CN103148811A - High-speed airflow dispersion-based stem content and threshing parameter detection and removing method - Google Patents

Abstract

The invention discloses a high-speed airflow dispersion-based stem content and threshing parameter detection and removing method, which includes the following steps: a belt conveyer is utilized to spread and convey tobacco leaves into a high-speed air duct; high-speed airflow is utilized to further uniformly disperse the tobacco leaves in the high-speed air duct, and at the same time, the tobacco leaves are sent into a detection area along with the airflow; a high-definition tobacco leave perspective is acquired in the detection area; the textural features of the tobacco leaves are extracted from the perspective, an algorithm is utilized to identify the laminae and the stems in the tobacco leaves, and the tobacco leaves containing the stems are removed; the shapes and sizes of the tobacco leaves are analyzed, so that tobacco leaf areas are obtained, and threshing parameters are calculated. By analyzing the acquired tobacco leave perspective image, the textural features are extracted, a binary tree is adopted to accurately distinguish the stems from the laminae, consequently, the invention solves the current complex problems of redrying factories, such as off-line stem content detection and long detection period, and thereby the product quality and efficiency of redrying factories can be remarkably increased.

Description

Reach based on the discrete Ye Zhonghan stalk of high velocity air and beat leaf parameter detecting elimination method

Technical field

The invention belongs to contain in the tobacco leaf leaf and obstruct the recognition detection technical field, be specifically related to a kind of reaching based on the discrete Ye Zhonghan stalk of high velocity air and beat leaf parameter detecting elimination method.

Background technology

Ye Zhonghan stalk rate is very important index in beating and double roasting process, (the offal diameter＞1.5mm) rate is higher for the Ye Zhonghan stalk, cause leaf quality to reduce, and because offal still contains higher moisture after redrying finishes, cause the blade of smoke box inside in blade storage and alcoholization process to go mouldy or fester, and Redrying Factory is difficult to accomplish that the leaf stalk separates fully when beating and double roasting, therefore secondary detection is rejected and is just seemed particularly important, diameter in blade need to be rejected out greater than the offal more than 1.5mm, to improve the blade product quality.For a long time, it is very backward that means are rejected in the detection of offal, mainly relies on manually-operated, and not only inefficiency, and quality and stability are difficult to guarantee.

Application number is that 201110213062.7 Chinese invention patent discloses a kind of Ye Zhonghan stalk based on the combined light perspective and contained stalk rate visual identity detection method, it points out to utilize the gray scale morphology method that image is operated with the window that is of a size of N * N, obtain offal by continuous corrosion and expansion, fairly simple being easy to of this method realized, yet practical problems is more: 1, parameter problem, here the window of morphological operation N * more difficult selection of N size, because tobacco leaf is not of uniform size, the selection of window is comparatively complicated, need to have self-adaptation to adjust the scheme of window; 2, interference is more after morphological operation, needs further cut apart with threshold value, and this moment, threshold value was chosen comparatively complexity, only was difficult to be effective with histogram; 3, fail to explicitly point out how to calculate the tobacco leaf area, its number by tobacco leaf pixel in statistical picture is estimated the area of tobacco leaf according to a preliminary estimate, this means and to produce following situation: suppose that tobacco leaf only covers three pixels (as shown in Figure 1), distance between pixel is assumed to be 1, this moment, its area that draws tobacco leaf was 3, and reality as can be known this tobacco leaf area be 0.5, be less object due to what process here, error is more obvious.Because it can't obtain accurate tobacco leaf area, thereby the various complexity that obtain are beaten the leaf parameter, as Percentage of large and middle size strips, all there is larger error in small pieces rates etc., what adopt due to it when calculating contains the stalk rate is Morphology Algorithm, be difficult to obtain real offal length, tobacco stalk quality calculates may exist larger error; In addition, no matter which kind of light source shining this patent with appoints and so keeps away unavoidable the problems referred to above.

Application number is that 200720018753.0 and 200910233734.3 Chinese utility model patent and Chinese invention patent disclose respectively reason stalk device and stalk knot eliminating device before tobacco stems impurity removing machine, offal chopping, yet its technological means is still more elementary.Application number is that 200920036277.4 Chinese utility model patent discloses a kind of cigarette paper and detects device for eliminating, this device comprises the conveying device that is supported on frame, video camera is settled in the top of conveying device, homonymy is provided with lighting source, the exit is mounted with the blow gun that is controlled by solenoid valve, the signal output part map interlinking of video camera is as the signals collecting end of identifying processing circuit, the control output end of image recognition processing circuit connects the controlled end of solenoid valve, during use, this cigarette paper is detected device for eliminating be placed in the end that residual cigarette is produced.Although this scheme has adopted comparatively advanced detection means, is subjected to its structural limitations, and is not suitable for the rejecting of offal.Application number is that 201220269009.9 Chinese utility model patent discloses a kind of intelligent offal device for eliminating, it is also only to rely on initial threshold that tobacco leaf is identified, operate comparatively simple, can't distinguish the tobacco leaf under complex situations, and it does not provide the method that tobacco leaf is evenly spread out.

