CN104198396A - Method for diagnosing nitrogen, phosphorus and potassium deficiency of crops by using polarization-hyperspectral technique - Google Patents
Method for diagnosing nitrogen, phosphorus and potassium deficiency of crops by using polarization-hyperspectral technique Download PDFInfo
- Publication number
- CN104198396A CN104198396A CN201410370826.7A CN201410370826A CN104198396A CN 104198396 A CN104198396 A CN 104198396A CN 201410370826 A CN201410370826 A CN 201410370826A CN 104198396 A CN104198396 A CN 104198396A
- Authority
- CN
- China
- Prior art keywords
- polarization
- phosphorus
- image
- nitrogen
- potassium
- 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.)
- Granted
Links
Landscapes
- Investigating Or Analysing Materials By Optical Means (AREA)
Abstract
The invention discloses a method for diagnosing nitrogen, phosphorus and potassium deficiency of crops by using a polarization-hyperspectral technique, belonging to the technical field of rapid detection of nutrient element deficiency of crops. The method comprises the following steps: acquiring a polarization spectrum of tomato leaves by utilizing polarization spectroscopy, and calculating polarization degree characteristics; acquiring a hyperspectral image of the tomato leaves by utilizing a hyperspectral imaging system, extracting the characteristic wavelength and gray scale and texture characteristics of the tomato leaves under the characteristic wavelength, fusing characteristic layers through the extracted characteristics, and further establishing a prediction model of nutrient content of nitrogen, phosphorus and potassium in the tomato flowering period so as to a method basis for detecting the nutrient content of the crops by utilizing the polarization-hyperspectral technique. Compared with the conventional detection method, the method disclosed by the invention is high in detection speed and easy and convenient to operate; and compared with a single near infrared spectrum or computer vision technological means, the method disclosed by the invention has the advantages that the obtained information is more comprehensive, and the accuracy and stability of the detection result are improved.
Description
Technical field
The invention belongs to the rapid detection technical field that crop alimentary element wanes, refer in particular to a kind of method based on polarization-Gao spectral technique diagnosis chamber crop blade n p k nutrition content.
Background technology
Tomato (LycopersiconesculentumMill.) is one of vegetables crop of China's greenhouse production.Nitrogen, phosphorus, potassium be tomato growth essential be also most important nutrient, nitrogen, phosphorus, potassium nutrition wane and can make its physiology change, and directly have influence on height and the mouthfeel of output, and then affect economic benefit.The research of plant physiology is verified, blade is to one of the most responsive position of nutrition condition reflection, nutritional deficiency can cause the feature generation marked changes such as leaf color, texture, roughness and pore, therefore, the crop leaf of take is diagnosed nutrition condition as research object becomes the focus of Recent study, therefore, can, by the variation of the above-mentioned feature of observation, the nitrogen of tomato, phosphorus, potassium stress state be diagnosed.
At present, the lossless detection method based on spectral technique adopts point source sample mode conventionally, is difficult to embody the reflective character difference of whole leaf area, the abundant characteristic information of blade in the time of cannot fully characterizing crop alimentary and wane.And the shortcoming of diagnostic method based on computer vision technique is to obtain the information of reflection blade interior tissue physiology biochemical characteristic.Therefore, single detection means often accuracy of detection is high and lack universality, and many research is mainly that certain nutrient is lacked or not scarce identification, is difficult to realize the accurate quantitative analysis evaluation of nutrient stress.Polarized spectrum technology and hyper-spectral image technique have the advantage of spectral technique and image technique concurrently in the fusion of feature aspect, the features such as the color that can cause plant water deficit, texture, metamorphosis are carried out visual analyzing, can the anisotropy of plant leaf spectral characteristic be distributed and be evaluated again, and then can improve comprehensive, reliability and the sensitivity of crop nitrogen, phosphorus, potassium Non-Destructive Testing.Some scholars were applied to hyper-spectral image technique in the detection of agricultural product quality and crop pest both at home and abroad in recent years; Polarized spectrum technology is once for the chlorophyllous content of inverting plant leaf blade, but has no the situation of utilizing polarization-Gao spectral technique to diagnose crop nitrogen, phosphorus, potassium nutrition to wane.
Summary of the invention
The present invention, in order to overcome above-mentioned deficiency of the prior art, utilizes polarization spectrum to gather the polarization spectrum of tomato leaf, calculates degree of polarization feature; Utilize Hyperspectral imager to gather the high spectrum image of tomato leaf, and extract gray scale, the textural characteristics of tomato leaf under characteristic wavelength and characteristic wavelength, by the feature of extracting is carried out to the fusion of characteristic layer, and then set up tomato at florescence nitrogen, phosphorus, potassium nutrition content prediction model, for utilizing polarization-Gao spectral technique to detect crop alimentary content supplying method foundation.
The method that polarization-Gao spectral technique diagnosis crop nitrogen phosphorus potassium of the present invention wanes, according to following step, carry out:
(1) set up nitrogen (N), phosphorus (P), potassium (K) Nutrient Stress test sample, each nutrient is divided into five levels
Process,
(2) polarization spectrum collection,
(3) degree of polarization feature extraction,
(4) high spectrum image collection,
(5) image pre-service,
(6) extraction of image texture characteristic,
(7) extraction of spectral signature,
(8) model is set up,
(9) thus whether utilize above-mentioned model to detect crop nitrogen, phosphorus, Determination of Potassium diagnosis crop there is nitrogen, phosphorus, potassium nutrition and wanes.
Wherein in step (1), each nutrient is divided into five levels and processes, refer to according to nitrogen in normal recipe, phosphorus,
25%~150% (mass ratio) of the normal contents of potassium, forms respectively the sample of five kinds of different nutrition levels, is severe successively
Coercing 25% (mass ratio), moderate coerces 50% (mass ratio), slightly coerces 75% (mass ratio), appropriate 100%
(mass ratio), excessive 150% (mass ratio).
Wherein the described polarization spectrum collection of step (2) refers to the polarization spectrum that utilizes polarization spectrum acquisition system to gather greenhouse tomato blade.
