CN105403147A - Embedded bottle pre-form detection system and detection method - Google Patents

Abstract

The invention discloses an embedded bottle pre-form detection system and a detection method. A bottle pre-form sidewall detection camera, a bottle opening thread side edge detection camera, a bottle opening size detection camera, a bottle opening thread top detection camera, a bottle opening sealing ring detection camera, and a bottle pre-form bottom detection camera are arranged on a detection line. Noise reduction, binarization and location fitting are carried out on the images through an image filtering algorithm based on morphology. The images are converted from an annular form to a strip form through a random sampling consistency algorithm. A registration operation is carried out on the images. Finally an Adaboost classifier is employed for defect identification. The embedded bottle pre-form detection system and the detection method have the advantages that the bottle blowing quality monitoring is transited from artificial detection to intelligent detection, the liquid state whole line loss is reduced by 5%, and the sealing fault ratio is reduced by 20%.

Description

A kind of based on Embedded bottle embryo detection system and detection method

Technical field

The invention belongs to mechanical equipment technical field, relate to a kind of based on Embedded bottle embryo detection system and detection method.

Background technology

Be main mainly with PET (PolyethyleneTerephthalate, polyethylene terephthalate) material in industries such as food, beverage, medicine, cosmeticses of everyday use in Production and Packaging process.China is often only PET bottle beverage consumption amount and reaches 2,000 ten thousand tons, and with annual 14% speed increase.Bottle embryo is the blank of shaping PET bottle, belongs to pipe embryo.At present, PET bottle embryo still adopts traditional manual detection mode, and the relative machine detection speed of manual detection is slow, can not meet the production line run fast; Simultaneously by the restriction of people's vision subjectivity, the effect of manual detection is difficult to ensure.

Along with the development of Computer Control Technology and intelligent science theory, vision-based detection obtains successfully in commercial production context of detection.The research of recent domestic just detects three different aspects towards the detection of the fine structure of large scale sensing range, sub-pix and 3D and develops rapidly, improves detection speed simultaneously and detects intelligent.Along with popularization and the widespread use of embedded technology and product, the machine vision product packaging effects of the plate card type of traditional Based PC/industrial computer and the problem such as environmental suitability is poor, cost performance is low, it feature possessing high-speed data acquisition is not brought into play, based on embedded platform intelligent detection equipment due to the characteristics such as its equipment is light, cost performance is high become current machine visual field research focus.So the high speed development of relevant software and hardware two aspect is that PET bottle embryo on-line checkingi realizes automation and intelligentification and becomes possibility.

The present invention is directed to that line speed is high, vessel port defect is small, filling and the difficult problem such as closure cap applying apparatus narrow space, have studied combined imaging principle, 360 ° of bottleneck panorama microdefect imaging systems are innovated, introduce embedded development platform, devise a kind of based on Embedded Intelligent bottle embryo pick-up unit.

Summary of the invention

The object of the present invention is to provide a kind of based on Embedded bottle embryo detection system and detection method, solve manual detection mode poor efficiency, false drop rate is high, existing system is difficult to carry out the problem that PET bottle embryo quick high accuracy detects.

The technical solution adopted in the present invention comprises reason fetus device, reason fetus device is used for carrying out arrangement reason embryo to the bottle embryo after injection molded, and bottle embryo is sent into input rotating disk, input dial rotation drives bottle embryo to enter the first detection rotating disk, the first bottle embryo detecting rotating disk transports track side and is provided with bottle embryo sidewall detection camera and whorl of bottleneck side detection camera, first detects rotating disk sends into the second detection rotating disk by the bottle embryo on track, the second transport track side detecting rotating disk is provided with bottleneck size detection camera, whorl of bottleneck top detection camera, bottle sealing ring detection camera, bottle embryo floor detection camera, all detection camera all connect embedded image processing platform, the image of embedded image processing platform to camera calibration processes, distinguish defective bottle embryo, second track detecting rotating disk is provided with rejecting mouth, embryo mouth is provided with out in track end, embedded image processing platform carries out judgement process to the image gathered, distinguish defective bottle embryo, defective bottle embryo is rejected from rejecting mouth, normal bottle embryo is sent from going out embryo mouth.