Summary of the invention

Goal of the invention: the object of the invention is to for the deficiencies in the prior art, provide a kind of reaching based on the discrete Ye Zhonghan stalk of high velocity air to beat leaf parameter detecting elimination method, can accurately identify blade and offal in tobacco leaf, accurate Calculation tobacco leaf area obtains beating the leaf parameter, thereby improves the production efficiency of beating and double roasting process flow process.

Technical scheme: the invention provides a kind of reaching based on the discrete Ye Zhonghan stalk of high velocity air and beat leaf parameter detecting elimination method, comprise the following steps:

(1) utilize make thinner tobacco leaf and tobacco leaf is transferred in the high speed air channel of belt feeder;

(2) blower fan is as source of the gas, produces high velocity air and makes tobacco leaf further uniform discrete in the high speed air channel, and tobacco leaf is admitted to detection zone with air-flow simultaneously;

(3) in detection zone, adopt light source that tobacco leaf is shone in a side of tobacco leaf, apart from large, therefore produce different penetrabilitys due to the density difference of blade and offal, obtain the skeleton view of high-resolution tobacco leaf with the opposite side of the tobacco leaf camera relative with light source position;

(4) extract the textural characteristics of tobacco leaf from the skeleton view that obtains, treatment step comprises:

1., the X-Y scheme of tobacco leaf perspective is I ₀, make its all pixels to the upper left of original position, upper, upper right, left and right, lower-left, under, the bottom right respectively is offset a position, obtains I ₁(x-1, y-1), I ₂(x, y-1), I ₃(x+1, y-1), I ₄(x-1, y), I ₅(x+1, y), I ₆(x-1, y+1), I ₇(x, y+1) and I ₈(x+1, y+1), wherein, x, y are I ₀In horizontal ordinate and the ordinate of arbitrary pixel, the image that is offset after 8 directions is respectively I ₁, I ₂, I ₃, I ₄, I ₅, I ₆, I ₇And I ₈, with the X-Y scheme I that produces ₁, I ₂, I ₃, I ₄, I ₅, I ₆, I ₇And I ₈In the gray-scale value I of each pixel p _i(p) deduct I ₀The gray-scale value I of upper correspondence position pixel ₀(p), i.e. I ' _i(p)=I _i(p)-I ₀(p), according to the difference I ' of gray-scale value _i(p) obtain corresponding diagram I ' _i, i=1 wherein, 2 ..., 8, this moment I ' _iBe the pixel value magnitude relationship between former figure and offset pixel, it is carried out binaryzation, in 3 * 3 zones, at I ₀In around to put pixel value be 1 when larger than intermediate point pixel value, otherwise be 0, for I ' _iMiddle any point p has:

$\mathrm{sig}\left(L_i^{\&prime;}\left(p\right)\right)=\left\{\begin{array}{cc}0& I_i^{\&prime;}\left(p\right)<0\\ 1& I_i^{\&prime;}\left(p\right)\&GreaterEqual;0\end{array}\right.,p\&Element;I_i^{\&prime;};$

2., I ' _iOnly can reflect that surrounding pixel is selected and the pixel value magnitude relationship of intermediary image vegetarian refreshments does not still reflect the pixel position relationship, and the textural characteristics of tobacco leaf need to comprise position correlation, here operate by weight matrix W: make f (p)=WX, wherein X is matrix $\left[\begin{array}{ccc}\mathrm{sig}\left({{\mathrm{<figure-callout\; id="1"\; label="I"\; filenames="HDA00002884894100021.png,HDA00002884894100022.png"\; state="\{\{state\}\}">I</figure-callout>}}_1}^{\&prime;}\left(p\right)\right)& \mathrm{sig}\left({{\mathrm{<figure-callout\; id="2"\; label="I"\; filenames="HDA00002884894100021.png,HDA00002884894100022.png"\; state="\{\{state\}\}">I</figure-callout>}}_2}^{\&prime;}\left(p\right)\right)& \mathrm{sig}\left({{\mathrm{<figure-callout\; id="3"\; label="I"\; filenames="HDA00002884894100022.png,HDA00002884894100041.png"\; state="\{\{state\}\}">I</figure-callout>}}_3}^{\&prime;}\left(p\right)\right)\\ \mathrm{sig}\left({{\mathrm{<figure-callout\; id="4"\; label="I"\; filenames="HDA00002884894100042.png"\; state="\{\{state\}\}">I</figure-callout>}}_4}^{\&prime;}\left(p\right)\right)& 0& \mathrm{sig}\left({{\mathrm{<figure-callout\; id="5"\; label="I"\; filenames="HDA00002884894100021.png,HDA00002884894100041.png"\; state="\{\{state\}\}">I</figure-callout>}}_5}^{\&prime;}\left(p\right)\right)\\ \mathrm{sig}\left({{\mathrm{<figure-callout\; id="6"\; label="I"\; filenames="HDA00002884894100041.png,HDA00002884894100042.png"\; state="\{\{state\}\}">I</figure-callout>}}_6}^{\&prime;}\left(p\right)\right)& \mathrm{sig}\left({{\mathrm{<figure-callout\; id="7"\; label="I"\; filenames="HDA00002884894100041.png,HDA00002884894100042.png"\; state="\{\{state\}\}">I</figure-callout>}}_7}^{\&prime;}\left(p\right)\right)& \mathrm{sig}\left({I_8}^{\&prime;}\left(p\right)\right)\end{array}\right]$ , W is matrix $\left[\begin{array}{ccc}1& 2& 4\\ 8& 0& 16\\ 32& 64& 128\end{array}\right];$ In the new feature figure that obtain this moment, element size is between 0～255, i.e. 0≤f (p)≤255, general fluoroscopy images size is 32 * 32, this moment, textural characteristics was 1024 dimensions, the textural characteristics dimension is far longer than 256 and is not easy to analyze, therefore at 0～255 enterprising column hisgram, making textural characteristics is only 256 dimensions to it.

