CN106934786B

CN106934786B - Method for realizing image processing software of mold monitor

Info

Publication number: CN106934786B
Application number: CN201511010200.6A
Authority: CN
Inventors: 李运秀
Original assignee: SHENZHEN PORCHESON TECHNOLOGY CO LTD
Current assignee: SHENZHEN PORCHESON TECHNOLOGY CO LTD
Priority date: 2015-12-31
Filing date: 2015-12-31
Publication date: 2022-04-26
Anticipated expiration: 2035-12-31
Also published as: CN106934786A

Abstract

An image processing software algorithm for realizing a mold monitor can detect a plastic part and a mold cavity simultaneously, and by increasing comparison with an over-exposure and under-exposure template, the problem of image false detection caused by image pixel value saturation (reaching areas near 255 and 0) of a conventional mold protector is solved, or an easily false detection area is added into an neglected area by manual intervention, so that the manual intervention is reduced, meanwhile, the problem of brightness change can be solved by structural similarity judgment, the original outline of the redundant template is compared with the outline of a differential image, the possibility of false detection is further reduced, and the times of the manual intervention are greatly reduced. Meanwhile, the image processing speed is increased through variance ratio comparison, and only the detected suspected NG area is subjected to multi-template matching, so that the detection speed is increased, and the influence on the production period is shortened.

Description

Method for realizing image processing software of mold monitor

Technical Field

The invention relates to the field of image processing software algorithms, in particular to an image processing software algorithm for realizing a mold monitor.

Background

The mould is used as the most important forming equipment for processing injection products in the manufacturing industry, and the quality of the mould is directly related to the quality of the products. In addition, the mold occupies a large proportion in the injection molding production cost, and the service life of the mold directly influences the product cost. Therefore, the quality of the die is improved, the die is properly maintained and maintained, the service cycle of the die is prolonged, and the die is an important weight for reducing the cost and improving the efficiency in the processing industry of injection molding products. In actual production, due to the fact that the molds are frequently replaced, when the injection molding machine runs, the molds with high value in each production period are possibly damaged due to product residues or sliding block dislocation, and the mold protector can effectively deal with various potential problems, so that shutdown mold repair is avoided, production cost is reduced, product quality is improved, and delivery construction period is guaranteed.

The existing mold protector mainly comprises the following methods:

1. image matching technology based on standard template images: the method comprises the steps of collecting a plurality of qualified template images as reference images, obtaining a current image after the mold opening is in place each time, and comparing the current image with all the template images in sequence, wherein the current template image is considered to be qualified as long as the current template image is basically similar to one of the reference images, and otherwise, the current template image is not qualified. The method is easy to realize, but if the position of the moving mold is not accurate, factors such as environmental illumination change and the like can cause image difference to cause false alarm, and in addition, the processing speed is slowed down if more matched templates exist, so that the production efficiency is influenced.

2. Judging the range of pixel color values: and collecting a plurality of qualified template images as reference images, wherein if the color value of each pixel point of the to-be-detected image is between the color value of the corresponding pixel point of the minimum image and the color value of the corresponding pixel point of the maximum image, the pixel point of the to-be-detected image is a qualified point, and otherwise, the pixel point is a unqualified point. The method is easy to generate false alarm when the brightness of the surrounding illumination is changed greatly or the vicinity of the saturated area of the image pixel point.

3. Binocular vision technology based on mechanism light: the technology can improve the detection precision, but the cost is expensive, the detection speed is influenced to a certain extent, and the user acceptance is difficult to obtain in the middle and low-end market.

Disclosure of Invention

The technical problem to be solved by the embodiments of the present invention is to provide an image processing software algorithm for implementing a mold monitor, which is characterized by comprising the following steps:

1) adjusting a camera image, acquiring a clear camera image, and indicating the definition degree of the image by using a focusing index;

2) setting an automatic light measuring area, automatically adjusting the image brightness, acquiring a normal brightness template image, and calculating the relationship between the camera gain and the image brightness;

3) setting an overexposure region and an underexposure region of an image, and acquiring overexposure template images and underexposure template images;

4) acquiring an image to be detected, comparing the image to be detected with the brightness of the current template image, and adjusting the image to be detected to be within an allowable range of the brightness difference with the template;

5) respectively calculating outline angular points of the template and the image to be detected, and registering the template according to the set allowable position deviation range;

6) solving a difference image between the image to be detected and the current template image, and performing gray level stretching treatment;

7) acquiring an ignoring alarm point, an ignoring area and an ignored burr processing area, and ignoring the area set when the image matching is calculated in the next step;

8) solving an edge image of the stretched image by using a canny operator, and expanding to enable edges to be communicated as much as possible;

9) removing the edge of the part overlapped with the original outline, and preventing extra edge noise caused by illumination change;

10) taking out a suspected defect area, solving the variance, if the variance is greater than a set threshold value, determining the area as a defect, if the variance is less than the set threshold value, calculating the structural similarity of the image to be detected and the corresponding area of the template image, if the similarity is greater than the set similarity threshold value, determining the area as an OK area, and if not, determining the area as an NG area;

