CN106841054A - Seed variety recognition methods and device - Google Patents
Seed variety recognition methods and device Download PDFInfo
- Publication number
- CN106841054A CN106841054A CN201710005012.7A CN201710005012A CN106841054A CN 106841054 A CN106841054 A CN 106841054A CN 201710005012 A CN201710005012 A CN 201710005012A CN 106841054 A CN106841054 A CN 106841054A
- Authority
- CN
- China
- Prior art keywords
- seed
- sample seed
- sample
- predetermined
- disaggregated model
- 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
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
- G01N21/25—Colour; Spectral properties, i.e. comparison of effect of material on the light at two or more different wavelengths or wavelength bands
- G01N21/27—Colour; Spectral properties, i.e. comparison of effect of material on the light at two or more different wavelengths or wavelength bands using photo-electric detection ; circuits for computing concentration
Landscapes
- Physics & Mathematics (AREA)
- Engineering & Computer Science (AREA)
- Mathematical Physics (AREA)
- Theoretical Computer Science (AREA)
- Spectroscopy & Molecular Physics (AREA)
- Health & Medical Sciences (AREA)
- Life Sciences & Earth Sciences (AREA)
- Chemical & Material Sciences (AREA)
- Analytical Chemistry (AREA)
- Biochemistry (AREA)
- General Health & Medical Sciences (AREA)
- General Physics & Mathematics (AREA)
- Immunology (AREA)
- Pathology (AREA)
- Investigating Or Analysing Materials By Optical Means (AREA)
- Image Analysis (AREA)
Abstract
The invention discloses a kind of seed variety recognition methods and device, belong to image classification field.The method includes obtaining P high spectrum image of each sample seed under P wave band in test set;For sample seed each described, characteristic parameter is obtained according to the P high spectrum image;The characteristic parameter is input into disaggregated model, the prediction kind of each sample seed is obtained;Predetermined sample seed is selected from the test set according to the prediction kind, and the disaggregated model is updated according to the predetermined sample seed;The kind of sample seed described in the test set is identified using the disaggregated model after renewal;Solve the problems, such as seed when same breed due to when time difference causes identification kind the degree of accuracy it is not high;Reach and automatically updated disaggregated model, influence when reducing the time to identification seed variety has improved the effect of the degree of accuracy of identification kind.
Description
Technical field
The present embodiments relate to image classification field, more particularly to a kind of seed variety recognition methods and device.
Background technology
Corn is one of staple crops of agricultural production, is the main source of grain, feed, fuel and the raw material of industry.
As the extensive use of hybrid seed technology, the kind of corn seed are more and more, classified by the kind of corn seed
Become increasingly complex.
The advantage of machine vision and near infrared spectrum is combined due to high light spectrum image-forming, can simultaneously reflect the inside of seed
Feature and surface, high light spectrum image-forming technology is widely used in seed classification.
However, the seed of the same breed in different year plantation, tillage condition, soil environment bar due to different year
Part is different with weather conditions, and the hyperspectral information of the seed of the same breed of different year can change, and cause to utilize bloom
The degree of accuracy of the seed variety that spectrogram picture and disaggregated model are identified is not high.
The content of the invention
In order to solve problem of the prior art, a kind of seed variety method and device is the embodiment of the invention provides.The skill
Art scheme is as follows:
First aspect, there is provided a kind of seed variety recognition methods, the method includes:
Obtain P high spectrum image of each sample seed under P wave band in test set;
For sample seed each described, characteristic parameter is obtained according to the P high spectrum image;
The characteristic parameter is input into disaggregated model, the prediction kind of each sample seed is obtained;
Predetermined sample seed is selected from the test set according to the prediction kind, and according to the predetermined sample kind
Son updates the disaggregated model;
The kind of sample seed described in the test set is identified using the disaggregated model after renewal.
Optionally, it is described that predetermined sample seed is selected from the test set according to the prediction kind, and according to institute
State predetermined sample seed and update the disaggregated model, including:
The sample seed is divided into by several classifications according to the prediction kind;Sample seed in each described classification
The prediction kind it is identical;
Calculate the class center of each classification;
Calculate the distance between each described sample seed in each classification and described class center;
For classification each described, the order arrangement by the sample seed by the distance from small to large;
Using the preceding n sample seed in each described classification as the predetermined sample seed, and by the predetermined sample
The corresponding characteristic parameter of seed is deleted from whole characteristic parameters;
The predetermined sample seed is added into training set and obtains the new training set;
The disaggregated model is updated using the new training set;
Whether detection predetermined condition is set up;
If the predetermined condition is invalid, perform it is described by the characteristic parameter be input into disaggregated model, obtain each institute
The step of stating the prediction kind of sample seed;
Wherein, the predetermined condition isMore than or equal to predetermined threshold, or, the predetermined condition is institute
State number of times and reach pre-determined number;NumjJ-th sample of classification is all determined when () is ith iteration i and during the i-th -1 time iteration
The quantity of seed, Numj(i-1) it is confirmed as the quantity of the sample seed of j-th classification when being the ith iteration.
