CN106650819A - Soil classification method through combination of multi-layer perceptron neural networks with spectral characteristic parameters - Google Patents

Soil classification method through combination of multi-layer perceptron neural networks with spectral characteristic parameters Download PDF

Info

Publication number
CN106650819A
CN106650819A CN201611254381.1A CN201611254381A CN106650819A CN 106650819 A CN106650819 A CN 106650819A CN 201611254381 A CN201611254381 A CN 201611254381A CN 106650819 A CN106650819 A CN 106650819A
Authority
CN
China
Prior art keywords
soil
spectral
soil sample
classification
sample
Prior art date
Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
Pending
Application number
CN201611254381.1A
Other languages
Chinese (zh)
Inventor
刘焕军
张小康
张新乐
王翔
窦欣
秦乐乐
杨皓轩
Current Assignee (The listed assignees may be inaccurate. Google has not performed a legal analysis and makes no representation or warranty as to the accuracy of the list.)
Northeast Agricultural University
Original Assignee
Northeast Agricultural University
Priority date (The priority date 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 date listed.)
Filing date
Publication date
Application filed by Northeast Agricultural University filed Critical Northeast Agricultural University
Priority to CN201611254381.1A priority Critical patent/CN106650819A/en
Publication of CN106650819A publication Critical patent/CN106650819A/en
Pending legal-status Critical Current

Links

Classifications

    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F18/00—Pattern recognition
    • G06F18/20—Analysing
    • G06F18/24—Classification techniques
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
    • G06N3/00—Computing arrangements based on biological models
    • G06N3/02—Neural networks

Landscapes

  • Engineering & Computer Science (AREA)
  • Theoretical Computer Science (AREA)
  • Physics & Mathematics (AREA)
  • Data Mining & Analysis (AREA)
  • Life Sciences & Earth Sciences (AREA)
  • Artificial Intelligence (AREA)
  • General Physics & Mathematics (AREA)
  • General Engineering & Computer Science (AREA)
  • Evolutionary Computation (AREA)
  • Bioinformatics & Computational Biology (AREA)
  • Computational Linguistics (AREA)
  • Computer Vision & Pattern Recognition (AREA)
  • Bioinformatics & Cheminformatics (AREA)
  • Health & Medical Sciences (AREA)
  • Biomedical Technology (AREA)
  • Biophysics (AREA)
  • Evolutionary Biology (AREA)
  • General Health & Medical Sciences (AREA)
  • Molecular Biology (AREA)
  • Computing Systems (AREA)
  • Mathematical Physics (AREA)
  • Software Systems (AREA)
  • Investigating Or Analysing Materials By Optical Means (AREA)

Abstract

The invention discloses a soil classification method through combination of multi-layer perceptron neural networks with spectral characteristic parameters and belongs to the technical field of soil classification. The method aims at solving the problem that the original physical significance of spectral characteristics cannot be reserved through utilization of the existing soil classification method which employs soil spectral reflectivity. The method comprises the steps of collecting soil samples and obtaining reference reflection spectral data of each soil sample; carrying out spectral resampling on the reference reflection spectral data of each soil sample by taking 10nm as an interval through utilization of a Gaussian model; carrying out envelope line removal on the spectral resampling data, thereby obtaining the envelope line removed data which highlights the absorption and reflection characteristics of reflection spectral curves; extracting m spectral characteristic parameters from the envelope line removed data; carrying out standard processing on the extracted m spectral characteristic parameters, thereby obtaining m soil classification indexes; and classifying the soil samples according to the soil classification indexes through utilization of the multi-layer perceptron neural networks. The method is used for soil classification.