11) if an NG area is detected in the step 9), judging whether the NG area gray level histogram has overexposure or underexposure, if the exposure is normal, proceeding to the step 11), if the area image is dark, proceeding to the step 12), and if the area image is bright, proceeding to the step 13):

12) traversing all normal exposure templates, comparing NG areas once again, considering the area NG if all NG areas are NG, determining that the area is OK if a certain area is OK, automatically adjusting the template matched with OK currently to the first order, and automatically sorting the rest templates;

13) overexposing the current image, traversing all overexposure templates, comparing the NG areas for one time only, considering the NG areas if all the NG areas are NG, determining that the areas are NG if the comparison result shows that a certain area is OK, automatically adjusting the template which is matched with OK currently to the first order of the sequence, and automatically sequencing the rest templates;

14) underexposing the current image, traversing all underexposed templates, comparing NG areas once again, considering the areas as NG if the images are NG, determining the areas as OK if a certain area is OK, automatically adjusting the currently matched OK template to the first order of the sequence, and automatically sequencing the rest templates;

15) if the detection result is NG, the user can add the current image as a template, the system automatically and simultaneously adds the overexposure template and the underexposure template, and simultaneously defaults the current detection result as OK;

16) if the detection result is NG, the current alarm point can be added to the neglected area, and meanwhile, the current detection result is defaulted to be OK;

17) in the algorithm integration process, when NG occurs, if manual intervention is not needed, the current detection result can be directly defaulted, and the next detection is continued.

The result of the invention is filtered and screened layer by layer through algorithms of each level, such as variance ratio, structural similarity, multi-template local matching, automatic brightness adjustment, overexposure and underexposure template comparison and the like, so that the accuracy and stability of the detection result are ensured, and meanwhile, a plurality of templates are added, alarm points are ignored and automatic exposure is carried out through manual intervention, so that the environmental adaptability of the algorithm is further improved, and the robustness of the system is enhanced. Meanwhile, after the gray level stretching is carried out on the differential image, the detection capability of the system under a low-contrast environment (white defects under a white background and black defects under a black background) is improved after edge noise is removed, the NG defect part is adopted to carry out multi-layer algorithm filtering, and a multi-template effective sequencing strategy is adopted, so that the detection efficiency of false detection confirmation is improved, and a complete solution is provided for various problems in the mold protector. The algorithm adopts C + + language modularization programming, can be well integrated on various platforms, and provides various solutions of the mold monitor.

Drawings

In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the drawings used in the description of the embodiments or the prior art will be briefly described below, it is obvious that the drawings in the following description are only some embodiments of the present invention, and for those skilled in the art, other drawings can be obtained according to the drawings without creative efforts.

FIG. 1 is an initialization flowchart

FIG. 2 is a block diagram of an algorithm software flow

FIG. 3 is an image adjustment interface

FIG. 4 is a view showing an interface for setting a photometric area

FIG. 5 is a fitting to a photometer

FIG. 6 is a histogram of an under-exposed template

FIG. 7 is a histogram of an overexposed template

FIG. 8 is a drawing of an underexposed stencil

FIG. 9 is an overexposure template

FIG. 10 is a normal exposure template

FIG. 11 shows the output results

FIG. 12 is a test interface

FIG. 13 is a test observation window for the test results

FIG. 14 is a parameter setting interface

FIG. 15 shows the new profile caused by the change of light

FIG. 16 shows NG interface as the detection result

Detailed Description

The invention is further illustrated with reference to the following figures and examples.

The technical scheme of the invention is further described below by combining the attached drawings, and as shown in figures 1 and 2, normal exposure, underexposure and overexposure images of a plurality of qualified images are taken as templates of an algorithm;

template overexposure and underexposure strategies: exposure adjustment is carried out by adopting a gray value [5-15], [240- & 250] brightness interval which can be set to carry out histogram distribution;

the exposure adjustment obtains the relation between the brightness of the current template and the gain through automatic learning, and simultaneously, the gain range is set to prevent the adjustment from crossing the boundary.

When the gain is out of range, adjusting the gain back to the normal range by adjusting the exposure time;

image template ordering strategy: sorting the newly added templates in the top order, and sorting the subsequent boards according to the matching probability;

adopting a difference image of the test image and the template image, stretching the gray level of the image, improving the contrast of the defect, highlighting the defect characteristic, then carrying out salt-pepper noise filtering, carrying out edge extraction, expanding the edge as much as possible, solving an edge area, and calculating the variance ratio and the structural similarity of the area.

And (3) judging the variance ratio: when the variance ratio is larger than a set threshold value, directly judging to be NG, otherwise, further judging the structural similarity;

for the nonlinear problem of illumination change at the edge, eliminating the nonlinear problem through edge removal and structural similarity;

judging the structural similarity: and comparing the structural similarity with the template, wherein if the structural similarity is greater than a set threshold value, the structural similarity is qualified, and if the structural similarity is less than the threshold value, the structural similarity is considered as a suspected NG area.

