CN104316473A - Gender determination method for chicken hatching egg incubation early embryo based on hyperspectral image - Google Patents

Abstract

The invention relates to a gender determination method for a chicken hatching egg incubation early embryo based on a hyperspectral image, and belongs to the non-destructive testing technology in the poultry propagation industry. The gender determination method comprises the following steps: utilizing a hyperspectral imager to obtain a transmission hyperspectral image on the tenth day of chicken embryo hatching, analyzing difference of spectral response between a male embryo and a female embryo, extracting a spectral value within a wavelength range of 600-900 nm as an input value of a neural network, and judging the gender of the chicken embryo. With adoption of the method, the gender of the chicken embryo can be accurately determined, manual detection after hatching is replaced, the cost in the poultry production industry is lowered, and the production efficiency is improved.

Description

A kind of based on the method for high spectrum image to breeding-egg hatching Early-stage judgment

Technical field

The present invention is a kind of hyper-spectral image technique at hatching early detection chicken kind egg Embryo method for distinguishing, belongs to the technical field that poultry propagation field hatching characteristic-nondestructive detects.

Background technology

In eggs are produced, the productivity effect of hen is far above cock, and in broiler production, cock is higher than food stuff for chicken laying utilization factor again, fast growth, and vitality is strong.The hatching process of egg approximately needs 21 days, be one consuming time, power consumption process.If can be early stage in hatching, detect kind of an egg sex, regulate and control male and female cultivation ratio better, meet the market demand of different times, just can save production cost, improve enterprises production efficiency.Plant egg sex abnormality and have a kind of the most original method, face shaping and physical behavior exactly by checking kind of egg identify sex, but this method mainly relies on the experience of people, and accuracy rate is lower.Therefore, seek a kind of effective detection hatching of breeding eggs method for distinguishing fast, significant to whole hatching industry.

The early stage sex identification of bird is mainly divided into eggshell to identify the qualification with egg content outward.The former is as methods such as kind of an egg external form, embryo line morphology, Gonadal Morphologies.[the Wang Jiapei such as Wang Jiapei, Xu Houqiang. utilize egg shape index to carry out sex identification Primary Study [J] to the little fragrant chicken kind egg in the southeast of Guizhou Province. Guizhou animal and veterinary, 2011,35 (4): 6-7] utilize egg shape index to carry out sex identification Primary Study to the little fragrant chicken kind egg in the southeast of Guizhou Province, find that egg shape index and sex have comparatively close correlativity.Tang Jianlin etc. [Tang Jianlin, Zhou Yulan. the early stage Sex judging of chicken embryo [J]. Guizhou animal and veterinary, 2001,25 (5): 29.] after kind of egg enters to incubate 3d, shine egg, find that the main blood vessel of male embryo is obvious, blood vessel is comparatively thick, is evenly distributed; Female embryo blood vessel is very thin, even thickness, and branch is more, in irregular, thus distinguishes kind of an egg sex.The latter is as molecular method, Hormonal methods.[the Griffiths R such as Griffiths, Double M C, Orr K, et al.A DNA test to sex most birds [J] .Molecular ecology, 1998,7 (8): 1071-1075.] describe and utilize round pcr to identify 5 ~ 7d chicken embryo method for distinguishing.Also there are some researches show, estrogen is the ideal indicator carrying out chicken embryo sex identification.Embrex company of the U.S. maintains the leading position in this regard, has applied for two patents (patent No. is respectively 6506570 and 6510811).Steiner [Steiner G, Bartels T, StellingA, et al.Gender determination of fertilized unincubated chicken eggs by infrared spectroscopic imaging [J] .Analytical and bioanalytical chemistry, 2011, 400 (9): 2775-2782.] etc. near infrared imaging system is utilized, the non-hatched blastocyst taken out from kind of egg is analyzed, result shows that near infrared imaging system has the potential ability differentiated as hatching eggs embryo, but need in testing process to destroy embryo, it not the method for Non-Destructive Testing.Through retrieving the patent of invention CN103472008A of application in 2013, proposing chicken embryo gender identification method in hatching near infrared kind in early stage egg, but there is no the concrete time of detection in the clear and definite incubation period, practical application difficulty.The patent CN1035487818A of application proposes a kind of for detecting the device whether egg is fertilized and is determined its sex, and this patent merely provides a platform detected, and the sex identification specifically how realizing chicken embryo does not relate to.Therefore, be necessary to develop a kind of qualification loseless method utilizing high spectrum image to realize the embryo hatching early sex of chicken, to improve the productivity effect of poultry propagation.