Optionally, the method also includes:
If the predetermined condition is set up, perform the disaggregated model using after updating and identify the test set
Described in sample seed kind the step of.
Optionally, the distance between the described sample seed and described class center calculated in each classification, including:
Using formulaCalculate the sample kind
It is sub the distance between with the class center;
Wherein,D is the dimension of vector.
Optionally, it is described that characteristic parameter is obtained according to the P high spectrum image, including:
Obtain high spectrum image of the sample seed under predetermined band;
Image segmentation is carried out to the high spectrum image under predetermined band, the profile for obtaining the sample seed is bent
Line;
The contour curve is projected into the P wave band, P seed profile is obtained from the P high spectrum image
Image;
According to the seed contour images, the spectrum average feature of the corresponding sample seed of each wave band, institute are obtained
State the average value that spectrum average is characterized in the reflective light intensity of all pixels point in the seed contour images;
Using the corresponding P spectrum average feature of the P wave band as characteristic parameter.
Second aspect, there is provided a kind of seed variety identifying device, the device includes:
High spectrum image acquisition module, for obtaining P EO-1 hyperion of each sample seed under P wave band in test set
Image;
Characteristic parameter acquisition module, for for sample seed each described, obtaining special according to the P high spectrum image
Levy parameter;
Prediction kind acquisition module, for the characteristic parameter to be input into disaggregated model, obtains each described sample seed
Prediction kind;
Model modification module, for selecting predetermined sample seed from the test set according to the prediction kind, and
The disaggregated model is updated according to the predetermined sample seed;
Variety ecotype module, for identifying sample kind described in the test set using the disaggregated model after renewal
The kind of son.
Optionally, the model modification module, specifically for:
The sample seed is divided into by several classifications according to the prediction kind;Sample seed in each described classification
The prediction kind it is identical;
Calculate the class center of each classification;
Calculate the distance between each described sample seed in each classification and described class center;
For classification each described, the order arrangement by the sample seed by the distance from small to large;
Using the preceding n sample seed in each described classification as the predetermined sample seed, and by the predetermined sample
The corresponding characteristic parameter of seed is deleted from whole characteristic parameters;
The predetermined sample seed is added into training set and obtains the new training set;
The disaggregated model is updated using the new training set;
Whether detection predetermined condition is set up;
If the predetermined condition is invalid, perform it is described by the characteristic parameter be input into disaggregated model, obtain each institute
The step of stating the prediction kind of sample seed;
Wherein, the predetermined condition isMore than or equal to predetermined threshold, or, the predetermined condition is institute
State number of times and reach pre-determined number;NumjJ-th sample of classification is all determined when () is ith iteration i and during the i-th -1 time iteration
The quantity of seed, Numj(i-1) it is confirmed as the quantity of the sample seed of j-th classification when being the ith iteration.
Optionally, the model modification module, is additionally operable to:
If the predetermined condition is set up, perform the disaggregated model using after updating and identify the test set
Described in sample seed kind the step of.
Optionally, the model modification module, is additionally operable to:
Using formulaCalculate the sample kind
It is sub the distance between with the class center;
Wherein,D is the dimension of vector.
Optionally, the characteristic parameter acquisition module, specifically for:
Obtain high spectrum image of the sample seed under predetermined band;
Image segmentation is carried out to the high spectrum image under predetermined band, the profile for obtaining the sample seed is bent
Line;
The contour curve is projected into the P wave band, P seed profile is obtained from the P high spectrum image
Image;
According to the seed contour images, the spectrum average feature of the corresponding sample seed of each wave band, institute are obtained
State the average value that spectrum average is characterized in the reflective light intensity of all pixels point in the seed contour images;
Using the corresponding P spectrum average feature of the P wave band as characteristic parameter.
The beneficial effect that technical scheme provided in an embodiment of the present invention is brought is:
By obtaining several sample seeds P high spectrum image respectively under P wave band in test set, according to bloom
Spectrogram picture obtains the characteristic parameter of each sample seed, and the prediction product of sample seed are obtained using characteristic parameter and disaggregated model
Kind, using kind selection predetermined sample seed is predicted, disaggregated model is updated according to predetermined sample seed, finally utilize after updating
Disaggregated model identification test set in whole seeds kind;The seed when same breed is solved because time difference causes to know
Degree of accuracy problem not high during other kind;Reach and automatically updated disaggregated model, shadow when reducing the time to identification seed variety
Ring, improve the effect of the degree of accuracy of identification kind.
Brief description of the drawings
Technical scheme in order to illustrate more clearly the embodiments of the present invention, below will be to that will make needed for embodiment description
Accompanying drawing is briefly described, it should be apparent that, drawings in the following description are only some embodiments of the present invention, for
For those of ordinary skill in the art, on the premise of not paying creative work, other can also be obtained according to these accompanying drawings
Accompanying drawing.
