CN110108644A - A kind of maize variety identification method based on depth cascade forest and high spectrum image - Google Patents
A kind of maize variety identification method based on depth cascade forest and high spectrum image Download PDFInfo
- Publication number
- CN110108644A CN110108644A CN201910400622.6A CN201910400622A CN110108644A CN 110108644 A CN110108644 A CN 110108644A CN 201910400622 A CN201910400622 A CN 201910400622A CN 110108644 A CN110108644 A CN 110108644A
- Authority
- CN
- China
- Prior art keywords
- corn
- forest
- image data
- image
- maize variety
- 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
Links
- 240000008042 Zea mays Species 0.000 title claims abstract description 105
- 235000002017 Zea mays subsp mays Nutrition 0.000 title claims abstract description 105
- 238000000034 method Methods 0.000 title claims abstract description 64
- 235000016383 Zea mays subsp huehuetenangensis Nutrition 0.000 title claims abstract description 31
- 235000009973 maize Nutrition 0.000 title claims abstract description 31
- 238000001228 spectrum Methods 0.000 title claims abstract description 30
- 235000005824 Zea mays ssp. parviglumis Nutrition 0.000 claims abstract description 74
- 235000005822 corn Nutrition 0.000 claims abstract description 74
- 238000004422 calculation algorithm Methods 0.000 claims abstract description 21
- 238000012549 training Methods 0.000 claims description 32
- 238000007637 random forest analysis Methods 0.000 claims description 30
- 239000013598 vector Substances 0.000 claims description 23
- 238000007689 inspection Methods 0.000 claims description 10
- 238000000605 extraction Methods 0.000 claims description 8
- 238000010276 construction Methods 0.000 claims description 6
- 238000012937 correction Methods 0.000 claims description 5
- 238000002790 cross-validation Methods 0.000 claims description 5
- 238000004364 calculation method Methods 0.000 claims description 4
- 238000003066 decision tree Methods 0.000 claims description 4
- 238000006243 chemical reaction Methods 0.000 claims description 3
- 238000012216 screening Methods 0.000 claims description 3
- 230000000717 retained effect Effects 0.000 claims description 2
- 238000010801 machine learning Methods 0.000 abstract description 3
- 235000019580 granularity Nutrition 0.000 description 7
- 238000001514 detection method Methods 0.000 description 6
- 238000005516 engineering process Methods 0.000 description 6
- 230000003595 spectral effect Effects 0.000 description 6
- 238000012706 support-vector machine Methods 0.000 description 5
- 238000012360 testing method Methods 0.000 description 5
- 235000013339 cereals Nutrition 0.000 description 4
- 239000000284 extract Substances 0.000 description 4
- 239000000203 mixture Substances 0.000 description 4
- 230000000750 progressive effect Effects 0.000 description 4
- 230000006740 morphological transformation Effects 0.000 description 3
- 240000007594 Oryza sativa Species 0.000 description 2
- 235000007164 Oryza sativa Nutrition 0.000 description 2
- 230000008901 benefit Effects 0.000 description 2
- 238000000701 chemical imaging Methods 0.000 description 2
- 238000007705 chemical test Methods 0.000 description 2
- 230000000694 effects Effects 0.000 description 2
- 238000011156 evaluation Methods 0.000 description 2
- 238000003709 image segmentation Methods 0.000 description 2
- 239000010977 jade Substances 0.000 description 2
- 238000004519 manufacturing process Methods 0.000 description 2
- 238000011160 research Methods 0.000 description 2
- 235000009566 rice Nutrition 0.000 description 2
- 208000003643 Callosities Diseases 0.000 description 1
- 241000450652 Crassula arborescens Species 0.000 description 1
- 241001269238 Data Species 0.000 description 1
- 206010020649 Hyperkeratosis Diseases 0.000 description 1
- BQCADISMDOOEFD-UHFFFAOYSA-N Silver Chemical compound [Ag] BQCADISMDOOEFD-UHFFFAOYSA-N 0.000 description 1
- 238000004458 analytical method Methods 0.000 description 1
- 238000009395 breeding Methods 0.000 description 1
- 230000001488 breeding effect Effects 0.000 description 1
- 230000007423 decrease Effects 0.000 description 1
- 230000007547 defect Effects 0.000 description 1
- 238000012850 discrimination method Methods 0.000 description 1
- 235000013305 food Nutrition 0.000 description 1
- 238000009415 formwork Methods 0.000 description 1
- 238000009499 grossing Methods 0.000 description 1
- 238000002329 infrared spectrum Methods 0.000 description 1
- 238000005259 measurement Methods 0.000 description 1
- 238000012986 modification Methods 0.000 description 1
- 230000004048 modification Effects 0.000 description 1
- 230000000877 morphologic effect Effects 0.000 description 1
- 238000012545 processing Methods 0.000 description 1
- 238000005070 sampling Methods 0.000 description 1
