CN103394472A - Method for detecting and grading greening potatoes based on machine vision - Google Patents

Abstract

The invention discloses a method for detecting and grading greening potatoes based on machine vision. The method comprises the following steps of: (1) collecting a complete potato image and filtering a background to obtain a potato split image; (2) calculating a figure center of a potato and removing a potato outline in proportion by taking the figure center as a standard; (3) scanning a target region of the potato and calculating an H value point by point; (4) comparing the calculated H value with a threshold value alpha, wherein if the H value is equal to the threshold value alpha, a pixel point is taken as a greening point, but if not, the pixel point is taken as a normal skin point; and (5) setting the threshold value and counting the quantity of greening points, and comparing the greening points with a threshold value beta for grading the potatoes. The method disclosed by the invention is low in computation complexity and easy to realize, is applicable to rapid and real-time grading of the potatoes, is strong in objectivity and high in efficiency, and has very good application prospect.

Description

A kind of potato of greening based on machine vision detects stage division

Technical field

The present invention relates to a kind of method that the greening potato is detected classification, be specifically related to a kind of based on the H(tone) passage gray value diagnostic method detects the method for classification to the potato of epidermis greening.

Background technology

Potato is a kind of nutritious, popular agricultural product that the grain dish has concurrently, has advantages of that output is high, purposes is wide and economic worth is high.Potato comes into the market as a kind of commodity with very big value, and can quality be its prerequisite that seize competitive advantage in market competition.

At present, processing the postpartum of domestic potato is mainly the detection of shape, size, maturity, blemish etc. to potato, and detects mainly by manually carrying out.The major defect of manual grading skill is: the amount of labour is large, and productivity ratio is low; Effectiveness of classification is unstable: grade scale acceptor viewing rings larger; The workman directly contacts with fruit, and the safe and sanitary and the classification that affect food are difficult to realize quick, accurate, harmless and intelligent.Utilize machine vision to carry out the potato detection, the interference of artificial subjective factor can be got rid of,, by the quantitative measurment of the characteristic indexs such as effects on surface defect, shape, comprehensive detection and the classification of a plurality of indexs of potato can be disposablely completed, effectively save time and the labour, improve sorting efficiency.

The External Defect of potato comprises that mainly diauxic growth, deformity, epidermis turn green, shriveling, mechanical damage, small holes caused by worms, damaged by rats, scab, dry rot or rot etc.Wherein, it is that the potato greening is very general that the epidermis of potato turns green, larger on detection and the classification impact of potato.Utilize machine vision to detect the greening of potato, to perfecting potato and afforesting potato and carry out classification, be mainly to carry out Digital Image Processing by the imagery exploitation computer to potato, extract the characteristic value of greening potato, and then whether epidermis turns green to differentiate potato.The extraction of greening potato image processing techniques and characteristic value is the core technology of most critical in whole visual rating scale, is directly connected to the feasibility that detects classification.

The color of the greening defect area of potato is not pure green, just with respect to brown potato body, looks like green.For example, the green pixel values that high pixel is obviously arranged in the image of the present machine vision of defect body aspect greening, yet, the spot that the potato surface-brightening turns white, higher green pixel values and high redness and blue value are also arranged, so the greening defect is difficult to detect.Several methods of inspection are arranged, and as Multivariate Discrimination method, neutral net and stochastic model method, this several method has been used for detection and the classification of green defect to be attempted, but the computational complexity of these algorithms is higher, is not suitable for online real-time detection classification.

Summary of the invention

The object of the invention is to the deficiency that detects stage division for existing, proposed a kind of gray value diagnostic method based on HSV color space H passage and detected the greening defect of potato.

For achieving the above object, the present invention adopts following technical scheme:

A kind of potato of greening based on machine vision detection method, comprise the steps:

(1) gather complete potato image, wiping out background, obtain potato and cut apart image;

(2) calculate the potato centre of form, remove in proportion the potato profile take the centre of form as benchmark,

At first calculate the potato centre of form:

$X=\frac{1}{N}\underset{i=1}{\overset{n}{\mathrm{\Σ}}}{x}_i,$ $Y=\frac{1}{N}\underset{i=1}{\overset{n}{\mathrm{\Σ}}}{y}_i$

In formula: N---the sum of potato pixel;

x _i---abscissa (i=1,2,3 of potato pixel ... n);

y _i---ordinate (i=1,2,3 of potato pixel ... n)

The coordinate of next convergent-divergent potato:

X _i＝K(x _i-X)+X?Y _i＝K(y _i-X)+Y

In formula: X _i---coordinate (i=1,2,3 after convergent-divergent ... n);

Y _i---coordinate (i=1,2,3 after convergent-divergent ... n);

K---zoom factor, K gets 0.95;

X---centre of form abscissa;

