CN106529523A - Image restoration and serial number identification method for aluminum-plastic foamed mask package medicines on high-speed conveyor belt - Google Patents

Abstract

The invention relates to an image restoration and serial number identification method for aluminum-plastic foamed mask package medicines on a high-speed conveyor belt, relates to the image restoration and image identification field and solves an image fuzziness problem of the aluminum-plastic foamed mask package medicines on the high-speed conveyor belt. A method system comprises an industrial digital camera, a light source, a photoelectric proximity switch and a photoelectric angle encoder. The method comprises steps that two intelligible continuous target object images are acquired by the camera at a largest frame rate when the conveyor belt is in low speed motion; image motion of the two images is calculated through utilizing a combined transformation correlation algorithm, and an average image motion value is acquired through multiple experiments; the conveyor belt moves in high speed to increase the exposure time of the camera to acquire fuzzy medicine images; a point diffusion function of the fuzzy medicine images is derived through utilizing the previously acquired average image motion value; image restoration is carried out through utilizing a Lucy-Richardson algorithm to acquire intelligible medicine images; serial number identification is carried out through utilizing a support vector machine method. The method is advantaged in that the image restoration effect is good, and identification accuracy is high.

Description

On a kind of High Speed Transfer band, the image restoration of aluminium-plastic blister package drugs and sequence number are known Other method

Technical field

The present invention relates to motion blur image restoration and field of image recognition, and in particular to plastic-aluminum on a kind of High Speed Transfer band The image restoration and sequence number recognition methods of blister package medicine.

Background technology

Plate containing medicines typically adopt aluminium-plastic bubble plate packing, and this packaged form advantage is more, wherein being some blister package aluminium Paper tinsel surface is printed on explanatory note and sequence number, improves the security of dispensing.

In process of production, medicine sequence number acts primarily as mark action.According to product batch number and the corresponding record of production, can To review the history of this batch of products material source, medicine forming process；After medicine forms finished product, according to sales figure, can be with Review the market whereabouts of medicine；The quality condition that medicine is entered behind market；Can control when needing or reclaim this batch of medicine Product.So will record to medicine sequence number during pharmaceutical production, packaging and dispensing etc., manual record mode efficiency Low, high labor intensive, is just replaced by machine vision technique at present step by step, replaces manual identified with the mode of image recognition.

Pharmaceutical production, packaging and dispensing etc. are usually to carry out on a moving belt, and this just brings difficulty for machine vision method Degree, because the image when speed of moving body is too fast relative to camera frame per second captured by camera will occur that motion blur is existing As.Existing method typically adopts high frame per second camera in the method for maximum frame per second collection image to avoid problem of image blurring.But It is to need extraneous additional light source to increase brightness of image when camera gathers image with maximum frame per second, otherwise image is partially dark or even complete It is black, but strong illumination can not be adopted on some special leechdom production lines, and extraneous polishing easily causes plastic-aluminum surface Mirror-reflection is so as to affecting IMAQ quality.

The content of the invention

The present invention is to solve the problems, such as that image blurring in medicine IMAQ and medicine obtains hardly possible in real time on high-speed conveyor, The image restoration and sequence number recognition methods of aluminium-plastic blister package drugs on a kind of High Speed Transfer band are provided.The present invention is by calculating Point spread function reflex grand master pattern poultice product image, beats high light when solving the problems, such as shooting using the method for increasing the time for exposure, And the sequence number on medicine image is recognized by support vector machine method.Comprise the following steps that：

(1) industrial digital camera and light source etc. are built on medicine conveyer belt, completes image capturing system, camera and transmission Band Relative vertical, installs photoelectric proximity switch and realizes that camera is triggered when medicine is occurred under camera lens obtains image, and pacify Dress photoelectric angular encoder is used for the movement velocity for measuring conveyer belt.

