CN101706445A - Beef marbling grade scoring method and device - Google Patents

Abstract

The invention discloses a beef marbling grade scoring method and a device, aiming at solving the problem that the manual scoring method has large subjectivity and arbitrariness. The method comprises the following steps: intaking a cross section image of cattle eye muscle; segmenting the eye muscle region on the cross section of the cattle eye muscle; extracting the marbling in the eye region and scoring on the grade of the marbling in the eye muscle, wherein, the grade scoring of the marbling in the eye region comprises that an index system of the grade scoring of the beef marbling and the grade scoring of the marbling in the eye region are established and four parameters of area density Sd and number N, distribution coefficient CVr and size uniformity CVs of the marbling are selected as scoring indexes. The invention also provides a beef marbling grade scoring device comprising a light source, an image acquisition card, a main control computer, a data acquisition card, a camera and a photoelectric sensor; and the beef marbling grade scoring method and the device are used for productive enterprises and inspection departments.

Description

Beef marbling grade scoring method and device

Technical field

The present invention relates to a kind of methods of marking and device that is used for beef manufacturing enterprise and examination and test of products department to beef quality grade robotization detection range based on the beef marbling grade of computer vision technique.

Background technology

The beef marbling score value is the important evidence of beef ranking.At present, most employing valuation officers manually mark in the production.The valuation officer marks to beef marbling grade according to beef marbling standard master drawing and valuation officer's oneself experience during scoring.Artificial methods of marking has bigger subjectivity and randomness, and the accuracy of appraisal result and authority are challenged.

Summary of the invention

Technical matters to be solved by this invention is to have overcome artificial methods of marking to have bigger subjectivity and random defective, and a kind of beef marbling grade scoring method and device of the robotization detection range based on computer vision technique is provided.

For solving the problems of the technologies described above, the present invention adopts following technical scheme to realize: beef marbling grade scoring method includes in eye muscle Region Segmentation on picked-up buphthalmos flesh cross-sectional view picture, the buphthalmos flesh square section, the eye muscle zone, buphthalmos flesh square section marbling grade scoring in the marbling extraction and eye muscle zone, cross-ox-eye flesh square section, and marbling grade scoring comprises the steps: in the eye muscle zone, described cross-ox-eye flesh square section

1. set up the beef marbling grade scoring index system

Choose marbling number N, marbling area density S _d, decorative pattern distribution coefficient CV _rWith decorative pattern size uniformity coefficient CV _sTotally 4 parameters are as the scoring index system, wherein:

1) marbling number N index computing formula is as follows:

N＝max(t) (1)

In the formula: N---the marbling number,

T---marblized index marker, t=1,2 ... ..., N.

2) marbling area density S _dThe index computing formula is as follows:

$S_d=\left(\underset{t=1}{\overset{N}{\mathrm{\Σ}}}{S}_t\right)/S_A,---\left(2\right)$

In the formula: S _d---the marbling area density,

S _A---the area in eye muscle zone, buphthalmos flesh square section, the .pixel of unit,

S _t---t marblized area, the .pixel of unit.

3) decorative pattern size uniformity coefficient CV _sThe index computing formula is as follows:

${\mathrm{CV}}_S=\sqrt{\frac{\underset{t=1}{\overset{N}{\mathrm{\Σ}}}(S_d\left(t\right)-\stackrel{\‾}{S})}{N-1}}/\stackrel{\‾}{S}---\left(3\right)$

In the formula: CV _s---marbling size uniformity coefficient,

S _d(t)---t marbling area density, S _d(t)=S _t/ S _A,

S---S _d(t) mean value.

4) decorative pattern distribution coefficient CV _rThe index computing formula is as follows:

${\mathrm{CV}}_r=\sqrt{\frac{\underset{i=1}{\overset{m}{\mathrm{\Σ}}}(d_i-\stackrel{\‾}{d})}{m-1}/\stackrel{\‾}{d}}---\left(4\right)$

In the formula: CV _r---the marbling distribution coefficient,

d _i---the interior effectively ratio of eye muscle area of marbling area and this subregion in i the subregion, I=1,2 ... ..., m,

R _i---i the subregion in eye muscle zone,

S _Ri---i effective eye muscle area that subregion is interior, the .pixel of unit, i=1,2 ... ..., m,

D---d _iMean value,

M---subregion number.

2. marbling grade scoring in the eye muscle zone, buphthalmos flesh square section

Neural network intelligent scoring model is adopted in the scoring of marbling grade in the eye muscle zone, buphthalmos flesh square section, model is selected 3 layers of BP neural network structure for use, hidden layer unit number is 15, and the training method of network selects for use speed of convergence to intend Newton's algorithm faster, and learning rate is 0.05; With marbling area density S _d, marbling number N, decorative pattern distribution coefficient CV _rAnd decorative pattern size uniformity coefficient CV _sFour characteristic parameters are as the input vector of network model, and marbling grade is evaluated in the eye muscle zone, cross-ox-eye flesh square section.