Fig. 1 is a kind of flow chart of the seed variety recognition methods according to an exemplary embodiment;
Fig. 2 is a kind of flow chart of the seed variety recognition methods according to another exemplary embodiment;
Fig. 3 A are a kind of partial schematic diagrams of the high spectrum image according to another exemplary embodiment;
Fig. 3 B are a kind of partial schematic diagrams of the high spectrum image according to another exemplary embodiment;
Fig. 3 C are a kind of partial schematic diagrams of the contour curve of the sample seed according to another exemplary embodiment;
Fig. 3 D are a kind of seed profile diagram of sample seed according to another exemplary embodiment under P wave band
Picture;
Fig. 3 E are a kind of spectrum averages of the corresponding sample seed of each wave band according to another exemplary embodiment
The curve synoptic diagram of feature;
Fig. 4 is a kind of block diagram of the seed variety identifying device according to another exemplary embodiment.
Specific embodiment
To make the object, technical solutions and advantages of the present invention clearer, below in conjunction with accompanying drawing to embodiment party of the present invention
Formula is described in further detail.
Fig. 1 is refer to, the flow chart of the seed variety recognition methods provided it illustrates one embodiment of the invention.As schemed
Shown in 1, the seed variety recognition methods may comprise steps of:
Step 101, obtains P high spectrum image of each sample seed under P wave band in test set.
Such as:Test is concentrated with N number of sample seed, then get N*P high spectrum image.
Specifically, the whole sample seeds in test set are prevented putting in high spectrum image acquisition system, collects each
P high spectrum image of the sample seed under P wave band.
Optionally, the sample seed of test set China can be placed in high spectrum image acquisition system by several times, be gathered
P high spectrum image of each sample seed under P wave band.
Test set includes the seed of several kinds, and the seed of each kind includes several times.
Step 102, for each sample seed, characteristic parameter is obtained according to P high spectrum image.
For each the sample seed in test set, obtained according to P high spectrum image of the sample seed under P wave band
Take the sample seed for characteristic parameter.
One sample seed has a characteristic parameter, and the corresponding characteristic parameter of sample seed is special by P spectrum average
Composition is levied, a sample seed has a corresponding spectrum average feature under a wave band.
Step 103, disaggregated model is input into by characteristic parameter, obtains the prediction kind of each sample seed.
Preliminary classification model is generated according to training set, and kind and the test set of the seed that training set includes include
Seed kind it is identical, the time of the seed of kind corresponding from training set is different in test set.
Seed in training set includes multiple kinds, and the seed of each kind includes multiple times.Seed in test set
Kind when it is unknown.
Such as:Seed in test set is A kinds, B kinds, C kinds, the seed in training set is also A kinds, B kinds,
C kinds, but the seed of A kinds is different from the time of the seed of A kinds in training set in test set, B kinds in test set
Seed is different from the time of the seed of B kinds in training set, in test set in the seed of C kinds and training set C kinds seed
Time it is different.
Optionally, disaggregated model is least square method supporting vector machine disaggregated model.
Step 104, predetermined sample seed is selected according to prediction kind from test set, and according to predetermined sample seed more
New disaggregated model.
Predetermined sample seed is selected from the whole sample seeds in test set according to prediction kind, what is selected is predetermined
Quantity of the quantity of sample seed less than whole sample seeds in test set.
Optionally, predetermined sample seed can be repeatedly obtained from test set, the predetermined sample seed for selecting every time is not
Together;A predetermined sample seed is often selected, disaggregated model is once updated using the predetermined sample seed for selecting.
Step 105, the kind of sample seed in test set is identified using the disaggregated model after renewal.
Optionally, the kind of whole sample seeds in test set is identified using the disaggregated model after renewal, or, utilize
Disaggregated model after renewal identifies the kind of test concentrated part sample seed.
In sum, seed variety recognition methods provided in an embodiment of the present invention, by obtaining each sample in test set
Seed P high spectrum image respectively under P wave band, the characteristic parameter of each sample seed is obtained according to high spectrum image,
The prediction kind of sample seed is obtained using characteristic parameter and disaggregated model, predicts that kind chooses predetermined sample seed utilizing,
Disaggregated model is updated according to predetermined sample seed, finally using the product of sample seed in the disaggregated model identification test set after updating
Kind;Solve the problems, such as seed when same breed due to when time difference causes identification kind the degree of accuracy it is not high;Reach certainly
Dynamic to update disaggregated model, influence when reducing the time to identification seed variety improves the effect of the degree of accuracy of identification kind.
Fig. 2 is refer to, the flow chart of the seed variety recognition methods provided it illustrates another embodiment of the present invention.Such as
Shown in Fig. 2, the seed variety recognition methods may comprise steps of:
Step 201, obtains P high spectrum image of each sample seed under P wave band in test set.
The step is set forth in a step 101, repeats no more here.
Step 202, for each sample seed, characteristic parameter is obtained according to P high spectrum image.
If there is N number of sample seed, N number of characteristic parameter is got.
