CN114862794A - A bubble detection system for glassware - Google Patents

Abstract

The invention discloses a bubble detection system for glassware, relates to the technical field of glass, and solves the problem that in the process of bubble detection in the prior art, the convolutional neural network model requires a large number of training sets, the data processing volume is large, and the The feedback of the detection results locates the cause of the bubbles, which leads to the technical problem of poor detection efficiency; the present invention generates an image data set by collecting image data of different angles of the glassware, and divides the image data set into two parts, one part is used for rough detection, and the other part is used for rough detection. It is used for precise detection, and combined with intelligent classification model to determine the cause of bubbles, so as to achieve high-efficiency detection of glass bubbles in glassware; the position of the camera and the light source in each group of the image acquisition equipment in the present invention The angle and angle are adjusted separately, and the acquisition angle is set according to the size and shape of the glassware to obtain the image data of the glassware in different directions, which can ensure the accuracy of bubble positioning and identification in the glassware.

Description

A bubble detection system for glassware

technical field

The invention belongs to the field of glass production and relates to an intelligent detection technology for glassware, in particular to a bubble detection system for glassware.

Background technique

Bubble is the most common glass defect, accompanying the whole process of glass production, and it is difficult to judge and solve the bubble. Bubble not only affects the quality of glassware, but also affects production efficiency and economic benefits. Therefore, it is an arduous task for glass workers to explore a fast and accurate bubble detection method.

The prior art (invention patent with publication number CN109191440A) discloses a method for detecting and counting glass bubbles, obtaining a convolutional neural network model through sample set training, and then obtaining the covering position and glass bubbles of the glass bubbles through the convolutional neural network model. The number of glass bubbles can be accurately detected and counted. In the process of bubble detection in the prior art, the convolutional neural network model requires a large number of training sets, the data processing volume is large, and the cause of the bubble cannot be located according to the feedback of the detection result, resulting in poor glass bubble detection efficiency; therefore, there is an urgent need for a method. Air bubble detection system for glassware.

SUMMARY OF THE INVENTION

The present invention aims to solve at least one of the technical problems existing in the prior art; for this purpose, the present invention proposes a bubble detection system for glassware, which is used to solve the problem of convolution neural network in the process of bubble detection in the prior art. The network model requires a large number of training sets, the amount of data processing is large, and the cause of the bubble cannot be located according to the feedback of the detection result, resulting in the technical problem of poor glass bubble detection efficiency.

The invention generates an image data set by collecting image data of different angles of glassware, and divides the image data set into two parts, one part is used for rough detection, and the other part is used for fine detection, so as to obtain relevant parameters of glass bubbles, and then combined with intelligent classification The model judges the causes of bubbles to achieve high-efficiency detection of glass bubbles in glassware.

In order to achieve the above object, a first aspect of the present invention provides a bubble detection system for glassware, comprising a data analysis module, and a data acquisition module connected thereto, the data acquisition module is connected with an image acquisition terminal;

Data acquisition module: collect glassware from different angles through the image acquisition terminal to obtain a number of image data; integrate a number of the image data to generate an image data set, and send it to the data analysis module;

Data analysis module: divide the image data set into data one and data two, analyze and locate the bubble position on the data one, and obtain the bubble parameters in combination with the data two; wherein, the data one includes a piece of image data; as well as

The bubble characteristic data is generated according to the number of bubbles and the bubble parameters, and the cause of the bubbles is determined in combination with the intelligent classification model; wherein, the intelligent classification model is established based on the artificial intelligence model.

Preferably, the data analysis module is in communication and/or electrical connection with the data acquisition module, and the data acquisition module is in communication and/or electrical connection with the image acquisition terminal.

Preferably, the image acquisition terminal includes at least one group of image acquisition devices, each group of the image acquisition devices includes a camera and a light source, and both the camera and the light source focus on the glassware.

Preferably, the positions and angles of the cameras and the light sources in each group of the image acquisition devices are adjusted independently, and the illumination modes of the light sources include backlight illumination, front illumination, and split albedo.

Preferably, when the image data of the glassware is collected by the image collection terminal, at least two collection angles are set according to the shape and size of the glassware, and the image collection terminal is automatically adjusted by the set collection angles.

Preferably, the data collection module integrates a plurality of the image data to generate the image data set, including:

Verifying a plurality of the image data, and obtaining a plurality of identification labels of the image data corresponding to the glassware after the verification is passed; wherein, the identification labels are used to identify the glassware;

Sorting a plurality of the image data according to the acquisition sequence, and associating the identification tag with the sorted plurality of the image data to generate the image data set.