(2) conveyer belt low speed V₁Motion, camera are adjusted to maximum frame per second i.e. time for exposure P₁, camera exposure time and conveyer belt Movement velocity meets below equation：

Wherein K is camera pixel size, and β is image enlargement ratio, in this case camera shooting clear image, is opened Light intensity light source, photoelectric proximity switch are closed, camera outer signal generator, and signal generator produces trigger of the cycle for T Triggering camera obtains two continuous target picture rich in detail I₁, I₂, adjacent two targets are calculated using joint transform related algorithm Body image slices shifting amount Δ x, Δ y,

(Δ x, Δ y)=JTC (I₁, I₂)

Wherein JTC is the related computational algorithm of joint transform, and joint transform is related to have good noise robustness, for The image of low signal-to-noise ratio can also obtain the image motion measurement of sub-pixel precision, obtain low when being adjusted to very low between solving upon exposure The problem of signal-to-noise ratio image.The mean value of repeat step (2) n computations picture shifting amountWithFormula is as follows：

(3) conveyer belt is with speed V during normal work₂Motion, weakens light source, increase camera exposure time to P₂, camera exposure Time should be adjusted to the half that the image intensity value captured by camera is maximum gradation value, open photoelectric proximity switch.Now because The increase with the camera exposure time is improved for conveyer belt speed, the aluminium-plastic blister package drugs image that camera is photographed is mould Paste image I_b。

(4) step (3) middle mold poultice product image I is extrapolated according to the picture shifting amount mean value obtained in step (2)_bPoint expand Scattered function h (x, y), computing formula are as follows：

Wherein L is that blurred length and θ are fuzzy angle, if blurred length is more than 10 pixels, often beyond recovery The limit, now can suitably reduce time for exposure P₂.Lucy-Richardson algorithm restored images are finally utilized, clear aluminium is obtained Modeling blister package medicine image.

(5) on the clear aluminium-plastic blister package drugs image obtained using the method identification step (4) of SVMs Sequence number, in order to preferably extract feature, first has to carry out character normalization, and the feature for sequence number character recognition has periphery Profile and thick meshed feature, realize disturbing training sample by Bagging (Bootstrap Aggregating) algorithm afterwards It is dynamic, so as to produce the base grader with otherness, then carry out the integrated study of base grader, finally with the support for training to Amount machine grader recognizes the sequence number on clear aluminium-plastic blister package drugs image.

Beneficial effects of the present invention：The present invention a kind of High Speed Transfer band on aluminium-plastic blister package drugs image restoration and Sequence number recognition methods, only make use of an industrial camera just to complete image restoration, and is not required in medicine shooting process Intense light source is wanted, solves the problems, such as that some special medicines can not be easily caused mirror-reflection by strong illumination and polishing.The present invention Image collecting device it is simple, image restoration effect is good, and medicine comment on aluminium-plastic blister package drugs etc. is clearly presented, And by the information of support vector machine method identification sequence number content obtaining medicine.

Description of the drawings

Fig. 1 is the image restoration of aluminium-plastic blister package drugs and sequence number identification side on a kind of High Speed Transfer band of the present invention The flow chart of method；

Fig. 2 be joint transform related algorithm in set up joint image schematic diagram；

Fig. 3 is the total system assembling schematic diagram of the present invention；

Fig. 4 is SVMs identification sequence number flow chart.

Specific implementation step

With reference to the accompanying drawings and in conjunction with the embodiments to aluminium-plastic blister package drugs on a kind of High Speed Transfer band of the present invention Image restoration and sequence number recognition methods are described further.

Step one：Referring to Fig. 3, industrial digital camera (1) and light source (2) etc. are built on medicine conveyer belt (5), figure is completed As acquisition system, camera (1) and conveyer belt (5) Relative vertical, photoelectric proximity switch (3) is installed and is realized when medicine occurs in camera (1) image is obtained when under camera lens, and photoelectric angular encoder (4) is installed for measuring the movement velocity of conveyer belt (5).