Description

The classification of soils of spectral signature parameter is combined using multilayer perceptron neural network model Method
Technical field
The present invention relates to the use of classification of soils method of the multilayer perceptron neural network model with reference to spectral signature parameter, category In classification of soils technical field.
Background technology
Carrying out the classification of soils using soil spectrum reflection characteristic can finely chart offer technical support to accelerate soil.At present Both at home and abroad in the method that the classification of soils is carried out using soil spectrum reflectivity, more with PCA process spectrum number According to, principal component is extracted as the input quantity of disaggregated model, set up disaggregated model using methods such as K-means, SVMs.But It is that the variable that principal component analytical method is obtained does not have clear and definite physical significance, without comparativity between different results of study.
Multilayer perceptron neural network model (Multi-layer perceptron neural networks, MLP Neural networks) as a kind of strong learning system, human brain thinking process is simulated, input layer, defeated can be realized Go out the Nonlinear Mapping of layer, be used widely in many fields, and obtain good result.Multilayer perceptron neutral net It is made up of an input layer, a hidden layer and an output layer, the activation primitive of hidden layer and output layer is being respectively hyperbolic just Cut function and softmax functions.
ASD companies of the U.S. are global foremost field ground feature spectrometer manufacturers, and its product is remote sensing and association area Most authoritative measuring apparatus and working stamndard.ASD spectrometers are widely used in global countries and regions more than 70, are widely used in The fields such as remote sensing science research, mining industry, Forestry Ecological, agricultural and material, are global most widely used spectrometers.
The content of the invention
The invention aims to the method for solving existing utilization soil spectrum reflectivity and carrying out the classification of soils cannot retain Spectrum thing levies the problem of original physical significance, there is provided one kind combines spectral signature using multilayer perceptron neural network model The classification of soils method of parameter.
Utilization multilayer perceptron neural network model of the present invention combines the classification of soils method of spectral signature parameter, it Comprise the following steps：
Step one：N soil sample of collection, n is the integer more than or equal to 3；With spectrometer to n soil-like one's duty Reflectance spectrum test is not carried out, for each soil sample：10 curves of spectrum are carried out arithmetic and are put down by 10 curves of spectrum of collection , the baseline reflectance spectrum data of each soil sample are obtained；
Step 2：By the baseline reflectance spectrum data of each soil sample with 10nm as interval, carried out using Gauss model Spectral resampling method；
Step 3：Spectral resampling method data are carried out with envelope removal, the absorption of prominent reflection spectrum curve and anti-is obtained That penetrates feature goes envelop data；
Step 4：Extract in envelop data is gone and obtain m spectral signature parameter；
Step 5：The m spectral signature parameter to extracting is standardized respectively, obtains m Soil reference；
Step 6：Using multilayer perceptron neural network model, soil sample is classified according to Soil reference.
Advantages of the present invention：The present invention can realize the quick and precisely classification to soil.It removes bag using spectral reflectivity Characteristic parameter after winding thread, such as wave band slope, the area of paddy is absorbed as directly utilizing ripple in classification indicators, with prior art Section is extracted principal component and is compared as classification indicators, and spectral signature parameter is extracted simple and easy to operate, it is most important that remain spectrum The original physical significance of feature, each spectral signature parameter is affected by specific soil physico-chemical property.
The present invention combines the generalization ability of multilayer perceptron neural network model and there is spectral signature parameter physics to anticipate The advantage of justice, using multilayer perceptron neural network model, with reference to the index of spectral signature parameter, judges the soil class of the object Type.It overcomes the defect that the classification indicators of conventional principal component analysis extraction do not have clear and definite physical significance so that Multilayer Perception Device neural network model is preferably applied to the classification of soils, not only increases the accuracy and speed of classification, and specify that classification The physical significance of index.Meanwhile, it is to accelerate soil finely to chart offer technical support.
Description of the drawings
The curve of spectral signature parameter is extracted in the step of Fig. 1 is the specific embodiment of the invention four in envelop data is gone Figure.
Specific embodiment
Specific embodiment one：With reference to Fig. 1 explanation present embodiments, described in present embodiment multilayer perceptron is utilized Neural network model combines the classification of soils method of spectral signature parameter, and it is comprised the following steps：
Step one：N soil sample of collection, n is the integer more than or equal to 3；With spectrometer to n soil-like one's duty Reflectance spectrum test is not carried out, for each soil sample：10 curves of spectrum are carried out arithmetic and are put down by 10 curves of spectrum of collection , the baseline reflectance spectrum data of each soil sample are obtained；
Step 2：By the baseline reflectance spectrum data of each soil sample with 10nm as interval, carried out using Gauss model Spectral resampling method；
Step 3：Spectral resampling method data are carried out with envelope removal, the absorption of prominent reflection spectrum curve and anti-is obtained That penetrates feature goes envelop data；