Y---centre of form ordinate;

x _i---abscissa (i=1,2,3 of potato pixel ... n);

y _i---abscissa (i=1,2,3 of potato pixel ... n);

Then with (X _i, Y _i), as the new coordinate of potato, obtain removing the potato after profile;

(3) point by point scanning potato target area, measure R, G and the three-channel pixel value of B of each pixel of potato target area, and it be transformed between 0-1, obtains r, g and b; Wherein r=R/255, G=g/255, B=b/255;

(4) r, g and b are converted into tone H value;

(5) pixel H value and the threshold value that will calculate, threshold alpha is compared, if the H value equals threshold alpha 127, this pixel is the greening point, if the H value is not equal to 127, pixel is the normal epidermis point;

(6) setting threshold and statistics greening are counted, and will afforest a little and compare potato is carried out classification with threshold value beta.

In order further to guarantee accurately wiping out background of step 1, complete reservation potato feature, the inventor has also made restriction to the employing of complete potato image, specifically be captured under the image acquisition region of potato automatic grading system and complete, described background is black, grey, blueness, black, grey or partially blue etc. partially partially.Above-mentioned background is conducive to the filtering of background, has guaranteed stability and the accuracy of detection method of the present invention.

Stage division of the present invention, in step 1, each potato is moving through not stall in the process of image acquisition region, and by the image that continuous acquisition at least 3 width differences show, coverage rate surpasses 90%.

In addition, in step 1, following technical scheme is adopted in the filtering of background: first measure and obtain potato complete image R, G and the three-channel pixel value of B, the global threshold domain division method wiping out background of recycling G, B passage gray value differences.Described global threshold domain division method is: setting threshold r is fixed value 25, if the G of pixel, B passage gray value difference are less than threshold value, think background dot, the R of this pixel, G and B value are made as 0, otherwise think potato, do not change R, G and the B channel value of this pixel and carry out statistical analysis.Detection stage division of the present invention, adopt the standard of HSV color space H passage as stage division of the present invention, the inventor finds in long-term a large amount of experimental study, H value can intuitively be reacted potato greening situation accurately, for the accurate classification of potato detection provides objective guarantee.The method is simply effective, is fit to online real-time graded and detects, and the False Rate that simultaneously normal potato is mistaken for the greening potato is extremely low, guarantees the potato effectiveness of classification.

In step 4, H is the value of HSV color space H passage, and its computing formula is:

1. at first calculate h, the h computing formula is as follows:

Wherein

$\mathrm{\θ}=\arccos \left\{\frac{\frac{1}{2}\[(r-b)+(r-b)\]}{{\[{(r-g)}^2+(r-b)(g-b)\]}^{1/2}}\right\}$

The h of this moment is between [0,360 °].

2. H=(h/360) * 255, the H of this moment is between [0,255].

Calculate H this moment between [0,360 °]; (use this formula H value between [0,1], the H value that calculates also multiply by 255, so the H value in (5) step is between [0,255])

Step 5:

With pixel H value and the threshold value calculated, threshold alpha is compared, if the H value equals threshold alpha, is 127, and this pixel is the greening point, if the H value is not equal to 127, pixel is the normal epidermis point;

Wherein, threshold alpha 127 is namely that the inventor verify and obtains based on the lot of experiments basis, can accurately to afforesting to put, judge.

Step 6: setting threshold β, will afforest a little and compare with threshold value beta, potato is carried out classification;

Wherein, threshold value beta be 9-12, be preferably 10.

Particularly, described step 6 is: will afforest a little and compare with threshold value beta, and when the greening point judges that greater than threshold value beta this potato is the greening potato, otherwise be judged as normal potato.

Or potato is carried out finer classification, described step 6 is:, when the greening point judges that greater than threshold value beta this potato is the greening potato, when the greening point, between the 1/2-1 of threshold value beta, be judged as not obvious greening potato; When less than 1/2 of threshold value beta, being judged as normal potato.

In the present invention, each potato does not roll through not stopping in the process of image acquisition region, by continuous acquisition,, to three width different surfaces images, is covered more than 90% of whole potato surface, can more intactly reflect the potato surface information.In the three width images that the hierarchical detection system photographs each potato in the present invention, as long as there is a sub-picture to judgment result is that in image, greening is arranged, think that namely this potato is the greening potato.Select 40 potato samples to afforest potato and detect test, wherein normal potato is 24,16 of greening potatos.If the greening point number that every single potato detects is Num_Green, the greening detection threshold is that Th_Green(is as 10), afforest the potato detection model and be:

These 40 potato samples are carried out 6 greenings detect test, its testing result is as shown in table 1.