Summary of the invention

Technical matters

In view of above-mentioned state-of-the-art, object of the present invention cannot realize the difficult problem of chicken embryo at hatching early sex Nondestructive Identification mainly for prior art, and the quick nondestructive method that exploitation high spectrum image detects meets the active demand in poultry propagation field.By utilizing high light spectrum image-forming technology, analyzing the spectral information difference of chicken embryo male and female embryo between the whole incubation period, extracting the characteristic parameter of response, build the qualification model of chicken embryo gender.Method of the present invention also may be used in the sex Nondestructive Identification application of other avian embryonic.

Technical scheme

1., based on the method for high spectrum image to breeding-egg hatching Early-stage judgment, its device constitutive characteristic is,

1) system composition comprises high light spectrum image-forming unit, sample holder, electric platforms, line source, light box, computer and image capture software composition, and whole device is placed in airtight black box.Wherein, high light spectrum image-forming unit is by video camera (Imperx, ICL-B1620, wavelength band is 400 ~ 1000nm, spectral resolution is 2.8nm), spectrometer (Specim, ImSpector, V10E) and focal length variable lens composition, light box is the tungsten halogen lamp of 150W, complete transmission by 1 linear optical fiber conduit, computer model is CPU E5800,3.2GHz, internal memory 2G, video card 256M GeForce GT240; Image capture software is the Spectral Image software of independent development;

2) signals collecting is transmission mode, and lens are 20cm from chicken kind egg sample distance, and sample is close to line source and is placed, and the intensity of light source is 75W, gathers time shutter 55ms, picking rate 1.6mm/s, image resolution ratio 804 × 440 pixel;

2., based on the method for high spectrum image to breeding-egg hatching Early-stage judgment, its detecting step is,

1) by being in, temperature is 37.8 DEG C, relative humidity is that 65% condition descends the hatching chicken kind egg of 10 days to take out, and is positioned in high spectrum image detection system, obtains high spectrum image;

2) utilize following formula to the correct image obtained, obtain the high spectrum image after correcting:

$\mathrm{Rc}=\frac{R_0-D}{W-D}$

Wherein: Rc is the EO-1 hyperion transmission image after correcting, R ₀for original EO-1 hyperion transmission image, W be by reflectivity be 99.99% reference white correction plate, be placed on directly over light source, scanning transmission blank obtains entirely white uncalibrated image, and D is by lens cap on lens cap, gathers entirely black uncalibrated image;

3) area-of-interest of position, middle 100 × 100, egg region pixel size in image is selected, extract the spectrum average of all pixels in this region each wavelength in 600-900nm wavelength band, and as the input value of the neural network model built, Output rusults is the sex of chicken embryo, the neural network model parameter wherein built is: input layer is 207, the hiding number of plies is 1, and hidden layer nodes is 14, and hidden layer activation function is tanh; Output layer number is 2, namely male and female, output layer activation function Softmax.

Beneficial effect

The present invention utilizes the detection to high spectrum image instrument response signal; when can not destroy embryo in egg integrality; by the EO-1 hyperion response characteristic difference of male and female embryo; within during incubation the 10th day, judge the sex of embryo; the gender information of hatched blastocyst can be provided for the producer; avoid slaughtering after hatching, cut the waste, reduce the worry of humanistic Protector.Detect relative to traditional destructiveness, not only save time, and avoid the use of chemical reagent.This techniques and methods is novel, achievement in research not only may be used for express-analysis and the detection in laboratory, and can by exploitation online detection instrument and portable instrument, chicken Embryo sexing in producing for industrial automation, also for the detection of other class fowl hatched blastocyst sexes provides useful reference.

