CN104867144A - IC element solder joint defect detection method based on Gaussian mixture model - Google Patents

Abstract

The invention discloses an IC element solder joint defect detection method based on a Gaussian mixture model. The method comprises following steps: initializing template quantities, weights, the mean value and the variance of the Gaussian mixture model, and establishing a frequency distribution graph according to the height and width of a training picture; acquiring an IC solder joint training picture from a training sample, and updating the Gaussian mixture model and the frequency distribution graph; determining whether the training sample finishes training, and calculating a defect degree threshold of the training sample if the training sample finishes training; collecting a picture of an IC element solder joint to be detected, then calculating the defect degree of the picture with the cooperation of the trained Gaussian mixture model and the frequency distribution graph; and comparing the defect degree of the picture and the defect degree threshold of the training sample so as to obtain a detection result of the IC element solder joint. The method is less in calculation amount, rapid in detection speed and high in accuracy, can effectively detect insufficient solder defects of the IC element solder joint, and can be widely applied to the field of IC element solder joint defect detection.

Description

Based on the IC element welding point defect detection method of mixed Gauss model

Technical field

The present invention relates to Digital Image Processing application, particularly based on the IC element welding point defect detection method of mixed Gauss model.

Background technology

Printing board PCB (Printed Circuit Board) defects detection is the focus direction that automatic optics inspection (automaticoptical inspection, AOI) is applied, and is more and more paid close attention in recent years.The main mode adopted detects the laggard row relax of image of PCB element by CCD thus realizes defects detection at present.In actual use, the situation that printed circuit board runs into is very complicated, often there is change in various degree and irregular phenomenon in the PCB part drawing picture that CCD collects, such as: intensity of illumination is uneven, lighting angle changes, and the image of CCD camera collection has the deflection of certain angle, and component size is more and more less, in pcb board, component density is increasing etc., and these problems make PCB welding point defect detect and become quite difficulty.And the size of IC element solder joint is much smaller than the size of general CHIP element solder joint, rosin joint and normal solder joint closely similar on image, this make the rosin joint of IC element solder joint detect be the difficult problem being difficult in defects detection capture always.

The existing comparatively ripe great majority of the detection method to IC element welding point defect are the method for feature based.Defects detection is divided into two steps by this method: extract characteristic sum classification.At extraction feature stage, select color gradient, region area, girth, hydraulic radius etc. characteristic feature; At sorting phase, select comparatively ripe sorter, such as neural network, AdaBoost, SVM etc., the feature extracted is classified.These methods achieve good effect at CHIP element solder joint.But because IC element welding spot size is little, solder joint closeness is large, and rosin joint solder joint sample is difficult to collect, and makes the current method based on sorter be difficult to obtain good classifying quality in IC element solder joint rosin joint detects.In addition, online test method is strict to time requirement, and these comparatively ripe classifier calculated amounts of neural network are large, are difficult to meet on-line monitoring requirement in time.Simultaneously, although also someone proposes the IC element solder joint detection method of the pixel modeling based on single Gauss model, and the detection speed of this method is fast, and accuracy rate is low, cannot apply in actual production.Generally speaking, current detection method cannot detect the welding point defect of IC element effectively, accurately and rapidly.

Summary of the invention

In order to solve above-mentioned technical matters, the object of this invention is to provide the IC element welding point defect detection method based on mixed Gauss model.

The technical solution adopted for the present invention to solve the technical problems is:

Based on the IC element welding point defect detection method of mixed Gauss model, comprising:

The template number of S1, initialization mixed Gauss model, weight, average and variance, set up histogram according to the height of training picture and width simultaneously;

S2, obtain from training sample IC solder joint training picture, mixed Gauss model and histogram are upgraded;

It is complete whether S3, training of judgement sample have trained, if the degree of imperfection threshold value of then calculation training sample, otherwise returns and perform step S2;

S4, gather IC element solder joint to be detected picture after, the mixed Gauss model that combined training is good and histogram calculate the degree of imperfection of this picture;

S5, the degree of imperfection threshold value of the degree of imperfection of this picture and training sample is compared after obtain the testing result of IC element solder joint.

Further, described step S1, comprising:

S11, the template number of mixed Gauss model is initialized as M, wherein M is natural number and 3≤M≤5;

S12, the weight of M template is all initialized as 1/M, and the variance of M template, as the average of M template, is initialized as particular value V by Stochastic choice numeral simultaneously from [0,255] interval ₀, wherein 3≤V ₀≤ 10;

S13, set up matrix that a size is H*W and after all elements value of this matrix is initialized as 1, using this matrix as histogram, wherein H represents the height of training picture, and W represents the width of training picture.

