CN103544473A - Electronic connector detection method based on machine vision - Google Patents

Abstract

The invention discloses an electronic connector detection method based on machine vision, and belongs to the technical field of computer program design comprising image acquisition, analysis and processing. Actual targeted images are processed by various methods of image smoothing, image segmentation, image matching, edge extraction and color recognition, complete geometrical characteristics of defects of the electronic connector are acquired, and automatic detection can be realized. According to experiments, system detecting results is consistent with visual detecting results, accuracy of the detection is guaranteed, and detecting efficiency of the electronic connector is improved greatly.

Description

A kind of electric power connector detection method based on machine vision

Technical field

The invention belongs to the technical field of computer programming, relate in particular to the electric power connector context of detection of machine vision.

Background technology

Electric power connector is a kind of organ with Electric connection characteristic, its major function is between device and assembly, assembly and rack, system and subsystem, to play a part electrical connection and signal transmission, is to form circuitry system electrical to connect one of requisite base components.It has been widely used in the fields such as aviation/space flight, military equipment, communication, computing machine, automobile, industry, household electrical appliance, and existing oneself develops into one of mainstay industry of current electronics and information infrastructure product.

In recent years along with low-cost area is shifted in Consumption of China electronics, network design, telecom terminal product output rapid growth and the production of world's electric power connector element, global link device productive capacity constantly shifts to China, at present, oneself increases fast and the most maximum market through becoming global link device China, and the Si great connector company of whole world maximum has all set up factory in China's Mainland successively since 20 end of the centurys.

With Guangdong Wei Li， the said firm of company of Electronics Factory, be mainly engaged in and produce various electric power connectors, the electric power connector great majority of its production export to abroad for automobile making, therefore very high to the quality requirements of electric power connector.In order to guarantee the quality of electric power connector, the said firm has dropped into a big chunk human and material resources in the quality testing of electric power connector, therefore develop a set of electric power connector detection system based on machine vision the human and material resources of saving to a great extent the said firm are enhanced productivity simultaneously, promote the automated process of the sector.

Electric power connector is of a great variety, be mainly because electric wire and splicing ear of a great variety, but manufacture process is consistent substantially, generally can be divided into three phases: stripping glue, punching press, assembling.

Stripping glue: the rubber at electric wire two ends is peeled off, exposed copper core bundle;

Punching press: splicing ear is stamped into electric wire two ends;

Assembling: the good electric wire of punching press is inserted on plastics cassette holder.

In the process of stripping glue, copper core may be cut off to the electric conductivity have influence on connector, in the process of punching press, splicing ear is easy to that punching press is not in place or punching press is bad, has influence on the quality of connector.If have sequence requirement (electric wire of the corresponding particular color of each jack) in the process of assembling, before must make correct judgement to the color of electric wire.

Summary of the invention

The present invention is mainly in order to replace artificial vision to check product quality by machine vision, thereby saves more human and material resources, enhances productivity.

An electric power connector detection method for machine vision, comprise image collection, number copper core, whether punching press puts in place and differentiates electric wire color to detect splicing ear.

When several copper core, first by the profile of copper core area of beam, obtain the approximate region of copper core bundle, then by fixed threshold method, copper core bundle is split and stretch processing is carried out in the region splitting, spacing between copper core and copper core is widened, finally the image after stretching is cut apart with watershed segmentation algorithm.

When detecting splicing ear punching press put in place, first by determining based on the relevant Fast Match Algorithm of normalizing eliminate indigestion, need the station that detects, then by extracting edge, judge that connection works as side-play amount and the deflection angle of front terminal, whether punching press puts in place.

When differentiating electric wire color, first build the neural network of three layers and train, then by the sorter of training, electric wire color is differentiated.

Its main functional module can be divided into 3 aspects:

1, visual information collection: gather original visual information by visual information acquisition system (camera, camera lens, light source etc.), for ensuing Vision information processing and analysis are prepared.

2, Vision information processing and analysis: according to the concrete feature of product and actual production environment, by studying following correlation technique content, comprise: pre-service, the image of visual pattern cut apart, images match, feature extraction, pattern discrimination etc., propose effective technical scheme, thereby judge that whether current production quality is qualified.

3, synchronize with Machinery Control System: detection system coordinates with Machinery Control System, thereby complete more rationally and effectively whole production run.