Further, described bottleneck size detection camera, whorl of bottleneck top detection camera, bottle sealing ring detection camera adopt ring surface light source, and level is taken from the top down in bottle plumular axis line; Bottle embryo floor detection camera level is taken from the top down in bottle plumular axis line; Bottle embryo sidewall detection camera, whorl of bottleneck side detection camera are taken from the side perpendicular to bottle plumular axis line.

A kind of detection method step based on embedded bottle embryo detection system is as follows:

Step 1: collecting bottle embryo side wall image, bottle embryo bottom diagram picture, whorl of bottleneck side view picture, bottleneck size image, whorl of bottleneck top graph picture and bottle sealing ring detected image, image is adopted and carries out denoising based on morphologic Image filter arithmetic, opening operation removes the relative bright distributed architecture that image does not match with the form of structural element, retains the part matched simultaneously; The relative dark distributed architecture that closed operation then blank map picture does not match with structural element, retains the part matched simultaneously;

Step 2: carry out binary conversion treatment to image, according to industry spot optical illumination darkness adjustment intensity of illumination, and adjusts dynamic threshold, obtains the image of binaryzation;

Step 3: the image detected for bottleneck size, whorl of bottleneck top and bottle sealing ring, positions process of fitting treatment, adopt RANSAC algorithm, flow process is as follows:

(1) may be bottle preform mouth edge point set P by edge detection method collection, its sample number:

∑P＞3；

(2) from P, randomly draw 3 points form subset S, calculate the round model M that these 3 points are determined;

(3) be less than with M distance in complementary set Sc=P-S, Sc to the point of set a distance t form in point set, interior point set and S jointly form the consistent of S and collect S ^*;

(4) number m is allowed, if ∑ (S for given minimum consistent collection sample ^*) > m, then use a most young waiter in a wineshop or an inn

Multiplication utilizes S ^*ask and obtain new model M ^*, return successfully; Otherwise judge whether iterations reaches designated value, if reach, return failure, otherwise return (2).

The conversion of bottle germ ring, is converted into polar coordinates rectangular coordinate original for binaryzation bottle germ ring image, if certain point coordinate is (x, y) on original image, then after conversion, corresponding point coordinate is (r, θ), has:

$\left\{\begin{array}{c}r=\sqrt{x^3+y^2}\\ \mathrm{\θ}=\arctan (y/x)\end{array}\right.$

Image is made to be lath-like morphologies from annular Morphological Transitions;

Step 4: to bottleneck size, whorl of bottleneck top and bottle sealing ring image carry out registration operation, make three width image informations set up contact mutually to utilize, wavelet pyramid is utilized to carry out image registration process: to adopt wavelet transformation to set up image pyramid respectively, the bottom is original image, upwards successively carry out wavelet transformation and retain low frequency region, form Pyramid, extract target size characteristic sum Corner Feature successively to mate, under the basis of this coupling enters pyramid, one deck mates again, by that analogy until enter bottom original image to complete final coupling,

Step 5: to bottle embryo side wall image, bottle embryo bottom diagram picture, whether whorl of bottleneck side view picture, bottleneck size that registration is good, whorl of bottleneck top and bottle sealing ring image adopt Adaboost sorter to carry out defect recognition, qualified to differentiate corresponding bottle embryo.

Further, in described step 1, opening operation and closed operation are completed by combination image being carried out to etching operation and expansive working, if the image after image I etching operation is E (I), image after expansive working is D (I), then the opening operation of image realizes by first corroding the mode expanded afterwards, is namely completed by following formula:

OPEN(I)＝D(E(I))

Closed operation is realized by the mode of the post-etching that first expands, and is namely completed by following formula:

CLOSE(I)＝E(D(I))

For each sub-picture of input, first carry out the opening operation of suitable yardstick, the closed operation then carrying out more large scale obtains processing rear image, finally by process afterwards image and original image carry out and operation, obtain image after denoising.