Further, described step S2, comprising:

S21, obtain from training sample IC solder joint training picture, for each pixel of this training picture, it is carried out matching treatment with M Gauss model respectively, and obtains Gauss's matching value of this pixel and each Gauss model;

S22, recurrence renewal is carried out to mixed Gauss model;

S23, calculate the weight of M corresponding to this pixel Gauss model and the ratio of standard deviation respectively, and according to ratio, M Gauss model is sorted;

S24, select according to following formula before B Gauss model point as a setting, and then judge whether this pixel mates with a front B Gauss model, if so, judge that this pixel is background dot, otherwise judge that this pixel is foreground point:

$B=\arg \underset{b}{\min }(\underset{i=1}{\overset{b}{\mathrm{\Σ}}}{w}_i>\mathrm{Th})$

In above formula, Th is default background threshold and 0.75≤Th≤0.90, and i represents sequence number, w ⁱrepresent the weight of i-th Gauss model;

S25, binaryzation assignment is carried out to this pixel, and renewal frequency distribution plan;

S26, travel through all pixels after obtain the binary image of this training picture.

Further, described step S21, it is specially:

IC solder joint training picture is obtained from training sample, for each pixel of this training picture, it is carried out matching treatment with a corresponding M Gauss model respectively, judge whether to meet following condition, if so, then judge that this pixel mates with corresponding Gauss model, both Gauss's matching values are 1, otherwise both do not mate, Gauss's matching value is 0:

|X _k-μ _i|<2.5σ _i

In above formula, X _krepresent the pixel value of this pixel, μ _irepresent the average of i-th Gauss model, σ _irepresent the variance of i-th Gauss model.

Further, described step S22, comprising:

S221, according to following formula, recurrence renewal is carried out to the weight of mixed Gauss model:

w _i,k＝(1-α)*w _i,k-1+α*H _i,k

S222, to Gauss model unmatched with this pixel, directly perform step S223, to the Gauss model mated with this pixel, after its average and standard deviation being upgraded according to following formula, terminate upgrade:

$\left\{\begin{array}{c}{\mathrm{\&mu;}}_{i,k}=(1-\mathrm{\&alpha;})*{\mathrm{\&mu;}}_{i,k-1}+\mathrm{\&alpha;}*H_{i,k}\\ {\mathrm{\&sigma;}}_{i,k}^2=(1-\mathrm{\&alpha;})*{\mathrm{\&sigma;}}_{i,k-1}^2+\mathrm{\&alpha;}*H_{i,k}\end{array}\right.$

S223, judge that M the Gauss model whether this pixel is corresponding with it does not all mate, if so, then replace the minimum Gauss model of weighted value according to the following formula, otherwise terminate to upgrade:

$\left\{\begin{array}{c}{w}_{\mathrm{sl},k}=1/M\\ {\mathrm{\&mu;}}_{\mathrm{sl},k}=X_k\\ {\mathrm{\&sigma;}}_{\mathrm{sl},k}=V_0\end{array}\right.$

Wherein, w _i,k, μ _i,kwith represent the weight of i-th Gauss model of this pixel, average and variance respectively, w _{i, k-1}, μ _{i, k-1}with represent the weight of i-th Gauss model of the pixel of a training picture and this pixel correspondence position, average and variance respectively, H _i,krepresent Gauss's matching value of this pixel and i-th Gauss model, α represents default learning efficiency value, w _{sl, k}, μ _{sl, k}and σ _{sl, k}represent the weight of the Gauss model that weighted value is minimum, average and standard deviation respectively.

Further, described step S25, it is specially:

If this pixel is background dot, is then 0 by its assignment, otherwise its assignment is 255 and the value of the histogram of this pixel correspondence position is added 1.

Further, the degree of imperfection threshold value of calculation training sample described in described step S3, it is specially:

According to following formula respectively each training picture of calculation training sample degree of imperfection and obtain the degree of imperfection threshold value of maximal value as training sample:

$E_m=\frac{1}{\underset{i=1}{\overset{H}{\mathrm{\&Sigma;}}}\underset{j=1}{\overset{W}{\mathrm{\&Sigma;}}}\frac{m^2}{f{(x,y)}^2}}\&CenterDot;\underset{i=1}{\overset{H}{\mathrm{\&Sigma;}}}\underset{j=1}{\overset{W}{\mathrm{\&Sigma;}}}\frac{m^2}{f^2(x,y)\&CenterDot;b(x,y)}$

In above formula, E _mrepresent the degree of imperfection of training picture, m represents the training picture number of training sample, and f (x, y) represents histogram, and b (x, y) represents the binary image of training picture.