The workflow of system, after the cementing bundle of stripping, by one group of collecting device, gather present image, then by the processing of present image and analysis have been judged whether to copper core is cut off, and judged result is fed back to Machinery Control System, last Machinery Control System is according to the result decision-making of feedback, execution corresponding actions.After stamped terminals finishes, by another group collecting device, gather present image, then by Treatment Analysis judge splicing ear whether punching press put in place and electric wire color whether correct, and judged result is fed back to Machinery Control System to make next step action.

Beneficial effect of the present invention is:

1. built the electric power connector detection method framework based on machine vision, for the Machine Vision Detection of multitude of different ways is laid a good foundation;

2. in order to count accurately the number of copper core, the present invention utilizes watershed algorithm combining image stretching scheduling algorithm to Image Segmentation Using, makes segmentation effect more accurate.

3. in order accurately judge terminal, whether punching press puts in place, and first the present invention needs by determining based on the relevant Fast Match Algorithm of normalizing eliminate indigestion the station that detects, and whether punching press puts in place then by extraction edge, to judge splicing ear.Its fast operation, accuracy of judgement.

4. in order accurately to judge the color of electric wire, first the present invention builds the neural network of three layers and trains, and then by the sorter of training, electric wire color is differentiated.

The technology providing by us, our designed system has reached the object that improves the accuracy rate of eye state analysis and the speed of whole fatigue detecting.

Accompanying drawing explanation

The workflow of Fig. 1 system;

Fig. 2 copper core segmentation effect figure;

The three-layer neural network figure that Fig. 3 builds.

Embodiment

When several copper core, first we obtain the approximate region of copper core bundle by the profile of copper core area of beam, and using this region as area-of-interest.Then by fixed threshold method, copper core bundle is split, stretch processing is carried out in the region splitting, the spacing between copper core and copper core is widened, for preparing ensuing cutting apart.Image after stretching is cut apart to (as shown in Figure 2) with watershed segmentation algorithm.Watershed segmentation method, it is a kind of dividing method of the mathematical morphology based on topological theory, its basic thought is that image is regarded as to a pair " topomap ", wherein the more intense regional pixel value of brightness is larger, and darker regional pixel value is smaller, by finding " ”He“ watershed divide, catchment basin boundary ", to Image Segmentation Using.Watershed algorithm is often to connect together to be difficult to cut apart this class problem for the treatment of the target object in image, and conventionally can obtain reasonable effect.Cut apart after end, if still have not separated copper core bundle, by area, judge.

When detecting bonder terminal punching press puts in place, first first we adopted the Fast Match Algorithm based on normalizing eliminate indigestion relevant (NCC), and matching score is the normalized crosscorrelation coefficient between masterplate image t and image i to be matched:

$\mathrm{score}=\frac{1}{n}\underset{(x,y)}{\mathrm{\Σ}}\frac{t(x,y)-m_t}{\sqrt{s_t^2}}*\frac{i(s,y)-m_i}{\sqrt{s_i^2}},$

Wherein n is pixel number, m _tand m _ibe respectively the average of masterplate image and image to be matched,

with be respectively the variance of masterplate image and image to be matched.Allow image to be detected mate to obtain corresponding deflection angle and displacement with template image.

Normalized crosscorrelation (NCC) is used as estimating of 2 image similarities conventionally.The advantage of normalized crosscorrelation is insensitive to the illumination change in 2 width images.In order to improve the speed of coupling, this project is first converted to gray-scale map by colored RGB image, and then mates.Compare with the matching algorithm based on shape, the image matching algorithm based on normalizing eliminate indigestion relevant (NCC) more has superiority in this project.Because the matching algorithm based on shape mates by profile, and the profile of splicing ear more complicated often, this can make matching score very low conventionally.In addition, due to often deflection to some extent of splicing ear, so that coupling is often inaccurate.

By mating us, can obtain side-play amount and the deflection angle when front terminal, the side-play amount that utilization obtains and deflection angle are done affined transformation to photo current just can be made it the terminal correction in present image to align with the terminal in template image.

When side-play amount and the deflection angle of front terminal, the side-play amount of establishing in horizontal direction and vertical direction is respectively dx and dy, and deflection angle is θ, and the pixel (x, y) of original image is transformed to through affined transformation

?