Step 4：Extract in envelop data is gone and obtain m spectral signature parameter；
Step 5：The m spectral signature parameter to extracting is standardized respectively, obtains m Soil reference；
Step 6：Using multilayer perceptron neural network model, soil sample is classified according to Soil reference.
The multilayer perceptron neural network model has input layer, hidden layer and output layer, the activation primitive of hidden layer For hyperbolic tangent function, the activation primitive of output layer is softmax functions.
Gather in step one after n soil sample, indoors each soil sample is ground respectively, air-dried and mistake 2mm is sieved, and then carries out reflectance spectrum test to each soil sample.
In step 5 to extract spectral signature parameter be standardized the formula for adopting for：
Wherein：ZjI () is the standardization result of i-th sample, j-th index, XjI () is i-th sample, j-th index Value, max [Xj(i)] be i-th sample, j-th index maximum, min [Xj(i)] be i-th sample, j-th index minimum Value, i=1,2,3 ... ... n, j=1,2,3 ... ..., m.
The spectral investigator that present embodiment is used is U.S.'s analysis spectral instrument company (analytical spectral Devices, ASD) production3 portable spectrometers, its spectral range is 350～2500nm；350～1000nm Between spectrum sample at intervals of 1.4nm, spectral resolution 3nm；The sampling interval is 2nm between 1000～2500nm, spectrally resolved Rate 10nm；Data resampling is finally 1nm by spectrometer.
Spectral resampling method data are further processed using envelope null method in step 3, obtain effectively prominent anti- Envelop data is gone in the absorption and reflectance signature for penetrating the curve of spectrum；
Step 5 is that the impact of classification results is carried out in order to eliminate different spectral signature parameter magnitudes.
Specific embodiment：
Step one：In the domestic collection 0～20cm topsoil soil samples in 13 cities and counties such as Beian City, Baiquan County, totally 138, the institute of table 1 Show, it is indoor soil sample to be ground, is air-dried, crossing 2mm sieves, then reflectance spectrum test is carried out to these samples.
Table 1
Great soil group Black earth Chernozem Sand soil It is total
Number of samples 35 74 29 138
Utilize3 portable spectrometers can control to carry out soil sample in the darkroom of illumination condition at one Spectrum test.Soil sample is respectively placed in diameter 12cm, the sample-containing dish of depth 1.8cm, is struck off on soil sample surface with ruler.Light Source is the Halogen lamp LED that power is 1000W, away from pedotheque surface 100cm, 30 ° of zenith angle, there is provided almost parallel to soil sample Light, for reducing the impact that soil roughness causes shade.It is placed in from soil-like using the sensor probe of 8 ° of angles of visual field The vertical direction of this surface 15cm.The impact of dark current in radiation intensity is first removed before test, is then demarcated with blank. Each soil sample gathers 10 curves of spectrum, and the baseline reflectance spectrum data of the soil sample are obtained after arithmetic average；
Step 2：To soil reflective spectrum data resulting in step one with 10nm as interval, entered using Gauss model Row spectral resampling method, this process is carried out in ENVI5.1；
Step 3：Take envelope null method to be further processed the resampling data in step 2, obtain effectively The absorption of prominent reflection spectrum curve and reflectance signature go envelop data, and this process is carried out in ENVI5.1；
Step 4：Envelop data of going in using step 3 extracts required spectral signature parameter, as shown in figure 1, and Determine classification indicators：V1, V2, V3, V4, V5 represent respectively S in 5 absorption paddy, figures and represent L generations in slope between wave band, figure in figure Table each absorb paddy minimum point be in corresponding absorption position, figure A represent absorb paddy area；The area of first absorption paddy Area with the first two absorbs paddy, is designated as respectively A1, A1+A2；Second absorbs paddy position, is designated as L2；500～600nm wave bands The slope of envelope is removed, S3 is designated as.
Step 5：In order to eliminate impact of the different spectral signature parameter magnitudes to classification results, to selected Spectral Properties Levy parameter to be standardized, obtain classification indicators.
Step 6：Determine an input layer, a hidden layer and an output layer of multilayer perceptron neutral net, hide The activation primitive of layer and output layer is respectively hyperbolic tangent function and softmax functions, and this process is carried out in SPSS 22.0；
The classification of soils is carried out using multilayer perceptron neural network model combining classification index.Wherein, 70% use of soil sample In modeling, 30% is used for precision test.
To present embodiment from production precision, user's precision, four angles of overall accuracy and Kappa coefficient analysis to dividing Class result carries out precision evaluation.
Production precision refers to that every class sum is deducted and Lou divides again divided by batch total；User's precision refers to that every class sum deducts mistake Divide again divided by batch total；Overall accuracy refers to correct batch total divided by soil sample sum.Kappa coefficients are using a kind of discrete The confusion matrix of polytechnics, it is contemplated that all factors of matrix, be it is a kind of calculating nicety of grading index, its computing formula For：
In formula：R is total ordered series of numbers, i.e., total classification number in error matrix；XiiIt is the i-th row in error matrix, the upper soil sample of the i-th row Quantity, i.e., the number of correct classification；Xi+And X+iIt is the total soil sample quantity on the i-th row and the i-th row；N is total for accuracy evaluation Soil sample quantity.
The present embodiment combines spectral signature parameter to black earth, chernozem and wind sand with multilayer perceptron neural network model Soil is classified, and precision evaluation result is：Production precision is respectively：100%th, 100%, 85.71；User's precision is respectively： 100%th, 89.47%, 100%；Overall accuracy is：95.35%；Kappa coefficients are 0.93, and disclosure satisfy that soil finely charts needs Ask.