Further, described step S4, comprising:

S41, gather IC element solder joint to be detected picture after, for each pixel of this picture, calculate the weight of M corresponding to this pixel Gauss model and the ratio of standard deviation respectively, and according to ratio, M Gauss model sorted;

S42, select according to following formula before B Gauss model point as a setting, and then judge whether this pixel mates with a front B Gauss model, if so, judge that this pixel is background dot, otherwise judge that this pixel is foreground point:

$B=\arg \underset{b}{\min }(\underset{i=1}{\overset{b}{\mathrm{\&Sigma;}}}{w}_i>\mathrm{Th})$

In above formula, Th is default background threshold 0.75≤Th≤0.90, and i represents sequence number, w ⁱrepresent the weight of i-th Gauss model;

S43, binaryzation assignment is carried out to this pixel;

S44, travel through all pixels after obtain the binary image of this picture;

S45, calculate the degree of imperfection of this picture according to following formula:

$E_m=\frac{1}{\underset{i=1}{\overset{H}{\mathrm{\&Sigma;}}}\underset{j=1}{\overset{W}{\mathrm{\&Sigma;}}}\frac{m^2}{f{(x,y)}^2}}\&CenterDot;\underset{i=1}{\overset{H}{\mathrm{\&Sigma;}}}\underset{j=1}{\overset{W}{\mathrm{\&Sigma;}}}\frac{m^2}{f^2(x,y)\&CenterDot;b(x,y)}$

In above formula, E _mrepresent the degree of imperfection of picture, m represents the training picture number of training sample, and f (x, y) represents histogram, and b (x, y) represents the binary image of this picture.

Further, described step S5, it is specially:

Judge whether this degree of imperfection is greater than the degree of imperfection threshold value of training sample, if so, then judge that this IC element solder joint is rosin joint solder joint, otherwise, judge that this IC element solder joint is normal solder joint.

The invention has the beneficial effects as follows: the IC element welding point defect detection method that the present invention is based on mixed Gauss model, comprise: the template number of initialization mixed Gauss model, weight, average and variance, set up histogram according to the height of training picture and width simultaneously; From training sample, obtain IC solder joint training picture, mixed Gauss model and histogram are upgraded; It is complete whether training of judgement sample has trained, if the degree of imperfection threshold value of then calculation training sample; After gathering the picture of IC element solder joint to be detected, the mixed Gauss model that combined training is good and histogram calculate the degree of imperfection of this picture; The testing result of IC element solder joint is obtained after the degree of imperfection threshold value of the degree of imperfection of this picture and training sample being compared.Compared to existing technology, computation amount, detection speed is fast, and accuracy rate is high, effectively can detect the rosin joint defect of IC element solder joint for this method.

Accompanying drawing explanation

Below in conjunction with drawings and Examples, the invention will be further described.

Fig. 1 is the process flow diagram of the IC element welding point defect detection method based on mixed Gauss model of the present invention.

Embodiment

With reference to Fig. 1, the invention provides a kind of IC element welding point defect detection method based on mixed Gauss model, comprising:

The template number of S1, initialization mixed Gauss model, weight, average and variance, set up histogram according to the height of training picture and width simultaneously;

S2, obtain from training sample IC solder joint training picture, mixed Gauss model and histogram are upgraded;

It is complete whether S3, training of judgement sample have trained, if the degree of imperfection threshold value of then calculation training sample, otherwise returns and perform step S2;

S4, gather IC element solder joint to be detected picture after, the mixed Gauss model that combined training is good and histogram calculate the degree of imperfection of this picture;

S5, the degree of imperfection threshold value of the degree of imperfection of this picture and training sample is compared after obtain the testing result of IC element solder joint.

Be further used as preferred embodiment, described step S1, comprising:

S11, the template number of mixed Gauss model is initialized as M, wherein M is natural number and 3≤M≤5;

S12, the weight of M template is all initialized as 1/M, and the variance of M template, as the average of M template, is initialized as particular value V by Stochastic choice numeral simultaneously from [0,255] interval ₀, wherein 3≤V ₀≤ 10;

S13, set up matrix that a size is H*W and after all elements value of this matrix is initialized as 1, using this matrix as histogram, wherein H represents the height of training picture, and W represents the width of training picture.

Be further used as preferred embodiment, described step S2, comprising:

S21, obtain from training sample IC solder joint training picture, for each pixel of this training picture, it is carried out matching treatment with M Gauss model respectively, and obtains Gauss's matching value of this pixel and each Gauss model;

S22, recurrence renewal is carried out to mixed Gauss model;

S23, calculate the weight of M corresponding to this pixel Gauss model and the ratio of standard deviation respectively, and according to ratio, M Gauss model is sorted;

S24, select according to following formula before B Gauss model point as a setting, and then judge whether this pixel mates with a front B Gauss model, if so, judge that this pixel is background dot, otherwise judge that this pixel is foreground point:

$B=\arg \underset{b}{\min }(\underset{i=1}{\overset{b}{\mathrm{\&Sigma;}}}{w}_i>\mathrm{Th})$

In above formula, Th is default background threshold and 0.75≤Th≤0.90, and i represents sequence number, w ⁱrepresent the weight of i-th Gauss model;

S25, binaryzation assignment is carried out to this pixel, and renewal frequency distribution plan;

S26, travel through all pixels after obtain the binary image of this training picture.