Preferably, before dividing the image data set, the data analysis module performs image processing on a plurality of the image data in the image data set; wherein, the image processing includes image correction, image segmentation and grayscale transformation.

Preferably, the data analysis module analyzes the first data to determine the position of air bubbles in the glassware, including:

Statistical analysis is performed on the grayscale of the pixel points in the data one, and the number of bubbles and the corresponding bubble positions in the data one are determined according to the statistical analysis result.

Preferably, when there are glass bubbles in the first data, the bubble parameters of the glass bubbles are obtained through the corresponding second data, including:

Positioning the glass bubble in combination with the bubble position in the second data;

The bubble parameters of the glass bubbles are obtained by performing statistical analysis on the pixel points corresponding to the glass bubbles and in combination with the statistical analysis results of the first data; wherein the bubble parameters include bubble size and bubble shape.

Preferably, the data analysis module determines the cause of the bubble in combination with the intelligent classification model, including:

Processing the bubble quantity and the bubble parameter corresponding to the same glassware to generate bubble characteristic data;

calling the intelligent classification model;

Input the bubble feature data into the intelligent classification model, obtain the output reason label, and determine the bubble reason according to the reason label; wherein, the reason label and the bubble reason are in one-to-one correspondence.

Compared with the prior art, the beneficial effects of the present invention are:

1. The present invention generates an image data set by collecting image data from different angles of glassware, and divides the image data set into two parts, one part is used for rough detection, and the other part is used for fine detection, so as to obtain relevant parameters of glass bubbles, and then combine them. The intelligent classification model judges the cause of bubbles to achieve high-efficiency detection of glass bubbles in glassware.

2. In the present invention, the image acquisition terminal includes at least one group of image acquisition devices, the positions and angles of the camera and the light source in each group of the image acquisition devices are adjusted independently, and when image data acquisition is performed, according to the glass At least two acquisition angles are set for the size and shape of the vessel, and the acquisition angle is automatically adjusted to obtain image data of the glass vessel in different directions, which can ensure the accuracy of the location and identification of bubbles in the glass vessel.

Description of drawings

In order to explain the embodiments of the present invention or the technical solutions in the prior art more clearly, the following briefly introduces the accompanying drawings that need to be used in the description of the embodiments or the prior art. Obviously, the accompanying drawings in the following description are only These are some embodiments of the present invention. For those of ordinary skill in the art, other drawings can also be obtained according to these drawings without creative efforts.

Figure 1 is a schematic diagram of the working steps of the present invention.

Detailed ways

The technical solutions of the present invention will be clearly and completely described below with reference to the embodiments. Obviously, the described embodiments are only a part of the embodiments of the present invention, rather than all the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those of ordinary skill in the art without creative efforts shall fall within the protection scope of the present invention.

The prior art (invention patent with publication number CN109191440A) discloses a method for detecting and counting glass bubbles, obtaining a convolutional neural network model through sample set training, and then obtaining the covering position and glass bubbles of the glass bubbles through the convolutional neural network model. The number of glass bubbles can be accurately detected and counted. In the process of bubble detection in the prior art, the convolutional neural network model requires a large number of training sets, the requirements for the accuracy and data volume of the training set are very high, the data processing volume is very large, and the cause of the bubbles cannot be located according to the feedback of the detection results. Lead to poor glass bubble detection efficiency;

The invention generates an image data set by collecting image data of different angles of glassware, and divides the image data set into two parts, one part is used for rough detection, and the other part is used for fine detection, so as to obtain relevant parameters of glass bubbles, and then combined with intelligent classification The model judges the causes of bubbles to achieve high-efficiency detection of glass bubbles in glassware.

Referring to FIG. 1, an embodiment of the first aspect of the present application provides a bubble detection system for glassware, including a data analysis module, and a data acquisition module connected to it, and the data acquisition module is connected to an image acquisition terminal;

Data acquisition module: collect glassware from different angles through the image acquisition terminal to obtain several image data; integrate several image data to generate an image data set, and send it to the data analysis module;

Data analysis module: divide the image data set into data one and data two, analyze and locate the bubble position on data one, and obtain bubble parameters in combination with data two; and

Generate bubble feature data based on the number of bubbles and bubble parameters, and determine the cause of the bubbles combined with an intelligent classification model.

In this application, the high-efficiency and high-precision detection of glass bubbles is realized through simple image recognition technology, the data analysis module communicates and/or electrical connection with the data acquisition module, and the data acquisition module communicates and/or electrical connection with the image acquisition terminal.

The first data in this application includes one piece of image data, the second data includes at least one piece of image data, and the second data may also include the remaining image data in the image data set except for the first data.