- 230000011218 segmentation Effects 0.000 description 1
- 229910052709 silver Inorganic materials 0.000 description 1
- 239000004332 silver Substances 0.000 description 1
- 239000000126 substance Substances 0.000 description 1
- 235000019587 texture Nutrition 0.000 description 1
- 238000011426 transformation method Methods 0.000 description 1
Classifications
-
- G—PHYSICS
- G01—MEASURING; TESTING
- G01N—INVESTIGATING OR ANALYSING MATERIALS BY DETERMINING THEIR CHEMICAL OR PHYSICAL PROPERTIES
- G01N21/00—Investigating or analysing materials by the use of optical means, i.e. using sub-millimetre waves, infrared, visible or ultraviolet light
- G01N21/17—Systems in which incident light is modified in accordance with the properties of the material investigated
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/24—Classification techniques
- G06F18/243—Classification techniques relating to the number of classes
- G06F18/24323—Tree-organised classifiers
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T5/00—Image enhancement or restoration
- G06T5/70—Denoising; Smoothing
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T5/00—Image enhancement or restoration
- G06T5/80—Geometric correction
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T7/00—Image analysis
- G06T7/10—Segmentation; Edge detection
- G06T7/11—Region-based segmentation
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T7/00—Image analysis
- G06T7/10—Segmentation; Edge detection
- G06T7/155—Segmentation; Edge detection involving morphological operators
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- G06V10/00—Arrangements for image or video recognition or understanding
- G06V10/20—Image preprocessing
- G06V10/25—Determination of region of interest [ROI] or a volume of interest [VOI]
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- G06V20/00—Scenes; Scene-specific elements
- G06V20/10—Terrestrial scenes
- G06V20/188—Vegetation
-
- G—PHYSICS
- G01—MEASURING; TESTING
- G01N—INVESTIGATING OR ANALYSING MATERIALS BY DETERMINING THEIR CHEMICAL OR PHYSICAL PROPERTIES
- G01N21/00—Investigating or analysing materials by the use of optical means, i.e. using sub-millimetre waves, infrared, visible or ultraviolet light
- G01N21/17—Systems in which incident light is modified in accordance with the properties of the material investigated
- G01N2021/1765—Method using an image detector and processing of image signal
- G01N2021/177—Detector of the video camera type
-
- G—PHYSICS
- G01—MEASURING; TESTING
- G01N—INVESTIGATING OR ANALYSING MATERIALS BY DETERMINING THEIR CHEMICAL OR PHYSICAL PROPERTIES
- G01N21/00—Investigating or analysing materials by the use of optical means, i.e. using sub-millimetre waves, infrared, visible or ultraviolet light
- G01N21/17—Systems in which incident light is modified in accordance with the properties of the material investigated
- G01N2021/178—Methods for obtaining spatial resolution of the property being measured
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T2207/00—Indexing scheme for image analysis or image enhancement
- G06T2207/10—Image acquisition modality
- G06T2207/10032—Satellite or aerial image; Remote sensing
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- G06V20/00—Scenes; Scene-specific elements
- G06V20/60—Type of objects
- G06V20/68—Food, e.g. fruit or vegetables
Landscapes
- Engineering & Computer Science (AREA)
- Physics & Mathematics (AREA)
- General Physics & Mathematics (AREA)
- Theoretical Computer Science (AREA)
- Computer Vision & Pattern Recognition (AREA)
- General Health & Medical Sciences (AREA)
- Life Sciences & Earth Sciences (AREA)
- Data Mining & Analysis (AREA)
- Health & Medical Sciences (AREA)
- Multimedia (AREA)
- Evolutionary Biology (AREA)
- Evolutionary Computation (AREA)
- General Engineering & Computer Science (AREA)
- Bioinformatics & Computational Biology (AREA)
- Bioinformatics & Cheminformatics (AREA)
- Artificial Intelligence (AREA)
- Chemical & Material Sciences (AREA)
- Analytical Chemistry (AREA)
- Biochemistry (AREA)
- Immunology (AREA)
- Pathology (AREA)
- Image Analysis (AREA)
Abstract
The present invention provides a kind of maize variety identification methods based on depth cascade forest and high spectrum image, comprising: pre-processes to the corn image data, obtains all band image data;Choose the corn image data of effective wave band;And using the corn image data of effective wave band as feature set, category of model is carried out using depth cascade forest model.Maize variety identification method of the invention makes full use of hyperspectral technique and depth to cascade forest model and carries out corn classification, and compared to common machine learning algorithm, it is higher to identify precision.