Be further used as preferred embodiment, described step S21, it is specially:

IC solder joint training picture is obtained from training sample, for each pixel of this training picture, it is carried out matching treatment with a corresponding M Gauss model respectively, judge whether to meet following condition, if so, then judge that this pixel mates with corresponding Gauss model, both Gauss's matching values are 1, otherwise both do not mate, Gauss's matching value is 0:

|X _k-μ _i|<2.5σ _i

In above formula, X _krepresent the pixel value of this pixel, μ _irepresent the average of i-th Gauss model, σ _irepresent the variance of i-th Gauss model.

Be further used as preferred embodiment, described step S22, comprising:

S221, according to following formula, recurrence renewal is carried out to the weight of mixed Gauss model:

w _i,k＝(1-α)*w _i,k-1+α*H _i,k

S222, to Gauss model unmatched with this pixel, directly perform step S223, to the Gauss model mated with this pixel, after its average and standard deviation being upgraded according to following formula, terminate upgrade:

$\left\{\begin{array}{c}{\mathrm{\&mu;}}_{i,k}=(1-\mathrm{\&alpha;})*{\mathrm{\&mu;}}_{i,k-1}+\mathrm{\&alpha;}*H_{i,k}\\ {\mathrm{\&sigma;}}_{i,k}^2=(1-\mathrm{\&alpha;})*{\mathrm{\&sigma;}}_{i,k-1}^2+\mathrm{\&alpha;}*H_{i,k}\end{array}\right.$

S223, judge that M the Gauss model whether this pixel is corresponding with it does not all mate, if so, then replace the minimum Gauss model of weighted value according to the following formula, otherwise terminate to upgrade:

$\left\{\begin{array}{c}{w}_{\mathrm{sl},k}=1/M\\ {\mathrm{\&mu;}}_{\mathrm{sl},k}=X_k\\ {\mathrm{\&sigma;}}_{\mathrm{sl},k}=V_0\end{array}\right.$

Wherein, w _i,k, μ _i,kwith represent the weight of i-th Gauss model of this pixel, average and variance respectively, w _{i, k-1}, μ _{i, k-1}with represent the weight of i-th Gauss model of the pixel of a training picture and this pixel correspondence position, average and variance respectively, H _i,krepresent Gauss's matching value of this pixel and i-th Gauss model, α represents default learning efficiency value, w _{sl, k}, μ _{sl, k}and σ _{sl, k}represent the weight of the Gauss model that weighted value is minimum, average and standard deviation respectively.

Be further used as preferred embodiment, described step S25, it is specially:

If this pixel is background dot, is then 0 by its assignment, otherwise its assignment is 255 and the value of the histogram of this pixel correspondence position is added 1.

Be further used as preferred embodiment, the degree of imperfection threshold value of calculation training sample described in described step S3, it is specially:

According to following formula respectively each training picture of calculation training sample degree of imperfection and obtain the degree of imperfection threshold value of maximal value as training sample:

$E_m=\frac{1}{\underset{i=1}{\overset{H}{\mathrm{\&Sigma;}}}\underset{j=1}{\overset{W}{\mathrm{\&Sigma;}}}\frac{m^2}{f{(x,y)}^2}}\&CenterDot;\underset{i=1}{\overset{H}{\mathrm{\&Sigma;}}}\underset{j=1}{\overset{W}{\mathrm{\&Sigma;}}}\frac{m^2}{f^2(x,y)\&CenterDot;b(x,y)}$

In above formula, E _mrepresent the degree of imperfection of training picture, m represents the training picture number of training sample, and f (x, y) represents histogram, and b (x, y) represents the binary image of training picture.