In a preferred embodiment, the image acquisition terminal includes at least one group of image acquisition devices, each group of image acquisition devices includes a camera and a light source, and both the camera and the light source focus on the glassware.

In this embodiment, the image acquisition terminal is mainly used to collect image data of glassware, so a camera needs to be set. Meanwhile, in order to ensure the quality of image data, this embodiment also configures at least one light source, such as an LED light source, for each camera.

In this embodiment, the positions and angles of the cameras and the light sources in each group of image acquisition devices are individually adjusted, and the illumination modes of the light sources include backlight illumination, front illumination, and split-reflection illumination.

The backlight illumination method of front perspective can have clear and clear edges, but it is easy to be interfered by other optical fibers; the backlight illumination method of oblique perspective can transform dark defect features on a bright background into bright defect features on a dark background; spectroscopic albedo The method increases the contrast through the background plate; the realization process of the above-mentioned irradiation methods can refer to the paper "Application Research of Image Processing Technology in Glass Defect Detection" by Wang Pingshun of Yanshan University, and select the appropriate light source irradiation method according to the actual inspection environment.

It is worth noting that when collecting image data of glassware through the image collecting terminal, at least two collecting angles are set according to the shape and size of the glassware, and the image collecting terminal is automatically adjusted by the set collecting angles.

When identifying glass bubbles, at least two photos from different angles are required to obtain the bubble parameters of the glass bubbles. Some glass bubbles with irregular shapes require more image data. Therefore, the angle of the camera and the light source can be adjusted. Before the detection, the collection angle can be set according to the shape and size of the glassware and the common glass bubble types, and the collection angle can be automatically adjusted according to the subsequent detection.

It can be understood that the present application can be used in combination with glassware or other glass product production lines as a quality inspection link.

In a preferred embodiment, the data acquisition module integrates several image data to generate an image data set, including:

Verifying a number of image data, and obtaining the identification label of the glassware corresponding to the image data after the verification is passed;

Sort a number of image data according to the acquisition sequence, associate the identification tag with the sorted number of image data, and generate an image data set.

This embodiment integrates several collected image data. Before the integration, the image data needs to be verified, that is, to verify that the image data can meet the quality requirements. When all the image data can meet the quality requirements, it is combined with the glass The identification labels of the vessels are integrated together to generate an image dataset.

It can be understood that the identification label is used to identify the glassware, and the digital number of the glassware can be used as the identification label to facilitate the management of image data and the tracking of the glassware.

In a preferred embodiment, the data analysis module analyzes the first data to determine the position of air bubbles in the glassware, including:

Statistical analysis is performed on the grayscale of the pixel points in the data one, and the number of bubbles in the data one and the corresponding bubble positions are determined according to the statistical analysis results.

In this embodiment, a preliminary judgment is mainly made on the glassware through the first data to determine the position and number of bubbles in the glassware, and then the second data is used to make an accurate judgment on the basis of the preliminary judgment.

This embodiment mainly uses image recognition technology. According to the above papers and experience, if (glass) bubbles appear in glassware or glass products, the gray value of the corresponding area will be larger, and the gray value of the surrounding pixels will be significantly different. The identification purpose can be achieved by setting a reasonable boundary value, and the boundary value can be set according to actual experience.

It can be understood that the data analysis module performs image processing on several image data in the image data set before dividing the image data set, and the image processing includes image correction, image segmentation and grayscale transformation to facilitate subsequent (glass) bubbles. identify.

In a specific embodiment, when there are glass bubbles in the first data, the bubble parameters of the glass bubbles are obtained through the corresponding data two, including:

Position the glass bubble in combination with the bubble position in the second data;

The bubble parameters of the glass bubble are obtained by statistical analysis of the pixel points corresponding to the glass bubble, combined with the statistical analysis results of data one.

After the number of bubbles in the glassware and the corresponding bubble position are determined through data 1, the size and shape of the bubbles, that is, the bubble parameters, need to be determined through data 2, which is different from the acquisition angle of data 1. The analysis of data 2 is still through image recognition technology, which is convenient and fast.

In a preferred embodiment, the data analysis module combines the intelligent classification model to determine the cause of the bubble, including:

The number of bubbles and bubble parameters corresponding to the same glassware are processed to generate bubble characteristic data;

Call the intelligent classification model;

Input the bubble feature data into the intelligent classification model, obtain the output cause label, and determine the cause of the bubble according to the cause label.

The intelligent classification model in this embodiment is obtained through an artificial intelligence model, including:

Get standard training data;

The constructed artificial intelligence model is trained through standard training data, and the trained artificial intelligence model is marked as an intelligent classification model and stored in the data analysis module.

