DE102009040944A1 - Method for identification and determination of condition of plants or other objects, involves receiving reflected spectrum of upper surface of plant or another object - Google Patents

Abstract

The method involves receiving a reflected spectrum of an upper surface of a plant or another object, and adjusting a wavelength dependent function at the spectrum by numeric variation by parameters of the function comparisons of the spectrum. The function is applied and is adjusted by adaptation of the parameters at the spectrum, until the deviations fall below a given value. Independent claims are also included for the following: (1) a computer program comprises a programming code unit; (2) a program integrated electronic component comprises a digital and analog component; and (3) a measuring device for a computer program.

Description

The present invention relates to a method for determining the state of plants by remote sensing with the features of the preamble of claim 1.

Remote sensing techniques are used for non-contact measurement of objects and their characterization based on electromagnetic radiation. As signal types, a distinction is made between reflected radiation, fluorescence radiation and thermally induced radiation. The most common measurement is the solar radiation reflected by the measurement object. The measurement methodology includes various multispectral and hyperspectral sensor technologies with wavelengths in the spectral range from 400 nm to 1000 nm and 2500 nm, respectively. The technology covers portable handheld sensors, ground-based systems and aircraft and satellite-based systems. After processing the sensor signal, a specific spectral signature is obtained over the measured wavelength range. Spectral signatures are the result of the reflected radiation from the measured object and show a characteristic pattern, which is composed of the reflection intensity as a function of the wavelength and which varies with the type and condition of the object.

For an object recorded by a corresponding sensor, there are multiple states, which are represented in different signatures ( Guyot, 1990 ). Loud Bushman (1993) A detection mode for a clear image over the measurement object is not enough. Therefore, the interpretation of the signatures often requires additional attributes from other areas.

The general problem is to create a link or classification between the purely technical measurement result and the measured object. If the object is known, the spectral curves represent defined states that describe in their characteristic a specific state of the object.

• 1. Erste Ableitung der spektralen Signatur zur Verdeutlichung sensitiver und signifikanter Wellenlängenbereiche und zur Eliminierung von Bodensignalen, bzw. Hintergrundstörgrößen;
• 2. Zerlegung der Spektren und Transformationsanalyse der relativen Differenzen von z. B. gesundem zu befallenem Pflanzengewebe;
• 3. Partial least square Regression (PLS) Techniken basieren auf multifaktoriellen Regressionsfunktionen oder Splines mit optimierten Variablenzahlen;
• 4. Anwendung multivariater Verfahren;
• 5. Wavelet Transformation als Weiterentwicklung einer Fourier-Transformation;
• 6. Distanzmessung zum Vergleich zweier Spektren, Zerlegung des Spektrum auf objektrelevante Wellenlängen;
• 7. Physikalisch begründete Modelle der Umweltbedingungen; sowie
• 8. Indexbildung.

For analysis and object classification, the following techniques are used in the prior art:

• 1. First derivative of the spectral signature to clarify sensitive and significant wavelength ranges and to eliminate ground signals or background noise;
• 2. decomposition of the spectra and transformation analysis of the relative differences of z. B. healthy plant tissue to be affected;
• 3. Partial least square regression (PLS) techniques are based on multifactorial regression functions or splines with optimized variable numbers;
• 4. application of multivariate procedures;
• 5. wavelet transformation as a further development of a Fourier transformation;
• 6. Distance measurement for comparing two spectra, decomposition of the spectrum to object-relevant wavelengths;
• 7. Physically justified models of environmental conditions; such as
• 8. Index formation.

Mostly, a reflection spectrum analysis is performed by indices. Usually this means a reduction of the information to a few wavebands or information reduction to some (significant) spectral regions.

These indices include:
Multispectral indices: SR (simple ratio), NDVI (normalized difference vegetation index), SAVI (soil-adjusted vegetation index) and many others;
Hyperspectral vegetation indices: RE, (Red Edge), Modified Chlorophyll Absorption in Reflectance Index (MCARI): (ratio of wavelengths of 700, 670, 550 nm;
Normalized ratio indices: Ratio of all possible combinations that could be relevant to a problem.