$\left(\begin{array}{c}\tilde{x}\\ \tilde{y}\\ 1\end{array}\right)=H\left(\begin{array}{c}x\\ \mathrm{<figure-callout\; id="1"\; label="y"\; filenames="130831175941.png"\; state="\{\{state\}\}">y</figure-callout>}\\ 1\end{array}\right),H=\left(\begin{array}{ccc}\cos \mathrm{\&theta;}& -\sin \mathrm{\&theta;}& \mathrm{dx}\\ \sin \mathrm{\&theta;}& \cos \mathrm{\&theta;}& \mathrm{dy}\\ 0& 0& 1\end{array}\right).$

Due to the regional location that has needed in clear and definite template image before this project to detect when creating masterplate, so in present image, the detection position of terminal has also just been determined.Location adopts different detection methods for different surveyed areas after finishing.

Detect splicing ear whether punching press to put in place be mainly that length by judgement copper core realizes, due to some gap often between copper core, these gap incident lights are difficult to therefrom reflect, and copper core is metal material, more intense to reflection of light ability, what therefore in image, show is chequered with black and white striped.For this feature, we adopt the method for edge extracting to extract the edge that direction is consistent with copper core direction, owing to having corresponding relation between edge and copper core, therefore can judge that by the edge extracting whether copper core is long or too short.The method of extracting edge is a lot, for example Roberts operator: $\left(\begin{array}{cc}1& 0\\ 0& -1\end{array}\right)\left(\begin{array}{cc}1& 0\\ 0& -1\end{array}\right),$ Sobel operator: $\left(\begin{array}{ccc}-1& -2& -1\\ 0& 0& 0\\ 1& 2& 1\end{array}\right)\left(\begin{array}{ccc}1& 0& -1\\ 2& 0& -2\\ 1& 0& -1\end{array}\right)$ Deng, comparing Canny with them uses strict mathematical method to analyze this problem, the best edge of deriving by 4 linear combination of exponential functions forms extracts operator net, the essence of its algorithm is to do level and smooth computing with an accurate Gaussian function, then with the single order differential location derivative maximal value with direction, Canny operator edge detection is a kind of practical edge detection operator, has good rim detection performance.Canny edge detection method utilizes the single order differential of Gaussian function, and it can obtain good balance between squelch and rim detection.Here we adopt Canny operator extraction edge.

When judgement electric wire color, we have adopted the neural network of multilayer perceptron to carry out color classification.Neural network is that a kind of application class is similar to the mathematical model that structure that cerebral nerve cynapse connects is carried out information processing.Neural network is a kind of operational model, by a large amount of node (or claim neuron) and between be coupled to each other formation.Each node represents a kind of specific output function, is called excitation function.Every two internodal connections all represent that one is referred to as weight for by the weighted value of this connection signal, and this is equivalent to the memory of artificial neural network.The output of network is according to the connected mode of network, the difference of weighted value and excitation function and difference.And network self is all to the approaching of certain algorithm of nature or function conventionally, may be also the expression to a kind of logic strategy.

First we build the neural network (as shown in Figure 3) of three layers, and input node has three, is respectively the RGB three-component of color, and implicit this project of node is made as 20, the kind number that output node number is all colours, the numbering that Output rusults is color.

Then the sample learning by some show that a desirable neural network identifies for ensuing color.Study is an important content of neural network research, and its adaptability realizes by study.According to the variation of environment, weights are adjusted, improve the behavior of system.Effectively learning algorithm, makes neural network to construct the intrinsic representation of objective world by connecting the adjustment of weights, forms characteristic information processing method, and information Storage and Processing is embodied in the connection of network.Different according to academic environment, the mode of learning of neural network can be divided into supervised learning and unsupervised learning.Here we adopt supervised learning mode, in supervised learning, the data of training sample are added to network input end, corresponding desired output is compared with network output simultaneously, obtain error signal, with this, control the adjustment of weights strength of joint, through repeatedly converging to definite weights after training.When sample situation changes, through study, can revise weights to adapt to new environment.

Claims (3)