Further, in described step 2, binary processing method is for establishing certain point (x on image, y) gray-scale value is f (x, y), and after binaryzation, the gray-scale value of this point is f ' (x, y), the pixel number needing ROI to be processed in image is N, and binaryzation is biased to δ, then binary-state threshold T is set as:

$T=\frac{\underset{(x,y)\∈ROI}{\Σ}f(x,y)}{N}+\mathrm{\δ}$

$f^{\&prime;}(x,y)=\{\begin{array}{cc}0& f(x,y)<T\\ 255& f(x,y)\&GreaterEqual;T\end{array}.$

The invention has the beneficial effects as follows and PET bottle embryo defect (as dirty in bottle preform mouth distortion, bottleneck breakage, the breakage of bottle germ ring, bottle germ wall, bottle preform mouth etc.) can be detected quickly and accurately, accuracy of detection is high, detection speed is fast, stability is high.

Achieve bottle blowing quality monitoring by the transition of manual detection to Intelligent Measurement, reduce liquid whole line loss consumption 5%, poor sealing rate declines 20%.

Accompanying drawing illustrates l

The structural representation of the Embedded bottle embryo detection system that Fig. 1 provides for one embodiment of the invention;

Fig. 2 is the detection algorithm flow process operated on embedded image processing platform.

In figure, 1, reason fetus device; 2, rotating disk is inputted; 3, first rotating disk is detected; 4, second rotating disk is detected; 5, bottle embryo sidewall detection camera; 6, whorl of bottleneck side detection camera; 7, bottleneck size detection camera; 8, whorl of bottleneck top detection camera; 9, bottle sealing ring detection camera; 10, bottle embryo floor detection camera; 11, embedded image processing platform; 12, mouth is rejected; 13 go out embryo mouth.

Embodiment

Below in conjunction with embodiment, the present invention is described in detail.

The structure of Embedded bottle embryo detection system of the present invention as shown in Figure 1, reason fetus device 1 is for carrying out arrangement reason embryo to the bottle embryo after injection molded, and bottle embryo is sent into input rotating disk 2, input rotating disk 2 rotates and drives bottle embryo to enter the first detection rotating disk 3, the first bottle embryo detecting rotating disk 3 transports track side and is provided with bottle embryo sidewall detection camera 5 and whorl of bottleneck side detection camera 6, first detects rotating disk 3 sends into the second detection rotating disk 4 by the bottle embryo on track, the second transport track side detecting rotating disk 4 is provided with bottleneck size detection camera 7, whorl of bottleneck top detection camera 8, bottle sealing ring detection camera 9, bottle embryo floor detection camera 10, all detection camera all connect embedded image processing platform 11, the image of embedded image processing platform 11 pairs of camera calibrations processes, distinguish defective bottle embryo, second track detecting rotating disk 4 is provided with rejects mouth 12, embryo mouth 13 is provided with out in track end, embedded image processing platform 11 carries out judgement process to the image gathered, distinguish defective bottle embryo, defective bottle embryo is rejected from rejecting mouth 13, normal bottle embryo is sent from going out embryo mouth 12.