Wherein the described degree of polarization feature extraction of step (3) refer to according to tomato leaf in Stokes formula calculation procedure (2) the degree of polarization of polarization spectrum.
Wherein the described high spectrum image collection of step (4) refers to the high spectrum image that utilizes high spectrum image acquisition system to gather greenhouse tomato blade.
Wherein the described image pre-service of step (5) refers to the high spectrum image in step (4) is carried out wave band screening, filtering and utilizes mask to Image Segmentation Using.
Wherein the extraction of the image texture characteristic described in step (6) refers to high spectrum image after pretreatment in step (5), first by principal component analysis (PCA), obtain the sensitive wave length of nitrogen, phosphorus, potassium, then under sensitive wave length, carry out the texture feature extraction based on second order probability statistical filtering.
Wherein the extraction of the spectral signature described in step (7) refers to the textural characteristics extracting in step (6), utilizes interval partial least square method-genetic algorithm preferred feature variable, and spectral signature variable is extracted.
Wherein the described model foundation of step (8) refers to and adopts support vector machine to set up tomato nitrogen in florescence, phosphorus, potassium nutrition content prediction model, specifically according to following step, carry out: (1) is normalized pre-service to sample, determine the quantity of input feature vector;
(2) based on grid search method (GS) and particle swarm optimization algorithm (PSO), to carrying out support vector machine recurrence (SVR) parameter, carry out optimizing respectively;
(3) method characteristic layer of degree of polarization feature, textural characteristics and spectral signature employing SVR being merged is set up the forecast model of tomato nitrogen in florescence, phosphorus, potassium nutrition content.
Beneficial effect of the present invention: the diagnostic method that utilizes the multidimensional optical information tomato nutrient based on polarization-Gao spectrum to coerce, take greenhouse tomato as research object, by soilless culture, cultivate nitrogen, phosphorus, the potassium of different nutrition levels and coerce alternately sample, the polarization spectrum that utilizes the polarization spectrum acquisition system collection tomato leaf of independent research, extracts degree of polarization feature and carries out based on degree of polarization feature in conjunction with appropriate chemometrics method; Utilize hyper-spectral image technique to extract respectively image texture characteristic and spectral signature, then, adopt mode identification method to carry out characteristic layer fusion these three kinds of features, obtain to greatest extent the proper vector that can give full expression to tomato leaf inside and outside integrated information, set up the tomato nutrient of the many information fusion of polarization-Gao spectrum and coerce quantitative model.
The present invention obtains the integrated information that can give full expression to the inner nutrient of tomato leaf and formalness to greatest extent, realizes and having complementary advantages, and has expanded quantity of information, solves monotechnics means diagnosis tomato nutrient and coerces that precision is low, the incomplete problem of information.The present invention compares with conventional sense method, fast, the easy and simple to handle convenience of detection speed; Compare with single near infrared spectrum or computer vision technique means, more comprehensively, accuracy and the stability of testing result all increase the information obtaining.The diagnostic method that multidimensional optical information tomato nutrient based on polarization-Gao spectrum provided by the invention is coerced, can realize nutritional information quick detection in process of crop growth.This invention, for science precision irrigation provides reference, has direct significance to improving intelligent management level, crop yield and raising crop quality.
Accompanying drawing explanation
Fig. 1. Hyperspectral imager,
Wherein: 1. light box; 2. light source; 3. controller; 4. computing machine; 5. near infrared camera; 6. imaging spectrometer; 7. stepper motor; 8. glass optical fiber lamp.
Fig. 2. polarization spectrum acquisition system,
Wherein: 1. light source; 2. luminous source optical fiber lamp; 3. detection optical fiber; 4. wheel measuring mechanism; 5. sample stage; 6. spectrometer; 7. microgalvanometer; 8. computing machine.
Embodiment
Tomato take below as example, by reference to the accompanying drawings the present invention is explained in further detail.
The high spectrum image acquisition system adopting in the specific embodiment of the invention is consulted Fig. 1.Utilize the high spectrum image acquisition system shown in Fig. 1 to gather greenhouse tomato blade high spectrum image, it comprises near infrared camera 5 (XEVA-FPA-1.7-320, XenICs, Leuven, Belgium), spectral range 900-1700nm, imaging spectrometer 6 (ImspectorN17E, Spectral ImagingLtd., Finland), resolution is 5nm, the direct current tunable light source 2 (2900-ER+9596-E of 150W halogen tungsten lamp, Illumination Technologies, Inc., EastSyracuse, NY, USA), displacement unit is by stepper motor 7 (MTS120, Beijing Optical Instrument Factory, Beijing, Chinese) and controller 3 (SC100, Beijing Optical Instrument Factory, Beijing, China) form, imaging spectrometer can gather image transmitting to computing machine 4 (DELL Inspiron530s, USA) in, the collection that glass optical fiber lamp 8 is image provides necessary illumination.Near infrared camera 5, imaging spectrometer 6, stepper motor 7 and glass optical fiber lamp 8 are positioned at light box 1.The present invention tests year September in March, 2012 to 2012 in the Venlo type greenhouse of Jiangsu University's modern agriculture equipment and key lab of technical education portion.Cultivating kind is the L-402 that Liaoning Academy Of Agricultural Sciences Vegetable Research Institute cultivates.For guaranteeing that the validity feature that the basic research in early stage can divide tomato accurately extracts, the present invention adopts cultivation technique without soil to carry out sample cultivation.In the situation that guaranteeing other nutritive element balances, nitrogen, phosphorus, potassium are accurately controlled, to obtain pure Nutrient Stress sample.Nutrient solution pH value is that 6-6.5, EC value are 1.2ms/cm.After planting first, water normal nutrient solution, for avoiding in perlite nutritional labeling residual, when tomato growth to the phase in strong sprout proceeds to nutritional deficiency breeding phase, carry out secondary transplanting.