1. A method for detecting electronic connectors based on machine vision, including image collection, counting copper cores, detecting whether the connecting terminals are punched in place and distinguishing the color of the wires, characterized in that: When counting copper cores, first obtain the approximate area of the copper core bundle through the profile of the copper core bundle section, and then segment the copper core bundle through the fixed threshold method and stretch the segmented area so that the copper core and the copper core The distance between them is widened, and finally the stretched image is segmented with the watershed segmentation algorithm; When detecting whether the connecting terminal is punched in place, first determine the station to be detected through the fast matching algorithm based on normalized product correlation, and then judge whether the offset and deflection angle of the current terminal are connected by extracting the edge, and whether the punching is in place; When distinguishing the color of wires, a three-layer neural network is constructed and trained first, and then the color of wires is judged by the trained classifier. 2. detection method according to claim 1, it is characterized in that based on the fast matching algorithm of normalized product correlation, the matching score is the normalized cross-correlation coefficient between the template image t and the image i to be matched: $\mathrm{<span\; class="notranslate">\; score</span>}<span\; class="notranslate">\; =</span>\frac{\mathrm{<span\; class="notranslate">\; 1</span>}}{\mathrm{<span\; class="notranslate">\; no</span>}}\underset{<span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; x</span>}<span\; class="notranslate">\; ,</span>\mathrm{<span\; class="notranslate">\; the\; y</span>}<span\; class="notranslate">\; )</span>}{\mathrm{<span\; class="notranslate">\; \&Sigma;</span>}}\frac{\mathrm{<span\; class="notranslate">\; t</span>}<span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; x</span>}<span\; class="notranslate">\; ,</span>\mathrm{<span\; class="notranslate">\; the\; y</span>}<span\; class="notranslate">\; )</span><span\; class="notranslate">\; -</span>{\mathrm{<span\; class="notranslate">\; m</span>}}_{\mathrm{<span\; class="notranslate">\; t</span>}}}{\sqrt{{\mathrm{<span\; class="notranslate">\; the\; s</span>}}_{\mathrm{<span\; class="notranslate">\; t</span>}}^{\mathrm{<span\; class="notranslate">\; 2</span>}}}}<span\; class="notranslate">\; *</span>\frac{\mathrm{<span\; class="notranslate">\; i</span>}<span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; the\; s</span>}<span\; class="notranslate">\; ,</span>\mathrm{<span\; class="notranslate">\; the\; y</span>}<span\; class="notranslate">\; )</span><span\; class="notranslate">\; -</span>{\mathrm{<span\; class="notranslate">\; m</span>}}_{\mathrm{<span\; class="notranslate">\; i</span>}}}{\sqrt{{\mathrm{<span\; class="notranslate">\; the\; s</span>}}_{\mathrm{<span\; class="notranslate">\; i</span>}}^{\mathrm{<span\; class="notranslate">\; 2</span>}}}}<span\; class="notranslate">\; ,</span>$ Where n is the number of pixels, m _t and m _i are the mean values of the template image and the image to be matched respectively,

and

are the variances of the template image and the image to be matched, respectively. 3. The detection method according to claim 1, characterized in that the offset and deflection angle of the current terminal, assuming that the offset in the horizontal direction and the vertical direction are respectively dx and dy, the deflection angle is θ, the original image The pixel point (x, y) is transformed into

but $\left(\begin{array}{c}\stackrel{<span\; class="notranslate">\; ~</span>}{\mathrm{<span\; class="notranslate">\; x</span>}}\\ \stackrel{<span\; class="notranslate">\; ~</span>}{\mathrm{<span\; class="notranslate">\; the\; y</span>}}\\ \mathrm{<span\; class="notranslate">\; 1</span>}\end{array}\right)<span\; class="notranslate">\; =</span>\mathrm{<span\; class="notranslate">\; h</span>}\left(\begin{array}{c}\mathrm{<span\; class="notranslate">\; x</span>}\\ \mathrm{<span\; class="notranslate">\; the\; y</span>}\\ \mathrm{<span\; class="notranslate">\; 1</span>}\end{array}\right)<span\; class="notranslate">\; ,</span>\mathrm{<span\; class="notranslate">\; h</span>}<span\; class="notranslate">\; =</span>\left(\begin{array}{ccc}\mathrm{<span\; class="notranslate">\; cos</span>}\mathrm{<span\; class="notranslate">\; \&theta;</span>}& <span\; class="notranslate">\; -</span>\mathrm{<span\; class="notranslate">\; sin</span>}\mathrm{<span\; class="notranslate">\; \&theta;</span>}& \mathrm{<span\; class="notranslate">\; dx</span>}\\ \mathrm{<span\; class="notranslate">\; sin</span>}\mathrm{<span\; class="notranslate">\; \&theta;</span>}& \mathrm{<span\; class="notranslate">\; cos</span>}\mathrm{<span\; class="notranslate">\; \&theta;</span>}& \mathrm{<span\; class="notranslate">\; dy</span>}\\ \mathrm{<span\; class="notranslate">\; 0</span>}& \mathrm{<span\; class="notranslate">\; 0</span>}& \mathrm{<span\; class="notranslate">\; 1</span>}\end{array}\right)<span\; class="notranslate">\; .</span>$

Images