Description
Technical field
The present invention relates to the discrimination methods of corn variety, more particularly to a kind of depth that is based on to cascade forest and high spectrum image
Maize variety identification method.
Background technique
Corn is important one of the cereal crops in China, accounts for 30% of total output of grain or more, production of hybrid seeds amount is every year more than 10
Hundred million tons.With the corn diversification of demand, China's corn variety also increases year by year.In breeding process, there is illegal retailer to fill with secondary
It is good, pretend to be high-grade maize kind using corn variety inferior, corn yield is caused to decline, to national food production and agricultural safety
Bring huge hidden danger, therefore how precise and high efficiency nondestructively identifies the kind of corn seed is of great significance.
The method for identifying corn seed at present mainly has artificial detection, Physico-chemical tests, Computer Vision Detection and EO-1 hyperion
Technology detection.The features such as size, form and color of the artificial detection mainly by naked eyes based on corn judge, this mode by
The subjective factor influence of people is very big, and not only detection accuracy is low, speed is slow, but also large labor intensity, it is difficult to form unified standard.
Physico-chemical tests are mainly based upon the biochemical characteristic of different cultivars corn to realize and identify, and such methods accuracy rate is high, still
It is merely able to be detected by sampling, damages and process is complicated, it is cumbersome, it is difficult to meet the market demand.Computer vision inspection
The morphological feature information of the near infrared light spectrum information and visible light that are based primarily upon seed is surveyed, application image mode identification technology is built
Vertical kind discrimination model, has the advantages that instant, efficient, lossless and accurate;But the characteristic information that such method is extracted is less, and
And be the evaluation based on surface mostly, the feature for characterizing its internal component can not be obtained, testing result reliability is by shadow
It rings;In addition, the crossover phenomenon of seed characteristics is serious as seed variety data increase, the separability of model is caused to be deteriorated.
Hyperspectral technique is to integrate the technology of spectrum and image, compared to machine vision technique and near infrared spectrum point
For analysis technology, hyperspectral technique can not only obtain the image information of object, and can obtain spectral information, can effective earth's surface
Levy the inner structural features and Chemical Composition Characteristics of observation object;In addition, the technological merit that hyperspectral technique is quick with its, lossless
Solve the problems, such as that detection of agricultural products field is cumbersome, the application in terms of nondestructive measuring method of the farm product is increasingly extensive.
Summary of the invention
For the defect of above method, the invention proposes a kind of corns based on depth cascade forest and high spectrum image
Variety discriminating method provides theoretical foundation in corn variety automatic identification field for high spectrum image.
A kind of maize variety identification method based on depth cascade forest and high spectrum image of the invention, comprising: step
1: corn image data being pre-processed, all band image data is obtained;Step 2: choosing the corn image number of effective wave band
According to;And step 3: using the corn image data of effective wave band as feature set, model is carried out using depth cascade forest model
Classification.
In above-mentioned maize variety identification method, step 1 includes: to acquire corn image by high spectrum image acquisition system
Data;Corn image data is corrected;Extract area-of-interest (ROI) image data of corn;And the jade to extraction
The area-of-interest image data of rice carries out spectrum correction, obtains all band image data.
In above-mentioned maize variety identification method, corn image data is acquired by high spectrum image acquisition system, and
Only it is retained in the corn image data of 146 wave bands equidistantly selected in wavelength band 533-893.4nm;
Corn image data is corrected using updating formula, updating formula are as follows:
Area-of-interest (ROI) image of each corn is extracted using the label fractional spins based on range conversion
Data;
Using smooth (S-G the is smooth) algorithm of Savitzky-Golay to corn area-of-interest (ROI) image number of extraction
According to progress spectrum correction.
In above-mentioned maize variety identification method, all band image is screened using Boruta algorithm, is selected wherein
Effective wave band image.
In above-mentioned maize variety identification method, the process of Boruta algorithm is as follows:
It is mixed into training set for each feature construction shadow feature, and removes being associated with for these features and classification;
The training random forest grader on the training set for being mixed into shadow feature;
Average loss and standard deviation calculation Z-score using feature;
Delete feature of those Z-score than shadow feature difference;And
When all features are identified or algorithm reaches the number of iterations of setting, algorithm stops.