Nutrient Stress test sample is divided into three groups, nitrogen (N), phosphorus (P), potassium (K), in every group, each nutrient is divided into five levels and processes, according to 25%~150% of the normal contents of nitrogen in normal recipe, phosphorus, potassium, forming respectively the sample of five kinds of different nutrition levels, is that severe water stress 25%, moderate are coerced 50%, slightly coerced 75%, appropriate 100%, excessive 150% successively.(being mass ratio)
The mensuration work of chemical score content is synchronizeed and is carried out with spectroscopic test, and the sample of cultivation sorts in the valve bag of numbering, and puts professional plant antistaling box into, takes back immediately spectrographic laboratory, starts polarized reflectance spectrum experiments of measuring and high spectrum image collection.After collection finishes, blade is put into baking oven, dry to constant weight for 80 ℃ and be placed in exsiccator.Adopt Kjeldahl's method (GB/T5009.5-1985) to measure the total nitrogen content of sample, instrument is Auto Analyzer3 type Continuous Flow Analysis instrument (the Seal Analytical Instruments Co. that Britain SEAL company produces, Ltd, England).Adopt molybdenum-antimony anti-spectrophotometric method (GB11893-1989) to measure the content of phosphorus in sample, instrument is VARIAN Oncology Systems's ultraviolet-visible pectrophotometer (Varian Inc., Palo Alto, USA; Model Cary100).Adopt flame photometry (GB/T18633-2002) to measure the content of potassium in sample, instrument is BWB-XP multielement flame photometer (BWB Co., British).After chemical score assay, for model below, set up and proofread and correct.
(1) polarization spectrum collection:
Polarized reflectance spectrum measuring and analysis system is seminar's independent research, this apparatus measures wavelength coverage 350-1000nm, as shown in Figure 2, the luminous source optical fiber lamp 2 that light source 1 picks out is arranged on the wheel measuring mechanism 4 in left side, the detection optical fiber 3 that spectrometer 6 picks out is arranged on the wheel measuring mechanism 4 on right side, sample stage 5 is for placing testing sample, and microgalvanometer 7 connects respectively spectrometer 6 and operating computer 8.
Polarized reflectance spectrum and high spectrum image collection guarantee to carry out fast successively within the short time of trying one's best, to guarantee the unitarity of sample.Before data acquisition, two cover instruments all need to carry out preheating and Hei Chang and white demarcation, the systematic error causing to eliminate environmental factor, and every blade is measured 3 times, averages as final measurement.Spectrum experiment sample is every group each 120, nitrogen, phosphorus, potassium, acquisition time is point in mornings 8, and unified seven leaves of every strain of selecting, the Ye Kuanying of blade is greater than 2cm, once adopting, pack rapidly that valve bag is sealed into and carry out scene numbering down, put professional plant antistaling box into, take back immediately spectrographic laboratory, start polarized reflectance spectrum experiments of measuring and high spectrum image collection.In experiment, for preventing the interference of external environment light, polarized reflection light spectrometry is carried out in darkroom; High spectrum image gathers in light box.After measurement finishes, blade is put into baking oven, dry to constant weight for 80 ℃ and be placed in exsiccator the use in order to chemical score mensuration.
(2) degree of polarization calculates:
Degree of polarization P is the ratio of intensity and this light wave total intensity of full polarized component, proves that thus the degree of polarization P of light beam can try to achieve by formula (1):
Natural atmosphere background and target object to the polarization effect of light source incident in, circularly polarized component is few, less in the scope that circular component can detect at instrument, with respect to error, can ignore.For the reflected light on tomato leaf surface, its circular component is ignored the impact of result, therefore the V component hypothesis in its Stokes vector is about 0, has only used I, Q, and tri-parameters of U, degree of polarization now can be write as:
Thereby, only need measure the light intensity of the linear polarization component of light on 0 °, 90 ° ,+45 ° ,-45 ° directions, just can pass through formula (2) and determine the polarization state of Ray Of Light completely, according to the research contents of a upper joint, under optimum combination condition, extract respectively the average polarization spectrum in detector place polaroid each tomato leaf samples region on 0 °, 90 ° ,+45 ° ,-45 ° directions.Correlation analysis is carried out in degree of polarization data pointwise to the horizontal tomato leaf sample of different n p k nutritions, obtain N, the P of 350~1000nm wavelength band tomato, the correlativity curve of K content, according to the result selection of correlation analysis and N, P, degree of polarization feature that K content correlativity is high.
Owing to there is extremely strong correlativity in the degree of polarization feature of adjacent band, suppose to choose at random and must have repetition from adjacent wave length, make the characteristic wavelength selected or the combination of characteristic wavelength can not possess representativeness, therefore need in the extremely significant sensitive band of correlativity, choose the weak wavelength of correlativity as sensitive wave length, that is to say that sensitive wave length must meet the following conditions: in the wavelength band that (1) draws in correlation analysis; (2) between the wave band of guaranteeing to elect, there is weak dependence; (3) in order to meet the needs of follow-up study, select N, P, the total sensitive wave length of K, can select N, P, K all than more sensitive wavelength as far as possible.Concrete implementation step is: the sensitive band first correlation analysis being drawn is divided into several small bands, and each small band contains 10 wavelength; Then according to large young pathbreaker's small band of related coefficient, sort, form band subset Ui; The wave band of related coefficient maximum is put into and selects Band Set U
s; According to front and back order, from small band subset Ui, select not at U successively
sin wave band, if it meets and U
sin the coefficient R of all wave bands be all less than 0.8, put into U
sin, otherwise rejected.According to screening conditions, select and can represent that degree of polarization and tomato leaf N, P, K content have the characteristic wavelength of remarkable relation: 655.41nm, 744.48nm, 850.58nm is the total sensitive wave length of N, P, K; And N, P, the distinctive sensitive wave length of K are respectively 380.49nm, 914.56nm, 488.42nm.
(3) high spectrum image collection:
The collection of high spectrum image data is based on SpectralCube (Spectral Imaging Ltd., Finland) software platform; The spectral range of actual acquisition is 871.6~1766.3nm, and spatial resolution is 62.5um, and sampling interval is 3.5nm, once gathers to obtain in sampled light spectral limit, to take independently high spectrum image of 256 width that 3.5nm is interval.