Be further used as preferred embodiment, described step S4, comprising:

S41, gather IC element solder joint to be detected picture after, for each pixel of this picture, calculate the weight of M corresponding to this pixel Gauss model and the ratio of standard deviation respectively, and according to ratio, M Gauss model sorted;

S42, select according to following formula before B Gauss model point as a setting, and then judge whether this pixel mates with a front B Gauss model, if so, judge that this pixel is background dot, otherwise judge that this pixel is foreground point:

$B=\arg \underset{b}{\min }(\underset{i=1}{\overset{b}{\mathrm{\&Sigma;}}}{w}_i>\mathrm{Th})$

In above formula, Th is default background threshold 0.75≤Th≤0.90, and i represents sequence number, w ⁱrepresent the weight of i-th Gauss model;

S43, binaryzation assignment is carried out to this pixel;

S44, travel through all pixels after obtain the binary image of this picture;

S45, calculate the degree of imperfection of this picture according to following formula:

$E_m=\frac{1}{\underset{i=1}{\overset{H}{\mathrm{\&Sigma;}}}\underset{j=1}{\overset{W}{\mathrm{\&Sigma;}}}\frac{m^2}{f{(x,y)}^2}}\&CenterDot;\underset{i=1}{\overset{H}{\mathrm{\&Sigma;}}}\underset{j=1}{\overset{W}{\mathrm{\&Sigma;}}}\frac{m^2}{f^2(x,y)\&CenterDot;b(x,y)}$

In above formula, E _mrepresent the degree of imperfection of picture, m represents the training picture number of training sample, and f (x, y) represents histogram, and b (x, y) represents the binary image of this picture.

Be further used as preferred embodiment, described step S5, it is specially:

Judge whether this degree of imperfection is greater than the degree of imperfection threshold value of training sample, if so, then judge that this IC element solder joint is rosin joint solder joint, otherwise, judge that this IC element solder joint is normal solder joint.

Below in conjunction with a specific embodiment, the invention will be further described.

With reference to Fig. 1, a kind of IC element welding point defect detection method based on mixed Gauss model, comprising:

The template number of S1, initialization mixed Gauss model, weight, average and variance, set up histogram according to the height of training picture and width simultaneously; This step specifically comprises step S11 ~ S13:

S11, the template number of mixed Gauss model is initialized as M, wherein M is natural number and 3≤M≤5; Therefore in the present embodiment, mixed Gauss model has M Gauss model;

S12, the weight of M template is all initialized as 1/M, and the variance of M template, as the average of M template, is initialized as particular value V by Stochastic choice numeral simultaneously from [0,255] interval ₀, wherein 3≤V ₀≤ 10;

S13, set up matrix that a size is H*W and after all elements value of this matrix is initialized as 1, using this matrix as histogram, wherein H represents the height of training picture, and W represents the width of training picture.

S2, obtain from training sample IC solder joint training picture, mixed Gauss model and histogram are upgraded; Concrete renewal process comprises step S21 ~ S26:

S21, obtain from training sample IC solder joint training picture, for each pixel of this training picture, it is carried out matching treatment with M Gauss model respectively, and obtains Gauss's matching value of this pixel and each Gauss model, it is specially:

IC solder joint training picture is obtained from training sample, for each pixel of this training picture, it is carried out matching treatment with a corresponding M Gauss model respectively, judge whether to meet following condition, if so, then judge that this pixel mates with corresponding Gauss model, both Gauss's matching values are 1, otherwise both do not mate, Gauss's matching value is 0:

|X _k-μ _i|<2.5σ _i

In above formula, X _krepresent the pixel value of this pixel, μ _irepresent the average of i-th Gauss model, σ _irepresent the variance of i-th Gauss model.

S22, recurrence renewal is carried out to mixed Gauss model, comprises step S221 ~ S223:

S221, according to following formula, recurrence renewal is carried out to the weight of mixed Gauss model:

w _i,k＝(1-α)*w _i,k-1+α*H _i,k

S222, to Gauss model unmatched with this pixel, directly perform step S223, to the Gauss model mated with this pixel, after its average and standard deviation being upgraded according to following formula, terminate upgrade:

$\left\{\begin{array}{c}{\mathrm{\&mu;}}_{i,k}=(1-\mathrm{\&alpha;})*{\mathrm{\&mu;}}_{i,k-1}+\mathrm{\&alpha;}*H_{i,k}\\ {\mathrm{\&sigma;}}_{i,k}^2=(1-\mathrm{\&alpha;})*{\mathrm{\&sigma;}}_{i,k-1}^2+\mathrm{\&alpha;}*H_{i,k}\end{array}\right.$

S223, judge that M the Gauss model whether this pixel is corresponding with it does not all mate, if so, then replace the minimum Gauss model of weighted value according to the following formula, otherwise terminate to upgrade:

$\left\{\begin{array}{c}{w}_{\mathrm{sl},k}=1/M\\ {\mathrm{\&mu;}}_{\mathrm{sl},k}=X_k\\ {\mathrm{\&sigma;}}_{\mathrm{sl},k}=V_0\end{array}\right.$

Wherein, w _i,k, μ _i,kwith represent the weight of i-th Gauss model of this pixel, average and variance respectively, w _{i, k-1}, μ _{i, k-1}with represent the weight of i-th Gauss model of the pixel of a training picture and this pixel correspondence position, average and variance respectively, H _i,krepresent Gauss's matching value of this pixel and i-th Gauss model, α represents default learning efficiency value and is steady state value, and in the present invention, α is 3/trm, and trm is the number of samples of mixed Gauss model, w _{sl, k}, μ _{sl, k}and σ _{sl, k}represent the weight of the Gauss model that weighted value is minimum, average and standard deviation respectively.