Blackburn ( Blackburn, GA, 2006, Hyperspectral Remote Sensing of Plant Pigments, J. Experim. Botany, 58, 4, 855-867. ), Bushman ( Buschmann, C., 1993, Remote Sensing of Plants, Propagation, Health and Productivity, Science 80, 439-453 ) or Jensen ( Jensen, John R., 2007, Remote sensing of the environment: an earth resource perspective (Prentice Hall series in geographic information science), 2nd ed. 592 pp ) have summarized the majority of these indices and described their possible uses.

The classification is done by regression from selected index to measured object-related size, z. B. Biomass in weight units. The use of specific tapes leads to loss of information and the dissolution of the information that could be realized according to the current state of sensor technology is neglected in the indices.

From the German patent DE 100 02 880 C1 For example, a method for determining a plant state is known, in which the analysis of the reflection of passive and active illumination, ie sunlight and illumination with an artificial light source, is evaluated. This procedure requires a gear tray supporting the light source for active lighting. It is generally required that for each possible combination of plant / pathogen / environmental interaction, a reference spectrum be present on the microprocessor in order to make an evaluation. This seems unrealistic with the given technical possibilities, because with the multitude of possible combinations, in particular the storage in the processor far exceeds the storage space currently available and in the foreseeable future. It is also not stated in this document in which form the reference spectrum is present and how the comparison is carried out. For a remote sensing method, this method is not suitable.

It is therefore an object of the present invention to provide a method for determining the state of plants or other objects by evaluating the reflected spectrum, which provides improved information over the prior art. In particular, the new method should be able to quantify the recorded spectra, i. H. to numerical values to make them analyzable and comparable. This object is achieved by a method having the features of claim 1.

kann durch Änderung der Parameter A, B₁ bis B_n, α₁ bis α_n und β₁ bis β_n obige Funktion an das Spektrum angepasst werden. Es wird ein Parametervektor ermittelt, der eine vereinfachte Übermittlung des Ergebnisses und eine einfache Auswertung durch Vergleich mit anderen Parametervektoren ermöglicht. Die Anpassung ermöglicht eine sehr gute Unterscheidbarkeit verschiedener Pflanzenzustände oder deren Klassifikation. Diese. Analyse verwendet die Gesamtinformation der momentan technisch möglichen Auflösung entsprechender Sensoren.Because the spectrum of the wavelength nm recorded in the range of visible and near-infrared light is described by the following function:

can be adapted to the spectrum by changing the parameters A, B ₁ to B _n , α ₁ to α _n and β ₁ to β _n above function. A parameter vector is determined which enables a simplified transmission of the result and a simple evaluation by comparison with other parameter vectors. The adaptation allows a very good distinctness of different plant states or their classification. These. Analysis uses the total information of the currently technically possible resolution of corresponding sensors.

This is especially true for a function with n = 2, which is meaningful in the range up to 850 nm. A further improvement is achieved if n = 3 or n = 4. For spectra up to about 2500 nm, functions with n = 5 or n = 6 can be advantageous.

As an example for n = 2, ie for the wavelength range from 400 to 850 nm, ie the visible and the near infrared range, the characteristic signatures with respect to the measured reflections can be described as follows:

The approach is based on a series of additively linked, double Weibull functions. Simple Weibull functions are part of standard statistical works. Thus, the total information of the spectral signature interval between 400 and about 850-1000 nm is compressed to 11 parameters, using the entire information range of the sensor data. The additional consideration of the mid-infrared range requires a corresponding extension by n = 3, 4,..., N, each with 5 additional parameters for each amplitude. The function can be adapted to the spectral data with each parameter estimator.

Parameter Description:

nm is the wavelength in the measurement interval. Parameter A describes the reflection baseline of an object [Unit:% Reflection]. B ₁ and B ₂ are used for model fitting on the Y axis [unit:% reflection]. The four parameters nm _α1 , nm _β1 , nm _α2 , nm _β2 describe the inflection points [unit: nm], the power parameters α1, α2, β1, β2 the degree of slopes of the function [unit: dimensionless].

The parameters of the function can be estimated from the measured spectra using the usual methods (least-deviation squares, trust-region algorithm, maximum likelihood).

The present invention will be described in more detail below with reference to two exemplary embodiments. Show it:

Tables 1 to 4: measured spectra in the range 400 nm-1000 nm and adapted functions as well as the determined parameter sets for healthy vegetation and different stages of the damage; such as

Table 5: Illustration of a spectral signature in the extended wavelength range up to 2500 nm.