Eye muscle Region Segmentation on the buphthalmos flesh square section described in the technical scheme comprises the steps:

1. the buphthalmos flesh square section original image to picked-up carries out the HSV conversion, selects parameter S, V to form new parameter 2 * S-V, carries out automatic threshold according to this new parameter and cuts apart the most of background of removal.

2. select mathematical morphology operators compute gradient image.

3. after utilizing opening and closing operation to carry out denoising, intactly recover the shape of object with reconstruction algorithm.

4. adopt connected component labeling and range conversion to obtain the prospect mark and the context marker of image respectively.

5. based on prospect mark and context marker, on gradient image, carry out force minimum reconstruct, obtain new gradient image.

6. on the gradient image after the reconstruct, carry out watershed transform and obtain eye muscle region contour line.

7. obtain eye muscle zone, buphthalmos flesh square section by outline line being carried out Boundary Extraction and seed filling;

Marbling extracts and comprises the steps: in the eye muscle zone, buphthalmos flesh square section described in the technical scheme

1. with carrying out logic and operation with the eye muscle image that extracts after the original image binaryzation, obtain noisy marbling image.

2. noisy marbling image is carried out binary conversion treatment, and finish connected component labeling, on marking image, carry out filtering, obtain the marbling image after the denoising.

3. marbling image after the denoising is carried out analytical calculation, obtain the marblized characteristic parameter in eye muscle zone, buphthalmos flesh square section;

Picked-up buphthalmos flesh cross-sectional view described in the technical scheme looks like to comprise the steps:

1. tested buphthalmos muscle is placed on the detection platform that is installed on the ox slaughter line, spacer pin is located the position of tested buphthalmos muscle on detection platform, to keep constant with the distance of camera, puts down the soft curtain on the detection case left side.

2. after postponing 1 second, photoelectric sensor sends trigger pip and absorbs buphthalmos flesh cross-sectional view picture through the collecting image of computer system that data collecting card enters in the main control computer startup main control computer.

3. buphthalmos flesh square section view data enters main control computer through image pick-up card and offers computer intelligence image analysis software system and carry out intellectual analysis;

Beef marbling grade scoring device described in the technical scheme includes worktable, detection platform, soft curtain, detection case, light source, image pick-up card, main control computer, data collecting card, camera, spacer pin and photoelectric sensor.

Detection case is placed on the worktable, and detection platform is placed in the left side in the detection case, and light source is installed in upper right side and the front and back, right side in the detection case, and the detection case left end opens wide, and horizontal edge is fixed one and started the soft curtain that puts down easily on the detection case left end.Camera is installed in the through hole on the vertical tank wall of detection case right-hand member, and camera is vertical with buphthalmos flesh square section, detection platform) the right side fixedly mount a spacer pin, photoelectric sensor vertically is installed in detection platform and is placed the below of buphthalmos muscular work face.Camera is connected with image pick-up card by data line, image pick-up card is connected with main control computer by computer interface by data line, photoelectric sensor is connected with data collecting card through data line, and data collecting card enters main control computer through line by computer interface.

Compared with prior art the invention has the beneficial effects as follows:

1. beef marbling grade scoring method of the present invention is to adopt computer vision technique and the automatic Recognition Theory of intelligence that beef marbling grade is marked, not only beef marbling quantizes accurately, the evaluation index distribution characteristics is obvious, scoring rapidly, with low cost, can also overcome the various defectives of artificial scoring.

2. the beef marbling grade scoring apparatus structure that beef marbling grade scoring method of the present invention adopted is simple, processing ease, price are low, is easy to promote the use of.

Description of drawings

The present invention is further illustrated below in conjunction with accompanying drawing:

Fig. 1 is the structural principle block diagram of beef marbling grade scoring device of the present invention;

Fig. 2 is the FB(flow block) that adopts beef marbling grade scoring method of the present invention and device cross-ox-eye flesh square section eye muscle Region Segmentation;

Fig. 3 is the FB(flow block) that adopts beef marbling grade scoring method of the present invention and device cross-ox-eye flesh square section marbling to extract;

Fig. 4 adopts beef marbling grade scoring method of the present invention and device to draw the marbling grade scoring FB(flow block) according to marbling scoring index system by the characteristic parameter of buphthalmos flesh square section marbling figure;

Fig. 5-a is buphthalmos flesh square section (original) image that adopts the collecting image of computer system picked-up in beef marbling grade scoring method of the present invention and the device main control computer;

Fig. 5-b adopts beef marbling grade scoring method of the present invention and device to select 2 * S-V parameter to carry out the image that Threshold Segmentation produces in the hsv color space;

Fig. 5-c adopts beef marbling grade scoring method of the present invention and device to select mathematical morphology operators to calculate to obtain gradient image;