In above-mentioned maize variety identification method, it includes two stages that depth, which cascades forest model: more granularity scannings and grade
Join the forest stage.
In above-mentioned maize variety identification method, each layer of structure of forest is cascaded by being made of 500 decision trees
Four random forest compositions, wherein four random forests include two completely random forests and two common random forests, wherein
The first two represents completely random forest, and remaining two represent common random forest, then divides screening split vertexes by gini.
In the cascade forest stage, every layer of four random forests all use k and roll over cross validation method, one layer in cascade forest
After completing training, one inspection set is predicted with this model, if the accuracy rate of current layer is than the accuracy rate of preceding layer
Height then continues next layer of construction cascade forest, until current layer to the accuracy rate of inspection set compared with preceding layer accuracy rate no longer
It is promoted, training will terminate, therefore model determines.
In above-mentioned maize variety identification method, more granularity scannings are using various sizes of sliding window on corn image
It slides, obtains primitive character information.
In above-mentioned maize variety identification method, slided on corn image using the window of 7,9 and 11 3 kind of scale,
Classification is three classes.
In above-mentioned maize variety identification method, window 7 will generate 19 7 × 7 training sub-images, these data are defeated
Enter a completely random forest and a common random forest training, each sub-image generate the classification possibility of one 3 dimension to
Amount generates the feature vector of one 114 dimension after cascade, window 9 will obtain the feature vector of 108 dimensions, and window 11 will obtain
To the feature vector of 102 dimensions, the cascade forest of the feature vector training first order of 114 dimensions, the feature vector training second of 108 dimensions
The cascade forest of grade, the cascade forest of the feature vector training third level of 102 dimensions, repeats this process, until essence on inspection set
Degree convergence, a possibility that taking the last layer of cascade forest vector classification average value, be maximized later as final pre-
Survey result.
The present invention provides a kind of maize variety identification methods based on depth cascade forest and high spectrum image, sufficiently benefit
Corn classification is carried out with hyperspectral technique and depth cascade forest model, compared to common machine learning algorithm, identifies precision
It is higher.
Detailed description of the invention
It, below will be to specific in order to illustrate more clearly of the specific embodiment of the invention or technical solution in the prior art
Embodiment or attached drawing needed to be used in the description of the prior art are briefly described.
Fig. 1 is a kind of maize variety identification method based on depth cascade forest and high spectrum image according to the present invention
Techniqueflow chart.
Fig. 2 a is the cascade forest in depth cascade forest model corn disaggregated model;Fig. 2 b is depth cascade forest model
More granularities scanning cascade forest in corn disaggregated model.
Fig. 3 is corn variety sample.
Fig. 4 is corn area-of-interest (ROI) extraction process.
Fig. 5 is corn classification results.
Fig. 6 is depth cascade forest model classification results.
Fig. 7 is classification results of the depth cascade forest under different parameters.
Specific embodiment
The present invention provides a kind of maize variety identification methods based on depth cascade forest and high spectrum image, comprising:
Corn image data is pre-processed, all band image data is obtained;Choose the corn image data of effective wave band;And it will
The corn image data of effective wave band carries out category of model as feature set, using depth cascade forest model.
A kind of technology of maize variety identification method cascading forest and high spectrum image based on depth provided by the invention
Process is as shown in Figure 1, mainly include three parts: data prediction;Effective waveband selection;Category of model.
1.1 data prediction
1.1.1 corn image collection
Using high spectrum image acquisition system acquire corn image data, wherein high spectrum image acquisition system by light source,
Mobile platform and acquisition unit composition.Image acquisition units include image light spectrometer, charge-coupled device (CCD) detector
(1392 × 1040 area array CCD cameras) and camera lens (F/2.4, MCT, C-mount).The slit of high spectrum image spectroscopic system is wide
Degree is 30 μm, and spectral region is 400~1000nm, spectral resolution 2.8nm.Since there are signals to make an uproar in data acquisition
Sound only retains the corn image number of 146 wave bands of wavelength band 533.1-893.4nm to guarantee corn image quality
According to, wherein 146 wave bands are equidistantly selected in wavelength band 533.1-893.4nm.
1.1.2 corn adjustment of image
Due to light source light uneven, influence of dark current in addition according to intensity distribution, image can have many noises, therefore need
It is corrected to weaken or eliminate the interference of noise, updating formula are as follows:
Wherein, ISFor original Hyperspectral imaging, IDFor blackboard calibration image, IWFor blank calibration image, R is after correcting
Hyperspectral imaging.