Specifically, to each sample seed, characteristic parameter is obtained according to following steps 2021 to step 2025.
Step 2021, obtains high spectrum image of the sample seed under predetermined band.
Optionally, predetermined band is high spectrum image most clearly wave band.Optionally, predetermined band is 782.59nm.
Step 2022, image segmentation is carried out to the high spectrum image under predetermined band, and the profile for obtaining sample seed is bent
Line.
Optionally, before image segmentation is carried out to high spectrum image, figure enhancing is carried out to high spectrum image.
Optionally, image segmentation is carried out to the high spectrum image under predetermined band using threshold segmentation method, obtains sample
The contour curve of this seed.
Step 2023, P wave band is projected to by contour curve, and P seed profile diagram is obtained from P high spectrum image
Picture.
Using the method for outline projection, contour curve is projected under P wave band, from the corresponding high-spectrum of each wave band
A seed contour images are isolated as in.
One wave band one seed contour images of correspondence.One sample seed has P seed contour images.
Step 2024, according to seed contour images, obtains the spectrum average feature of the corresponding sample seed of each wave band.
Spectrum average is characterized in the average value of the reflective light intensity of all pixels point in seed contour images.
According to the corresponding seed contour images of each wave band, the reflective light intensity of all pixels point in seed contour images is obtained
Average value.
In a wave band to that should have a spectrum average feature, a sample seed has P to one sample seed in P wave band
Individual spectrum average feature.
Step 2025, using the corresponding P spectrum average feature of P wave band as characteristic parameter.
Such as:P spectrum average feature of one sample seed is respectively p1、p2、p3、……、pp, then the sample seed
Characteristic parameter F=[p1,p2,p3,……,pp]。
Step 203, disaggregated model is input into by characteristic parameter, obtains the prediction kind of each sample seed.
Optionally, by characteristic parameter input disaggregated model in sample seed, the prediction kind of each sample seed is obtained.
Step 204, several classifications are divided into according to prediction kind by sample seed.
The prediction kind of the sample seed in each classification is identical.
If it should be noted that have in step 203 it is unidentified go out to predict the sample seed of kind, this step divide class
When other, do not consider unidentified to go out to predict the sample seed of kind.
In each iteration, the quantity of the classification for marking off can be differed.
Step 205, calculates the class center of each classification.
Assuming that sample seed is T classification, the T class center of classification is calculated for [c1,...,ci,...,cT]。
Step 206, calculates the distance between each sample seed in each classification and class center.
The distance between each sample seed and class center in each classification are calculated using Pearson correlation coefficient.
Specifically, using formulaCalculate sample
The distance between this seed and class center;
Wherein,A sample seed in each classification is represented,The class center of each classification is represented,D is the dimension of vector.
It should be noted thatValue it is bigger, illustrate represent sample seed vector sum represent class center
Vector it is more similar, namely sample seed is smaller with the distance between class center.
Step 207, for each classification, the order arrangement by sample seed by distance from small to large.
Step 208, using the preceding n sample seed in each classification as predetermined sample seed, and by predetermined sample seed
Corresponding characteristic parameter is deleted from whole characteristic parameters.
If the corresponding characteristic parameter of predetermined sample seed is deleted from whole characteristic parameters, in next iteration, suddenly
The predetermined sample seed of characteristic parameter is slightly deleted in test set, namely the predetermined sample seed of deleted characteristic parameter is no longer joined
With the renewal of disaggregated model.
Step 209, is added into predetermined sample seed training set and obtains new training set.
Step 210, disaggregated model is updated using new training set.
Whether step 211, detection predetermined condition is set up.
Wherein, predetermined condition isMore than or equal to predetermined threshold, or, predetermined condition is pre- for number of times reaches
Determine number of times;NumjJ-th quantity of the sample seed of classification is all determined when () is ith iteration i and during the i-th -1 time iteration,
Numj(i-1) it is confirmed as j-th quantity of the sample seed of classification when for ith iteration.
Optionally, predetermined threshold is the value for pre-setting.WhenDuring more than or equal to predetermined threshold, explanation
After last iteration disaggregated model predict the outcome it is very high with the similitude that predicts the outcome of current iteration disaggregated model, namely classification
Model is not almost updated, it is necessary to stop iteration.
Optionally, pre-determined number is maximum iteration.Maximum iteration and the sample seed obtained from test set
Quantity, test set in the quantity of seed variety that includes it is relevant.Optionally, if obtain P wave band under P high-spectrum
As when by several times obtain, then maximum iteration also with obtain image when every time obtain sample seed quantity it is relevant.
If predetermined condition is invalid, step 203 is held.
When predetermined condition is invalid, the disaggregated model in step 203 is the disaggregated model updated according to new training set,
The characteristic parameter after the corresponding characteristic parameter of predetermined sample seed will be deleted and be input into disaggregated model.
If predetermined condition is set up, step 212 is performed.
Step 212, the kind of whole seeds in test set is identified using the disaggregated model after renewal.