Fig. 5-d adopts beef marbling grade scoring method of the present invention and device through opening and closing operation and the image rebuild;

Fig. 5-e adopts beef marbling grade scoring method of the present invention and device to use prospect is carried out the prospect marking image that the method for connected component labeling obtains;

Fig. 5-f adopts beef marbling grade scoring method of the present invention and device with after the gradient image binaryzation, the context marker that obtains by range conversion;

Fig. 5-g adopts beef marbling grade scoring method of the present invention and device based on prospect and background gradient image to be carried out the force minimum reconstructed image;

Fig. 5-h adopts beef marbling grade scoring method of the present invention and device to carry out the image that watershed segmentation obtains eye muscle region contour line on the gradient image after the reconstruct;

Fig. 6-a is the position of eye muscle zone in original image shown in Fig. 5-a, buphthalmos flesh square section of adopting beef marbling grade scoring method of the present invention and device to extract.

Fig. 6-b is the position of marbling in original image shown in Fig. 5-a of adopting beef marbling grade scoring method of the present invention and device to extract by logical operation.

Fig. 7 estimates the evaluation map that adopts beef marbling grade scoring method of the present invention and device cross-ox-eye flesh square section eye muscle zone marbling segmentation effect.

Fig. 8 is the evaluation result and artificial evaluation result comparison diagram that adopts beef marbling grade scoring method of the present invention and device utilization neural network intelligent scoring method.

Among the figure: 1. worktable, 2. detection platform, 3. buphthalmos muscle, 4. soft curtain, 5. detection case, 6. light source, 7. image pick-up card, 8. main control computer, 9. data collecting card, 10. camera, 11. spacer pins, 12. photoelectric sensors.

Embodiment

Below in conjunction with accompanying drawing the present invention is explained in detail:

Consult Fig. 1, the beef marbling grade scoring device based on computer vision technique of the present invention includes worktable 1, detection platform 2, soft curtain 4, detection case 5, light source 6, image pick-up card 7, main control computer 8, data collecting card 9, camera 10, spacer pin 11 and photoelectric sensor 12.Wherein, light source uses efficient LED Flame Image Process special light source, and is adjustable according to illumination requirement.

Detection case 5 is placed on the table top of worktable 1, and detection platform 2 is placed on the table top of detection case 5 left side worktable 1.Light source 6 is installed in the upper right side and the front and back, right side of detection case 5, detection case 5 left ends be non-sealing be convenient to pick and place buphthalmos muscle 3, during work in order to prevent that light leak horizontal edge on detection case 5 left sides from fixing one and can start the soft curtain 4 that puts down easily.Camera 10 is installed in the through hole on the detection case 5 vertical tank walls, and guarantee camera 10 be with upright position, buphthalmos muscle 3 square section.The right side of detection platform 2 fixedly mounts a spacer pin 11, by the position of spacer pin 11 location buphthalmos muscle 3 xsects, to determine that the buphthalmos muscle 3 and the distance of camera 10 are constant.Photoelectric sensor 12 is installed in the below that detection platform 2 is placed buphthalmos muscle 3 workplaces, and vertical with workplace.Camera 10 links to each other with image pick-up card 7 by data line, and image pick-up card 7 is connected with main control computer 8 by computer interface by data line.Obtain the cross section image of buphthalmos flesh by high precision camera 10 and enter main control computer 8 through image pick-up card 7.Photoelectric sensor 12 links to each other with data collecting card 9 through data line, and data collecting card 9 enters main control computer 8 through line by the main control computer interface.Built-in data collecting card 9 also can directly be connected main control computer 8 by the computer PCI slot with built-in picture capture card 7.The analysis and the calculating of the buphthalmos flesh cross-sectional view picture that high precision camera 10 is obtained are to be finished by computer intelligence image analysis software system.

Consult Fig. 1 to Fig. 4, beef marbling grade scoring method of the present invention is main according to obtaining view data by image capturing system, carry out obtaining buphthalmos flesh square section eye muscle region contour according to new parameter in the HSV space then based on the watershed segmentation of connected component labeling, utilize Boundary Extraction and se ed filling algorithm to obtain eye muscle zone, accurate buphthalmos flesh square section again, use the image logical operation to extract marbling at last. take all factors into consideration marblized total amount and distribution situation in the eye muscle zone, image buphthalmos flesh square section, choose the beef marbling area density, the decorative pattern number, the decorative pattern distribution coefficient, four indexs of decorative pattern size uniformity coefficient are the scoring yardstick. the step of following detailed description beef marbling grade scoring method:

One. picked-up buphthalmos flesh cross-sectional view picture (consulting Fig. 1)

1. at first tested buphthalmos muscle 3 is placed on when working on the detection platform 2 in the beef marbling grade scoring device of the present invention that is installed on the ox slaughter line, the tested position (can not surmount spacer pin 11 can not break away from spacer pin 11) of buphthalmos muscle 3 on detection platform 2, spacer pin 11 location, distance with maintenance and camera 10 is a constant, puts down the soft curtain 4 on detection case 5 left sides.