After analyzing the tobacco leaf fluoroscopy images from texture, contain as can be known the tobacco leaf of offal obvious with the tobacco leaf difference on textural characteristics that does not contain offal, utilize this characteristics, algorithm for design detects blade and the offal in the identification tobacco leaf and offal is rejected, and described algorithm comprises the following steps:

The described textural characteristics of I, tobacco leaf is totally 256 dimensions, is designated as F ₀, F ₁, F ₂..., F ₂₅₅

II, build a binary tree with textural characteristics, at first the textural characteristics of a large amount of blades and offal is made statistical study before building binary tree, at F ₀, F ₁, F ₂..., F ₂₅₅The obvious one group of key feature points of textural characteristics difference of middle discovery blade and offal, with this group key feature points each node as binary tree, described key feature points characteristic of correspondence value is the correlative value of each node of binary tree traversal; From the root node of the binary tree that constructs, whether test the eigenwert of this node greater than described correlative value, the branch corresponding according to given next key feature points moves down, then repeat on the subtree take new node as root, can arrive leafy node at last, mark example tobacco leaf to be identified this moment is blade or offal, and identifying after binary tree traversal is blade or offal.

By to a large amount of blades and the histogrammic statistical study of offal textural characteristics, draw at F ₇, F ₁₅, F ₃₁And F ₂₅₅The time blade and offal textural characteristics difference obvious, with F ₇, F ₁₅, F ₃₁And F ₂₅₅As the node of binary tree, corresponding correlative value is designated as a ₁, a ₂, a ₃And a ₄, particularly, with F ₁₅As tree root, the F of test tobacco leaf to be identified ₁₅Whether eigenwert is greater than a ₂, be to move to leaf F ₂₅₅, otherwise move to branch F ₃₁Then test F ₂₅₅Whether eigenwert is greater than a ₄, be can reach a conclusion to be offal, the no blade that draws, the F that tests simultaneously ₃₁Whether eigenwert is greater than a ₃, be to reach a conclusion to be blade, otherwise need to continue judgement F ₇Whether eigenwert is greater than a ₁, be can reach a conclusion to be blade, the no offal that draws is on basis based on a large amount of statisticss the shortcut calculation of trying one's best when guaranteeing accurately to distinguish offal and blade according to the purpose of this kind Structural Tectonics binary tree.

(5) according to skeleton view, the shape and size of tobacco leaf are analyzed, obtained the tobacco leaf area, calculate and beat the leaf parameter: Percentage of large and middle size strips, small pieces rate and Ye Zhonghan stalk rate.

Beat the leaf parameter in order further to calculate, the concrete steps in step (5) comprise:

A, employing Canny edge detection operator or Sobel edge detection operator look like to carry out the tobacco leaf Boundary Extraction to schematic perspective view diagram;

B, according to the sealing area Φ of tobacco leaf feature modeling boundary profile, be the tobacco leaf area;

The leaf parameter is beaten in c, calculating: the surface density of known blade is σ, the mass M of tobacco leaf _t=σ * Φ distinguishes big-and-middle tobacco leaf and small pieces tobacco leaf, big-and-middle leaf quality M by threshold area T is set _B=σ * Φ (Φ〉T), leaflet protonatomic mass M _S=σ * Φ (Φ≤T), Percentage of large and middle size strips η _B=M _B/ M _T, small pieces rate η _s=M _s/ M _T, contain stalk rate η _g=(M-M _t)/M _t, wherein M is the quality of tobacco leaf weighing before practical operation, M-M _tBe the quality of offal,, comprised simultaneously the area of offal in the area calculating of blade here, can ignore here and the offal surface area is very little.