Embodiment 1

Table 1

The parameter estimators for the nm _α1 , nm _β1 parameters give a consistent image and correspond approximately to the absorption ranges of anthocyanin and carotenoids, and the upper absorption maxima of chlorophyll a and b. nm _α2 describes the so-called "blue edge" region at 716 nm.

The absolute values of the parameters A, B ₁ , B ₂ change with the scale level (here leaf, plant, stock) or type and / or variety or other factors and serve to classify an object. In the healthy leaf state, the power parameters have value ranges of the upper approximation, higher values lead to marginal changes in the function profile. The parameter β2 addresses the reflection of the water content which is absorbed by the sensors in this wavelength range.

To table 2:

Slight stress symptoms are described for the time being by a reduction of the power parameters α1, β1 and an increase of the parameter B. Thus, changes first appear in the reflection of the visible light. This allows early identification of stress symptoms in the parameters and quantification via stress levels at the level of relative parameter changes (Table 2).

Table 3:

Increasing stress leads to a change in the signature process (Table 3) and thus to an altered parameter vector. The tendency of decreasing values for the potency parameters increases as the stress state of the plant increases. In addition, there are changes in the parameter ranges concerning the near-infrared range.

To table 4:

If the intensity of a stress is lethal, the destruction of the leaf structure is also reflected in the parameter values. The physiologically justifiable parameters of the inflection points, which have so far reacted little to the stress-induced changes, are shifted to value ranges which no longer correspond to the photosynthetically active regions. The deviations in the parameter values reflect the degree of cell structure destruction.

In the given example, the value of the parameter α1 becomes smaller, in addition both power parameters β approximate 1. The change in value describes the transition from incipient stress to the massive signs of stress and death and further destruction and dissolution of the tissue structure.

Embodiment 2

Table 5: Illustration of the spectral signature in the wavelength range up to 2500 nm, data extended from Mahlein et al., 2009 , The example of a healthy sugar beet leaf shows how the model with n = 6 can be adapted to the sensor data. Even using 31 parameters, the system can still be numerically evaluated.

The model behaves consistently over the temporal course of spectral signatures and is able to cover the course of an entire vegetation period of a plant from seed to maturity and senescence.

advantages

Classifications can be made across the set of parameters. It is assumed that the individuality of a parameter vector with its 11 parameters at n = 2 can be used for a classification of stocks, species, varieties and a higher selectivity is achieved. Matching an estimated parameter set with spectral databases can overcome the difficulties of similar classifications.

The vegetation curves represent the overall state of an object. Ie. it summarizes multiple factors in their overall effect. Deviations in some specific parameters allow the quantification of the deviation compared to the normal healthy state at a given time.

Object-related, graduated analysis options:

For the scale level Inventory, Classification of Culture
Culture determination by the culture-specific parameter set, consisting of the 11 parameters (for n = 2 and given physiological time), z. B. for a differentiation of forest and grassland. A deposit of culture-specific databases allows real-time or online analysis;

For a given scale level:
Stress diagnosis by changing the parameters, ie which parameters change in which order of magnitude, examples: nutrient stress, water stress, pathogen stress;

For a given diagnosis, ie stressor is known:
Intensity of stress, acquisition and quantification over time series, parameter values must be interpreted relative to age structure of the object;

Statistical analysis

Comparison of individual parameters over the variance / confidence intervals of the estimators can be used for the statistical analysis of treatment effects. The parameter comparison allows a statistically significant demarcation and allows comparability of signatures on the significance of the statistical tests.

The extended form of the analysis is the comparison of the spectrum over the entire wavelength range as a special method of function comparisons. The absolute and relative change of direction (+/-) of a parameter allows the quantification of a treatment effect and the relative assignment to a possible stressor characterization.

The analysis method is applicable to all spectral reflectance signatures and not reduced to the analysis of vegetation curves. The simple algorithm with a simple, fast solution is the basis for real-time and high-throughput analysis. The conditions of use are given by the existing sensor techniques. The algorithm is applicable to portable hand-held sensors, ground-based systems, to aircraft and satellite-based sensors. It is suitable for all commercially and scientifically used sensors and for use in statistical program packages and geoinformation systems.