2. after postponing 1 second, photoelectric sensor 12 sends trigger pip and absorbs buphthalmos flesh cross-sectional view picture through the collecting image of computer system that data collecting card 9 enters in the main control computer 8 startup main control computers 8.

3. buphthalmos flesh square section view data enters main control computer 8 through image pick-up card 7 and offers computer intelligence image analysis software system and carry out intellectual analysis.

Two. the eye muscle Region Segmentation (consulting Fig. 2) on the buphthalmos flesh square section

To the buphthalmos flesh square section original image of picked-up select to make new advances in the hsv color space parameter (2 * S-V), use threshold method to remove most of background on this basis.

2. selection mathematical morphology operators compute gradient image obtains the gradient image based on mathematical morphology operators.

3. utilize opening and closing operation that fine texture is rejected together with noise, adopt reconstruction algorithm to recover the shape of object exactly.

4. adopt connected component labeling and range conversion method to obtain the prospect and the context marker of image.

5. by the force minimum technology gradient image is carried out reconstruct based on prospect and context marker, obtain the gradient image after the reconstruct.

6. on the gradient image after the reconstruct, carry out watershed transform, obtain eye muscle region contour line.

7. by outline line being carried out Boundary Extraction and carries out seed filling to obtain the eye muscle zone.

Three. marbling extracts (consulting Fig. 3) in the eye muscle zone, buphthalmos flesh square section

1. with carrying out logic and operation with the eye muscle zone that extracts after the original image binaryzation, obtain noisy marbling image.

2. noisy marbling image is carried out binary conversion treatment, and finish connected component labeling, on marking image, carry out filtering, obtain the marbling image after the denoising.

3. marbling image after the denoising is carried out analytical calculation, obtain the marblized characteristic parameter in eye muscle zone, buphthalmos flesh square section.

Four. set up marbling scoring in the eye muscle zone, marbling grade scoring index system cross-ox-eye flesh square section

(1) sets up the beef marbling grade scoring index system

Choose marbling area density S _d, marbling number N, decorative pattern distribution coefficient CV _rWith decorative pattern size uniformity coefficient CV _sTotally 4 parameters are as the scoring index system.Marbling area density S _dThe ratio of expression marbling in eye muscle zone, buphthalmos flesh square section.Marbling area density S _dIt is fatty what one of the index of deposition in the reflection eye muscle zone, buphthalmos flesh square section; Marbling number N has reflected the feature of fat mass in the eye muscle zone, buphthalmos flesh square section to a certain extent, is the important indicator of marbling grade evaluation; Marbling distribution coefficient CV _rWhat reflect also is marblized distribution situation, is the important indicator of judging the marbling distributing homogeneity; Marbling size uniformity coefficient CV _sBe the parameter of expression marbling difference in size, characterize with the coefficient of dispersion of marbling area density.4 scoring index computing formula are as follows:

1. marbling number N

N＝max(t) (1)

In the formula: N---the marbling number,

T---marblized index marker, t=1,2 ... ..., N;

2. marbling area density S _d

$S_d=\left(\underset{t=1}{\overset{N}{\mathrm{\Σ}}}{S}_t\right)/S_A---\left(2\right)$

In the formula: S _d---the marbling area density,

S _A---the area in eye muscle zone, the .pixel of unit,

S _t---t marblized area, the .pixel of unit;

3. decorative pattern size uniformity coefficient CV _s

${\mathrm{CV}}_S=\sqrt{\frac{\underset{t=1}{\overset{N}{\mathrm{\Σ}}}(S_d\left(t\right)-\stackrel{\‾}{S})}{N-1}}/\stackrel{\‾}{S}---\left(3\right)$

In the formula: CV _s---marbling size uniformity coefficient,

S _d(t)---t marbling area density, S _d(t)=S _t/ S _A,

S---S _d(t) mean value;

4. decorative pattern distribution coefficient CV _r

${\mathrm{CV}}_r=\sqrt{\frac{\underset{i=1}{\overset{m}{\mathrm{\Σ}}}(d_i-\stackrel{\‾}{d})}{m-1}}/\stackrel{\‾}{d}---\left(4\right)$

In the formula: CV _r---the marbling distribution coefficient,

d _i---the interior effectively ratio of eye muscle area of marbling area and this subregion in i the subregion,

I=1,2 ... ..., m,

R _i---i the subregion in eye muscle zone,

S _Ri---i effective eye muscle area that subregion is interior, the .pixel of unit, i=1,2 ... ..., m

D---d _iMean value,

M---subregion number.