Owing to calculating each dozen leaf parameter for the more difficult identification tobacco leaf of the overlapping situation of tobacco leaf area, so thereby spread out to make thinner by belt feeder before and can avoid the overlapping situation of tobacco leaf to calculate the required leaf parameter of beating with high velocity air.

the mathematical method of calculating the sealing area Φ of tobacco leaf profile in step b is preferably: suppose that profile is is that 1 pixel consists of by gray-scale value, make the labeling method of position relationship between adjacent two pixels be: the right side is designated as 0, the upper right is designated as 1, on be designated as 2, the upper left is designated as 3, a left side is designated as 4, the lower-left is designated as 5, under be designated as 6, the bottom right is designated as 7, according to described labeling method since a closed outline any point record, position relationship between all each pixels of profile one obtains one section path chain code, read the sealing area Φ that described path chain code calculates the tobacco leaf profile.

Beneficial effect: 1, directly utilize the skew of pixel, pixel size is concerned binaryzation, and obtain position relationship between pixel by weight matrix, pixel and the size of surrounding pixel point and the textural characteristics that position relationship obtains tobacco leaf in the skeleton view of analysis tobacco leaf, avoid choosing N * N window and carry out morphological operation extraction shape and structure feature, need not to obtain the texture difference that filtering image can be analyzed blade and offal; 2, add up in a large number tobacco leaf textural characteristics F ₀, F ₁, F ₂..., F ₂₅₅Draw the obvious key feature points F of eigenwert difference of blade and offal ₇, F ₁₅, F ₃₁And F ₂₅₅, according to choosing correlative value in the eigenwert disparity range of these four some blades and offal, set up binary tree, with each tobacco leaf F ₇, F ₁₅, F ₃₁And F ₂₅₅Characteristic of correspondence value and correlative value compare can distinguish blade and offal, and recognition methods is simply effective, the disturbing effect of avoiding speckle that texture is distinguished; 3, the labeling method of position relationship between determined pixel point, obtain path chain code on the tobacco leaf outline line by this labeling method, calculate the tobacco leaf area according to the path chain code, the area calculated value is accurate, adopts the method to calculate the tobacco leaf area that only covers three pixels described in background technology, can obtain exact value 0.5, therefore even tobacco leaf is less object, area value does not have error yet, and each dozen leaf parameter that therefore obtains is accurate, helps to improve the quality of beating and double roasting process; 4, belt feeder is spread tobacco leaf out and is made thinner, with blower fan, tobacco leaf is transported to the high speed air channel and further makes tobacco leaf homogenising discretize, greatly reduce overlapping tobacco leaf, be difficult for overlapping phenomenon after the detection zone of tobacco leaf access to plant, the fluoroscopy images of camera shooting tobacco leaf is more clear; 5, the tobacco leaf fluoroscopy images that obtains by analysis, textural characteristics is extracted, adopt binary tree accurately to distinguish offal and blade, use the method to coordinate device for eliminating accurately to reject the tobacco leaf that contains offal, simultaneously accurately calculate online the various leaf parameters of beating, comprise Percentage of large and middle size strips, small pieces rate and Ye Zhonghan stalk rate etc., solved and contain stalk and sense cycle in present Redrying Factory offline inspection leaf and the challenge such as grow, thereby obviously improved Redrying Factory product quality and efficient.

Description of drawings

Fig. 1 is the situation schematic diagram that tobacco leaf described in background technology of the present invention only covers three pixels;

Fig. 2 (a) is the skeleton view of blade;

Fig. 2 (b) is the skeleton view of offal;

Fig. 3 (a) is textural characteristics histogram corresponding to Fig. 2 (a);

Fig. 3 (b) is textural characteristics histogram corresponding to Fig. 2 (b);

Fig. 4 is for being used for the binary tree of classification offal and blade;

Fig. 5 (a) is boundary profile figure corresponding to Fig. 2 (a);

Fig. 5 (b) is boundary profile figure corresponding to Fig. 2 (b);

Fig. 6 is the labeling method schematic diagram of position relationship between adjacent two pixels;

Fig. 7 is the path chain code schematic diagram of tobacco leaf boundary profile;

Fig. 8 is the structural representation that offal described in embodiment detects the identification device for eliminating;

Fig. 9 is the partial enlarged drawing that offal described in embodiment detects detection zone in the identification device for eliminating.

Embodiment

The below is elaborated to technical solution of the present invention, but protection scope of the present invention is not limited to described embodiment.

Embodiment: a kind of reaching based on the discrete Ye Zhonghan stalk of high velocity air of the present invention beaten leaf parameter detecting elimination method, comprises the following steps:

(1) with make thinner tobacco leaf and tobacco leaf is transferred in the high speed air channel of belt feeder;

(2) blower fan is as source of the gas, produces high velocity air and makes tobacco leaf further uniform discrete in the high speed air channel, and tobacco leaf is admitted to detection zone with air-flow simultaneously;

(3) in detection zone, a side at tobacco leaf adopts light source that tobacco leaf is shone, and because the density difference distance of blade and offal is large, therefore produces different penetrabilitys, obtain the skeleton view of high-resolution tobacco leaf with the opposite side of the tobacco leaf camera relative with light source position, as Fig. 2 (a), 2(b) as shown in;

(4) extract the textural characteristics of tobacco leaf from the skeleton view that obtains, extraction step comprises:

1., the X-Y scheme of tobacco leaf perspective is I ₀, make its all pixels to the upper left of original position, upper, upper right, left and right, lower-left, under, the bottom right respectively is offset a position, obtains I ₁(x-1, y-1), I ₂(x, y-1), I ₃(x+1, y-1), I ₄(x-1, y), I ₅(x+1, y), I ₆(x-1, y+1), I ₇(x, y+1) and I ₈(x+1, y+1), wherein, x, y are I ₀In horizontal ordinate and the ordinate of arbitrary pixel, the image that is offset after 8 directions is respectively I ₁, I ₂, I ₃, I ₄, I ₅, I ₆, I ₇And I ₈, with the X-Y scheme I that produces ₁, I ₂, I ₃, I ₄, I ₅, I ₆, I ₇And I ₈In the gray-scale value I of each pixel p _i(p) deduct I ₀The gray-scale value I of upper correspondence position pixel ₀(p), i.e. I ' _i(p)=I _i(p)-I ₀(p), according to the difference I ' of gray-scale value _i(p) obtain corresponding diagram I ' _i, i=1 wherein, 2 ..., 8, this moment I ' _iBe the pixel value magnitude relationship between former figure and offset pixel, it is carried out binaryzation, in 3 * 3 zones, at I ₀In around to put pixel value be 1 when larger than intermediate point pixel value, otherwise be 0, for I ' _iMiddle any point p has:

$\mathrm{sig}\left(L_i^{\&prime;}\left(p\right)\right)=\left\{\begin{array}{cc}0& I_i^{\&prime;}\left(p\right)<0\\ 1& I_i^{\&prime;}\left(p\right)\&GreaterEqual;0\end{array}\right.,p\&Element;I_i^{\&prime;};$

2., I ' _iOnly can reflect that surrounding pixel is selected and the pixel value magnitude relationship of intermediary image vegetarian refreshments does not still reflect the position relationship of pixel, and the textural characteristics of tobacco leaf need to comprise position correlation, here operate by weight matrix W: make f (p)=WX, wherein X is matrix $\left[\begin{array}{ccc}\mathrm{sig}\left({{\mathrm{<figure-callout\; id="1"\; label="I"\; filenames="HDA00002884894100021.png,HDA00002884894100022.png"\; state="\{\{state\}\}">I</figure-callout>}}_1}^{\&prime;}\left(p\right)\right)& \mathrm{sig}\left({{\mathrm{<figure-callout\; id="2"\; label="I"\; filenames="HDA00002884894100021.png,HDA00002884894100022.png"\; state="\{\{state\}\}">I</figure-callout>}}_2}^{\&prime;}\left(p\right)\right)& \mathrm{sig}\left({{\mathrm{<figure-callout\; id="3"\; label="I"\; filenames="HDA00002884894100022.png,HDA00002884894100041.png"\; state="\{\{state\}\}">I</figure-callout>}}_3}^{\&prime;}\left(p\right)\right)\\ \mathrm{sig}\left({{\mathrm{<figure-callout\; id="4"\; label="I"\; filenames="HDA00002884894100042.png"\; state="\{\{state\}\}">I</figure-callout>}}_4}^{\&prime;}\left(p\right)\right)& 0& \mathrm{sig}\left({I_5}^{\&prime;}\left(p\right)\right)\\ \mathrm{sig}\left({I_6}^{\&prime;}\left(p\right)\right)& \mathrm{sig}\left({I_7}^{\&prime;}\left(p\right)\right)& \mathrm{sig}\left({I_8}^{\&prime;}\left(p\right)\right)\end{array}\right]$ , W is matrix $\left[\begin{array}{ccc}1& 2& 4\\ 8& 0& 16\\ 32& 64& 128\end{array}\right];$

3., element size is between 0～255 in this moment new feature figure of obtaining, because of sig (I ' _i(p)) value is 0 or 1, therefore 0≤f (p)≤255, general fluoroscopy images size is 32 * 32, this moment, textural characteristics was 1024 dimensions, the textural characteristics dimension is far longer than 256 and is not easy to analyze, thus to it at 0～255 enterprising column hisgram, the tobacco leaf textural characteristics histogram in Fig. 2 (a) is obtained Fig. 3 (a), tobacco leaf textural characteristics histogram in Fig. 2 (b) is obtained Fig. 3 (b), and making textural characteristics is only 256 dimensions.