Optionally, using the characteristic parameter of the sample seed for needing to recognize kind in high spectrum image acquisition test set, will
Disaggregated model after needing the characteristic parameter input of the sample seed for recognizing kind to update in test set, identifies sample seed
Kind.
Optionally, it is necessary to the sample seed for recognizing kind is the whole sample seeds in test set, or, it is necessary to recognize product
The sample seed planted is test concentrated part sample seed.
The sample seed for needing identification can be the sample kind that characteristic parameter is deleted during disaggregated model is updated
Son.
In sum, seed variety recognition methods provided in an embodiment of the present invention, by obtaining each sample in test set
Seed P high spectrum image respectively under P wave band, the characteristic parameter of each sample seed is obtained according to high spectrum image,
The prediction kind of sample seed is obtained using characteristic parameter and disaggregated model, predicts that kind chooses predetermined sample seed utilizing,
Disaggregated model is updated according to predetermined sample seed, finally using the product of sample seed in the disaggregated model identification test set after updating
Kind;Solve the problems, such as seed when same breed due to when time difference causes identification kind the degree of accuracy it is not high;Reach certainly
Dynamic to update disaggregated model, influence when reducing the time to identification seed variety improves the effect of the degree of accuracy of identification kind.
In an exemplary example, it is assumed that the quantity of the sample seed of kind to be identified is 50 in test set, is obtained
P high spectrum image of each sample seed under P wave band in test set.50 seed specimens are obtained in 782.59nm wave bands
Under high spectrum image, Fig. 3 A show the partial schematic diagram of the high spectrum image;Image enhaucament is carried out to high spectrum image, is schemed
3B shows the partial schematic diagram of enhanced high spectrum image;Using thresholding method, 50 profiles of sample seed are obtained
Curve, Fig. 3 C show 50 partial schematic diagrams of the contour curve of sample seed;The contour curve that will be got is projected to P
Wave band, obtains seed contour images of each sample seed under P wave band, as shown in Figure 3 D;Further according to seed contour images,
The spectrum average feature of the corresponding sample seed of each wave band is obtained, each sample seed is obtained according to P spectrum average feature
Characteristic parameter, as shown in FIGURE 3 E.
Assuming that disaggregated model is least square method supporting vector machine disaggregated model Mlssvm, initial MlssvmGiven birth to according to training set S
Into.The characteristic parameter of 50 sample seeds is input into M respectivelylssvm, obtain the prediction label of each sample seed;Assuming that there is 47
Sample seed have identified prediction label, then 47 sample seeds are classified according to prediction label, the sample of prediction label
Seed is assigned in same classification, and prediction label has 4 classes, respectively A classes, B classes, C classes, D classes;A classes, B classes, C are calculated respectively
Class, D Lei Lei centers, then calculate each sample to the distance at the class center of respective generic for each class;By each
Used as predetermined sample seed, the set of predetermined sample seed can be expressed as the nearest n samples seed in distance-like center in classification
L, training set S, training set S=S+L are added into by predetermined sample seed L;Predetermined sample seed in test set is ignored, accordingly
The characteristic parameter of predetermined sample seed is deleted on ground, and test set can be expressed as set U, U=U-L;Updated using training set S
Mlssvm;Whether detection predetermined condition is set up, if predetermined condition is invalid, then performs the characteristic parameter of the sample seed in U
Input Mlssvm, in acquisition U the step of the prediction kind of each sample seed, until predetermined condition is set up, then utilize after updating
MlssvmThe kind of the sample seed in identification test set.
Following is apparatus of the present invention embodiment, can be used for performing the inventive method embodiment.For apparatus of the present invention reality
The details not disclosed in example is applied, the inventive method embodiment is refer to.
Fig. 4 is refer to, the block diagram of the seed variety identifying device provided it illustrates one embodiment of the invention.
The seed variety identifying device can by software, hardware or both be implemented in combination with turn into it is above-mentioned provide seed variety know
The all or part of the terminal of other method.The device includes:
High spectrum image acquisition module 410 is high for obtaining in test set P of each sample seed under P wave band
Spectrum picture;
Characteristic parameter acquisition module 420, for for each sample seed, feature ginseng being obtained according to P high spectrum image
Number;
Prediction kind acquisition module 430, for characteristic parameter to be input into disaggregated model, obtains the prediction of each sample seed
Kind;
Model modification module 440, for selecting predetermined sample seed from test set according to prediction kind, and according to pre-
Random sample this seed renewal disaggregated model;
Variety ecotype module 450, the kind for identifying sample seed in test set using the disaggregated model after renewal.
In sum, seed variety identifying device provided in an embodiment of the present invention, by obtaining each sample in test set
Seed P high spectrum image respectively under P wave band, the characteristic parameter of each sample seed is obtained according to high spectrum image,
The prediction kind of sample seed is obtained using characteristic parameter and disaggregated model, predicts that kind chooses predetermined sample seed utilizing,
Disaggregated model is updated according to predetermined sample seed, finally using the product of sample seed in the disaggregated model identification test set after updating
Kind;Solve the problems, such as seed when same breed due to when time difference causes identification kind the degree of accuracy it is not high;Reach certainly
Dynamic to update disaggregated model, influence when reducing the time to identification seed variety improves the effect of the degree of accuracy of identification kind.