Four, accompanying drawing explanation

Fig. 1: the device of Hyperspectral imager kind egg sex identification

Fig. 2: ROI region position is selected

Fig. 3: different brooding time kind egg middle parts transmission spectrum curve

Fig. 4: different sexes kind egg embryonic development aspect graph

Fig. 5: the 10 day center section male and female kind egg averaged spectrum curve

Five, embodiment

Based on the method for high spectrum image to breeding-egg hatching Early-stage judgment, embodiment is as follows:

1. test material

Material is white shell chicken wings & legs with brown sauce kind egg, is purchased from Zhejiang Province's Rui'an City rare bird breeding field.Kind of egg 35 DEG C, soak disinfection in 5 minutes in 0.1% bromogeramine solution after, put into incubator, incubation temperature be 37.8 DEG C, relative humidity is 65%.By number high spectrum image collection is carried out to 94 kind eggs, within every 2 days, gather 1 time, continuous detecting 12 days.Wherein the sample of 2/3 is intensive 62 of model training, and remaining 1/3 sample is intensive 32 of model prediction.Experiment hatching of breeding eggs, by the 16th day, is opened kind of egg and is carried out destructiveness detection, artificial judgment embryo gender.The accuracy rate of EO-1 hyperion system judged result by drawing compared with artificial cognition.

2. EO-1 hyperion transmission image acquisition system

Hyperspectral imager inspection kind egg sex device as shown in Figure 1.Transmission mode is adopted in this research.In this experiment, horizontal placement is taked to kind of an egg, gather image.

The hardware of the Hyperspectral imager used in experiment forms primarily of high spectrum image unit, direct current tunable light source, sample arm platform, conveyer, computing machine and image capture software.Wherein high spectrum image unit is by CCD camera (Imperx, ICL-B1620), significant wave segment limit is 400 ~ 1000nm, and spectral resolution is 2.8nm, image light spectrometer (Specim, ImSpector V10E) and focal length variable lens are formed.Direct current tunable light source is made up of 150W tungsten halogen lamp and controller (Illumination Technologies Inc., 3900).Inspected object is placed on electric platforms, arranges suitable transfer rate by parametric controller, makes detection sample by spectrum imaging device in the mode of line sweep, record space and spectral information.

3. high spectrum image collection and correction

Hyperspectral imager image capture software Spectral Image (isuzuoptics, tw, China) adopted herein is provided by Taiwan five bell optics incorporated company.The collection of this software primary responsibility high spectrum image, storage, correction and initial analysis.It also has the functions such as imaging resolution setting, the setting of camera exposure time controling, electric platforms speed, black and white correction in addition.

For ensureing to collect clear, undistorted and facilitate post processing image, need to carry out relevant setting to parameters such as the speed of camera time shutter, light intensity focal length, image storage size and conveyer.Through experiment, finally determine that parameter is: lens are 20cm from chicken kind egg sample distance, and gather time shutter 55ms, the intensity of light source is 75W, picking rate 1.6mm/s, image resolution ratio 804 × 440 pixel.Due to the impact of dark current existence and extraneous factor in camera, cause image to there is certain noise, therefore to correct image, must affect to eliminate partial noise ^[12].Test is first by contrast kind of egg sample spectra value, and select the reference white correction plate (reflectivity is 99.99%) of suitable thickness, be placed on directly over light source, scanning transmission blank obtains entirely white uncalibrated image W; Then cover lens cap, gather entirely black uncalibrated image D; Finally calculate the relative image Rc after correction according to formula (1).

$\mathrm{Rc}=\frac{R_0-D}{W-D}---\left(1\right)$

(1) in formula, R ₀for original EO-1 hyperion transmission image; D is entirely black uncalibrated image; W is entirely white uncalibrated image; Rc is for demarcating rear EO-1 hyperion transmission image.

4. Mathematical Modeling Methods

For fully more different modeling methods is on the impact of estimated performance, partial least squares discriminant analysis method, support vector machine method and artificial neural network has been used to carry out modeling analysis to data

5. different brooding time kind egg curve of spectrum situation of change

Choose the representative of ROI (representative region ofinterest) region as sample.In order to better study kind of egg different parts to the impact of sex identification accuracy rate, choose ROI region as kind of an egg representative at the round end of kind of egg, centre, tip respectively, concrete region as shown in Figure 2.Plant egg embryo to germinate from kind of an egg round end, extend gradually, just can observe obvious air chamber at hatching of breeding eggs to when about the 8th day to tip, it is more suitable in 1/4 position of kind of egg diameter therefore to select during selection round end ROI region.Fig. 3 represents 400 ~ 1000nm averaged spectrum curve of the kind egg middle part ROI region in different brooding time, can find out to there is obvious difference between the curve of spectrum, and illustrate that the hatching of breeding eggs time different, the curve of spectrum is also different.Along with the prolongation of brooding time, the transmitted light spectrum of planting egg is more and more lower.