Texture dimension according to said extracted, after analyzing the tobacco leaf fluoroscopy images from texture, containing as can be known the tobacco leaf of offal differs greatly on textural characteristics with the tobacco leaf that does not contain offal, utilize this characteristics, after the acquisition tobacco leaf image, the tobacco leaf textural characteristics is extracted, and with this textural characteristics, offal is detected and rejects; Specifically comprise the following steps: the textural characteristics of tobacco leaf is totally 256 dimensions, is designated as F ₀, F ₁, F ₂..., F ₂₅₅, build a binary tree with textural characteristics, at first the textural characteristics of a large amount of blades and offal is made statistical study before the structure binary tree, the histogrammic statistical study by to a large amount of blades and offal textural characteristics draws at F ₇, F ₁₅, F ₃₁And F ₂₅₅The time blade and offal textural characteristics difference obvious, the tobacco leaf numerical value of these four some correspondences, offal numerical value and difference value thereof are as shown in table 1:

Table 1

F	F 7	F 15	F 31	F 255
					Blade numerical value	21693	42394	26392	188839
Offal numerical value	5896	14134	9055	329739
					Difference value	15797	28260	17337	140900

Obtain thus, with F ₇, F ₁₅, F ₃₁And F ₂₅₅As the node of binary tree, determine corresponding correlative value a according to difference value ₁Get 20000, a ₂Get 40000, a ₃Get 25000, a ₄Get 200000, particularly, as shown in Figure 4, with F ₁₅As tree root, the F of test tobacco leaf to be identified ₁₅Whether eigenwert is greater than a ₂, be to move to leaf F ₂₅₅, otherwise move to branch F ₃₁Test simultaneously F ₃₁And F ₂₅₅, F ₂₅₅Eigenwert whether greater than a ₄, be can reach a conclusion to be offal, the no blade that draws simultaneously, moves to branch F ₃₁The eigenwert of testing judges that whether it is greater than a ₃, be to reach a conclusion to be blade, otherwise need to continue judgement F ₇Whether eigenwert is greater than a ₁, be can reach a conclusion to be blade, the no offal that draws.Be on basis based on a large amount of statisticss the shortcut calculation of trying one's best when guaranteeing accurately to distinguish offal and blade according to the purpose of this kind Structural Tectonics binary tree.In the present embodiment, according to histogram 3(a) and 3(b) in binary tree in specified number value traversing graph 4 obtain: the tobacco leaf in Fig. 2 (a) is divided into blade, and the tobacco leaf in Fig. 2 (b) is divided into offal.

(5) according to skeleton view, the shape and size of tobacco leaf are analyzed, adopt the Canny edge detection operator to look like to carry out the tobacco leaf Boundary Extraction to schematic perspective view diagram, the contour pattern that the tobacco leaf in Fig. 2 (a), Fig. 2 (b) is extracted is as shown in Fig. 5 (a), Fig. 5 (b), sealing area Φ according to tobacco leaf feature modeling boundary profile, suppose that profile is is that 1 pixel consists of by gray-scale value, make the labeling method of position relationship between adjacent two pixels be illustrated in figure 6 as: right is designated as 0, the upper right side is to being designated as 1, upper direction is designated as 2, the upper left side is to being designated as 3, left is to being designated as 4, the lower left is to being designated as 5, lower direction is designated as 6, the lower right is to being designated as 7, according to described labeling method since a closed outline any point record, position relationship between all adjacent pixels of profile one obtains one section path chain code, one section path chain code as shown in Figure 7 is 320200766444, can obtain chain code FMa and FMb corresponding to Fig. 5 (a) with 5(b), can calculate the sealing area Φ of profile by basic mathematical knowledge, here method is that profile is divided into rectangle and triangle, and a pair of its ask for area, and summation at last can get the tobacco leaf area.The surface density of known blade is σ, the mass M of tobacco leaf _t=σ * Φ distinguishes big-and-middle tobacco leaf and small pieces tobacco leaf, big-and-middle leaf quality M by threshold area T is set _B=σ * Φ (Φ〉T), leaflet protonatomic mass M _S=σ * Φ (Φ≤T), Percentage of large and middle size strips η _B=M _B/ M _T, small pieces rate η _s=M _s/ M _T, contain stalk rate η _g=(M-M _t)/M _t, wherein M is the quality of tobacco leaf weighing before practical operation, M-M _tBe the quality of offal,, comprised simultaneously the area of offal in the area calculating of blade here, can ignore here and the offal surface area is very little.Do not add up complicated offal length here, namely can obtain numerous leaf parameters of beating, computing method are simple.

Adopt the offal of above-mentioned detection elimination method to detect the identification device for eliminating as shown in Figure 8, source of the gas is that blower fan 7 produces high velocity air, carry by passage 2, throughput direction such as figure upward arrow represent, the pipe passage becomes the square tube passage before material enters detection zone 3, then after separating by the material device 6 with gas, solid buffer action through detection zone 3, the air channel is got back to blower fan 7 and is completed closed loop.At first, the shake groove 1 of smoked sheet by belt feeder spread out evenly enters pipeline, then after the wind that produces by source of the gas is further beaten evenly, deliver to detection zone 3, here, due to for the more difficult identification tobacco leaf of the overlapping situation of tobacco leaf area, be difficult to calculate each dozen leaf parameter, so spread the situation that to avoid tobacco leaf overlapping that blows of making thinner with high velocity air out by belt feeder shake groove 1 before; The offal of rejecting through detection zone 3 is gentle flows to output after gas lock 4, and normal blade continues to be transported to feeder 6, then exports by belt feeder 5 and transports.