Optionally, model modification module, specifically for:
Sample seed is divided into by several classifications according to prediction kind;The prediction kind phase of the sample seed in each classification
Together;
Calculate the class center of each classification;
Calculate the distance between each sample seed in each classification and class center;
For each classification, the order arrangement by sample seed by distance from small to large;
Using the preceding n sample seed in each classification as predetermined sample seed, and by the corresponding spy of predetermined sample seed
Parameter is levied to be deleted from whole characteristic parameters;
Predetermined sample seed is added into training set and obtains new training set;
Disaggregated model is updated using new training set;
Whether detection predetermined condition is set up;
If predetermined condition is invalid, performs and characteristic parameter is input into disaggregated model, obtain the prediction of each sample seed
The step of kind;
Wherein, predetermined condition isMore than or equal to predetermined threshold, or, predetermined condition is pre- for number of times reaches
Determine number of times;NumjJ-th quantity of the sample seed of classification is all determined when () is ith iteration i and during the i-th -1 time iteration,
Numj(i-1) it is confirmed as j-th quantity of the sample seed of classification when for ith iteration.
Optionally, model modification module, is additionally operable to:
If predetermined condition is set up, the kind that sample seed in test set is identified using the disaggregated model after renewal is performed
The step of.
Optionally, model modification module, is additionally operable to:
Using formulaCalculate sample seed with
The distance between class center;
Wherein,D is the dimension of vector.
Optionally, characteristic parameter acquisition module, specifically for:
Obtain high spectrum image of the sample seed under predetermined band;
Image segmentation is carried out to the high spectrum image under predetermined band, the contour curve of sample seed is obtained;
Contour curve is projected into P wave band, P seed contour images are obtained from P high spectrum image;
According to seed contour images, the spectrum average feature of the corresponding sample seed of each wave band is obtained, spectrum average is special
Levy the average value of the reflective light intensity for being all pixels point in seed contour images;
Using the corresponding P spectrum average feature of P wave band as characteristic parameter.
It should be noted that:The seed variety identification computing device that above-described embodiment is provided is performing seed variety identification side
During method, only carried out with the division of above-mentioned each functional module for example, in practical application, can be as needed and by above-mentioned functions
Distribute and completed by different functional module, will the internal structure of equipment be divided into different functional modules, retouched with completing the above
The all or part of function of stating.In addition, the seed variety identifying device of above-described embodiment offer and seed variety recognition methods
Embodiment belongs to same design, and it implements process and refers to embodiment of the method, repeats no more here.
The embodiments of the present invention are for illustration only, and the quality of embodiment is not represented.
One of ordinary skill in the art will appreciate that realizing that all or part of step of above-described embodiment can be by hardware
To complete, it is also possible to instruct the hardware of correlation to complete by program, described program can be stored in a kind of computer-readable
In storage medium, storage medium mentioned above can be read-only storage, disk or CD etc..
The foregoing is only presently preferred embodiments of the present invention, be not intended to limit the invention, it is all it is of the invention spirit and
Within principle, any modification, equivalent substitution and improvements made etc. should be included within the scope of the present invention.
Claims (10)
1. a kind of seed variety recognition methods, it is characterised in that methods described includes:
Obtain P high spectrum image of each sample seed under P wave band in test set;
For sample seed each described, characteristic parameter is obtained according to the P high spectrum image;
The characteristic parameter is input into disaggregated model, the prediction kind of each sample seed is obtained;
Predetermined sample seed is selected from the test set according to the prediction kind, and according to the predetermined sample seed more
The new disaggregated model;
The kind of sample seed described in the test set is identified using the disaggregated model after renewal.
2. method according to claim 1, it is characterised in that described to be selected from the test set according to the prediction kind
Predetermined sample seed is taken out, and the disaggregated model is updated according to the predetermined sample seed, including:
The sample seed is divided into by several classifications according to the prediction kind;The institute of the sample seed in each described classification
State prediction kind identical;
Calculate the class center of each classification;
Calculate the distance between each described sample seed in each classification and described class center;
For classification each described, the order arrangement by the sample seed by the distance from small to large;
Using the preceding n sample seed in each described classification as the predetermined sample seed, and by the predetermined sample seed
The corresponding characteristic parameter is deleted from whole characteristic parameters;
The predetermined sample seed is added into training set and obtains the new training set;
The disaggregated model is updated using the new training set;
Whether detection predetermined condition is set up;
If the predetermined condition is invalid, perform it is described by the characteristic parameter be input into disaggregated model, obtain each described sample
The step of prediction kind of this seed;
Wherein, the predetermined condition isMore than or equal to predetermined threshold, or, the predetermined condition is described time
Number reaches pre-determined number;NumjJ-th sample seed of classification is all determined when () is ith iteration i and during the i-th -1 time iteration
Quantity, Numj(i-1) it is confirmed as the quantity of the sample seed of j-th classification when being the ith iteration.