And (3) taking out the suspected NG area, judging the grey value of the histogram of the suspected NG area, comparing an underexposed image with the corresponding underexposed template if the range is close to the interval range near 255, comparing an overexposed image with the overexposed template if the range is close to the interval range near 0, and comparing the area in the normal exposure range with the normal exposure template.

Adjusting a lens, enabling the image definition to be maximum according to the picture focusing index, simultaneously obtaining an image with moderate brightness by the system through an automatic exposure algorithm, displaying a brightness histogram, enabling the brightness distribution of the image to be clear through the histogram, and taking the image as a template image as shown in FIG. 3;

setting the photometric area as shown in fig. 4, the brightness of the whole image is adjusted based on the brightness of the area of the template, so that the difference with the template image caused by the environmental concern change can be reduced.

And setting an ROI (region of interest) and an neglected region, wherein in actual operation, the final detection region is a region obtained by removing the neglected region, the burr neglected region and the neglected alarm point region from the ROI, and the image outside the region is not detected.

And performing photometric fitting, as shown in fig. 5, finding a relationship between camera gain and image brightness, and when the illumination change of the later-period brightness environment is large, using the relationship as a reference value for image brightness adjustment, adjusting the current image brightness to a range within which the brightness error of the plate image is reasonable, setting an adjustment range of the gain, adjusting the exposure time within a range of 20-50% once the range is exceeded, wherein the range can be set, and then adjusting the gain again, so that the system can increase the adaptability to the large-range light ratio change.

Acquiring overexposure and underexposure templates: for the underexposure template, mainly in order to prevent that image details cannot be reflected after the image pixel value reaches 255 and is good, in the area near the 255-value pixel, the brightness change is slightly increased, and the image details are submerged, so in order to accurately detect the area, the image brightness needs to be reduced, the gray value of the image pixel in the area is far away from the 255 value, so that a larger safe distance exists, and the overexposed area is adjusted to proper brightness through underexposure; similarly, the overexposure template solves the problem that the details of the dark part of the image are lost when the gray value of the pixel is close to 0. As shown in the histograms of fig. 6 and 7, the current example is set to [5, 10], and [245, 250] interval pixel distribution is 0, and exposure adjustment is performed, so as to obtain an under-exposed template shown in fig. 8, an over-exposed template map shown in fig. 9, and a normal exposure template shown in fig. 10, respectively.

The operation result output window is shown in fig. 11, and the result is stored in a local excel file, and fig. 12 shows a normal detection result. As shown in fig. 13, the output window adjusts various parameters by adjusting the parameter adjustment window shown in fig. 14, and the meaning of each parameter is as follows:

area threshold: the detectable minimum area is set, so that the method can adapt to different application scenes and adjust the detection sensitivity;

the result similarity is as follows: the judgment standard threshold values of OK and NG are too high or too low, the OK is easy to be judged as NG by mistake if the judgment standard threshold values are too high, and the NG is easy to be judged as OK by mistake if the judgment standard threshold values are too low, so the judgment standard threshold values need to be set according to the specific situation of the site;

salt and pepper noise threshold: the size of a salt and pepper noise filtering value;

variance ratio: when the variance is greater than the set value, the detection result is directly judged to be NG, and the next operation is not carried out, so that the detection efficiency is improved;

area of structural similarity: the minimum area of the structural similarity calculation is adjusted, the structural similarity value is prevented from being higher due to too few local features, and meanwhile, the redundant interference contours caused by illumination change shown in fig. 15 can be removed by the structural similarity, so that the stretching threshold value 1 and the stretching threshold value 2 are removed: when the contrast of the differential image is small, false detection is easy to occur, and the contrast is improved by adjusting the 2 values, so that the defect with small contrast is detected;

edge coefficient: adjusting the size of the burr detection.

It should be understood that the detailed description of the present invention is only for illustrating the present invention and is not limited by the technical solutions described in the embodiments of the present invention, and those skilled in the art should understand that the present invention can be modified or substituted equally to achieve the same technical effects; as long as the use requirements are met, the method is within the protection scope of the invention.

Claims

1. A method of implementing image processing software for a mold monitor, comprising the steps of:

7) acquiring an ignoring alarm point, an ignoring region and an ignored rough edge processing region, and ignoring a set of the ignoring region and the ignored rough edge processing region when the image matching is calculated in the next step;

11) if an NG area is detected in the step 10), judging whether the NG area gray level histogram has overexposure or underexposure, if the exposure is normal, proceeding to the step 12), if the area image is dark, proceeding to the step 13), and if the area image is bright, proceeding to the step 14):

12) traversing all normal exposure templates, comparing NG areas once again, considering the area NG if all NG areas are NG, determining that a certain area is OK if the comparison result is OK, automatically adjusting the template matched with OK currently to the first order, and automatically sorting the rest templates;

14) underexposing the current image, traversing all underexposed templates, comparing NG areas once again, considering the areas as NG if the images are all NG, determining that the areas are OK if a certain area is OK, automatically adjusting the currently matched OK template to the first order of the sequence, and automatically sequencing the rest templates;