Determine that the time shutter of near infrared camera is to guarantee the clear of image, the speed of definite displacement platform is to avoid the distortion of picture size and spatial resolution simultaneously.Relatively determine that by analysis the time shutter is for being 20ms, the translational speed of displacement platform is 1.25mm/s.During data acquisition, first carry out Hei Chang and white demarcation, set reflectivity range, and then utilize the fertile thatch wave filter of second order Bart to carry out digital filtering, remove noise.
(4) image pre-service:
For fear of the illumination of sensor dark current and light source skewness and make image contain larger noise under each wave band, or cause the larger brightness value difference under different wave length different, first institute's original sample image is demarcated.Complete white uncalibrated image W obtains by scanning barium sulphate standard white plate; Complete black uncalibrated image B collects after covering the camera lens of video camera.According to formula (3), the absolute image I collecting is become to relative image R.
The scope of the curve of spectrum of the high spectrum of tomato leaf gathering is 390~1050nm.The curve of spectrum exists obvious noise with the above region of 950nm below 450nm, therefore, in the data handling procedure in later stage, chooses within the scope of 450~950nm, and the high-spectral data of totally 388 wave bands carries out subsequent analysis research.
The present invention adopts the median filtering method of 5 * 5 windows to carry out filtering to image, and the medium filtering of 5 * 5 windows is noise reduction but also undistorted not only, to subsequent characteristics, extracts more favourable.By image being cut apart to the bianry image obtaining at 700nm place, set up mask, mask is a kind of special image in ENVI, be one by 0 and 1 binary picture forming.When having mask to participate in a Hyperspectral imagery processing, 1 value region is processed, 0 value region conductively-closed, be that background is black, the part of mask will not participate in follow-up computing, white portion is the tomato leaf region splitting, and participates in subsequent treatment, has reduced greatly the impact of background on feature extraction.
(5) extraction of image texture characteristic:
Principal component analysis (PCA) (PCA) is method the most frequently used in high-spectrum image dimensionality reduction.Its target is to seek a kind of conversion, and raw data is mapped to a new space.Obtain respectively the weight coefficient curve map of the first five major component of nitrogen, phosphorus, potassium after PCA conversion, the weight coefficient curve of the nitrogen of drawing out, phosphorus, high the first five major component of spectrum of potassium.According to extracting the main crest of weight coefficient curve and the corresponding wavelength in trough place as sensitive wave length.The after comparing wherein characteristic wavelength of N element is respectively: 464.91nm, 566.29nm, 696.28nm, 724.66nm; The characteristic wavelength of P element is respectively: 474.85nm, 567.54nm, 693.71nm, 738.89nm; The characteristic wavelength of K element is respectively: 565.03nm, 691.14nm, 733.71nm, 766.14nm.
Under characteristic wavelength, adopt texture feature extraction intermediate value, covariance, homogeney, entropy, diversity, second moment, contrast and the correlativity based on second order probability statistical filtering.
(1) intermediate value (Mean):
The median operation is here similar to convolution, but what calculate is not weighted sum, but the pixel in neighborhood is sorted by gray level, then select the intermediate value of this group as output, its major function is that the larger pixel of difference of transference surrounding pixel gray-scale value changes to take the value approaching with pixel value around.
(2) covariance (Variance):
In formula:
What generally covariance represented intuitively is the expectation of two variable global errors.Here covariance is for weighing the global error of two variablees.
(3) homogeney (Homogeneity):
Claim again unfavourable balance distance, it can measure the number of image texture localized variation.It is less that its value greatly illustrates that image texture changes between zones of different, and part is very even, and the bright local distribution of value novel is inhomogeneous.
(4) entropy:
The tolerance of the quantity of information that image has.It has reflected confusion degree and the unordered degree of image, has represented heterogeneity and the complicacy of texture in image, and entropy more large texture is more complicated, and the less texture of entropy is more even.
(5) diversity (Dissimilarity):
Diversity size is for the different degree between all elements in the direction of the direction considering element and be expert at or row, change index for other row or other row, if otherness is larger in certain direction of image, the DIS value of this direction will be greater than the DIS value of other directions.
(6) second moment (AngularSecondMoment):
Be called again energy, it is by the calculating of gray level co-occurrence matrixes element value quadratic sum, the degree of thickness between the texture particle that reflection gradation of image distributes.The higher explanation texture of value of second moment is thicker, and the lower explanation texture of the value of second moment is thinner.
(7) contrast (Contrast):
Change depth degree and clear picture degree that index has contrasted texture rill.If rill is darker, CON value is larger, and visual effect is more clear and image is more obvious; CON value is less, shows that rill is shallow, and effect is fuzzy.
(8) relevant (Correlation):
In formula:
This index for the element of weighing gray level co-occurrence matrixes in the row direction, or the similarity on column direction.It can reflect the correlativity of image local gray scale.If the similarity degree of the higher explanation element of COR value on the line direction of image or on column direction is higher, otherwise lower.
Then these eight characteristic parameters that count are carried out to correlation analysis with the measured value of tomato leaf N, K, P content respectively, the results are shown in Table 1, table 2 and table 3.
The related coefficient of textural characteristics and tomato N content under table 1 sensitive wave length
The related coefficient of textural characteristics and tomato P content under table 2 sensitive wave length
The related coefficient of textural characteristics and tomato K content under table 3 sensitive wave length
Under four sensitive wave lengths the correlativity of the middle value tag of textural characteristics and N, P, K three's content all a little less than, illustrate that it is invalid feature, should give up.According to table 1, table 2 and table 3, the feature that preferably correlativity is high is as the characteristic variable of setting up for model, in order to facilitate follow-up research, from related coefficient, find out six with the correlativity of nitrogen, phosphorus, potassium higher feature all, the common high spectral signature as nitrogen phosphorus potassium, is respectively: VAR
693.71, CON
566.29, DIS
693.71, ENT
733.71, ASM
566.29, COR
733.71.From remaining feature, select respectively again two features that correlativity is the highest as the distinctive high spectral signature of nitrogen phosphorus potassium, respectively be: nitrogen ASM
464.91, COR
464.91; Phosphorus HOM
693.71, ENT
474.85; Potassium HOM
762.24, ENT
762.24.