The greening of table 1 potato detects test data

Unit: individual

The potato number m of the greening that does not detect in test ₁Be 11, the normal potato number m of erroneous judgement ₂Be 11, the total number n of potato ₁Be 240, the horse bell is afforested Detection accuracy p ₁:

$p_1=1-\frac{m_1+m_2}{n_1}\×100\%=1-\frac{11+11}{240}\×100\%=90.8\%$

Potato is afforested the False Rate p2 that detects:

p ₂＝1-p ₁＝1-90.8%＝9.2%

When the present invention was applied to the real-time graded of potato, in classification process, by to the drawing of sorting potato number and the statistics of sorting time, the efficiency of separation can reach 12/second.

The present invention has the following advantages:

(1) potato greening defects detection algorithm computation complexity is low, is easy to realize the quick real-time grading of applicable potato;

(2) computational methods credibility of the present invention and the degree of reliability are high;

(3) adopt the H passage to process and have directly perceived, easy characteristics, and reduced labour intensity;

(4) the present invention detects that the method objectivity of classification is strong, efficiency is high, and noncontact has good application prospect without injury.

(5) remove in proportion the method for potato profile take the centre of form as benchmark simple consuming time few, the assorted point background can also be removed when removing profile in.

Description of drawings

Fig. 1 is that the present invention detects the stage division schematic flow sheet;

Fig. 2 is that potato is cut apart image;

Fig. 3 is the potato figure after the removal profile.

The specific embodiment

Below in conjunction with drawings and Examples, the present invention is described further.

Embodiment 1

A kind of potato of greening based on machine vision detection method as shown in Figure 1, comprise the steps:

(1) do not stop to roll in image acquisition region, continuous acquisition three width different surfaces images, coverage rate is potato surface 95%; Mensuration obtains potato complete image R, G and the three-channel pixel value of B, and the global threshold domain division method wiping out background of recycling G, B passage gray value differences, obtain potato and cut apart the image (see figure 2);

(2) calculate the potato centre of form, remove in proportion the potato profile take the centre of form as benchmark,

At first calculate the potato centre of form, $X=\frac{1}{N}\underset{i=1}{\overset{n}{\mathrm{\Σ}}}{x}_i,$ $Y=\frac{1}{N}\underset{i=1}{\overset{n}{\mathrm{\Σ}}}{y}_i$

In formula: N---the sum of potato pixel;

x _i---abscissa (i=1,2,3 of potato pixel ... n);

y _i---ordinate (i=1,2,3 of potato pixel ... n)

The coordinate of next convergent-divergent potato,

X _i＝K(x _i-X)+X?Y _i＝K(y _i-X)+Y

In formula: X _i---coordinate (i=1,2,3 after convergent-divergent ... n);

Y _i---coordinate (i=1,2,3 after convergent-divergent ... n);

K---zoom factor, K gets 0.95;

X---centre of form abscissa;

Y---centre of form ordinate;

x _i---abscissa (i=1,2,3 of potato pixel ... n);

y _i---abscissa (i=1,2,3 of potato pixel ... n);

Then with (X _i, Y _i), as the new coordinate of potato, obtain removing the potato after profile;

(3) point by point scanning potato target area, measure R, G and the three-channel pixel value of B of each pixel of potato target area, and it be transformed between 0-1, obtains r, g and b; Wherein r=R/255, G=g/255, B=b/255; Calculating r is that [0,1], b are [0,1] for [0,1], g;

(4) r, g and b are converted into tone H value;

The concrete steps of calculating H are:

1. at first calculate h.The h computing formula is as follows:

Wherein

$\mathrm{\θ}=\arccos \left\{\frac{\frac{1}{2}\[(r-b)+(r-b)\]}{{\[{(r-g)}^2+(r-b)(g-b)\]}^{1/2}}\right\}$

The h of this moment is between [0,360 °].

2. H=(h/360) * 255, the H of this moment is between [0,255].

Calculate H this moment between [0,360 °]; (use this formula H value between [0,1], the H value that calculates also multiply by 255, so the H value in (5) step is between [0,255])

(5) pixel H value and the threshold value that will calculate, threshold alpha is compared, if the H value equals threshold alpha 127, this pixel is the greening point, if the H value is not equal to 127, pixel is the normal epidermis point; Can access final defect extraction effect figure this moment, sees Fig. 3.(namely carry out image segmentation, remove profile, calculate the H value H value is equaled 127 design sketch after thinking defect point, wherein the H value equals 127 point and thinks to afforest a little, for convenient effect of observing the extraction of greening defect, think that the point of afforesting defect is set to 127,127 is that the part of grey in a gray value figure three is exactly to afforest defect, and remaining part is set to 0,0th, black.So just can afforest as seen from the figure the extraction effect of defect)

(6) setting threshold β, will afforest a little and compare with threshold value beta, and potato is carried out classification., when the greening point is the greening potato greater than 10 these potatos of judgement of threshold value beta, when the greening point, between 5-10, be judged as not obvious greening potato; O'clock less than 5, be judged as normal potato when greening.