as shown in Figure 9, the vertical supervisor 3-0 that contains the bottom in and top out that is communicated with passage 2 in detection zone 3, supervisor's stage casing is the detector tube 3-8 that transparent material is made, one side of detector tube is settled infrared light supply 3-9, opposite side arranges reflective mirror 3-2, reflective mirror 3-2 will reflect on the horizon light alignment through detector tube, the top of reflective mirror 3-2 is provided with the catotropic camera 3-3 of mirror, the signal output part of camera 3-3 is by the controlled end of control circuit connected electromagnetic valve nozzle 3-5, solenoid valve nozzle 3-5 connects high-pressure air source 3-4, tobacco leaf through detection zone 3 is carried by air-flow, enter from detector tube 3-8 lower end, image is caught by camera 3-3 by reflective mirror 3-2 under the irradiation of infrared light supply 3-9, after this, camera 3-3 carries out texture feature extraction with the tobacco leaf information conveyance of Real-time Collection to the processing system, classification blade and offal also calculate and beat the leaf parameter, offal is blown to rejecting in branched bottom 3-10 by solenoid valve nozzle 3-5, if during blade, solenoid valve nozzle 3-5 does not work, upwards send into feeder 6 along supervisor 3-0, opposite side at the relative supervisor 3-0 of solenoid valve nozzle 3-5 is rejecting branched bottom 3-10, the exit of rejecting branched bottom 3-10 is provided with the gas lock 4 (being star-like discharging device) that the isolation air-flow is used, particularly, the barrel of this gas lock 4 is arranged on the outlet of rejecting branched bottom 3-10, uniform one group of radial blade of isolation on barrel, the external diameter of blade is suitable with the cross-sectional width of rejecting branched bottom 3-10, be divided into cavity between each blade.During work, when a certain cavity rotation extremely communicated with rejecting branched bottom 3-10, in material, isolated offal entered wherein, rotation is to this cavity and rejecting branched bottom 3-10 open position, play the effect of isolation air-flow, continue rotation extremely towards upper/lower positions, offal drops out under Action of Gravity Field.Can adopt other isolation airflow apparatus replacements such as gas curtain herein.Supervisor 3-0 is provided with cylinder 3-7, the lower end abutment of cylinder 3-7 is in the upper end of detector tube 3-8, its effect is promote or press down detector tube 3-8: when pressing down, guarantee line seal by the descending press seal pad of cylinder 3-7, during lifting, drive detector tube 3-8 and lower end web joint and disconnect, so that whole detector tube 3-8 can be drawn out along the upper end slideway, the convenient cleaning.

As mentioned above, although represented and explained the present invention with reference to specific preferred embodiment, it shall not be construed as the restriction to the present invention self.Under the spirit and scope of the present invention prerequisite that does not break away from the claims definition, can make in the form and details various variations to it.

Claims (3)

1. one kind is obstructed and beats leaf parameter detecting elimination method based on the discrete Ye Zhonghan of high velocity air, it is characterized in that: comprise the following steps:

(1) utilize make thinner tobacco leaf and tobacco leaf is transferred in the high speed air channel of belt feeder;

(2) make tobacco leaf further uniform discrete in the high speed air channel by high velocity air, tobacco leaf is admitted to detection zone with air-flow simultaneously;

(3) in detection zone, adopt light source that tobacco leaf is shone in a side of tobacco leaf, apart from large, therefore produce different penetrabilitys due to the density difference of blade and offal, obtain the skeleton view of high-resolution tobacco leaf with the opposite side of the tobacco leaf camera relative with light source position;

(4) extract the textural characteristics of tobacco leaf from the skeleton view that obtains, treatment step comprises:

1., the X-Y scheme of establishing tobacco leaf perspective is I ₀, make its all pixels to the upper left of original position, upper, upper right, left and right, lower-left, under, the bottom right respectively is offset a position, obtains I ₁(x-1, y-1), I ₂(x, y-1), I ₃(x+1, y-1), I ₄(x-1, y), I ₅(x+1, y), I ₆(x-1, y+1), I ₇(x, y+1) and I ₈(x+1, y+1), wherein, x, y are I ₀In horizontal ordinate and the ordinate of arbitrary pixel, the image that is offset after 8 directions is respectively I ₁, I ₂, I ₃, I ₄, I ₅, I ₆, I ₇And I ₈, with the X-Y scheme I that produces ₁, I ₂, I ₃, I ₄, I ₅, I ₆, I ₇And I ₈In the gray-scale value I of each pixel p _i(p) deduct I ₀The gray-scale value I of upper correspondence position pixel ₀(p), i.e. I ' _i(p)=I _i(p)-I ₀(p), according to the difference I ' of gray-scale value _i(p) obtain corresponding diagram I ' _i, i=1 wherein, 2 ..., 8, this moment I ' _iBe the pixel value magnitude relationship between former figure and offset pixel, it is carried out binaryzation, in 3 * 3 zones, at I ₀In around to put pixel value be 1 when larger than intermediate point pixel value, otherwise be 0, for I ' _iMiddle any point p has:

$\mathrm{sig}\left(L_i^{\&prime;}\left(p\right)\right)=\left\{\begin{array}{cc}0& I_i^{\&prime;}\left(p\right)<0\\ 1& I_i^{\&prime;}\left(p\right)\&GreaterEqual;0\end{array}\right.,p\&Element;I_i^{\&prime;};$

2., the textural characteristics of tobacco leaf comprises pixel value magnitude relationship and the pixel position relationship of pixel and neighboring pixel point, I ' _iThe magnitude relationship of reflection intermediary image vegetarian refreshments and surrounding pixel point makes f (p)=WX, and wherein X is matrix $\left[\begin{array}{ccc}\mathrm{sig}\left({I_1}^{\&prime;}\left(p\right)\right)& \mathrm{sig}\left({I_2}^{\&prime;}\left(p\right)\right)& \mathrm{sig}\left({I_3}^{\&prime;}\left(p\right)\right)\\ \mathrm{sig}\left({I_4}^{\&prime;}\left(p\right)\right)& 0& \mathrm{sig}\left({I_5}^{\&prime;}\left(p\right)\right)\\ \mathrm{sig}\left({I_6}^{\&prime;}\left(p\right)\right)& \mathrm{sig}\left({I_7}^{\&prime;}\left(p\right)\right)& \mathrm{sig}\left({I_8}^{\&prime;}\left(p\right)\right)\end{array}\right],$ W is weight matrix $\left[\begin{array}{ccc}1& 2& 4\\ 8& 0& 16\\ 32& 64& 128\end{array}\right],$ The position relationship of f (p) reflection intermediary image vegetarian refreshments and surrounding pixel point, in the new feature figure that obtains, element size is between 0～255, i.e. 0≤f (p)≤255, thus to the textural characteristics of tobacco leaf at 0～255 enterprising column hisgram, making textural characteristics is only 256 dimensions;

Contain the tobacco leaf of offal large with the tobacco leaf difference on textural characteristics that does not contain offal, blade and offal in algorithm for design identification tobacco leaf, the tobacco leaf that will contain offal is rejected, and described algorithm comprises the following steps:

The described textural characteristics of I, tobacco leaf is totally 256 dimensions, is designated as F ₀, F ₁, F ₂..., F ₂₅₅

II, build a binary tree with textural characteristics, the textural characteristics of a large amount of blades and offal is made statistical study, at F ₀, F ₁, F ₂..., F ₂₅₅Middle discovery is at F ₇, F ₁₅, F ₃₁And F ₂₅₅The time blade and offal the textural characteristics difference obviously, with F ₇, F ₁₅, F ₃₁And F ₂₅₅This group key feature points is as each node of binary tree, and described key feature points characteristic of correspondence value is the correlative value of each node of binary tree traversal, represents the classification of blade and offal after binary tree traversal with leafy node;

(5) according to skeleton view, the shape and size of tobacco leaf are analyzed, obtained the tobacco leaf area, calculate and beat the leaf parameter: Percentage of large and middle size strips, small pieces rate and Ye Zhonghan stalk rate.

2. according to claim 1 reaching based on the discrete Ye Zhonghan stalk of high velocity air beaten leaf parameter detecting elimination method, it is characterized in that: calculate the concrete steps of beating the leaf parameter in step (5) and comprise:

A, employing Canny edge detection operator or Sobel edge detection operator look like to carry out the tobacco leaf Boundary Extraction to schematic perspective view diagram;

B, according to the sealing area Φ of tobacco leaf feature modeling profile, be the tobacco leaf area;

The leaf parameter is beaten in c, calculating: the surface density of known blade is σ, the mass M of tobacco leaf _t=σ * Φ distinguishes big-and-middle tobacco leaf and small pieces tobacco leaf, big-and-middle leaf quality M by threshold area T is set _B=σ * Φ (Φ〉T), leaflet protonatomic mass M _S=σ * Φ (Φ≤T), Percentage of large and middle size strips η _B=M _B/ M _T, small pieces rate η _s=M _s/ M _T, contain stalk rate η _g=(M-M _t)/M _t, wherein M is the quality of tobacco leaf weighing before practical operation, M-M _tBe the quality of offal,, comprised simultaneously the area of offal in the area calculating of blade here, can ignore here and the offal surface area is very little.

3. according to claim 2 reaching based on the discrete Ye Zhonghan stalk of high velocity air beaten leaf parameter detecting elimination method, it is characterized in that: the mathematical method of calculating the sealing area Φ of tobacco leaf profile in step b is: suppose that profile is is that 1 pixel consists of by gray-scale value, make the labeling method of position relationship between adjacent two pixels be: the right side is designated as 0, the upper right is designated as 1, on be designated as 2, the upper left is designated as 3, a left side is designated as 4, the lower-left is designated as 5, under be designated as 6, the bottom right is designated as 7, according to described labeling method since a closed outline any point record, position relationship between all each pixels of profile one obtains one section path chain code, read the sealing area Φ that described path chain code calculates the tobacco leaf profile.

Images