As can be seen here, when this step upgrades mixed Gauss model, the weight of M Gauss model is all upgraded, then determine whether to upgrade the average of this Gauss model and variance according to the match condition of each Gauss model and pixel, if Gauss model mates with pixel, then its average and variance are upgraded, otherwise do not upgrade.Finally, if all Gauss models all do not mate with pixel, then replace the minimum Gauss model of weighted value.

S23, calculate the weight of M corresponding to this pixel Gauss model and the ratio of standard deviation respectively, and according to ratio, M Gauss model is sorted;

S24, select according to following formula before B Gauss model point as a setting, and then judge whether this pixel mates with a front B Gauss model, if so, judge that this pixel is background dot, otherwise judge that this pixel is foreground point:

$B=\arg \underset{b}{\min }(\underset{i=1}{\overset{b}{\mathrm{\&Sigma;}}}{w}_i>\mathrm{Th})$

In above formula, Th is default background threshold and 0.75≤Th≤0.90, and i represents sequence number, w ⁱrepresent the weight of i-th Gauss model;

S25, binaryzation assignment is carried out to this pixel, and renewal frequency distribution plan, it is specially:

If this pixel is background dot, is then 0 by its assignment, otherwise its assignment is 255 and the value of the histogram of this pixel correspondence position is added 1.

S26, travel through all pixels after obtain the binary image of this training picture.

It is complete whether S3, training of judgement sample have trained, if then according to following formula respectively each training picture of calculation training sample degree of imperfection and obtain the degree of imperfection threshold value of maximal value as training sample, otherwise return perform step S2:

$E_m=\frac{1}{\underset{i=1}{\overset{H}{\mathrm{\&Sigma;}}}\underset{j=1}{\overset{W}{\mathrm{\&Sigma;}}}\frac{m^2}{f{(x,y)}^2}}\&CenterDot;\underset{i=1}{\overset{H}{\mathrm{\&Sigma;}}}\underset{j=1}{\overset{W}{\mathrm{\&Sigma;}}}\frac{m^2}{f^2(x,y)\&CenterDot;b(x,y)}$

In above formula, E _mrepresent the degree of imperfection of training picture, m represents the training picture number of training sample, and f (x, y) represents histogram, and b (x, y) represents the binary image of training picture.

S4, gather IC element solder joint to be detected picture after, the mixed Gauss model that combined training is good and histogram calculate the degree of imperfection of this picture, in this step, the degree of imperfection of calculating chart sheet is same computing formula with what adopt in step S3, specifically comprises step S41 ~ S45:

S41, gather IC element solder joint to be detected picture after, for each pixel of this picture, calculate the weight of M corresponding to this pixel Gauss model and the ratio of standard deviation respectively, and according to ratio, M Gauss model sorted;

S42, select according to following formula before B Gauss model point as a setting, and then judge whether this pixel mates with a front B Gauss model, if so, judge that this pixel is background dot, otherwise judge that this pixel is foreground point:

$B=\arg \underset{b}{\min }(\underset{i=1}{\overset{b}{\mathrm{\&Sigma;}}}{w}_i>\mathrm{Th})$

In above formula, Th is default background threshold 0.75≤Th≤0.90, and i represents sequence number, w ⁱrepresent the weight of i-th Gauss model;

S43, binaryzation assignment is carried out to this pixel;

S44, travel through all pixels after obtain the binary image of this picture;

S45, calculate the degree of imperfection of this picture according to following formula:

$E_m=\frac{1}{\underset{i=1}{\overset{H}{\mathrm{\&Sigma;}}}\underset{j=1}{\overset{W}{\mathrm{\&Sigma;}}}\frac{m^2}{f{(x,y)}^2}}\&CenterDot;\underset{i=1}{\overset{H}{\mathrm{\&Sigma;}}}\underset{j=1}{\overset{W}{\mathrm{\&Sigma;}}}\frac{m^2}{f^2(x,y)\&CenterDot;b(x,y)}$

In above formula, E _mrepresent the degree of imperfection of picture, m represents the training picture number of training sample, and f (x, y) represents histogram, and b (x, y) represents the binary image of this picture.