Claims (4)

1. a kind of utilization multilayer perceptron neural network model combines the classification of soils method of spectral signature parameter, and its feature exists In it is comprised the following steps：
Step one：N soil sample of collection, n is the integer more than or equal to 3；N soil sample is entered respectively with spectrometer Row reflectance spectrum is tested, for each soil sample：10 curves of spectrum are carried out arithmetic average by 10 curves of spectrum of collection, Obtain the baseline reflectance spectrum data of each soil sample；
Step 2：By the baseline reflectance spectrum data of each soil sample with 10nm as interval, using Gauss model spectrum is carried out Resampling；
Step 3：Spectral resampling method data are carried out with envelope removal, the absorption and reflection for obtaining prominent reflection spectrum curve is special That what is levied goes envelop data；
Step 4：Extract in envelop data is gone and obtain m spectral signature parameter；
Step 5：The m spectral signature parameter to extracting is standardized respectively, obtains m Soil reference；
Step 6：Using multilayer perceptron neural network model, soil sample is classified according to Soil reference.
2. utilization multilayer perceptron neural network model according to claim 1 combines the classification of soils of spectral signature parameter Method, it is characterised in that the multilayer perceptron neural network model has input layer, hidden layer and output layer, hidden layer Activation primitive is hyperbolic tangent function, and the activation primitive of output layer is softmax functions.
3. utilization multilayer perceptron neural network model according to claim 2 combines the classification of soils of spectral signature parameter Method, it is characterised in that
Gather in step one after n soil sample, be ground, air-dry and cross 2mm to each soil sample respectively indoors and sieve, Then reflectance spectrum test is carried out to each soil sample.
4. utilization multilayer perceptron neural network model according to claim 3 combines the classification of soils of spectral signature parameter Method, it is characterised in that
In step 5 to extract spectral signature parameter be standardized the formula for adopting for：
Z j ( i ) = X j ( i ) - min [ X j ( i ) ] max [ X j ( i ) ] - min [ X j ( i ) ] ,
Wherein：ZjI () is the standardization result of i-th sample, j-th index, Xj(i) be i-th sample, j-th desired value, max [Xj(i)] be i-th sample, j-th index maximum, min [Xj(i)] be i-th sample, j-th index minimum of a value, i= 1,2,3 ... ... n, j=1,2,3 ... ..., m.
CN201611254381.1A 2016-12-30 2016-12-30 Soil classification method through combination of multi-layer perceptron neural networks with spectral characteristic parameters Pending CN106650819A (en)

Priority Applications (1)

Application Number Priority Date Filing Date Title
CN201611254381.1A CN106650819A (en) 2016-12-30 2016-12-30 Soil classification method through combination of multi-layer perceptron neural networks with spectral characteristic parameters

Applications Claiming Priority (1)

Application Number Priority Date Filing Date Title
CN201611254381.1A CN106650819A (en) 2016-12-30 2016-12-30 Soil classification method through combination of multi-layer perceptron neural networks with spectral characteristic parameters

Publications (1)

Publication Number Publication Date
CN106650819A true CN106650819A (en) 2017-05-10

Family

ID=58835543

Family Applications (1)

Application Number Title Priority Date Filing Date
CN201611254381.1A Pending CN106650819A (en) 2016-12-30 2016-12-30 Soil classification method through combination of multi-layer perceptron neural networks with spectral characteristic parameters

Country Status (1)

Country Link
CN (1) CN106650819A (en)

Cited By (4)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN107657264A (en) * 2017-06-09 2018-02-02 南京师范大学 One kind carries out soil profile kind identification method based on KNN classification
CN109740468A (en) * 2018-12-24 2019-05-10 核工业北京地质研究院 A kind of adaptive Gauss low-pass filtering algorithm for organic matter in black soil information extraction
CN113848191A (en) * 2021-10-26 2021-12-28 北京水云星晗科技有限公司 Intelligent sandstone classification method based on spectrum
CN114324216A (en) * 2022-01-06 2022-04-12 中国科学院南京土壤研究所 Soil numerical value classification method based on soil layer combination characteristics