The output data in the standard training data is consistent with the content attributes of the bubble feature data. An example of how to integrate the number of bubbles and bubble parameters into the bubble feature data:

Get the shape label corresponding to the bubble shape; for example, the shape label of a circle is 1, and the shape label of an irregular shape is 2;

The number of bubbles corresponding to the shape of the bubbles, and the proportion of the glass bubbles of this shape to the total area of the glassware (pixel ratio);

Then integrate bubble feature data such as [shape label, number of bubbles, proportion].

The output data corresponding to the input data in the standard training data is set according to the expert experience. If the experienced staff judges that the (glass) bubbles are caused by the glass melt bubbles, the reason label is set to 1, because the forming process is related. , the reason label is set to 2, and the reason label is set to 3 for multiple reasons, and the bubble reason can be obtained after obtaining the reason label.

The working principle of the present invention:

The data acquisition module collects the glassware from different angles through the image acquisition terminal, acquires several image data, integrates the several image data to generate an image data set, and sends it to the data analysis module.

The data analysis module divides the image data set into data 1 and data 2, analyzes the data 1 to locate the bubble position, and obtains the bubble parameters in combination with the data 2; then generates the bubble characteristic data according to the number of bubbles and the bubble parameters, and determines the cause of the bubble in combination with the intelligent classification model. .

The above embodiments are only used to illustrate the technical method of the present invention and not limit it. Although the present invention has been described in detail with reference to the preferred embodiments, those of ordinary skill in the art should understand that the technical method of the present invention can be modified or equivalently replaced. Without departing from the spirit and scope of the technical method of the present invention.

Claims (8)

1. a bubble detection system for glassware, comprising a data analysis module, and a data acquisition module connected with it, the data acquisition module is connected with an image acquisition terminal, and is characterized in that: Data acquisition module: collect glassware from different angles through the image acquisition terminal to obtain a number of image data; integrate a number of the image data to generate an image data set, and send it to the data analysis module; Data analysis module: divide the image data set into data one and data two, analyze and locate the bubble position on the data one, and obtain the bubble parameters in combination with the data two; wherein, the data one includes a piece of image data; as well as The bubble characteristic data is generated according to the number of bubbles and the bubble parameters, and the cause of the bubbles is determined in combination with the intelligent classification model; wherein, the intelligent classification model is established based on the artificial intelligence model. 2 . The bubble detection system for glassware according to claim 1 , wherein the image acquisition terminal includes at least one group of image acquisition devices, and each group of the image acquisition devices includes a camera and a light source, 3 . And both the camera and the light source are focused on the glassware. 3. A bubble detection system for glassware according to claim 2, wherein the positions and angles of the cameras and the light sources in each group of the image acquisition devices are adjusted independently, and the The illumination modes of the light source include backlight illumination, front illumination and spectral albedo. 4 . The bubble detection system for glassware according to claim 3 , wherein when collecting the image data of the glassware through the image acquisition terminal, at least two bubbles are set according to the shape and size of the glassware. A collection angle is set, and the image collection terminal is automatically adjusted through the set collection angle. 5. The bubble detection system for glassware according to claim 1, wherein the data acquisition module integrates a plurality of the image data to generate the image data set, comprising: Verifying a plurality of the image data, and obtaining a plurality of identification labels of the image data corresponding to the glassware after the verification is passed; wherein, the identification labels are used to identify the glassware; Sorting a plurality of the image data according to the acquisition sequence, and associating the identification tag with the sorted plurality of the image data to generate the image data set. 6. A bubble detection system for glassware according to claim 1, wherein the data analysis module analyzes the data one to determine the position of the bubbles in the glassware, comprising: Statistical analysis is performed on the grayscale of the pixel points in the data one, and the number of bubbles and the corresponding bubble positions in the data one are determined according to the statistical analysis result. 7. A bubble detection system for glassware according to claim 6, characterized in that, when glass bubbles exist in the data one, the bubble parameters of the glass bubbles are obtained through the corresponding data two, include: Positioning the glass bubble in combination with the bubble position in the second data; The bubble parameters of the glass bubbles are obtained by performing statistical analysis on the pixel points corresponding to the glass bubbles and in combination with the statistical analysis results of the first data; wherein the bubble parameters include bubble size and bubble shape. 8. The bubble detection system for glassware according to claim 7, wherein the data analysis module determines the cause of the bubble in combination with the intelligent classification model, comprising: Processing the bubble quantity and the bubble parameter corresponding to the same glassware to generate bubble characteristic data; calling the intelligent classification model; Input the bubble feature data into the intelligent classification model, obtain the output cause label, and determine the cause of the bubble according to the reason label.

Images