3. method according to claim 2, it is characterised in that methods described also includes:
If the predetermined condition is set up, perform the disaggregated model using after updating and identify institute in the test set
The step of stating the kind of sample seed.
4. method according to claim 2, it is characterised in that the sample seed and institute in the calculating each classification
The distance between Shu Lei centers, including:
Using formulaCalculate the sample seed with
The distance between described class center;
Wherein,D is the dimension of vector.
5. according to any described method of Claims 1-4, it is characterised in that described to be obtained according to the P high spectrum image
Characteristic parameter, including:
Obtain high spectrum image of the sample seed under predetermined band;
Image segmentation is carried out to the high spectrum image under predetermined band, the contour curve of the sample seed is obtained;
The contour curve is projected into the P wave band, P seed contour images are obtained from the P high spectrum image;
According to the seed contour images, the spectrum average feature of the corresponding sample seed of each wave band, the light are obtained
Spectrum characteristics of mean is the average value of the reflective light intensity of all pixels point in the seed contour images;
Using the corresponding P spectrum average feature of the P wave band as characteristic parameter.
6. a kind of seed variety identifying device, it is characterised in that described device includes:
High spectrum image acquisition module, for obtaining P high-spectrum of each sample seed under P wave band in test set
Picture;
Characteristic parameter acquisition module, for for sample seed each described, feature ginseng being obtained according to the P high spectrum image
Number;
Prediction kind acquisition module, for the characteristic parameter to be input into disaggregated model, obtains the pre- of each sample seed
Survey kind;
Model modification module, for selecting predetermined sample seed from the test set according to the prediction kind, and according to
The predetermined sample seed updates the disaggregated model;
Variety ecotype module, for identifying sample seed described in the test set using the disaggregated model after renewal
Kind.
7. device according to claim 6, it is characterised in that the model modification module, specifically for:
The sample seed is divided into by several classifications according to the prediction kind;The institute of the sample seed in each described classification
State prediction kind identical;
Calculate the class center of each classification;
Calculate the distance between each described sample seed in each classification and described class center;
For classification each described, the order arrangement by the sample seed by the distance from small to large;
Using the preceding n sample seed in each described classification as the predetermined sample seed, and by the predetermined sample seed
The corresponding characteristic parameter is deleted from whole characteristic parameters;
The predetermined sample seed is added into training set and obtains the new training set;
The disaggregated model is updated using the new training set;
Whether detection predetermined condition is set up;
If the predetermined condition is invalid, perform it is described by the characteristic parameter be input into disaggregated model, obtain each described sample
The step of prediction kind of this seed;
Wherein, the predetermined condition isMore than or equal to predetermined threshold, or, the predetermined condition is described time
Number reaches pre-determined number;NumjJ-th sample seed of classification is all determined when () is ith iteration i and during the i-th -1 time iteration
Quantity, Numj(i-1) it is confirmed as the quantity of the sample seed of j-th classification when being the ith iteration.
8. device according to claim 7, it is characterised in that the model modification module, is additionally operable to:
If the predetermined condition is set up, perform the disaggregated model using after updating and identify institute in the test set
The step of stating the kind of sample seed.
9. device according to claim 7, it is characterised in that the model modification module, is additionally operable to:
Using formulaCalculate the sample seed with
The distance between described class center;
Wherein,D is the dimension of vector.
10. it is specific to use according to any described device of claim 6 to 9, it is characterised in that the characteristic parameter acquisition module
In:
Obtain high spectrum image of the sample seed under predetermined band;
Image segmentation is carried out to the high spectrum image under predetermined band, the contour curve of the sample seed is obtained;
The contour curve is projected into the P wave band, P seed contour images are obtained from the P high spectrum image;
According to the seed contour images, the spectrum average feature of the corresponding sample seed of each wave band, the light are obtained
Spectrum characteristics of mean is the average value of the reflective light intensity of all pixels point in the seed contour images;
Using the corresponding P spectrum average feature of the P wave band as characteristic parameter.