1.1.3 corn area-of-interest (ROI) Extraction of Image
Before the spectral information and texture information for extracting each corn, need to automatically extract corn from original image
ROI image.The ROI image of each corn, the corn of extraction are extracted using the label fractional spins based on range conversion
ROI image is placed in 44 × 44 × 146 blank array, background value 0.
1.1.4 Pretreated spectra
Since spectral information has many noises, it is therefore desirable to carry out pretreatment and eliminate noise jamming raising model prediction ability
And stability, spectrum school is carried out to the corn ROI image of extraction using smooth (S-G the is smooth) algorithm of Savitzky-Golay herein
Just.Algorithmic procedure is as follows: by odd number (this paper setup parameter is 5), spectrum o'clock is as a window, using multinomial algorithm pair
Data in window do least square fitting, the first derivative and smoothed data value of calculation window central point, and moving window repeats
The above process does smoothing processing to whole picture image, obtains all band image data.
1.2 effective waveband selections
In order to reduce the higher-dimension of high-spectral data, need to choose feature set of effective wave band as category of model.Herein
146 wave bands are screened using Boruta algorithm, select effective wave band image data therein.Boruta algorithm is a kind of
Effective feature selecting algorithm mainly using the Z-score of feature as the measurement of feature importance, and then filters out important
Feature, algorithmic procedure are as follows:
(1) training set is mixed into for each feature construction shadow feature, and remove being associated with for these features and classification;
(2) the training random forest grader on the training set for being mixed into shadow feature;
(3) average loss of feature and standard deviation calculation Z-score are utilized;
(4) feature of those Z-score than shadow feature difference is deleted;
(5) when all features are identified or algorithm reaches the number of iterations of setting, algorithm stops.
1.3 category of model
The depth cascade forest model used herein includes two stages: more granularity scannings and cascade forest stage, such as being schemed
Shown in 2a and Fig. 2 b.
1.3.1 forest is cascaded
Cascade forest each layer (such as in attached drawing 2a and 2b, respectively with level 1, level 2 ... level m or
level 1A、level 1B、level 1C……level MA、level MB、level MCIndicate) be all made of decision tree
Forest composition, each layer are made of two different forests (being expressed as forest A and forest B), and the number of every kind of forest is 2,
Such result makes cascade forest more have diversity.In each layer of structure of cascade forest, the first two of Fig. 2 a is gloomy
Lin represents completely random forest, and remaining two forests represent common random forest, and completely random forest and common random forest are all
It is made of 500 decision trees, screening split vertexes is then divided by gini.Four of every layer are random gloomy in the cascade forest stage
Woods all uses k folding cross validation method, wherein k can be customized, such as k can be 3 or 5.One layer in cascade forest
After completing training, one inspection set is predicted with this model, if the accuracy rate of current layer is than the accuracy rate of preceding layer
Height then continues next layer of construction cascade forest, until current layer to the accuracy rate of inspection set compared with preceding layer accuracy rate no longer
It is promoted, training will terminate, therefore model determines.
1.3.2 more granularity scannings
Window slide has commonly used in the time series datas such as text, voice and image data, is inspired by this, more granularities are swept
It retouches and is slided on image data using various sizes of sliding window, obtain primitive character information, as shown in Figure 2 b.Assuming that adopting
It is slided on corn image with the window of 7,9 and 11 3 kind of scale, classification is 3 classes.Window 7 will generate 19 7 × 7 instructions
Practice sub-image, these data input a completely random forest and a common random forest training, and each sub-image generates one
The classification possibility vector of a 3 dimension, generates the feature vector of one 114 dimension after cascade, window 9 will obtain the spy of 108 dimensions
Vector is levied, window 11 will obtain the feature vector of 102 dimensions.The cascade forest of the 114 dimensional feature vectors training first order, 108 dimensions
The cascade forest of the feature vector training second level, the cascade forest of the 102 dimensional feature vectors training third level.This process is repeated,
Until precision convergence on inspection set.The classification average value of a possibility that taking cascade forest the last layer vector, is maximized later
As final prediction result.
Table 1: variable declaration
The realization of method
This research chooses the corn of agriculture China 213, silver jade 274 and beautiful 439 3 kinds of silver as research object, and dominant hue is equal
For yellow, each kind chooses 200, sample, and total sample number is 600, as shown in Figure 3.Every class corn according to 7:3 ratio
Random division is training set (420) and test set (180), and training set is used for precision evaluation for establishing model, test set.