Image captured by six high-speed industrial cameras imports embedded image processing platform into, owing to being inevitably subject to the interference of noise in shooting, transmitting procedure, so first the image that all cameras gather utilizes morphologic filtering to carry out image denoising process.Then for the ease of the detection of defect, binaryzation operation is carried out to all images, thus distinguish territory, bottle embryonic region, non-territory, bottle embryonic region and defect area.In order to evade the problem that brightness instability is brought, adopt the method for relative threshold binaryzation.Because target in bottleneck size, whorl of bottleneck top, bottle sealing ring three width image is positive circular in the picture, so first utilize the position of its shape facility to bottle preform mouth and screw thread to determine accurately before testing.Utilize RANSAC (RANdomSAmpleConsensus) circle fitting algorithm can obtain the positioning result that anti-noise ability is strong, precision is high.In addition, in order to improve accuracy of detection and stability, carry out analysis and distinguishing after the image that bottleneck size, whorl of bottleneck top and bottle sealing ring detect three cameras needs to carry out registration fusion simultaneously, adopt the method for registering images based on wavelet pyramid under the prerequisite ensureing registration accuracy, significantly can improve processing speed.Finally, each width image utilizes the Adaboost sorter after positive and negative sample training carry out defect recognition after acquisition characteristics, if recognition result is existing defects, then mark this bottle of embryo in software, rejected when this bottle of embryo moves to and reject mouth.

After the injection molded or crystallization of bottle embryo, advanced reasonable fetus device.After manager's embryo, the bottle embryo of marshalling is sent into the input rotating disk of pick-up unit by slideway, then detection rotating disk is sent to, detected by SPEED VISION system, unacceptable product by rejecting a mouthful automatic rejection, can detect outer to lack, in lack, run through, interiorly to lack, at the bottom of embryo and all types of defects of the bottle embryo such as body stain.

Embedded image processing platform 11 carries out judgement process to the image gathered, and as shown in Figure 2, its concrete steps are as described below for method:

Step 1: collecting bottle embryo side wall image, bottle embryo bottom diagram picture, whorl of bottleneck side view picture, bottleneck size image, whorl of bottleneck top graph picture and bottle sealing ring detected image pre-service.Because image inevitably exists noise in collection, transmitting procedure, so before carrying out processing and identification to image, first denoising will be carried out.Image is adopted based on morphologic Image filter arithmetic: opening operation can remove the relative bright distributed architecture that image does not match with the form of structural element, retains the part that those match simultaneously; Closed operation then can fill the relative dark distributed architecture that those images do not match with structural element, retains the part that those match simultaneously.Therefore they can be used for effectively extracting the level and smooth picture of characteristic sum.The opening operation of image and closed operation can be completed by combination image being carried out to etching operation and expansive working, if the image after image I etching operation is E (I), image after expansive working is D (I), then the opening operation of image can realize by first corroding the mode expanded afterwards, is namely completed by following formula:

OPEN(I)＝D(E(I))

Closed operation realizes by the mode of the post-etching that first expands, and is namely completed by following formula:

CLOSE(I)＝E(D(I))

For each sub-picture of input, first carry out the opening operation of suitable yardstick, the closed operation then carrying out more large scale obtains processing rear image, finally by process afterwards image and original image carry out and operation, obtain image after denoising.

Step 2: the details at utmost maintaining original image based on the image after morphologic image filtering, now carries out binary conversion treatment to image, is follow-up high speed processing key link.According to industry spot optical illumination darkness adjustment intensity of illumination, and adjust dynamic threshold, obtain the image of binaryzation, proceed to next step link respectively and detect.If certain point (x on image, y) gray-scale value is f (x, y), after binaryzation, the gray-scale value of this point is f ' (x, y), the pixel number needing ROI to be processed (area-of-interest, RegionOfInterest) in image is N, binaryzation is biased to δ, then binary-state threshold T is set as:

$T=\frac{\underset{(x,y)\&Element;ROI}{\&Sigma;}f(x,y)}{N}+\mathrm{\&delta;}$

$f^{\&prime;}(x,y)=\left\{\begin{array}{cc}0& f(x,y)<T\\ 255& f(x,y)\&GreaterEqual;T\end{array}\right.$

This approach simplify binarization method, reduce computation complexity, improve the speed of Image semantic classification.