Priority Applications (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201710005012.7A CN106841054B (en) | 2017-01-04 | 2017-01-04 | Seed variety recognition methods and device |
Applications Claiming Priority (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201710005012.7A CN106841054B (en) | 2017-01-04 | 2017-01-04 | Seed variety recognition methods and device |
Publications (2)
Publication Number | Publication Date |
---|---|
CN106841054A true CN106841054A (en) | 2017-06-13 |
CN106841054B CN106841054B (en) | 2019-06-07 |
Family
ID=59118401
Family Applications (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
CN201710005012.7A Active CN106841054B (en) | 2017-01-04 | 2017-01-04 | Seed variety recognition methods and device |
Country Status (1)
Country | Link |
---|---|
CN (1) | CN106841054B (en) |
Cited By (1)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
NL2027491A (en) | 2020-02-27 | 2021-09-30 | Univ Zhejiang | Method for rapidly identifying identities of pumpkin seeds |
Citations (6)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN102072883A (en) * | 2010-07-07 | 2011-05-25 | 北京农业智能装备技术研究中心 | Device and method for detecting comprehensive quality of crop seeds |
CN102621077A (en) * | 2012-03-30 | 2012-08-01 | 江南大学 | Hyper-spectral reflection image collecting system and corn seed purity nondestructive detection method based on same |
CN105117734A (en) * | 2015-07-28 | 2015-12-02 | 江南大学 | Corn seed hyper-spectral image classification identification method based on model on-line updating |
CN105224960A (en) * | 2015-11-04 | 2016-01-06 | 江南大学 | Based on the corn seed classification hyperspectral imagery model of cognition update method of clustering algorithm |
CN106203522A (en) * | 2016-07-15 | 2016-12-07 | 西安电子科技大学 | Hyperspectral image classification method based on three-dimensional non-local mean filtering |
CN106203452A (en) * | 2016-07-18 | 2016-12-07 | 江南大学 | Corn seed high spectrum image multiple features conversion method based on polyteny discriminant analysis |
-
2017
- 2017-01-04 CN CN201710005012.7A patent/CN106841054B/en active Active
Patent Citations (6)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN102072883A (en) * | 2010-07-07 | 2011-05-25 | 北京农业智能装备技术研究中心 | Device and method for detecting comprehensive quality of crop seeds |
CN102621077A (en) * | 2012-03-30 | 2012-08-01 | 江南大学 | Hyper-spectral reflection image collecting system and corn seed purity nondestructive detection method based on same |
CN105117734A (en) * | 2015-07-28 | 2015-12-02 | 江南大学 | Corn seed hyper-spectral image classification identification method based on model on-line updating |
CN105224960A (en) * | 2015-11-04 | 2016-01-06 | 江南大学 | Based on the corn seed classification hyperspectral imagery model of cognition update method of clustering algorithm |
CN106203522A (en) * | 2016-07-15 | 2016-12-07 | 西安电子科技大学 | Hyperspectral image classification method based on three-dimensional non-local mean filtering |
CN106203452A (en) * | 2016-07-18 | 2016-12-07 | 江南大学 | Corn seed high spectrum image multiple features conversion method based on polyteny discriminant analysis |
Cited By (1)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
NL2027491A (en) | 2020-02-27 | 2021-09-30 | Univ Zhejiang | Method for rapidly identifying identities of pumpkin seeds |
Also Published As
Publication number | Publication date |
---|---|
CN106841054B (en) | 2019-06-07 |
Similar Documents
Publication | Publication Date | Title |
---|---|---|
CN110148120B (en) | Intelligent disease identification method and system based on CNN and transfer learning | |
CN113160192B (en) | Visual sense-based snow pressing vehicle appearance defect detection method and device under complex background | |
CN109754017B (en) | Hyperspectral image classification method based on separable three-dimensional residual error network and transfer learning | |
CN106897673B (en) | Retinex algorithm and convolutional neural network-based pedestrian re-identification method | |
CN111027493A (en) | Pedestrian detection method based on deep learning multi-network soft fusion | |
CN110245678B (en) | Image matching method based on heterogeneous twin region selection network | |
CN111723860A (en) | Target detection method and device | |
CN110770752A (en) | Automatic pest counting method combining multi-scale feature fusion network with positioning model | |
CN111369540A (en) | Plant leaf disease identification method based on mask convolutional neural network | |
CN109829914A (en) | The method and apparatus of testing product defect | |
CN111553240B (en) | Corn disease condition grading method and system and computer equipment | |
CN109101934A (en) | Model recognizing method, device and computer readable storage medium | |
CN110929944A (en) | Wheat scab disease severity prediction method based on hyperspectral image and spectral feature fusion technology | |
CN111860537B (en) | Deep learning-based green citrus identification method, equipment and device | |
CN110310305B (en) | Target tracking method and device based on BSSD detection and Kalman filtering | |
CN112215795A (en) | Intelligent server component detection method based on deep learning | |
CN110097535A (en) | The nitrogenous quantity measuring method of plant leaf blade, device, computer equipment and storage medium | |
CN110348503A (en) | A kind of apple quality detection method based on convolutional neural networks | |
CN111340019A (en) | Grain bin pest detection method based on Faster R-CNN | |
CN107274425B (en) | A kind of color image segmentation method and device based on Pulse Coupled Neural Network | |
CN112149707B (en) | Image acquisition control method, device, medium and equipment | |
CN116740650A (en) | Crop breeding monitoring method and system based on deep learning | |
CN111091122A (en) | Training and detecting method and device for multi-scale feature convolutional neural network | |
CN109741380B (en) | Textile picture fast matching method and device | |
Zhao et al. | Rice seed size measurement using a rotational perception deep learning model |
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 | ||
GR01 | Patent grant | ||
GR01 | Patent grant |