Tabelle 2: Signaturbeispiel: leicht gestresstes Blatt (Daten nach Hillnhütter et al., 2009)

Tabelle 3: Signaturbeispiel: stark gestresstes Blatt (Daten nach Hillnhütter et al., 2009)

Tabelle 4: Signaturbeispiel: letal geschädigtes Blatt (Daten nach Hillnhütter et al., 2009)

Tabelle 5: Signaturbeispiel bis zu einem Wellenlängenbereich von 2500 nm (Daten erweitert nach Mahlein et al., 2009), Bsp. gesundes Zuckerrübenblatt, n = 6;

The remote sensing uses, among other things, the reflected radiation for non-contact detection of objects. This measurement method provides characteristic, spectral signatures in the wavelength range considered, which must be analyzed for the detection or classification of a measured object. The various curves of spectral signatures can be described in terms of additively linked double Weibull functions in their entirety. For the wavelength interval of 400 to 850-1000 nm, the total information is compressed to 11 parameters. The object classification takes place via the individuality of the parameter vector or via the relative change of individual parameters. The algorithm provides a basis for real-time and high-throughput analysis, as well as statistical comparison of signature curves in scientific analysis and estimation of treatment effects. Table 1: Signature example: healthy leaf of a sugar beet plant (data according to Hillnhütter et al., 2009)

Table 2: Signature example: slightly stressed sheet (data according to Hillnhütter et al., 2009)

Table 3: Signature example: severely stressed sheet (data according to Hillnhütter et al., 2009)

Table 4: Signature example: lethally damaged leaf (data according to Hillnhütter et al., 2009)

Table 5: Signature example up to a wavelength range of 2500 nm (data expanded according to Mahlein et al., 2009), eg healthy sugar beet leaf, n = 6;

QUOTES INCLUDE IN THE DESCRIPTION

This list of the documents listed by the applicant has been generated automatically and is included solely for the better information of the reader. The list is not part of the German patent or utility model application. The DPMA assumes no liability for any errors or omissions.

Cited patent literature

• DE 10002880 C1 [0010]

Cited non-patent literature

• Guyot, 1990 [0003]
• Bushman (1993) [0003]
• Blackburn, GA, 2006, Hyperspectral Remote Sensing of Plant Pigments, J. Experim. Botany, 58, 4, 855-867. [0008]
• Buschmann, C., 1993, Remote Sensing of Plants, Propagation, Health and Productivity, Natural Sciences 80, 439-453 [0008]
• Jensen, John R., 2007, Remote Sensing of the Environment: An Earth Resource Perspective (Prentice Hall Series in Geographic Information Science), 2nd ed. 592 pp [0008]
• Mahlein et al., 2009 [0027]

Claims (10)

Method for identifying and / or determining the state of plants or other objects by evaluating the reflected spectrum in the wavelength range of 400 nm to 2,500 nm, comprising the following steps: - recording a spectrum reflected from at least one surface of a plant or another object, - adapting a wavelength-dependent function F _(nm) to the spectrum by numerical variation of parameters of the function comparing the spectrum, characterized in that - the function

is used and this is adjusted by adjusting the parameters A, B ₁ to B _n , α ₁ to α _n and β ₁ to β _n to the spectrum until the deviations fall below a predetermined value. A method according to claim 1, characterized in that n = 2 or n = 3 for the wavelength range of 400 nm to 850 nm. Method according to one of the preceding claims, characterized in that n = 4 to n = 6 for the wavelength range from 400 nm to 2,500 nm. Method according to one of the preceding claims, characterized in that for a comparison of the spectrum with a database, a parameter vector comprising the parameters A, B ₁ to B _n , α ₁ to α _n and β ₁ to β _{n is} determined. Method according to one of the preceding claims, characterized in that the spectrum is a reflection spectrum of plants and plant stands. Method according to one of the preceding claims, characterized in that the determination of the parameters takes place according to the method of the smallest deviation squares. Computer program with program code means for carrying out a method according to claims 1 to 6. Programmed integrated electronic module with digital and analog components to carry out a method according to claims 1 to 6. Measuring device for executing a computer program according to claim 7. Measuring device for operating blocks according to claim 8.