Classification and Identification model is established using depth cascade forest algorithm under all band (FS) and effective wave band (ES), and is compared traditional
Progressive tree (GBDT) model of gradient, random forest (RF) model and support vector machines (SVM) model, utilize 5 folding cross validation sides
Method determines model parameter, is predicted on test set and calculates classification accuracy (indicating with %).
(1) corn ROI extracts (as shown in Figure 4)
Step 1: the smoothed out image of SG (that is, image after correction) setting threshold value is extracted into two-value image;
Step 2: the preposition region of corn seed is extracted using morphological transformation method, obtains morphological transformation image;
Step 3: Image Segmentation seed is extracted according to the result after morphological transformation;
Step 4: seed boundary is divided according to Image Segmentation seed, to obtain segmentation result;
(2) corn classification results
Step 1: respectively extract corn ROI all band and effective wave band spectral signature, using gradient it is progressive tree, with
Machine forest and support vector machines establish model to classify;
Step 2: corn ROI is inputted respectively in the spectrum image of all band and effective wave band, is built using depth cascade forest
Formwork erection type is classified;
Step 3: establishing model using 5 folding cross validation methods on training set, determine parameter, predict on test set,
View result;
We compare more granularity scanning (MGSCF) models, progressive tree (GBDT) model of gradient, random forest (RF) model
With classification results of support vector machines (SVM) model under all band (FS) and effective wave band (ES).As shown in figure 5, part 1
It is the classifying quality of the lower 4 kinds of models of all band, part 2 is the classifying quality of the lower 4 kinds of models of effective wave band.
(3) depth cascade forest classified parameter list compares
It can be further improved the performance of depth cascade forest model, such as more sliding window rulers by adjusting parameter
It is very little, simply final forest model (RF) is replaced with the progressive tree-model of gradient (GBDT).The depth cascade forest model point of acquisition
Class result is as shown in Figure 6.The present invention compare depth cascade forest model under different grain size and forest model effect (including
Effect under all band (FS) and effective wave band (ES)), as shown in table 2 and Fig. 7.
Window size under 2 different grain size of table
The superiority of this method:
Maize variety identification method provided by the invention based on depth cascade forest and high spectrum image, comprising: to jade
Rice image data is pre-processed, and all band image data is obtained;Choose the corn image data of effective wave band;And it will be effective
The corn image data of wave band carries out category of model as feature set, using depth cascade forest model.This method makes full use of
Hyperspectral technique and depth cascade forest model carry out corn classification, compared to common machine learning algorithm, identify precision more
It is high.
Above embodiments, only a specific embodiment of the invention, to illustrate technical solution of the present invention, rather than to it
Limitation, scope of protection of the present invention is not limited thereto, although the present invention is described in detail referring to the foregoing embodiments,
Those skilled in the art should understand that: anyone skilled in the art the invention discloses technology model
In enclosing, still it can modify to technical solution documented by previous embodiment or variation can be readily occurred in, or to it
Middle some technical characteristics are equivalently replaced;And these modifications, variation or replacement, do not make the essence of corresponding technical solution de-
Spirit and scope from technical solution of the embodiment of the present invention, should be covered by the protection scope of the present invention.Therefore, of the invention
Protection scope should be subject to the protection scope in claims.
Claims (10)
1. a kind of maize variety identification method based on depth cascade forest and high spectrum image, which is characterized in that including following
Step:
Step 1: corn image data being pre-processed, all band image data is obtained;
Step 2: choosing the corn image data of effective wave band;And
Step 3: using the corn image data of effective wave band as feature set, model being carried out using depth cascade forest model
Classification.
2. maize variety identification method according to claim 1, which is characterized in that wherein, the step 1 includes: to pass through
High spectrum image acquisition system acquires corn image data;The corn image data is corrected;The sense for extracting corn is emerging
Interesting region (ROI) image data;And spectrum correction is carried out to the area-of-interest image data of the corn of extraction, it obtains
The all band image data.
3. maize variety identification method according to claim 2, which is characterized in that wherein,
Corn image data is acquired by high spectrum image acquisition system, and is only retained in wavelength band 533-893.4nm
The corn image data of 146 wave bands equidistantly selected;
The corn image data is corrected using updating formula, the updating formula are as follows:
Area-of-interest (ROI) image number of each corn is extracted using the label fractional spins based on range conversion
According to;
Using smooth (S-G the is smooth) algorithm of Savitzky-Golay to corn area-of-interest (ROI) the image number of extraction
According to progress spectrum correction.