6. different sexes kind egg developmental morphology change

Hyper-spectral image technique mainly combines spectral detection and image detects respective advantage, can carry out visual analyzing to the internal feature of research object.And in kind of egg sex-screening, because male and female kind egg image difference is not obvious, as shown in Figure 4, therefore application image feature cannot carry out sex identification.Herein by analyzing kind of egg spectrum, sex identification is carried out to kind of egg.

7. the hatching of breeding eggs sex identification under the different model of all band curve of spectrum

When embryonic development was by the 8th day, testis and ovary can be distinguished by form.Right ovary in female is degenerated, and cause ovary both sides to be grown uneven, this phenomenon can be observed under low power magnifier; And male, both sides testis isogony.In order to determine whether spectral signature can predict the sex of kind of egg, adopt partial least squares discriminant analysis (Partial Least Squares Discriminant Analysis respectively, PLSDA), support vector machine (Support Vector Machine, and artificial neural network (Artificial Neural Network SVM), ANN) model is built, and compare, intend finding suitable model.PLSDA and SVM model construction differentiates at Matlab offset minimum binary respectively and completes in svmda tool box.ANN model construction completes in SPSS multilayer perceptron tool box.Wherein, SVM model parameter is: kernel function is radial basis function, and optimized parameter gamma value is 0.0316, and it is 100 that transmission and half transmitting pattern cost value are, and reflective-mode cost value is 3.1623.The neural network model parameter built is: input layer is 207, and the hiding number of plies is 1, and hidden layer nodes is 14, and hidden layer activation function is tanh; Output layer number is 2, namely male and female, output layer activation function Softmax.

The ROI averaged spectrum curve data extracted by all band original spectrum, as input variable, will plant egg sex as output variable, set up sex identification prediction SVM model, PLSDA model and ANN model.Modeling collection and the forecast set result of the SVM model of the different extracted parts gained, PLSDA model and ANN model all have difference.The SVM model set up of the ROI curve of spectrum data extracted with kind of egg round end position the 6th day and the 8th day accuracy rate the highest, the 6th day modeling collection and forecast set accuracy rate are respectively 82.26% and 65.63%.8th day modeling collection and forecast set accuracy rate are respectively 83.87% and 65.63%.The PLSDA model set up is the highest the 6th day accuracy rate, and modeling collection and forecast set accuracy rate are respectively 69.35% and 62.5%.The ANN model set up is the highest the 8th day accuracy rate, and modeling collection and forecast set accuracy rate are respectively 70.30% and 75.00%.Not quite, the detection model whole structure of the hatching of breeding eggs sex that ANN builds obviously is better than SVM model and PLSDA model for SVM model accuracy rate and PLSDA model accuracy rate difference.

The SVM model that the ROI curve of spectrum data extracted with kind of egg middle part are set up is the highest the 10th day accuracy rate, and modeling collection and forecast set accuracy rate are respectively 90.32% and 71.88%.The PLSDA model set up is the highest the 10th day accuracy rate, and modeling collection and forecast set accuracy rate are respectively 88.71% and 71.88%.The ANN model set up is the highest the 10th day accuracy rate equally, and modeling collection and forecast set accuracy rate are respectively 75.00% and 80.00%.Not quite, the detection model whole structure of the hatching of breeding eggs sex that ANN builds obviously is better than SVM model and PLSDA model for SVM model accuracy rate and PLSDA model accuracy rate difference.

The highest the 10th day accuracy rate to find the SVM model that the ROI curve of spectrum data extracted with kind of egg tip position are set up, modeling collection and forecast set accuracy rate are respectively 69.35% and 62.5%.The PLSDA model set up is the highest the 10th day accuracy rate equally, and modeling collection and forecast set accuracy rate are respectively 69.35% and 65.63%.The ANN model set up is the highest the 2nd day accuracy rate, and modeling collection and forecast set accuracy rate are respectively 79.70% and 73.30%.Not quite, the detection model successful of the hatching of breeding eggs sex that ANN builds is better than SVM model and PLSDA model for SVM model accuracy rate and PLSDA model accuracy rate difference.