S5, the degree of imperfection threshold value of the degree of imperfection of this picture and training sample is compared after obtain the testing result of IC element solder joint, it is specially:

Judge whether this degree of imperfection is greater than the degree of imperfection threshold value of training sample, if so, then judge that this IC element solder joint is rosin joint solder joint, otherwise, judge that this IC element solder joint is normal solder joint.

After tested, after mixed Gauss model being trained by adopting the training sample containing 400 training pictures, this method is adopted to carry out rosin joint defects detection, be 0% to the false drop rate of the rosin joint defect of IC element solder joint, loss is 0.90%, can effectively ensure higher accuracy rate, by the accuracy rate adopting this method greatly can improve the detection of IC element solder joint rosin joint, and detection speed is fast, can as the method solving an IC element solder joint detection difficult problem.

More than that better enforcement of the present invention is illustrated, but the invention is not limited to described embodiment, those of ordinary skill in the art also can make all equivalent variations or replacement under the prerequisite without prejudice to spirit of the present invention, and these equivalent modification or replacement are all included in the application's claim limited range.

Claims (9)

1., based on the IC element welding point defect detection method of mixed Gauss model, it is characterized in that, comprising:

The template number of S1, initialization mixed Gauss model, weight, average and variance, set up histogram according to the height of training picture and width simultaneously;

S2, obtain from training sample IC solder joint training picture, mixed Gauss model and histogram are upgraded;

It is complete whether S3, training of judgement sample have trained, if the degree of imperfection threshold value of then calculation training sample, otherwise returns and perform step S2;

S4, gather IC element solder joint to be detected picture after, the mixed Gauss model that combined training is good and histogram calculate the degree of imperfection of this picture;

S5, the degree of imperfection threshold value of the degree of imperfection of this picture and training sample is compared after obtain the testing result of IC element solder joint.

2. the IC element welding point defect detection method based on mixed Gauss model according to claim 1, it is characterized in that, described step S1, comprising:

S11, the template number of mixed Gauss model is initialized as M, wherein M is natural number and 3≤M≤5;

S12, the weight of M template is all initialized as 1/M, and the variance of M template, as the average of M template, is initialized as particular value V by Stochastic choice numeral simultaneously from [0,255] interval ₀, wherein 3≤V ₀≤ 10;

S13, set up matrix that a size is H*W and after all elements value of this matrix is initialized as 1, using this matrix as histogram, wherein H represents the height of training picture, and W represents the width of training picture.

3. the IC element welding point defect detection method based on mixed Gauss model according to claim 2, it is characterized in that, described step S2, comprising:

S21, obtain from training sample IC solder joint training picture, for each pixel of this training picture, it is carried out matching treatment with M Gauss model respectively, and obtains Gauss's matching value of this pixel and each Gauss model;

S22, recurrence renewal is carried out to mixed Gauss model;

S23, calculate the weight of M corresponding to this pixel Gauss model and the ratio of standard deviation respectively, and according to ratio, M Gauss model is sorted;

S24, select according to following formula before B Gauss model point as a setting, and then judge whether this pixel mates with a front B Gauss model, if so, judge that this pixel is background dot, otherwise judge that this pixel is foreground point:

$B=\arg \underset{b}{\min }(\underset{i=1}{\overset{b}{\mathrm{\&Sigma;}}}{w}_i>\mathrm{Th})$

In above formula, Th is default background threshold and 0.75≤Th≤0.90, and i represents sequence number, w _irepresent the weight of i-th Gauss model;

S25, binaryzation assignment is carried out to this pixel, and renewal frequency distribution plan;

S26, travel through all pixels after obtain the binary image of this training picture.

4. the IC element welding point defect detection method based on mixed Gauss model according to claim 3, it is characterized in that, described step S21, it is specially:

IC solder joint training picture is obtained from training sample, for each pixel of this training picture, it is carried out matching treatment with a corresponding M Gauss model respectively, judge whether to meet following condition, if so, then judge that this pixel mates with corresponding Gauss model, both Gauss's matching values are 1, otherwise both do not mate, Gauss's matching value is 0:

|X _k-μ _i|＜2.5σ _i

In above formula, X _krepresent the pixel value of this pixel, μ _irepresent the average of i-th Gauss model, σ _irepresent the variance of i-th Gauss model.