Embodiment 2

Compare with embodiment 1, distinctive points only is step 6, and the present embodiment step 6 is specially:

Setting threshold β is 10, will afforest a little and compare with threshold value beta, and potato is carried out classification.Be that 10 these potatos of judgement are the greening potato when the greening point greater than threshold value beta; O'clock less than 10, be judged as normal potato when greening.

Embodiment 3

Compare with embodiment 1, distinctive points is only 9 or 12 in the present embodiment threshold value beta.

Embodiment in above-described embodiment can further make up or replace; and embodiment is described the preferred embodiments of the present invention; not the spirit and scope of the present invention are limited; under the prerequisite that does not break away from design philosophy of the present invention; the various changes and modifications that in this area, the professional and technical personnel makes technical scheme of the present invention, all belong to protection scope of the present invention.

Claims (9)

1. the potato of the greening based on a machine vision detection method, comprise the steps:

(1) gather complete potato image, wiping out background, obtain potato and cut apart image;

(2) calculate the potato centre of form, remove in proportion the potato profile take the centre of form as benchmark,

At first calculate the potato centre of form:

$X=\frac{1}{N}\underset{i=1}{\overset{n}{\mathrm{\Σ}}}{x}_i,$ $Y=\frac{1}{N}\underset{i=1}{\overset{n}{\mathrm{\Σ}}}{y}_i$

In formula: N---the sum of potato pixel;

x _i---abscissa (i=1,2,3 of potato pixel ... n);

y _i---ordinate (i=1,2,3 of potato pixel ... n)

The coordinate of next convergent-divergent potato:

X _i＝K(x _i-X)+X?Y _i＝K(y _i-X)+Y

In formula: X _i---coordinate (i=1,2,3 after convergent-divergent ... n);

Y _i---coordinate (i=1,2,3 after convergent-divergent ... n);

K---zoom factor, K gets 0.95;

X---centre of form abscissa;

Y---centre of form ordinate;

x _i---abscissa (i=1,2,3 of potato pixel ... n);

y _i---abscissa (i=1,2,3 of potato pixel ... n);

Then with (X _i, Y _i), as the new coordinate of potato, obtain removing the potato after profile;

(3) point by point scanning potato target area, measure R, G and the three-channel pixel value of B of each pixel of potato target area, and it be transformed between 0-1, obtains r, g and b; Wherein r=R/255, G=g/255, B=b/255;

(4) r, g and b are converted into tone H value;

(5) pixel H value and the threshold value that will calculate, threshold alpha is compared, if the H value equals threshold alpha, is 127, and this pixel is the greening point, if the H value is not equal to 127, pixel is the normal epidermis point;

(6) setting threshold and statistics greening are counted, and will afforest a little and compare potato is carried out classification with threshold value beta.

2. detection method according to claim 1, it is characterized in that: complete under the image acquisition region that is captured in the potato automatic grading system of complete potato image in described step 1, described background is black, grey, blueness, inclined to one side black, inclined to one side grey or partially blue.

3. detection method according to claim 1 and 2, is characterized in that: in described step 1, first measure and obtain potato complete image R, G and the three-channel pixel value of B, the global threshold domain division method wiping out background of recycling G, B passage gray value differences.

4. according to claim 1-3 described detection methods of any one, it is characterized in that: described global threshold domain division method is: setting threshold r is fixed value 25, if the G of pixel, B passage gray value difference are less than threshold value, think background dot, the R of this pixel, G and B value are made as 0, otherwise think potato, do not change R, G and the B channel value of this pixel and carry out statistical analysis.

5. according to claim 1-4 described detection methods of any one, it is characterized in that: in described step 4, the computing formula of H is:

1. at first calculate h, the h computing formula is as follows:

Wherein

$\mathrm{\θ}=\arccos \left\{\frac{\frac{1}{2}\[(r-b)+(r-b)\]}{{\[{(r-g)}^2+(r-b)(g-b)\]}^{1/2}}\right\}$

②H＝(h/360)×255。

6. according to claim 1-5 described detection methods of any one, it is characterized in that: in described step 6, threshold value beta is 10.

7. according to claim 1-6 described detection methods of any one, it is characterized in that: described step 6 is: when the greening point judges that greater than threshold value beta this potato is the greening potato, otherwise be judged as normal potato.

8. according to claim 1-6 described detection methods of any one, it is characterized in that: described step 6 is: when the greening point judges that greater than threshold value beta this potato is the greening potato, between the 1/2-1 of threshold value beta, be judged as not obvious greening potato when the greening point; When less than 1/2 of threshold value beta, being judged as normal potato.

9. according to claim 1-6 described detection methods of any one, it is characterized in that: in described step 1, each potato is moving through not stall in process of image acquisition region, the image that is shown by continuous acquisition at least 3 width differences, coverage rate surpasses 90%, obtains the potato complete image.

Images