Plant egg middle part sex identification accuracy rate be better than kind of egg round end position and plant egg tip position.By finding the comparative analysis of data, the highest to kind of the accuracy rate that egg sex is predicted with the ROI curve of spectrum data Modling model of kind of egg center section extraction in the 10th day, identification result is as shown in table 1.

The kind egg sex identification SVM of table 1 400 ~ 1000nm response spectrum, PLSDA, ANN Model checking result

Therefore the data of the 10th day center section are extracted and do more deep analysis.Extract the ROI curve of spectrum of the kind egg of all 10th day center section and be divided into male and female two class, averaging respectively, drawing the 10th day center section male and female kind egg averaged spectrum curve data, as shown in Figure 5.Can find out that the transmitted light spectrum of male kind of egg is lower than female kind of egg, exists certain rule.

Can find that wavelength exists interfere information between 400-600nm and between 900-1000nm by observing the curve of spectrum, therefore the curve of spectrum between 600-900nm is intercepted, re-establish sex identification prediction SVM model, PLSDA model and ANN model, to obtaining better sex identification accuracy rate.Identification result is as shown in table 2.

The kind egg sex identification SVM of table 2 600 ~ 900nm response spectrum, PLSDA, ANN Model checking result

After intercepting the curve of spectrum between 600-900nm, SVM model and PLSDA model prediction collection differentiate that accuracy rate all rises to 75.00% from 71.88%, and accuracy rate improves 3.12%.ANN model prediction collection differentiates that accuracy rate rises to 82.86% from 80.00%, and accuracy rate improves 2.86%.Prove that removing the raising of interfere information to model accuracy rate has certain effect.

Claims (2)

1., based on the method for high spectrum image to breeding-egg hatching Early-stage judgment, its device constitutive characteristic is,

1) system composition comprises high light spectrum image-forming unit, sample holder, electric platforms, line source, light box, computer and the image capture software composition containing video camera, spectrometer and focal length variable lens, whole device is placed in airtight black box, wherein, video camera is Imperx, ICL-B1620, wavelength band is 400 ~ 1000nm, and spectral resolution is 2.8nm; Spectrometer is ImSpectorV10E; Light box is the tungsten halogen lamp of 150W, completes transmission by 1 linear optical fiber conduit; Computer model is CPU E5800,3.2GHz, internal memory 2G, video card 256M GeForce GT240; Image capture software is the Spectral Image software of independent development;

2) signals collecting uses transmission mode, and lens are 20cm from chicken kind egg sample distance, and sample is close to line source and is placed, and the intensity of light source is 75W, gathers time shutter 55ms, picking rate 1.6mm/s, image resolution ratio 804 × 440 pixel.

It is 2. as claimed in claim 1 that based on the method for high spectrum image to breeding-egg hatching Early-stage judgment, its detecting step is,

1) by being in, temperature is 37.8 DEG C, relative humidity is that 65% condition descends the hatching chicken kind egg of 10 days to take out, and is positioned in high spectrum image detection system as claimed in claim 1, obtains high spectrum image;

2) utilize following formula to the correct image obtained, obtain the high spectrum image after correcting:

Wherein: Rc is the EO-1 hyperion transmission image after correcting, R ₀for original EO-1 hyperion transmission image, W be by reflectivity be 99.99% reference white correction plate, be placed on directly over light source, scanning transmission blank obtains entirely white uncalibrated image, and D is by lens cap on lens cap, gathers entirely black uncalibrated image;

3) area-of-interest of position, middle 100 × 100, egg region pixel size in image is selected, extract the spectrum average of all pixels in this region each wavelength in 600-900nm wavelength band, and as the input value of the neural network model built, Output rusults is the sex of chicken embryo, the neural network model parameter wherein built is: input layer is 207, the hiding number of plies is 1, and hidden layer nodes is 14, and hidden layer activation function is tanh; Output layer number is 2, namely male and female, output layer activation function Softmax.