Step 3: the image detected for bottleneck size, whorl of bottleneck top and bottle sealing ring positions process of fitting treatment, adopt RANSAC algorithm (RANSAC), flow process is as follows:

(1) may be bottle preform mouth edge point set P by edge detection method collection, its sample number:

∑P＞3；

(2) from P, randomly draw 3 points form subset S, calculate the round model M that these 3 points are determined;

(3) be less than with M distance in complementary set Sc=P-S, Sc to the point of set a distance t form in point set, interior point set and S jointly form the consistent of S and collect S ^*;

(4) number m is allowed, if ∑ (S for given minimum consistent collection sample ^*) > m, then utilize S by least square method ^*ask and obtain new model M ^*, return successfully; Otherwise judge whether iterations reaches designated value, if reach, return failure, otherwise return (2).

The conversion of bottle germ ring, is converted into polar coordinates rectangular coordinate original for binaryzation bottle germ ring image, so that extract feature.If certain point coordinate is (x, y) on original image, then after conversion, corresponding point coordinate is (r, θ), has:

$\left\{\begin{array}{c}r=\sqrt{x^2+y^2}\\ \mathrm{\&theta;}=\arctan (y/x)\end{array}\right.$

Quadratic linear interpolation:

By above-mentioned formula, each bottle preform mouth image is lath-like morphologies from annular Morphological Transitions, identifies for step 4 registration and step 5.

Step 4: registration operation is carried out to three width bottle preform mouth images.Owing to obtaining three width bottle preform mouth images during image acquisition, bottleneck size, whorl of bottleneck top and bottle sealing ring image respectively, but the information of this three width image can be assisted mutually, bottleneck size image can observe bottle preform mouth mouth plane simultaneously, effectively can improve accuracy of detection than the image deflects display of counterpart plane and sealing ring correspondence position and reduce false drop rate; And for example thread measurement needs bottleneck size image determination precise thread region, to avoid the generation of a large amount of flase drop.So method is that three width bottle preform mouth images are carried out image registration preferably, contact can be set up to make three width image informations and mutually utilize.And the method calculated amount of normal image registration is excessive, be not suitable for the time requirement of high-speed production lines on-line checkingi.So utilize wavelet pyramid to carry out image registration process in the present invention.Adopt wavelet transformation to set up image pyramid respectively to three width bottle preform mouth images, the bottom is original image, upwards successively carries out wavelet transformation and retains low frequency region, form Pyramid.Extract target size characteristic sum Corner Feature successively to mate, under the basis of this coupling enters pyramid, one deck mates, more by that analogy until enter bottom original image to complete final coupling.Because upper layer images resolution is low, so registration speed can be very fast, significantly improve image registration speed thus.

Step 5: based on the defect recognition of Adaboost sorter.By front 2 steps, obtain a bottle embryo side wall image, bottle embryo bottom diagram picture, whorl of bottleneck side view picture, provide front 4 steps to obtain the good bottle preform mouth image of three width registrations.Finally by sorter, these images are carried out defect recognition, whether qualified to differentiate corresponding bottle embryo.In order to improve the recognition accuracy of system, Adaboost sorter is adopted to classify.The feature wherein adopted in bottle embryo side wall image comprises the dark connected domain information in territory, smooth bottle embryonic region in ROI, comprises area, length, width, length breadth ratio, area girth ratio and area grayscale etc.The feature that bottle embryo bottom diagram picture adopts is similar to bottle embryo side wall image, but first can evade the interference of injection hole on the basis of low level.Bottle embryo sidewall thread imagery exploitation screw thread number, screw thread shadow length and width and flight pitch feature differentiate.Bottle preform mouth checks by column to the bar shaped bottle preform mouth on each image after registration, whether there is the situation that blackspot, black line or longitudinal width are inconsistent, with the breakage of this corresponding bottle embryo, crackle and circle etc. defect, outside feature is identical with bottle embryo sidewall features, also need the not flatness of extraction flask embryo pig's tongue line, the degree that corresponding mouth is not justified, and the out-of-shape degree of screw thread, corresponding screw thread is damaged.