(6) extraction of spectral signature:
Genetic algorithm (GA) has been simulated organic sphere nature genetic mechanism and natural selection, wishes to solve the optimization problem between variable by simulation.GA is than complicated, the nonlinear optimal problem that are more suitable for solving, and conventional search methods effect when solving problems is not good enough, genetic algorithm at present oneself through being widely used in preferred near infrared spectrum feature wave number point.The thought of region offset minimum binary preferred feature spectrum district and genetic algorithm preferred feature wave number point is combined, adopt the preferably characteristic spectrum area of spectrum of interval partial least square-genetic algorithm (iPLS-GA).First with iPLS, cut the wave band of selecting modeling accuracy the best, then carry out again modeling after adopting the method for GA to optimize several variablees that can represent this wave band, wish that the characteristic variable by trying one's best few replaces full spectrum data can obtain good model accuracy again simultaneously.
IPLS-GA model optimizes 4 variablees on the basis of iPLS model, and the sensitive wave length of N element, is respectively 741.48nm, 755.74nm, 767.44nm, 784.37nm.IPLS-GA model optimizes 8 variablees on the basis of iPLS model, and the sensitive wave length of P element, is respectively 770.04nm, 779.16nm, 813.12nm, 824.92nm.Optimize 4 variablees, the sensitive wave length of K element, is respectively 618.23nm, 630.97nm, 645.00nm, 705.30nm.
(7) model is set up:
According to the research of front two chapters, six high spectral signature VAR of the tomato leaf of extraction
693.71, CON
566.29, DIS
693.71, ENT
733.71, ASM
566.29, COR
733.71; Four reflection spectrum characteristics are respectively: nitrogen 741.48nm, 755.74nm, 767.44nm, 784.37nm; Phosphorus 770.04nm, 779.16nm, 813.12nm, 824.92nm; Potassium 618.23nm, 630.97nm, 645.00nm, 705.30nm.Also comprise total degree of polarization feature 655.41nm, 744.48nm, 850.58nm, peculiar degree of polarization feature N, P, K are respectively 380.49nm, 914.56nm, 488.42nm; While forming many information combination feature space, adopt respectively total characteristic variable to add the form of distinctive characteristic variable, each element amounts to the foundation that 14 characteristic variables participate in model.N, P, K respectively have 96 tomato nutrients to coerce the sample of blade for the foundation of model.
To tomato leaf carry out SVR recurrence before, the numerical value of the gray scale textural characteristics extracting due to high spectrum differs greatly, the difficulty of numerical evaluation during for fear of training, first adopt maximum-minimum requirement method to training sample and checking sample are carried out to standardization, the characteristic value normalization of all samples is arrived in [0,1] scope, and to parameter (C, g), to after being optimized, set up the optimum prediction model of tomato leaf nutrient content.
1. in the nitrogen element S VR of GS parameter optimization model
First set the scope of (C, g), this research is selected, C ∈ [2
-8, 2
8], g ∈ [2
-8, 2
8] wherein C and g search step all length be 0.5, cross validation broken number is 10.Work as C=32, during g=0.35355, cross validation root-mean-square error is minimum.Therefore, the optimal parameter of finding based on grid search method is: C=32, g=0.35355, CVmse=0.46807 now, the model of setting up is with this understanding best, the model tuning collection RMSECV=0.1166% setting up, Rc=0.9562, forecast set Rp=0.9291, RMSEP=0.2217%.
2. the nitrogen element S VR parameter optimization model based on PSO
Utilize PSO to carry out the optimizing of SVR parameter, parameter c
1=1.5, c
2=1.7, evolutionary generation is 100, and population quantity pop is 20, and cross validation broken number is 10.The optimum of parameter optimization is: work as C=40.4315, during g=0.2849, cross validation square error is minimum, now CVmse=0.5303.The optimal parameter of utilizing PSO optimized algorithm to find is carried out SVR modeling to calibration set, the calibration set of the parameter optimization model based on SVR-PSO method and the regression fit result of forecast set are: model tuning collection RMSECV=0.1268%, Rc=0.9521, forecast set Rp=0.9289, RMSEP=0.2215%.
2. the SVR parameter optimization model based on GS
First set the scope of (C, g), during the modeling of P element, select C ∈ [2
-8, 2
8], g ∈ [2
-8, 2
8] wherein C and g search step all length be 0.5, cross validation broken number is 5.Work as C=4, during g=32, cross validation root-mean-square error is minimum, CVmse=2.5835 now, the best results of parameter optimization.The calibration set of the parameter optimization model of P element based on SVR-GS and the regression fit result of forecast set are: calibration set Rc=0.9987, RMSECV=0.0099%, forecast set Rp=0.8978, RMSEP=0.1950%.
3. the SVR parameter optimization model based on PSO
Utilize PSO to carry out the optimizing of SVR parameter, parameter c
1=1.5, c
2=1.7, evolutionary generation is 100, and population quantity pop is 20, and cross validation broken number is 5.The result of parameter optimization is: work as C=5.534, and during g=30.0753, cross validation square error CVmse=2.5831 now.The calibration set of the parameter optimization model based on SVR-PSO method and the regression fit result of forecast set are: model tuning collection Rc=0.9988, RMSECV=0.0100%, forecast set Rp=0.8998, RMSEP=0.1912%.
4. the SVR parameter optimization model based on GS
First set the scope of (C, g), during the modeling of K element, select C ∈ [2
-8, 2
8], g ∈ [2
-8, 2
8] wherein C and g search step all length be 0.5, cross validation broken number is 5.Work as C=5.6569, during g=16, cross validation root-mean-square error is minimum, CVmse=2.0921 now, and the model of setting up is with this understanding best.The regression fit result of the K element of setting up based on SVR-GS is: calibration set Rc=0.9988, and RMSECV=0.0098%, forecast set Rp=0.9101, RMSEP=0.1417%, has reached good prediction effect.