5. the IC element welding point defect detection method based on mixed Gauss model according to claim 4, it is characterized in that, described step S22, comprising:

S221, according to following formula, recurrence renewal is carried out to the weight of mixed Gauss model:

w _i,k＝(1-α)*w _i,k-1+α*H _i,k

S222, to Gauss model unmatched with this pixel, directly perform step S223, to the Gauss model mated with this pixel, after its average and standard deviation being upgraded according to following formula, terminate upgrade:

$\left\{\begin{array}{c}{\mathrm{\&mu;}}_{i,k}=(1-\mathrm{\&alpha;})*{\mathrm{\&mu;}}_{i,k-1}+\mathrm{\&alpha;}*H_{i,k}\\ {\mathrm{\&sigma;}}_{i,k}^2=(1-\mathrm{\&alpha;})*{\mathrm{\&sigma;}}_{i,k-1}^{*}+\mathrm{\&alpha;}*H_{i,k}\end{array}\right.$

S223, judge that M the Gauss model whether this pixel is corresponding with it does not all mate, if so, then replace the minimum Gauss model of weighted value according to the following formula, otherwise terminate to upgrade:

$\left\{\begin{array}{c}{w}_{\mathrm{sl},k}=1/M\\ {\mathrm{\&mu;}}_{\mathrm{sl},k}=X_k\\ {\mathrm{\&sigma;}}_{\mathrm{sl},k}=V_0\end{array}\right.$

Wherein, w _i,k, μ _i,kwith represent the weight of i-th Gauss model of this pixel, average and variance respectively, w _{i, k-1}, μ _{i, k-1}with represent the weight of i-th Gauss model of the pixel of a training picture and this pixel correspondence position, average and variance respectively, H _i,krepresent Gauss's matching value of this pixel and i-th Gauss model, α represents default learning efficiency value, w _{sl, k}, μ _{sl, k}and σ _{sl, k}represent the weight of the Gauss model that weighted value is minimum, average and standard deviation respectively.

6. the IC element welding point defect detection method based on mixed Gauss model according to claim 3, it is characterized in that, described step S25, it is specially:

If this pixel is background dot, is then 0 by its assignment, otherwise its assignment is 255 and the value of the histogram of this pixel correspondence position is added 1.

7. the IC element welding point defect detection method based on mixed Gauss model according to claim 3, it is characterized in that, the degree of imperfection threshold value of calculation training sample described in described step S3, it is specially:

According to following formula respectively each training picture of calculation training sample degree of imperfection and obtain the degree of imperfection threshold value of maximal value as training sample:

$E_m=\frac{1}{\underset{i=1}{\overset{H}{\mathrm{\&Sigma;}}}\underset{j=1}{\overset{W}{\mathrm{\&Sigma;}}}\frac{m^2}{f{(x,y)}^2}}\&CenterDot;\underset{i=1}{\overset{H}{\mathrm{\&Sigma;}}}\underset{j=1}{\overset{W}{\mathrm{\&Sigma;}}}\frac{m^2}{f^2(x,y)\&CenterDot;b(x,y)}$

In above formula, E _mrepresent the degree of imperfection of training picture, m represents the training picture number of training sample, and f (x, y) represents histogram, and b (x, y) represents the binary image of training picture.

8. the IC element welding point defect detection method based on mixed Gauss model according to claim 2, it is characterized in that, described step S4, comprising:

S41, gather IC element solder joint to be detected picture after, for each pixel of this picture, calculate the weight of M corresponding to this pixel Gauss model and the ratio of standard deviation respectively, and according to ratio, M Gauss model sorted;

S42, select according to following formula before B Gauss model point as a setting, and then judge whether this pixel mates with a front B Gauss model, if so, judge that this pixel is background dot, otherwise judge that this pixel is foreground point:

$B=\arg \underset{b}{\min }(\underset{i=1}{\overset{b}{\mathrm{\&Sigma;}}}{w}_i>\mathrm{Th})$

In above formula, Th is default background threshold and 0.75≤Th≤0.90, and i represents sequence number, w _irepresent the weight of i-th Gauss model;

S43, binaryzation assignment is carried out to this pixel;

S44, travel through all pixels after obtain the binary image of this picture;

S45, calculate the degree of imperfection of this picture according to following formula:

$E_m=\frac{1}{\underset{i=1}{\overset{H}{\mathrm{\&Sigma;}}}\underset{j=1}{\overset{W}{\mathrm{\&Sigma;}}}\frac{m^2}{f{(x,y)}^2}}\&CenterDot;\underset{i=1}{\overset{H}{\mathrm{\&Sigma;}}}\underset{j=1}{\overset{W}{\mathrm{\&Sigma;}}}\frac{m^2}{f^2(x,y)\&CenterDot;b(x,y)}$

In above formula, E _mrepresent the degree of imperfection of picture, m represents the training picture number of training sample, and f (x, y) represents histogram, and b (x, y) represents the binary image of this picture.

9. the IC element welding point defect detection method based on mixed Gauss model according to claim 1, it is characterized in that, described step S5, it is specially:

Judge whether this degree of imperfection is greater than the degree of imperfection threshold value of training sample, if so, then judge that this IC element solder joint is rosin joint solder joint, otherwise, judge that this IC element solder joint is normal solder joint.