Citations (2)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN101915753A (en) * 2010-07-30 2010-12-15 浙江师范大学 Genetic Neural NetworkQuantitative analysis method for laser induced breakdown spectroscopy based on gGenetic Neural Networkgenetic neural network
CN106124449A (en) * 2016-06-07 2016-11-16 中国科学院合肥物质科学研究院 A kind of soil near-infrared spectrum analysis Forecasting Methodology based on degree of depth learning art

Patent Citations (2)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN101915753A (en) * 2010-07-30 2010-12-15 浙江师范大学 Genetic Neural NetworkQuantitative analysis method for laser induced breakdown spectroscopy based on gGenetic Neural Networkgenetic neural network
CN106124449A (en) * 2016-06-07 2016-11-16 中国科学院合肥物质科学研究院 A kind of soil near-infrared spectrum analysis Forecasting Methodology based on degree of depth learning art

Non-Patent Citations (1)

* Cited by examiner, † Cited by third party
Title
康苒: "松嫩平原典型土壤有机质高光谱预测模型研究", 《万方数据》 *

Cited By (5)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN107657264A (en) * 2017-06-09 2018-02-02 南京师范大学 One kind carries out soil profile kind identification method based on KNN classification
CN109740468A (en) * 2018-12-24 2019-05-10 核工业北京地质研究院 A kind of adaptive Gauss low-pass filtering algorithm for organic matter in black soil information extraction
CN113848191A (en) * 2021-10-26 2021-12-28 北京水云星晗科技有限公司 Intelligent sandstone classification method based on spectrum
CN114324216A (en) * 2022-01-06 2022-04-12 中国科学院南京土壤研究所 Soil numerical value classification method based on soil layer combination characteristics
CN114324216B (en) * 2022-01-06 2023-08-01 中国科学院南京土壤研究所 Soil numerical classification method based on soil layer combination characteristics

Similar Documents

Publication Publication Date Title
Hong et al. Estimating lead and zinc concentrations in peri-urban agricultural soils through reflectance spectroscopy: Effects of fractional-order derivative and random forest
Shi et al. Development of a national VNIR soil-spectral library for soil classification and prediction of organic matter concentrations
CN106650819A (en) Soil classification method through combination of multi-layer perceptron neural networks with spectral characteristic parameters
Leone et al. Multivariate analysis of laboratory spectra for the assessment of soil development and soil degradation in the southern Apennines (Italy)
Zhang et al. Application of hyperspectral remote sensing for environment monitoring in mining areas
CN102426153B (en) A kind of Wheat plant moisture monitoring method based on canopy high spectral index
Adam et al. Multispectral and hyperspectral remote sensing for identification and mapping of wetland vegetation: a review
CN107478580B (en) Soil heavy metal content estimation method and device based on hyperspectral remote sensing
Delalieux et al. Hyperspectral indices to diagnose leaf biotic stress of apple plants, considering leaf phenology
CN101576485A (en) Analytical method of multi-source spectrum fusion water quality
Ahmad Spectral vegetation indices performance evaluated for Cholistan Desert
Aitkenhead et al. Digital RGB photography and visible-range spectroscopy for soil composition analysis
Richter et al. Free iron oxide determination in Mediterranean soils using diffuse reflectance spectroscopy
CN104596957A (en) Estimation method for content of copper in soil on basis of visible-light near-infrared spectrum technology
CN112634212A (en) Hyperspectral unmanned aerial vehicle-based disease latent tree detection method and system
CN104502288A (en) Soil lead content measurement method using visible and near-infrared spectroscopy technology
CN103076292B (en) System and method for quantitatively evaluating concentration of rare earth in water body based on ground reflection spectrum
CN102507474A (en) Method and system for identifying oil spilling target of ship
CN104596943A (en) Indoor spectral layered measuring method for poisonous and harmful elements at mine reclamation area
Gürsoy et al. Detecting of lithological units by using terrestrial spectral data and remote sensing image
CN106990056A (en) A kind of total soil nitrogen spectrum appraising model calibration samples collection construction method
Demattê et al. Weathering sequence of soils developed from basalt as evaluated by laboratory (IRIS), airborne (AVIRIS) and orbital (TM) sensors
CN111523587A (en) Woody plant species spectrum identification method based on machine learning
Brossard et al. Deconvolution of SWIR reflectance spectra for automatic mineral identification in hyperspectral imaging
Mitran et al. Retrieval of soil physicochemical properties towards assessing salt-affected soils using Hyperspectral Data

Legal Events

Date Code Title Description
PB01 Publication
PB01 Publication
SE01 Entry into force of request for substantive examination
SE01 Entry into force of request for substantive examination
RJ01 Rejection of invention patent application after publication

Application publication date: 20170510

RJ01 Rejection of invention patent application after publication