6. the SVR parameter optimization model based on PSO
Utilize PSO to carry out the optimizing of SVR parameter, algorithm parameter c
1=1.5, c
2=1.7, evolutionary generation is 100, and population quantity pop is 20, and cross validation broken number is 5.The result of parameter optimization is: work as C=7.5929, and during g=14.0114, cross validation square error CVmse=2.0828 now.The calibration set of the parameter optimization model based on SVR-PSO method and the regression fit result of forecast set are: model tuning collection Rc=0.9985, RMSECV=0.0096%, forecast set Rp=0.9036, RMSEP=0.1488%.
(8) utilize above-mentioned model to detect crop nitrogen, phosphorus, potassium, whether diagnosis crop nitrogen, phosphorus, potassium nutrition occur wanes:
After model is set up, adopt the spectral information of hyper-spectral data gathering tomato leaf to be measured, the above-mentioned model of substitution after treatment, can calculate nitrogen, phosphorus, the potassium nutrition content of tomato leaf, and whether diagnosis crop nutritional deficiency occurs.
The present invention be take tomato as research object as can be seen from the above-described embodiment, adopts polarization-Gao spectrum Dynamic Non-Destruction Measurement, detects nitrogen, phosphorus, the potassium nutrition content of tomato during florescence.First adopt the polarization spectrum acquisition system and the Hyperspectral imager that build voluntarily to gather tomato leaf high spectrum image data; Extract degree of polarization, image texture and spectrum and amount to 14 characteristic variables, for these 14 characteristic variables, adopt SVR method to set up regression model, the predicted value of model and the coefficient R of measured value are all higher, and model accuracy and stability are higher.
More than just in conjunction with a specific embodiment (take tomato nitrogen, phosphorus, potassium be example); the present invention is further understood in exemplary illustration and help; but embodiment detail is only for the present invention is described; do not represent that the present invention conceives lower whole technology embodiment; therefore should not be construed as the total technology embodiment of the present invention is limited; some are In the view of technician; not departing from the unsubstantiality of inventive concept changes; for example, to there is simple the change or replacement of technical characterictic of same or similar technique effect, all belong to protection domain of the present invention.
Claims (8)
1. the method that polarization-Gao spectral technique diagnosis crop nitrogen phosphorus potassium wanes, is characterized in that carrying out according to following step:
(1) polarization spectrum collection,
(2) degree of polarization feature extraction,
(3) high spectrum image collection,
(4) image pre-service,
(5) extraction of image texture characteristic,
(6) extraction of spectral signature,
(7) model is set up, and adopts support vector machine to set up crop flowers phase nitrogen, phosphorus, potassium nutrition content prediction model;
(8) utilize above-mentioned model to detect crop nitrogen, phosphorus, Determination of Potassium, thereby whether diagnosis crop nitrogen, phosphorus, potassium nutrition occur, wane.
2. polarization spectrum collection according to claim 1 refers to the polarization spectrum that utilizes polarization spectrum acquisition system to gather greenhouse tomato blade.
3. degree of polarization feature extraction according to claim 1 refers to the degree of polarization feature of calculating tomato leaf according to Stokes formula.
4. high spectrum image collection according to claim 1 refers to the high spectrum image that utilizes high spectrum image acquisition system to gather greenhouse tomato blade.
5. image pre-service according to claim 1 refers to image is carried out to wave band screening, filtering and utilize mask to Image Segmentation Using.
6. the extraction of image texture characteristic according to claim 1 refers to the sensitive wave length that first obtains nitrogen, phosphorus, potassium by principal component analysis (PCA), then under sensitive wave length, carries out the texture feature extraction based on second order probability statistical filtering.
7. the extraction of spectral signature according to claim 1 refers to and utilizes interval partial least square method-genetic algorithm preferred feature variable, and spectral signature variable is extracted.
8. model according to claim 1 is set up, and it is characterized in that specifically according to following step, carrying out: (1) is normalized pre-service to sample, determines the quantity of input feature vector;
(2) based on grid search method (GS) and particle swarm optimization algorithm (PSO), to carrying out support vector machine recurrence (SVR) parameter, carry out optimizing respectively;
(3) method characteristic layer of degree of polarization feature, textural characteristics and spectral signature employing SVR being merged is set up the forecast model of tomato nitrogen in florescence, phosphorus, potassium nutrition content;
(4) other sample data of simultaneously obtaining while utilizing sample collection, tests to set up model.