4. maize variety identification method according to claim 1, which is characterized in that using Boruta algorithm to all band shadow
As data are screened, effective wave band image data therein is selected.
5. maize variety identification method according to claim 4, which is characterized in that the process of the Boruta algorithm is such as
Under:
It is mixed into training set for each feature construction shadow feature, and removes being associated with for these features and classification;
The training random forest grader on the training set for being mixed into shadow feature;
Average loss and standard deviation calculation Z-score using feature;
Delete feature of the Z-score than shadow feature difference;And
When all features are identified or algorithm reaches the number of iterations of setting, algorithm stops.
6. maize variety identification method according to claim 1, which is characterized in that the depth cascades forest model and includes
Two stages: more granularity scannings and cascade forest stage.
7. maize variety identification method according to claim 6, which is characterized in that each layer of structure of the cascade forest
It is made of four random forests being made of 500 decision trees, wherein four random forests include two completely randoms
Forest and two common random forests, wherein the first two represents the completely random forest, remaining two represent it is described commonly with
Then machine forest divides screening split vertexes by gini;In the cascade forest stage, every layer of four random forests are all
Cross validation method is rolled over using k, after cascading one layer of completion training in forest, one inspection set is carried out with this model
Prediction continues next layer of construction cascade forest if the accuracy rate of current layer is higher than the accuracy rate of preceding layer, until current
Layer no longer promotes the accuracy rate of inspection set compared with the accuracy rate of preceding layer, and training will terminate, therefore model determines.
8. maize variety identification method according to claim 6, which is characterized in that more granularity scannings use different rulers
Very little sliding window slides on corn image, obtains primitive character information.
9. maize variety identification method according to claim 8, which is characterized in that using 7,9 and the window of 11 3 kind of scale
Mouth slides on corn image, and classification is three classes.
10. maize variety identification method according to claim 9, which is characterized in that window 7 will generate 19 7 × 7
Training sub-image, these data input a completely random forest and a common random forest training, and each sub-image generates
The classification possibility vector of one 3 dimension, generates the feature vector of one 114 dimension after cascade, window 9 will obtain 108 dimensions
Feature vector, window 11 will obtain the feature vector of 102 dimensions;The cascade of the feature vector training first order of 114 dimension is gloomy
Woods, the cascade forest of the feature vector training second level of 108 dimension, the grade of the feature vector training third level of 102 dimension
Join forest, repeat this process, until inspection set on precision convergence, take cascade forest the last layer a possibility that vector class
Other average value, is maximized later as final prediction result.
Priority Applications (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201910400622.6A CN110108644A (en) | 2019-05-15 | 2019-05-15 | A kind of maize variety identification method based on depth cascade forest and high spectrum image |
Applications Claiming Priority (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201910400622.6A CN110108644A (en) | 2019-05-15 | 2019-05-15 | A kind of maize variety identification method based on depth cascade forest and high spectrum image |
Publications (1)
Publication Number | Publication Date |
---|---|
CN110108644A true CN110108644A (en) | 2019-08-09 |
Family
ID=67490058
Family Applications (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
CN201910400622.6A Pending CN110108644A (en) | 2019-05-15 | 2019-05-15 | A kind of maize variety identification method based on depth cascade forest and high spectrum image |
Country Status (1)
Country | Link |
---|---|
CN (1) | CN110108644A (en) |
Cited By (5)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN110490114A (en) * | 2019-08-13 | 2019-11-22 | 西北工业大学 | Target detection barrier-avoiding method in a kind of unmanned plane real-time empty based on depth random forest and laser radar |
CN111929270A (en) * | 2020-07-07 | 2020-11-13 | 长江大学 | Wheat mutant identification method |
CN112161937A (en) * | 2020-11-04 | 2021-01-01 | 安徽大学 | Wheat flour gluten degree detection method based on cascade forest and convolutional neural network |