Step 2：The time for exposure of industrial camera (1) is reduced as far as possible to limit time P₁, i.e. camera (1) reaches largest frames Rate, increasing light source (2) brightness makes target (6) blur-free imaging on conveyer belt (3), opens conveyer belt (5) telecontrol equipment, now passes Band (5) is sent with low speed V₁Motion, camera (1) time for exposure and conveyer belt (5) movement velocity meet below equation：

Wherein K is camera pixel size, and β is image enlargement ratio, and camera can be with shooting clear target in this case Object image, photoelectric proximity switch (3) are closed, camera (1) outer signal generator, and signal generator produces triggering of the cycle for T Signal triggering camera (1) obtains two continuous target (6) picture rich in detail I₁, I₂, calculate adjacent using joint transform related algorithm Two target object image slices shifting amount Δ x, Δ y,

(Δ x, Δ y)=JTC (I₁, I₂)

Wherein JTC is joint transform related algorithm, and method is as follows：

It is assumed that target image I₂(x, y) is relative to reference picture I₁The displacement of (x, y) on x, y direction is Δ x, Δ respectively The center of the two as shown in Figure 2, is positioned over (- a, 0) with (a, 0), obtaining joint input image j (x, y) is by y：

J (x, y)=I₁(x+a, y)+I₂(x-a, y)=I₁(x+a, y)+I₁(x-a- Δ x, y- Δ y)

A Fourier transformation is carried out to which and its power spectrum is taken, calculation is derived by equation below：

| J (u, v) |²=2 | I '₁(u, v) |²+2|I′₁(u, v) |²cos[2π(Δx+2a)u+2πΔyv]

I ' in the formula₁(u, v) is I₁The Fourier transformation of (x, y), | * | are modulus computing.Fourier is done again to power spectrum Conversion, obtains joint correlation output：

C (x, y)=2C_rr(x, y)+C_rr(x+2a+ Δ x, y+ Δ y)+C_rr(x-2a- Δ x, y- Δ y)

C in the formula_rr(x, y) is I₁The auto-correlation function of (x, y), C_rrThe peak value of (x, y) be located at (0,0), and C_rr(x+2a+ Δ x, y+ Δ y) and C_rr(peak value of x-2a- Δ x, y- Δ y) respectively be located at (- 2a- Δ x ,-Δ y) and (2a+ Δ x, Δ y), according to C_rr(x+2a+ Δ x, y+ Δ y) and C_rr(two peak relations of x-2a- Δ x, y- Δ y) can just obtain target image phase For the displacement of reference picture, (Δ x, Δ y), i.e., picture between the two are moved, and the position coordinates of two correlation peaks passes through matter Center algorithm is calculated, and joint transform is related to have good noise robustness, can also obtain Asia for the image of low signal-to-noise ratio The image motion measurement of pixel precision, obtains low signal-to-noise ratio (SNR) images when being adjusted to very low between solving the problems, such as upon exposure well. The mean value of repeat step (2) n computations picture shifting amountWithFormula is as follows：

Step 3：Conveyer belt (5) is with speed V during normal work₂Motion, weakens light source (2), when increase camera (1) exposes Between arrive P₂, camera (1) time for exposure should be adjusted to the half that the image intensity value captured by camera (1) is close to maximum gradation value, Open photoelectric proximity switch (3).Now because conveyer belt (5) movement velocity improves the increase with camera (1) time for exposure, camera (1) aluminium-plastic blister package drugs (6) image for photographing is blurred picture I_b。

Step 4：Picture shifting amount mean value according to obtaining in step 2 extrapolates step 3 middle mold poultice product image I_bPoint Spread function h (x, y), computing formula are as follows：

Wherein L is that blurred length and θ are fuzzy angle, afterwards using Lucy-Richardson algorithm restored images, Lucy-Richardson algorithms assume that image obeys Poission distributions, are estimated using maximum likelihood, are that one kind is based on The iterative algorithm of Bayesian analysis, which estimates that iterative equation is as follows：

Wherein I_bFormer blurred picture is represented, g represents restored image, and h represents point spread function, and k represents iterations.Utilize MATLAB software rejuvenation images, then h=fspecial (' motion ', L, θ), g=deconvlucy (I_b, h, 5).This just completes Recovery operation to aluminum-plastic blister medicine image on conveyer belt.