Bottle embryo detection system of the present invention and method, bottle embryo common deficiency (as dirty in bottle preform mouth distortion, bottleneck breakage, the breakage of bottle germ ring, bottle germ wall, bottle preform mouth etc.) can be detected quickly and accurately, accuracy of detection is high, detection speed is fast, stability is high, efficiently solves the difficult problem that bottle embryo quick high accuracy detects.

The advantage of embedded bottle embryo detection system of the present invention is:

One, non-contact detecting.Adopt the mode of vision-based detection, to transparent vessel noncontact.Damage that container is caused and pollution can be avoided.

Two, accuracy of detection is high.Because this method imaging effect is good, it is little to disturb, stability is high, then by multiple camera parallel detection, accuracy of detection significantly promotes compared with classic method, meets the requirement of manufacturing enterprise to bottle embryo quality precision completely.

Three, detection speed is fast.Detection algorithm is simply effective, and calculated amount is little, thus computing velocity is fast.The detection speed being not less than 1200 bottles/minute can be supported, meet production line demand completely.

Four, production line loss and defective products rate can significantly be reduced.Achieve bottle blowing quality monitoring by the transition of manual detection to Intelligent Measurement, reduce liquid whole line loss consumption 5%, poor sealing rate declines 20%.

Five, can on-line checkingi on a production line.Due to the above-mentioned advantage of this method, add and can be installed on travelling belt easily, do not affect container transport, and debug convenient, so be very suitable for on-line checkingi.

The above is only to better embodiment of the present invention, not any pro forma restriction is done to the present invention, every any simple modification done above embodiment according to technical spirit of the present invention, equivalent variations and modification, all belong in the scope of technical solution of the present invention.

Claims (5)

1. an Embedded bottle embryo detection system, it is characterized in that: comprise reason fetus device (1), reason fetus device (1) is for carrying out arrangement reason embryo to the bottle embryo after injection molded, and bottle embryo is sent into input rotating disk (2), input rotating disk (2) is rotated and is driven bottle embryo to enter the first detection rotating disk (3), the first bottle embryo detecting rotating disk (3) transports track side and is provided with bottle embryo sidewall detection camera (5) and whorl of bottleneck side detection camera (6), first detects rotating disk (3) sends the bottle embryo on track into the second detection rotating disk (4), the second transport track side detecting rotating disk (4) is provided with bottleneck size detection camera (7), whorl of bottleneck top detection camera (8), bottle sealing ring detection camera (9), bottle embryo floor detection camera (10), all detection camera all connect embedded image processing platform (11), embedded image processing platform (11) image to camera calibration processes, distinguish defective bottle embryo, second track detecting rotating disk (4) is provided with rejects mouth (12), embryo mouth (13) is provided with out in track end, embedded image processing platform (11) carries out judgement process to the image gathered, distinguish defective bottle embryo, defective bottle embryo is rejected from rejecting mouth (13), normal bottle embryo is sent from going out embryo mouth (12).

2. according to Embedded bottle embryo detection system a kind of described in claim 1, it is characterized in that: described bottleneck size detection camera (7), whorl of bottleneck top detection camera (8), bottle sealing ring detection camera (9) adopt ring surface light source, and level is taken from the top down in bottle plumular axis line; Bottle embryo floor detection camera (10) level is taken from the top down in bottle plumular axis line; Bottle embryo sidewall detection camera (5), whorl of bottleneck side detection camera (6) are taken from the side perpendicular to bottle plumular axis line.