(2) beef marbling grade scoring

Consult Fig. 4, neural network intelligent scoring model is adopted in the scoring of beef marbling grade, and model is selected 3 layers of BP neural network structure for use, and hidden layer unit number is 15, the training method of network selects for use speed of convergence to intend Newton's algorithm faster, and learning rate is 0.05.With marbling area density S _d, marbling number N, decorative pattern distribution coefficient CV _rAnd decorative pattern size uniformity coefficient CV _sFour characteristic parameters are evaluated beef marbling grade as the input vector of network model.

Embodiment

One. picked-up buphthalmos flesh cross-sectional view picture

At first adopt beef marbling grade scoring device of the present invention picked-up buphthalmos flesh cross-sectional view picture, buphthalmos flesh square section view data enters main control computer 8 through image pick-up card 7 and offers computer intelligence image analysis software system and carry out intellectual analysis.

Two. buphthalmos flesh square section eye muscle Region Segmentation

1. the buphthalmos flesh cross-sectional view picture of picked-up is selected the parameter make new advances in the hsv color space, use threshold method to cut apart the eye muscle zone roughly, be beneficial to and produce the tangible image of gradient

Consult Fig. 5-a, in the beef marbling image, have eye muscle with accessory constituent around grow nonparasitically upon another plant " being connected bridge " between the flesh, the eye muscle and cut irregular square section, they are very unmanageable in image processing algorithm, because this a part of gray-scale value is between the gray-scale value of buphthalmos muscle and fat, when adopting the method for Threshold Segmentation, be easy to it is divided into the eye muscle part, so just will grow nonparasitically upon another plant flesh and eye muscle couples together, and can not cut apart correct eye muscle zone; When using Mathematical Morphology Method to cut apart, because the existence that this bridge-type connects, will have certain requirement to structural elements, even eye muscle successfully separates with the flesh of growing nonparasitically upon another plant the most at last, eye muscle zone after excessive erosion is expanded is being changed in shape, produce difference with the original area shape, influence segmentation effect.

HSV (tone, saturation degree, numerical value) color representation method is consistent to the perception of color with the people, and the independence of each component strengthens, and colouring information is lost few.The saturation degree of S representation in components color of image, and the depth of V representation in components color, because the computing of S and V component changes or the curvature on illuminated surface can keep relative stability when changing in illumination, can reduce the influence that factors such as individual difference, illumination condition and lighting angle cause.According to the color and the saturation degree characteristics thereof in eye muscle zone, the parameter that selection makes new advances uses threshold method to remove most of background, cut-out " connection bridge " and the more shallow accessory constituent of weakening eye muscle ambient color on this basis.Rgb value is to S, and the mapping process of V component is as follows.

V＝MAX(R，G，B) (2)

By S, the new parameter that the V component is formed is

I＝2×S-V (3)

Consult Fig. 5-b, on the new figure that parameter I constitutes, color image information has obtained distributing effectively, the contrast of target area and background increases, the more shallow zone of eye muscle ambient color, " connection bridge " zone and do not cut smooth square section and all correspondingly be removed or weaken, the eye muscle zone roughly " is isolated ", just can cut apart the eye muscle zone roughly by threshold method, both can eliminate most of background, be beneficial to the tangible image of generation gradient, also reduce calculated amount for follow-up watershed segmentation.

2. select mathematical morphology operators to calculate and obtain gradient image

Consult Fig. 5-c, gradient image can reflect the variation tendency of image better, facts have proved, the gradient of watershed segmentation algorithm and image has bigger getting in touch, and non-image own, we can say that the quality of watershed segmentation effect depends on gradient image to a great extent.Based on various edge detection operators to noise than the mathematical morphology operators sensitivity, select mathematical morphology operators compute gradient image.

(x, the computing formula of mathematical morphology gradient image y) is image f

$G\left[f(x,y)\right]=(f\⊕M)-\left(\mathrm{f\ΘM}\right)---\left(4\right)$

In the formula: M is a structural element,

Represent dilation operation, Θ represents erosion operation, and G[f (x, y)] be f (x, mathematical morphology gradient image y).

3. open and close and rebuild denoising

Consult Fig. 5-d, calculate the gradient image obtain, still have noise and unnecessary image detail, can produce the over-segmentation phenomenon so it is cut apart still by Mathematical Morphology Method.Though merely use the expansion of mathematical morphology and corrosion can remove some noises and junction, be difficult to return to original-shape.And the purpose that opens and closes re-establishing filter just is to eliminate in the gradient image because the local extremum that non-regular disturbance of gray scale and noise cause keeps important profile information.Compare and open and close part high gray scale and the low gray scale details of filtering in can only removal of images, open and close to rebuild again image is carried out having increased process of reconstruction on the basis of opening and closing operation, can recover that those do not have fully by the border of the composition of opening and closing operation institute filtering in the image.In opening and closing the filtering of reconstruction filter, fine texture is rejected by opening and closing operation together with noise, and contour of object is recovered in process of reconstruction, makes image keep the shape information of main object when simplifying, and recovers the shape of object exactly.