Step 5：Using on the clear aluminium-plastic blister package drugs image that the method identification step four of SVMs is obtained Sequence number, in order to preferably extract feature, first have to carry out character normalization, the feature for sequence number character recognition has outer Profile and thick meshed feature are enclosed, is realized to training sample by Bagging (Bootstrap Aggregating) algorithm afterwards Disturbance, so as to produce the base grader with otherness, then carries out the integrated study of base grader, finally with the support for training Vector machine classifier recognizes the sequence number on clear aluminium-plastic blister package drugs image.

Claims (4)

1. on a kind of High Speed Transfer band aluminium-plastic blister package drugs image restoration and sequence number recognition methods, it is characterised in that bag Include following steps：

Step 1, image collecting device is built on medicine conveyer belt (5), image collecting device includes an industrial digital camera (1) and light source (2), camera (1) and conveyer belt (5) Relative vertical, install photoelectric proximity switch (3) and realize when medicine (6) appearance Camera (1) synchronous photo taking is triggered when under camera (1) camera lens, and photoelectric angular encoder (4) is installed for measuring conveyer belt (5) fortune Dynamic speed；

Step 2, conveyer belt (5) low speed V₁Motion, camera (1) are adjusted to maximum frame per second i.e. time for exposure P₁, light intensity light source (2) is opened, Photoelectric proximity switch (3) is closed, camera (1) outer signal generator, and signal generator produces the cycle for the trigger triggering of T Camera (1) obtains two continuous target (6) pictures rich in detail, calculates between image as shifting amount, step 2 is repeated several times and tries to achieve image The mean value of x and y directions picture shifting amount is respectivelyWith；

Step 3, conveyer belt (5) are with speed V during normal work₂Motion, weakens light source (2), increase camera (1) time for exposure to P₂, Half of the medicine image intensity value captured by camera (1) for maximum gradation value, opens photoelectric proximity switch (3), obtains plastic-aluminum The blurred picture of blister package medicine；

Step 4, the point spread function that step 3 middle mold poultice product image is extrapolated according to the picture shifting amount obtained in step 2, restored map As obtaining clear aluminium-plastic blister package drugs image；

Sequence on step 5, the clear aluminium-plastic blister package drugs image obtained using the method identification step 4 of SVMs Number.

2. on a kind of High Speed Transfer band according to claim 1, the image restoration of aluminium-plastic blister package drugs and sequence number are known Other method, it is characterised in that：In described step 2, target is five great circle targets, is calculated between adjacent image as shifting amount method For joint transform related algorithm.

3. on a kind of High Speed Transfer band according to claim 1, the image restoration of aluminium-plastic blister package drugs and sequence number are known Other method, it is characterised in that：Described step 4 middle mold poultice product image point spread function computing formula is：

$L=\sqrt{{\left(m\stackrel{\‾}{\mathrm{\Δ}x}\right)}^2+{\left(m\stackrel{\‾}{\mathrm{\Δ}y}\right)}^2}$

$\mathrm{\θ}={\tan }^{-1}\frac{\stackrel{\‾}{\mathrm{\Δ}y}}{\stackrel{\‾}{\mathrm{\Δ}x}}$

$m=\frac{V_2\×P_2}{V_1\×T}$

Wherein L is blurred length, and θ is fuzzy angle, and h (x, y) is point spread function, and the method for restored image adopts Lucy- Richardson algorithms.

4. on a kind of High Speed Transfer band according to claim 1, the image restoration of aluminium-plastic blister package drugs and sequence number are known Other method, it is characterised in that：In described step 5, support vector machine method includes：Character normalization, feature extraction, support to The training of amount machine grader, the identification of support vector machine classifier.