3. a detection method for Embedded bottle embryo detection system, is characterized in that: described embedded image processing platform 11 carries out judging that the method step of process is as follows to the image gathered:

Step 1: collecting bottle embryo side wall image, bottle embryo bottom diagram picture, whorl of bottleneck side view picture, bottleneck size image, whorl of bottleneck top graph picture and bottle sealing ring detected image, image is adopted and carries out denoising based on morphologic Image filter arithmetic, opening operation removes the relative bright distributed architecture that image does not match with the form of structural element, retains the part matched simultaneously; The relative dark distributed architecture that closed operation then blank map picture does not match with structural element, retains the part matched simultaneously;

Step 2: carry out binary conversion treatment to image, according to industry spot optical illumination darkness adjustment intensity of illumination, and adjusts dynamic threshold, obtains the image of binaryzation;

Step 3: the image detected for bottleneck size, whorl of bottleneck top and bottle sealing ring, positions process of fitting treatment, adopt RANSAC algorithm, flow process is as follows:

(1) may be bottle preform mouth edge point set P by edge detection method collection, its sample number:

ΣP＞3；

(2) from P, randomly draw 3 points form subset S, calculate the round model M that these 3 points are determined;

(3) be less than with M distance in complementary set Sc=P-S, Sc to the point of set a distance t form in point set, interior point set and S jointly form the consistent of S and collect S ^*;

(4) number m is allowed for given minimum consistent collection sample, if Σ (S*) > is m, then utilize S by least square method ^*ask and obtain new model M ^*, return successfully; Otherwise judge whether iterations reaches designated value, if reach, return failure, otherwise return (2).

The conversion of bottle germ ring, is converted into polar coordinates rectangular coordinate original for binaryzation bottle germ ring image, if certain point coordinate is (x, y) on original image, then after conversion, corresponding point coordinate is (r, θ), has:

$\left\{\begin{array}{c}r=\sqrt{x^2+y^2}\\ \mathrm{\&theta;}=\arctan (y/x)\end{array}\right.$

Image is made to be lath-like morphologies from annular Morphological Transitions;

Step 4: to bottleneck size, whorl of bottleneck top and bottle sealing ring image carry out registration operation, make three width image informations set up contact mutually to utilize, wavelet pyramid is utilized to carry out image registration process: to adopt wavelet transformation to set up image pyramid respectively, the bottom is original image, upwards successively carry out wavelet transformation and retain low frequency region, form Pyramid, extract target size characteristic sum Corner Feature successively to mate, under the basis of this coupling enters pyramid, one deck mates again, by that analogy until enter bottom original image to complete final coupling,

Step 5: to bottle embryo side wall image, bottle embryo bottom diagram picture, whether whorl of bottleneck side view picture, bottleneck size that registration is good, whorl of bottleneck top and bottle sealing ring image adopt Adaboost sorter to carry out defect recognition, qualified to differentiate corresponding bottle embryo.

4. according to a kind of described in claim 3 detection method of Embedded bottle embryo detection system, it is characterized in that: in described step 1, opening operation and closed operation are completed by combination image being carried out to etching operation and expansive working, if the image after image I etching operation is E (I), image after expansive working is D (I), then the opening operation of image realizes by first corroding the mode expanded afterwards, is namely completed by following formula:

OPEN(I)＝D(E(I))

Closed operation is realized by the mode of the post-etching that first expands, and is namely completed by following formula:

CLOSE(I)＝E(D(I))

For each sub-picture of input, first carry out the opening operation of suitable yardstick, the closed operation then carrying out more large scale obtains processing rear image, finally by process afterwards image and original image carry out and operation, obtain image after denoising.

5. according to a kind of described in claim 3 detection method of Embedded bottle embryo detection system, it is characterized in that: in described step 2, binary processing method is for establishing certain point (x on image, y) gray-scale value is f (x, y), and after binaryzation, the gray-scale value of this point is f ' (x, y), the pixel number needing ROI to be processed in image is N, and binaryzation is biased to δ, then binary-state threshold T is set as:

$T=\frac{\underset{(x,y)\&Element;ROI}{\&Sigma;}f(x,y)}{N}+\mathrm{\&delta;}$

$f^{\&prime;}(x,y)=\left\{\begin{array}{cc}0& f(x,y)<T\\ 255& f(x,y)\&GreaterEqual;T\end{array}\right..$