If original image f is a mask, g is a mark, wherein

g＝fΘM (5)

Make h ₁=f, the iteration of then opening reconstruct is defined as

R _O： $h_{k+1}=(h_k\⊕M)\∩f---\left(6\right)$

In the formula: R ₀-Kai reconstructed image is worked as h _K+1=h _kThe time iteration stopping

Make j ₁=R ₀, the iteration of then closing reconstruct is defined as

R _C： $j_{k+1}=(j_k\⊕M)\∩R_O---\left(7\right)$

In the formula: R _c-close reconstructed image, work as j _K+1=j _kThe time iteration stopping

4. adopt connected component labeling and watershed segmentation

Consult Fig. 5-e, the method of employing zone marker obtains the inside homogeneous region of corresponding each object of image, can realize better segmentation effect, not need to carry out again follow-up merging processing and just can access comparatively ideal results, suppress the over-segmentation phenomenon effectively.Employing marks the local maximum in destination object (eye muscle zone), to obtain better foreground target.

Consult Fig. 5-f, the gradient image that will open and close again after rebuilding is converted into bianry image, again by range conversion (Euclidean distance), promptly tries to achieve retaining basin carries out background to the distance of watershed divide mark.

Consult Fig. 5-g, after the intact prospect background of mark, gradient image be reconstructed by the force minimum technology,

Consult Fig. 5-h, on the gradient image after the reconstruct, carry out watershed transform, obtain the region contour line.

Consult Fig. 6-a, final eye muscle zone is to obtain by the region contour line being carried out Boundary Extraction and carry out seed filling, and the eye muscle zone that extracts is in the place of the position in the original image shown in figure medium green color.

5. eye muscle Region Segmentation evaluation of result

Segmentation precision to image object (eye muscle zone) depends on the quality of cutting apart, and we have selected final measuring accuracy and mistake branch rate.Final measuring accuracy is round the final goal of image segmentation---obtains the accurate measurement of object feature value in the image is proposed, by measurements and calculations, cut apart quality according to it reflects and the performance of partitioning algorithm was made an appraisal object feature value.The mistake branch number of pixels that produces owing to segmentation errors for image segmentation result is an important images measuring quality index, mistake branch rate just is selected as evaluation index, it has promptly considered the situation that target is correctly cut apart, and has considered the situation that background is correctly cut apart again.

Final measuring accuracy: establish R _fThe primitive character value that representative obtains from image as a reference, and S _fThe actual characteristic value that obtains the image of representative after cutting apart then has

${\mathrm{RUMA}}_f=\frac{|R_f-S_f|}{R_f}\×100\%---\left(8\right)$

RUMA _fBe used for the quality of cutting apart of quantitative evaluation image object.F represents target signature, gets 2 common feature, i.e. eye muscle region area (A) and girth (P).

Mistake divides the computing formula of rate as follows

$M_E=\frac{\left|B_O{\∩B}_T\right|+|I_O\∩I_T|}{I_O}---\left(9\right)$

In the formula: B _O, B _TThe number of pixels of the background in-original image, the split image

I _O, I _TThe number of pixels of the target in-original image, the split image

The mean value of the final measuring accuracy of the area in eye muscle zone (A) is 96.98%, and standard deviation is 1.76; The mean value of the final measuring accuracy of girth (P) is 92.67%, and standard deviation is 6.72; The mean value of mistake branchs rate is 9.44%, and standard deviation is 3.54, shows the extraction in processing eye muscle zone and dividing method accurately and reliably.

Three. marbling extracts in the eye muscle zone, buphthalmos flesh square section

1. consult Fig. 6-b, on the basis of buphthalmos flesh square section eye muscle Region Segmentation, extract eye muscle zone marbling by logical operation, the eye muscle that extracts zone marbling is in the place of the position in the original image shown in figure medium green color.

2. the marbling segmentation result is estimated

Because marbling is not a complete zone, but a plurality of small sizes zone that disperses, select the segmentation evaluation index to want the number of consideration of regional, cut apart among resulting target (marbling) number and the former figure difference of target number and reflected to a certain extent and cut apart quality.Select final measuring accuracy of area and target number consistance as the evaluation index of estimating segmentation result.

Consult Fig. 7, what represent among the figure is final measuring accuracy of area and the target number consistance that marbling is cut apart, and the mean value of the final measuring accuracy of area is 93.41%, and standard deviation is 4.44; The mean value of the accuracy rate of measuring of target number is 91.27%, and standard deviation is 5.55.

