horizontal ordinate is represented certain year actual measurement nitrogen content.This chart open fire rice testing model y from the jointing to the pustulation period ₁=0.7405x+0.7405, testing accuracy R _A ²=0.866, relative root-mean-square error RRMSE=0.131, sample number N=180; The testing model y of wheat from the jointing to the pustulation period ₂=0.9795x+0.3672, testing accuracy R _A ²=0.8583, relative root-mean-square error RRMSE=0.169, sample number N=250.

Embodiment

Below in conjunction with accompanying drawing the present invention is done further explanation.

S101 measures plant nitrogen content, spectrum sample

Sample information is from the different cultivars rice wheat of different ecological point, handles and totally 8 in the rice wheat experimental plot of moisture processing through difference nitrogenous fertilizer.Rice varieties comprises military fragrant round-grained rice 9, magnificent round-grained rice 2, fine, the military fragrant round-grained rice 14 of Japan, 27123 and two line system; Wheat breed comprises peaceful wheat 9, raises wheat 12, Henan wheat 34, Huaihe River wheat 25 and Xu wheat 856.Nitrogenous fertilizer is handled (paddy rice: 0～405kghm ^-2, wheat: 0～270kghm ^-2).Moisture is handled (60 ,-40 ,-20,5cm water layer irrigate field capacity number percent 40%～45%, 60%～65%, 75%～80% and 100%).Ecosite comprises Nanjing and Huaian.The collection of sample information was from 2004～2009 years.

What this enforcement was monitored is the nitrogen content of rice wheat blade, when carrying out spectrum sample, the open-air high spectral radiometer of the FieldSpec Pro FR2500 type back hanging type that adopts U.S. Analytical Spectral Device (ASD) company to produce is sampled to rice wheat canopy.Be chosen in fine, calm during sampling or wind speed carries out when very little, the sampling time is 10:00～14:00h, and the spectral radiometer field angle is 25 °, and apart from the about 1.0m of the straight distance of canopy roof pendant, the ground field range is 0.44m ²3 observation stations of every cell measurement, 10 samplings of each observation station record spectrum represents this residential quarter spectral reflectivity with its mean value.

Mensuration and the spectrum observation of plant nitrogen content are synchronous, adopt the destructive sampling in field, and 10 strains are got in every residential quarter, green blade is pressed the leaf position separate, and are completing under 105 ℃ and are weighing after the oven dry down at 80 ℃.Adopt Kjeldahl to record nitrogen content, go out the actual measurement nitrogen content of canopy blade by various position leaves leaf dry weight weighted calculation.

S102 makes up mNDVI

$\mathrm{mNDVI}=\frac{(R_{\mathrm{nir}}-(R_{\mathrm{red}}-k\×R_{\mathrm{green}/\mathrm{blue}}))}{(R_{\mathrm{nir}}+(R_{\mathrm{red}}-k\×R_{\mathrm{green}/\mathrm{blue}}))};$

R wherein _NirBe the reflectivity of near-infrared band, R _RedBe the reflectivity of visible light red spectral band, R _Green/blueBe visible light blue green light wave band reflectivity, k is correction factor.

K is in (5,5) interval in simulation, under the step change at interval 0.5, and the correlativity of mNDVI and plant nitrogen content, when k=2, the susceptibility of mNDVI and blade nitrogen content is the highest, and then obtains:

$\mathrm{mNDVI}=\frac{(R_{\mathrm{nir}}-(R_{\mathrm{red}}-2\×R_{\mathrm{green}/\mathrm{blue}}))}{(R_{\mathrm{nir}}+(R_{\mathrm{red}}-2\×R_{\mathrm{green}/\mathrm{blue}}))}.$

S103 determines optimum R that should tested plant index _Nir, R _Red, R _Green/blue

Respectively near-infrared band (760nm～1000nm), visible light red spectral band (620～760nm), visible light blue green light wave band (gets R arbitrarily in 400～600nm) _Nir, R _Red, R _Green/blueRandom groups is built the linear regression model (LRM) of upright mNDVI and plant nitrogen content jointly, gets the higher coefficient of determination (R ²) corresponding wave band is as optimum R _Nir, R _Red, R _Green/blue, obtain λ _Nir=924nm, λ _Red=703nm, λ _Green/blue=423nm.Thereby obtain:

$\mathrm{mNDVI}=\frac{(R_{924}-(R_{703}-2\×R_{423}))}{(R_{924}+(R_{703}-2\×R_{423}))}.$

S104 utilizes mNDVI to set up the nitrogen content monitoring model;

Based on the mNDVI of above-mentioned new structure, utilize the test figure in 2004～2009 years each field experiment districts, make up the monitoring model Y that is applicable to rice wheat jointing stage to pustulation period blade nitrogen content ₁And Y ₂A plurality of mNDVI of test statistics are set up monitoring model Y=a * mNDVI+b, and adopt the monitoring coefficient of determination (R _Y ²) and standard error (SE) goodness of fit of monitoring model is assessed.As shown in Figure 3:

Paddy rice: Y ₁=4.4366 * mNDVI (R ₉₂₄, R ₇₀₃, R ₄₂₃)-1.0648; Monitoring coefficient of determination R _Y ²Be 0.870, standard error SE is 0.052;

Wheat: Y ₂=7.3912 * mNDVI (R ₉₂₄, R ₇₀₃, R ₄₂₃)-2.7266; Monitoring coefficient of determination R _Y ²Be 0.857, standard error SE is 0.148.