CN113049530A (en) * | 2021-03-17 | 2021-06-29 | 北京工商大学 | Single-seed corn seed moisture content detection method based on near-infrared hyperspectrum |
CN113433076A (en) * | 2021-05-18 | 2021-09-24 | 中国检验检疫科学研究院 | Hyperspectral imaging technology-based method for identifying aflatoxin in corn seeds |
Citations (1)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN106546541A (en) * | 2016-10-31 | 2017-03-29 | 浙江大学 | A kind of identifying device and method based on EO-1 hyperion transgenic corns seed |
-
2019
- 2019-05-15 CN CN201910400622.6A patent/CN110108644A/en active Pending
Patent Citations (1)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN106546541A (en) * | 2016-10-31 | 2017-03-29 | 浙江大学 | A kind of identifying device and method based on EO-1 hyperion transgenic corns seed |
Non-Patent Citations (4)
Title |
---|
M. LI ET AL: "Hyperspectral Image Classification Based on Deep Forest and Spectral-Spatial Cooperative Feature", 《IMAGE AND GRAPHICS》 * |
张凯: "高维小样本数据的互信息特征选择方法研究", 《中国优秀硕士学位论文全文数据库信息科技辑》 * |
杨小玲: "高光谱图像技术检测玉米种子品质研究", 《中国博士学位论文全文数据库农业科技辑》 * |
陈吕鹏 等: "基于深度森林算法的电力系统短期负荷预测", 《电力建设》 * |
Cited By (5)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN110490114A (en) * | 2019-08-13 | 2019-11-22 | 西北工业大学 | Target detection barrier-avoiding method in a kind of unmanned plane real-time empty based on depth random forest and laser radar |
CN111929270A (en) * | 2020-07-07 | 2020-11-13 | 长江大学 | Wheat mutant identification method |
CN112161937A (en) * | 2020-11-04 | 2021-01-01 | 安徽大学 | Wheat flour gluten degree detection method based on cascade forest and convolutional neural network |
CN113049530A (en) * | 2021-03-17 | 2021-06-29 | 北京工商大学 | Single-seed corn seed moisture content detection method based on near-infrared hyperspectrum |
CN113433076A (en) * | 2021-05-18 | 2021-09-24 | 中国检验检疫科学研究院 | Hyperspectral imaging technology-based method for identifying aflatoxin in corn seeds |
Similar Documents
Publication | Publication Date | Title |
---|---|---|
CN110108644A (en) | A kind of maize variety identification method based on depth cascade forest and high spectrum image | |
Trier et al. | Tree species classification in Norway from airborne hyperspectral and airborne laser scanning data | |
CN104990892B (en) | The spectrum picture Undamaged determination method for establishing model and seeds idenmtification method of seed | |
Guzmán et al. | Infrared machine vision system for the automatic detection of olive fruit quality | |
Bensaeed et al. | Oil palm fruit grading using a hyperspectral device and machine learning algorithm | |
CN110596280B (en) | Rapid detection method for wheat vomitoxin pollution level based on hyperspectral image and spectrum information fusion | |
CN105866056A (en) | Hybrid purity identification method based on near infrared spectroscopy | |
CN109253975A (en) | Apple slight damage hyperspectral detection method based on MSC-CFS-ICA | |
WO2018054091A1 (en) | Method for identifying components of exocarpium | |
CN108734205A (en) | A kind of simple grain for different cultivars wheat seed pinpoints identification technology | |
CN110163101B (en) | Method for rapidly distinguishing seeds of traditional Chinese medicinal materials and rapidly judging grades of seeds | |
CN112699756B (en) | Hyperspectral image-based tea origin identification method and system | |
CN107679569A (en) | Raman spectrum substance automatic identifying method based on adaptive hypergraph algorithm | |
CN110363125A (en) | Using the method for Model Transfer identification different cultivars Citrus Huanglongbing pathogen | |
CN108169165A (en) | Maltose mixture quantitative analysis method based on tera-hertz spectra and image information fusion | |
CN105628708A (en) | Quick nondestructive testing method for multi-parameter quality of south Xinjiang red dates | |
CN109883967A (en) | A kind of Eriocheir sinensis quality grade method of discrimination based on information fusion | |
Tamayo-Monsalve et al. | Coffee maturity classification using convolutional neural networks and transfer learning | |
CN114663821B (en) | Real-time nondestructive detection method for product quality based on video hyperspectral imaging technology | |
CN113920376A (en) | Method for identifying wheat seed varieties based on light-weight convolutional neural network | |
CN113607656A (en) | Leaf chlorophyll content monitoring method and system based on hyperspectral imaging | |
Wei et al. | Tea moisture content detection with multispectral and depth images | |
Monteiro et al. | Image processing to identify damage to soybean seeds | |
CN105866043A (en) | Method for detecting apple sour through hyperspectral technology | |
Ji et al. | Apple color automatic grading method based on machine vision |
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 | ||
RJ01 | Rejection of invention patent application after publication |
Application publication date: 20190809 |