Four. set up marbling scoring in the eye muscle zone, marbling grade scoring index system cross-ox-eye flesh square section

(1) sets up the beef marbling grade scoring index system

Have five category feature parameters to can be used as candidate's index of estimating beef marbling grade, they are respectively distribution characteristics parameter, subregion property parameters, particle size parameters, parameters for shape characteristic and polygon step response parameter.Therefrom select 4 characteristic parameters as the evaluation index of estimating beef marbling grade, be respectively marbling area density S _d, decorative pattern distribution coefficient CV _r, decorative pattern size uniformity coefficient CV _sAnd marbling number N.4 scoring index computing formula are as follows:

1) marbling number N

N＝max(t) (10)

In the formula: N---the marbling number

T---marblized index marker, t=1,2 ... ..., N;

2) marbling area density S _d

$S_d=\left(\underset{t=1}{\overset{N}{\mathrm{\Σ}}}{S}_t\right)/S_A,---\left(11\right)$

In the formula: S _d---the marbling area density

S _A---the area of buphthalmos muscle region, the .pixel of unit

S _t---t marblized area, the .pixel of unit

3) decorative pattern size uniformity coefficient CV _s

${\mathrm{CV}}_S=\sqrt{\frac{\underset{t=1}{\overset{N}{\mathrm{\Σ}}}(S_d\left(t\right)-\stackrel{\‾}{S})}{N-1}}/\stackrel{\‾}{S}---\left(12\right)$

In the formula: CV _s---marbling size uniformity coefficient

S _d(t)---t marbling area density, S _d(t)=S _t/ S _A

S---S _d(t) mean value

4) decorative pattern distribution coefficient CV _r

${\mathrm{CV}}_r=\sqrt{\frac{\underset{i=1}{\overset{m}{\mathrm{\Σ}}}(d_i-\stackrel{\‾}{d})}{m-1}}/\stackrel{\‾}{d}---\left(13\right)$

In the formula: CV _r---the marbling distribution coefficient

d _i---the interior effectively ratio of eye muscle area of marbling area and this subregion in i the subregion,

I=1,2 ... ..., m,

R _i---i the subregion in eye muscle zone,

S _Ri---i effective eye muscle area that subregion is interior, the .pixel of unit, i=1,2 ... ..., m,

D---d _iMean value,

M---subregion number (this test m=4).

(2) marbling grade scoring in the beef zone

Consult Fig. 4, select 3 layers of BP neural network structure for use, hidden layer unit number is 15, and the training method of network selects for use speed of convergence to intend Newton's algorithm faster, and learning rate is 0.05.With marbling area density S _d, decorative pattern distribution coefficient CV _r, decorative pattern size uniformity coefficient CV _sAnd four characteristic parameters of marbling number N are as the input vector of network model.Be checking appraisal result accuracy, by 3 experienced professional grading persons with reference to the beef marbling standard drawing, marbling grade is evaluated, final artificial evaluation result is determined with majority rule, carries out the training of network as the output vector of network with artificial judgement marbling grade.

Utilize institute's 52 groups of data that obtain as the neural metwork training collection, select other 38 groups of data to verify as verification msg set pair model.Evaluation result as shown in Figure 8.

The result calculates according to checking, and the evaluation of computer system is 94.74% to the scoring accuracy rate of marbling grade, and evaluation result is satisfactory, shows that this intelligent scoring method based on computer vision technique can replace the method for the artificial evaluation of tradition.

Claims (4)

1. beef marbling grade scoring method, include in eye muscle Region Segmentation on picked-up buphthalmos flesh cross-sectional view picture, the buphthalmos flesh square section, the eye muscle zone, buphthalmos flesh square section marbling grade scoring in the marbling extraction and eye muscle zone, cross-ox-eye flesh square section, it is characterized in that marbling grade scoring comprises the steps: in the eye muscle zone, described cross-ox-eye flesh square section

1) sets up the beef marbling grade scoring index system

Choose marbling number N, marbling area density S _d, decorative pattern distribution coefficient CV _rWith decorative pattern size uniformity coefficient CV _sTotally 4 parameters are as the scoring index system, wherein:

(1) marbling number N index computing formula is as follows:

N＝max(t) (1)

In the formula: N---the marbling number,

T---marblized index marker, t=1,2 ..., N;

(2) marbling area density S _dThe index computing formula is as follows:

$S_d=\left(\underset{t=1}{\overset{N}{\mathrm{\Σ}}}{S}_t\right)/S_A,---\left(2\right)$

In the formula: S _d---the marbling area density,

S _A---the area in eye muscle zone, buphthalmos flesh square section, the .pixel of unit,

S _t---t marblized area, the .pixel of unit;

(3) decorative pattern size uniformity coefficient CV _sThe index computing formula is as follows:

${\mathrm{CV}}_S=\sqrt{\frac{\underset{t=1}{\overset{N}{\mathrm{\Σ}}}(S_d\left(t\right)-\stackrel{\‾}{S})}{N-1}}/\stackrel{\‾}{S}---\left(3\right)$

In the formula: CV _S---marbling size uniformity coefficient,

S _d(t)---t marbling area density, S _d(t)=S _t/ S _A,

S---S _d(t) mean value;

(4) decorative pattern distribution coefficient CV _rThe index computing formula is as follows:

${\mathrm{CV}}_r=\sqrt{\frac{\underset{i=1}{\overset{m}{\mathrm{\Σ}}}(d_i-\stackrel{\‾}{d})}{m-1}}/\stackrel{\‾}{d}---\left(4\right)$

In the formula: CV _r---the marbling distribution coefficient,

d _i---the interior effectively ratio of eye muscle area of marbling area and this subregion in i the subregion,

I=1,2 ..., m,

R _i---i the subregion in eye muscle zone,

S _Ri---i effective eye muscle area that subregion is interior, the .pixel of unit, i=1,2 ..., m,

D---d _iMean value,

M---subregion number;

2) marbling grade scoring in the eye muscle zone, buphthalmos flesh square section

Neural network intelligent scoring model is adopted in the scoring of marbling grade in the eye muscle zone, buphthalmos flesh square section, model is selected 3 layers of BP neural network structure for use, hidden layer unit number is 15, and the training method of network selects for use speed of convergence to intend Newton's algorithm faster, and learning rate is 0.05; With marbling area density S _d, marbling number N, decorative pattern distribution coefficient CV _rAnd decorative pattern size uniformity coefficient CV _sFour characteristic parameters are as the input vector of network model, and marbling grade is evaluated in the eye muscle zone, cross-ox-eye flesh square section.

2. according to the described beef marbling grade scoring method of claim 1, it is characterized in that the eye muscle Region Segmentation on the described buphthalmos flesh square section comprises the steps:

1) the buphthalmos flesh square section original image to picked-up carries out the HSV conversion, selects parameter S, V to form new parameter 2 * S-V, carries out automatic threshold according to this new parameter and cuts apart the most of background of removal;

2) select mathematical morphology operators compute gradient image;

3) utilize opening and closing operation to carry out denoising after, intactly recover the shape of object with reconstruction algorithm;

4) adopt connected component labeling and range conversion to obtain the prospect mark and the context marker of image respectively;

5) based on prospect mark and context marker, on gradient image, carry out force minimum reconstruct, obtain new gradient image;

6) on the gradient image after the reconstruct, carry out watershed transform and obtain eye muscle region contour line;

7) obtain eye muscle zone, buphthalmos flesh square section by outline line being carried out Boundary Extraction and seed filling.

3. according to the described beef marbling grade scoring method of claim 1, it is characterized in that marbling extracts and comprises the steps: in the eye muscle zone, described buphthalmos flesh square section

1) with carrying out logic and operation with the eye muscle image that extracts after the original image binaryzation, obtains noisy marbling image;

2) noisy marbling image is carried out binary conversion treatment, and finish connected component labeling, on marking image, carry out filtering, obtain the marbling image after the denoising;

3) marbling image after the denoising is carried out analytical calculation, obtain the marblized characteristic parameter in eye muscle zone, buphthalmos flesh square section.

4. according to the described beef marbling grade scoring method of claim 1, it is characterized in that described picked-up buphthalmos flesh cross-sectional view looks like to comprise the steps:

1) tested buphthalmos muscle (3) is placed on the detection platform (2) that is installed on the ox slaughter line, the position of tested buphthalmos muscle (3) on detection platform (2), spacer pin (11) location, to keep constant, put down the soft curtain (4) on detection case (5) left side with the distance of camera (10);

2) postpone after 1 second, photoelectric sensor (12) sends trigger pip and absorbs buphthalmos flesh cross-sectional view picture through the collecting image of computer system that data collecting card (9) enters in main control computer (8) the startup main control computer (8);

3) buphthalmos flesh square section view data enters main control computer (8) through image pick-up card (7) and offers computer intelligence image analysis software system and carry out intellectual analysis.

5 one kinds of devices of implementing the described beef marbling grade scoring method of claim 1, it is characterized in that described beef marbling grade scoring device includes worktable (1), detection platform (2), soft curtain (4), detection case (5), light source (6), image pick-up card (7), main control computer (8), data collecting card (9), camera (10), spacer pin (11) and photoelectric sensor (12);

Detection case (5) is placed on the worktable (1), detection platform (2) is placed in the left side in the detection case (5), light source (6) is installed in upper right side and the front and back, right side in the detection case (5), detection case (5) left end opens wide, and fixes one at the last horizontal edge of detection case (5) left end and starts the soft curtain (4) that puts down easily; Camera (10) is installed in the through hole on the vertical tank wall of detection case (5) right-hand member, camera (10) is vertical with buphthalmos flesh square section, the right side of detection platform (2) fixedly mounts a spacer pin (11), and photoelectric sensor (12) vertically is installed in detection platform (2) and places the below of buphthalmos muscle (3) workplace; Camera (10) is connected with image pick-up card (7) by data line, image pick-up card (7) is connected with main control computer (8) by computer interface by data line, photoelectric sensor (12) is connected with data collecting card (9) through data line, and data collecting card (9) enters main control computer (8) through line by computer interface.

Images