Fig. 3 has shown the modelling effect figure of mNDVI of the present invention and rice wheat blade nitrogen content monitoring model, and ordinate is represented the blade nitrogen content, and horizontal ordinate is corresponding triband vegetation index.The result shows: the monitoring model Y=4.4366x-1.0648 of paddy rice from the jointing to the pustulation period, monitoring coefficient of determination R _Y ²=0.870, standard error SE=0.052, sample number N=183; The monitoring model Y=7.3912x-2.7266 of wheat from the jointing to the pustulation period, monitoring coefficient of determination R _Y ²=0.857, standard error SE=0.148, sample number N=228.

S105 check nitrogen content monitoring model.

The nitrogen content (LNC) that utilizes independent time testing data rice wheat blade to record is set up testing model y=c * LNC+d.Testing model adopts testing accuracy (R _A ²), relatively root-mean-square error (RRMSE) and slope (slope) be to carrying out comprehensive evaluation. $\mathrm{RRMSE}=\sqrt{\frac{1}{n}\×\underset{i=1}{\overset{n}{\mathrm{\Σ}}}{(P_i-O_i)}^2}\×\frac{100}{\stackrel{\‾}{O_i}},$ N is the model testing sample number, P _iBe model estimated value, O _iBe the experimental observation value.

The testing model of paddy rice from the jointing to the pustulation period: y ₁=0.7405x+0.7405, testing accuracy R _A ²=0.866, slope slope is 0.749, relatively root-mean-square error RRMSE=13.1%;

The testing model of wheat from the jointing to the pustulation period: y ₂=0.9795x+0.3672, testing accuracy R _A ²=0.8583, slope slope is 0.797, relatively root-mean-square error RRMSE=16.9%.

Result and the performance of existing spectrum vegetation index in the estimation of rice wheat blade nitrogen content of above-mentioned testing data are compared, these spectrum vegetation indexs comprise: the normalization index (mND) of triband index correction, ratio value index number (mSR), the enhancement mode vegetation index (EVI) revised, conversion hysteria pigment absorption spectrum index (TCARI), intermediate-resolution ground chlorophyll index (MTCI), two wave band vegetation index normalization difference red limit indexes (NDRE), canopy pigment content index (CCCI), RVI (R ₈₇₀, R ₆₆₀) etc., as shown in table 1:

The rice wheat blade nitrogen content monitoring model performance of the various vegetation indexs of table 1

The result shows that the rice wheat nitrogen content estimation model accuracy that triband spectrum vegetation index mNDVI provided by the present invention sets up is higher, and predictive ability is stronger, is better than existing triband exponential sum two band indexs.

The above only is preferred implementation of the present invention; be noted that for those skilled in the art; under the prerequisite that does not break away from the principle of the invention, can also make some improvements and modifications, these improvements and modifications also should be considered as protection scope of the present invention.

Claims (2)

1. the method based on triband spectrum index estimation plant nitrogen content is characterized in that comprising the steps:

(1) measures plant nitrogen content and carry out spectrum sample;

(2) make up triband spectrum index mNDVI, described $\mathrm{mNDVI}=\frac{(R_{\mathrm{nir}}-(R_{\mathrm{red}}-k\×R_{\mathrm{green}/\mathrm{blue}}))}{(R_{\mathrm{nir}}+(R_{\mathrm{red}}-k\×R_{\mathrm{green}/\mathrm{blue}}))},$

R wherein _NirBe the reflectivity of near-infrared band, R _RedBe the reflectivity of visible light red spectral band, R _Green/blueBe visible light blue green light wave band reflectivity, k is correction factor;

(3) get mNDVI and the plant nitrogen content correlativity k value when the highest;

(4) in near-infrared band 760nm～1000nm, visible light red spectral band 620～760nm, visible light blue green light wave band 400～600nm, get R arbitrarily respectively _Nir, R _Red, R _Green/blueRandom groups is built the linear regression model (LRM) of upright mNDVI and plant nitrogen content jointly, gets higher coefficient of determination R ²Corresponding wave band is as optimum R _Nir, R _Red, R _Green/blue

(5) utilize mNDVI to set up the monitoring model of this plant;

(6) monitoring model that obtains of checking procedure (5);

Set up monitoring model in the described step (5) and set up monitoring model Y=a * mNDVI+b by a plurality of mNDVI with test statistics in the certain hour, and monitoring model is assessed;

Described step (6) comprising: utilize the blade nitrogen content LNC that records of this plant of independent time to set up testing model y=c * LNC+d;

Monitoring coefficient of determination R is adopted in the assessment of described monitoring model _Y ²With standard error SE assessment models degree of fitting;

Described testing model adopts testing accuracy R _A ², relatively root-mean-square error RRMSE and slope slope carry out comprehensive evaluation.

2. a kind of method based on triband spectrum index estimation plant nitrogen content according to claim 1, it is characterized in that: simulation k is (5 in the described step (3), 5) in the interval, under 0.5 the step change, get the highest k value of mNDVI and plant nitrogen content correlativity at interval.

Images