Priority Applications (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201410370826.7A CN104198396B (en) | 2014-07-30 | 2014-07-30 | The method that polarization-hyperspectral technique diagnosis crop N P and K wanes |
Applications Claiming Priority (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201410370826.7A CN104198396B (en) | 2014-07-30 | 2014-07-30 | The method that polarization-hyperspectral technique diagnosis crop N P and K wanes |
Publications (2)
Publication Number | Publication Date |
---|---|
CN104198396A true CN104198396A (en) | 2014-12-10 |
CN104198396B CN104198396B (en) | 2016-08-31 |
Family
ID=52083721
Family Applications (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
CN201410370826.7A Expired - Fee Related CN104198396B (en) | 2014-07-30 | 2014-07-30 | The method that polarization-hyperspectral technique diagnosis crop N P and K wanes |
Country Status (1)
Country | Link |
---|---|
CN (1) | CN104198396B (en) |
Cited By (17)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN104677853A (en) * | 2015-01-27 | 2015-06-03 | 天津大学 | Fresco scaling damage assessment method based on near-infrared hyperspectrum |
CN104865194A (en) * | 2015-04-03 | 2015-08-26 | 江苏大学 | Detection apparatus and method for pesticide residues in vegetable based on near infrared, fluorescence and polarization multi-spectrum |
CN107315851A (en) * | 2017-05-19 | 2017-11-03 | 中国科学院长春光学精密机械与物理研究所 | A kind of robustness Film Design method of wide spectrum multiplayer films in EUV |
CN108064531A (en) * | 2017-12-05 | 2018-05-25 | 江苏大学 | Device is applied in a kind of light simplified greenhouse liquid manure filling |
CN108254396A (en) * | 2017-12-05 | 2018-07-06 | 江苏大学 | A kind of tomato seedling water stress detection method based on micro-CT and polarization-high light spectrum image-forming multiple features fusion |
CN108323295A (en) * | 2017-12-05 | 2018-07-27 | 江苏大学 | A kind of seedling stage crop liquid manure based on multiple dimensioned habitat information detects and controls method and device |
CN109521001A (en) * | 2018-11-19 | 2019-03-26 | 华南理工大学 | A kind of flying marking measuring method based on PSO and ε-SVR |
CN109580513A (en) * | 2018-12-18 | 2019-04-05 | 塔里木大学 | A kind of remote sensing jujube moisture content detecting method near the ground and device |
CN109827910A (en) * | 2019-01-22 | 2019-05-31 | 塔里木大学 | A kind of quick monitoring process method of orchard establishing data |
CN109961073A (en) * | 2019-03-29 | 2019-07-02 | 云南电网有限责任公司电力科学研究院 | The acquisition methods and device of a kind of transmission line of electricity and shaft tower information |
CN110455722A (en) * | 2019-08-20 | 2019-11-15 | 中国热带农业科学院橡胶研究所 | Rubber tree blade phosphorus content EO-1 hyperion inversion method and system |
CN110553982A (en) * | 2019-08-29 | 2019-12-10 | 安徽科技学院 | Polarization hyperspectral image acquisition system |
CN110579444A (en) * | 2019-09-28 | 2019-12-17 | 中国科学院合肥物质科学研究院 | Visible-near infrared light polarization spectrum detection probe and detection method thereof |
CN111220734A (en) * | 2020-02-26 | 2020-06-02 | 江苏大学 | Method for evaluating nutritional state of crops in latent nutrient deficiency period by using sensitive metabolites |
CN111257240A (en) * | 2020-01-17 | 2020-06-09 | 华中农业大学 | Near-earth hyperspectral data and integrated model-based rape nitrogen-phosphorus-potassium nutrient diagnosis method |
CN111948155A (en) * | 2020-07-30 | 2020-11-17 | 中国科学院西安光学精密机械研究所 | Fine full spectrum combined GS-SVR improved complex water body nitrate quantitative analysis method |
WO2021111449A1 (en) * | 2019-12-05 | 2021-06-10 | The State Of Israel, Ministry Of Agriculture & Rural Development, Agricultural Research Organization (Aro) (Volcani Center) | Device for measuring element concentrations in plant leaves and method of implementing the same |
Citations (1)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN103018180A (en) * | 2012-12-11 | 2013-04-03 | 江苏大学 | Method and device for detecting cotton diseases based on multisource light information technology |
-
2014
- 2014-07-30 CN CN201410370826.7A patent/CN104198396B/en not_active Expired - Fee Related
Patent Citations (1)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN103018180A (en) * | 2012-12-11 | 2013-04-03 | 江苏大学 | Method and device for detecting cotton diseases based on multisource light information technology |
Non-Patent Citations (5)
Title |
---|
张莉莉: "利用偏振高光谱反演植被叶绿素含量", 《中国优秀硕士学位论文全文数据库 基础科学辑》 * |
张金恒等: "水稻叶片反射光谱诊断氮素营养敏感波段的研究", 《浙江大学学报(农业与生命科学版)》 * |
杨燕: "基于高光谱成像技术的水稻稻瘟病诊断关键技术研究", 《中国博士学位论文全文数据库 农业科技辑》 * |
石吉勇: "基于高光谱图像技术的设施栽培作物营养元素亏缺诊断研究", 《中国博士学位论文全文数据库 信息科技辑》 * |
黄敏: "基于光谱及数据挖掘技术的油菜养分及品质信息的无损检测研究", 《中国博士学位论文全文数据库 农业科技辑》 * |
Cited By (24)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN104677853A (en) * | 2015-01-27 | 2015-06-03 | 天津大学 | Fresco scaling damage assessment method based on near-infrared hyperspectrum |
CN104865194A (en) * | 2015-04-03 | 2015-08-26 | 江苏大学 | Detection apparatus and method for pesticide residues in vegetable based on near infrared, fluorescence and polarization multi-spectrum |
CN107315851B (en) * | 2017-05-19 | 2020-06-23 | 中国科学院长春光学精密机械与物理研究所 | Robust film system design method of broad-spectrum extreme ultraviolet multilayer film |
CN107315851A (en) * | 2017-05-19 | 2017-11-03 | 中国科学院长春光学精密机械与物理研究所 | A kind of robustness Film Design method of wide spectrum multiplayer films in EUV |
CN108064531A (en) * | 2017-12-05 | 2018-05-25 | 江苏大学 | Device is applied in a kind of light simplified greenhouse liquid manure filling |
CN108254396A (en) * | 2017-12-05 | 2018-07-06 | 江苏大学 | A kind of tomato seedling water stress detection method based on micro-CT and polarization-high light spectrum image-forming multiple features fusion |
CN108323295A (en) * | 2017-12-05 | 2018-07-27 | 江苏大学 | A kind of seedling stage crop liquid manure based on multiple dimensioned habitat information detects and controls method and device |
US11406057B2 (en) | 2017-12-05 | 2022-08-09 | Jiangsu University | Multi-scale habitat information-based method and device for detecting and controlling water and fertilizer for crops in seedling stage |
WO2019109384A1 (en) * | 2017-12-05 | 2019-06-13 | 江苏大学 | Multi-scale habitat information-based method and device for detecting and controlling water and fertilizer for crops in seedling stage |
CN109521001A (en) * | 2018-11-19 | 2019-03-26 | 华南理工大学 | A kind of flying marking measuring method based on PSO and ε-SVR |
CN109580513A (en) * | 2018-12-18 | 2019-04-05 | 塔里木大学 | A kind of remote sensing jujube moisture content detecting method near the ground and device |
CN109580513B (en) * | 2018-12-18 | 2021-06-11 | 塔里木大学 | Near-ground remote sensing red date moisture content detection method and device |
CN109827910B (en) * | 2019-01-22 | 2021-05-04 | 塔里木大学 | Fast monitoring and processing method for orchard establishment data |
CN109827910A (en) * | 2019-01-22 | 2019-05-31 | 塔里木大学 | A kind of quick monitoring process method of orchard establishing data |
CN109961073A (en) * | 2019-03-29 | 2019-07-02 | 云南电网有限责任公司电力科学研究院 | The acquisition methods and device of a kind of transmission line of electricity and shaft tower information |
CN110455722A (en) * | 2019-08-20 | 2019-11-15 | 中国热带农业科学院橡胶研究所 | Rubber tree blade phosphorus content EO-1 hyperion inversion method and system |
CN110553982A (en) * | 2019-08-29 | 2019-12-10 | 安徽科技学院 | Polarization hyperspectral image acquisition system |
CN110579444A (en) * | 2019-09-28 | 2019-12-17 | 中国科学院合肥物质科学研究院 | Visible-near infrared light polarization spectrum detection probe and detection method thereof |
WO2021111449A1 (en) * | 2019-12-05 | 2021-06-10 | The State Of Israel, Ministry Of Agriculture & Rural Development, Agricultural Research Organization (Aro) (Volcani Center) | Device for measuring element concentrations in plant leaves and method of implementing the same |
EP4070075A4 (en) * | 2019-12-05 | 2023-01-18 | The State of Israel, Ministry of Agriculture & Rural Development, Agricultural Research Organization (ARO) (Volcani Center) | Device for measuring element concentrations in plant leaves and method of implementing the same |
CN111257240A (en) * | 2020-01-17 | 2020-06-09 | 华中农业大学 | Near-earth hyperspectral data and integrated model-based rape nitrogen-phosphorus-potassium nutrient diagnosis method |
CN111220734A (en) * | 2020-02-26 | 2020-06-02 | 江苏大学 | Method for evaluating nutritional state of crops in latent nutrient deficiency period by using sensitive metabolites |
CN111948155A (en) * | 2020-07-30 | 2020-11-17 | 中国科学院西安光学精密机械研究所 | Fine full spectrum combined GS-SVR improved complex water body nitrate quantitative analysis method |
CN111948155B (en) * | 2020-07-30 | 2022-05-10 | 中国科学院西安光学精密机械研究所 | Fine full spectrum combined GS-SVR improved complex water body nitrate quantitative analysis method |
Also Published As
Publication number | Publication date |
---|---|
CN104198396B (en) | 2016-08-31 |
Similar Documents
Publication | Publication Date | Title |
---|---|---|
CN104198396B (en) | The method that polarization-hyperspectral technique diagnosis crop N P and K wanes | |
CN102495005B (en) | Method for diagnosing crop water deficit through hyperspectral image technology | |
Yang et al. | Using reflectance to explain vegetation biochemical and structural effects on sun-induced chlorophyll fluorescence | |
US10712261B2 (en) | Systems and methods for determination and application of nitrogen fertilizer using crop canopy measurements | |
CN102788752B (en) | Non-destructive detection device and method of internal information of crops based on spectrum technology | |
CN103048266B (en) | Automatic recognizing method and device for nitrogen phosphorus and potassium stress of protected tomatoes | |
CN110874617B (en) | Method for establishing winter wheat leaf nitrogen content estimation model | |
CN107796764A (en) | A kind of construction method of the wheat leaf area index appraising model based on three wave band vegetation indexs | |
CN110567892B (en) | Summer corn nitrogen hyperspectral prediction method based on critical nitrogen concentration | |
Cao et al. | Hyperspectral inversion of nitrogen content in maize leaves based on different dimensionality reduction algorithms | |
CN107607486B (en) | Soil total nitrogen detection method and device | |
Zhao et al. | Effect of vertical distribution of crop structure and biochemical parameters of winter wheat on canopy reflectance characteristics and spectral indices | |
CN108520127A (en) | A kind of EO-1 hyperion inversion method of seeds leaf area index | |
Zhang et al. | Vis/NIR reflectance spectroscopy for hybrid rice variety identification and chlorophyll content evaluation for different nitrogen fertilizer levels | |
CN104198397A (en) | Method for detecting nutrient content of greenhouse crops under nitrogen, phosphorus and potassium interaction | |
Wang et al. | Rapid detection of chlorophyll content and distribution in citrus orchards based on low-altitude remote sensing and bio-sensors | |
CN113252583A (en) | Method for calculating alpine hay coverage based on hay vegetation index | |
Li et al. | Monitoring of leaf nitrogen content of winter wheat using multi-angle hyperspectral data | |
CN113670913B (en) | Construction method for inversion hyperspectral vegetation index of nitrogen content of rice | |
CN102788796B (en) | Nutrient diagnosis device and nutrient diagnosis method for nitrogen of crops based on multi-information integration of high spectral images and fluorescent images | |
Zibordi et al. | Automated quality control of AERONET-OC LWN data | |
US20240099179A1 (en) | Machine learning-based hyperspectral detection and visualization method of nitrogen content in soil profile | |
CN105158171A (en) | Spectral calibration method for crop nitrogen sensor | |
Kong et al. | An integrated field and hyperspectral remote sensing method for the estimation of pigments content of Stipa Purpurea in Shenzha, Tibet | |
Gitelson | Recent developments in remote estimation of crop biophysical and biochemical properties at various scales |
Legal Events
Date | Code | Title | Description |
---|---|---|---|
C06 | Publication | ||
PB01 | Publication | ||
C10 | Entry into substantive examination | ||
SE01 | Entry into force of request for substantive examination | ||
C14 | Grant of patent or utility model | ||
GR01 | Patent grant | ||
CF01 | Termination of patent right due to non-payment of annual fee | ||
CF01 | Termination of patent right due to non-payment of annual fee |
Granted publication date: 20160831 Termination date: 20180730 |