CN115452844B - Injection molding part detection method and system based on machine vision - Google Patents

Abstract

The invention discloses a method and system for detecting injection molded parts based on machine vision. The technical scheme of the invention processes two images irradiated by different light sources by collecting two images of the same ring-shaped injection molded part to be tested. According to the feature of enlarged reflective points on the part, the reflective points are filtered according to the brightness value of each highlighted position point in the two collected images, which eliminates the impact noise of the reflective points on the ring-shaped injection molded part on the recognition of the shape of the bubble defect, and overcomes the inability of the existing technology to analyze the injection molding The technical problem of accurately identifying the bubble defects on the injection molded parts, especially the bubble defects on the annular injection molded parts, realizes the accurate identification of the bubble defects on the injection molded parts, especially the bubble defect recognition of the annular injection molded parts, which can improve the detection of injection molded parts success rate and accuracy.

Description

A machine vision-based inspection method and system for injection molded parts

technical field

The invention relates to the technical field of image processing, in particular to a method and system for detecting injection molded parts based on machine vision.

Background technique

Injection molded parts are workpieces made of materials such as polypropylene and polyethylene by fusing a variety of organic solvents. The production principle is to inject plastic pellets into the model after high-temperature melting, extrude the machine and cool it to form a piece. The production process is greatly affected by other factors such as temperature, pressure and model, and the surface of the produced injection molded parts is large. There may be a lack of glue or too much glue in the probability, so the injection molded parts after production usually need to find defects through inspection.

At present, the types of defects in injection molded parts are common in: shrinkage, air marks, lack of material, drape, and line clamping. In the face of the above-mentioned common defect types, the existing inspection strategies for injection molded parts are basically based on manual inspection. The staff observe the surface conditions of the produced injection molded parts and classify the defective injection molded parts. The above-mentioned obvious defects may still be detected manually, but special defect types such as "bubbles" are encountered. Bubbles are formed because the material is injected too fast, or voids are caused by uneven volume shrinkage during molding, so that the bubbles exist inside the injection molded part. However, some high-precision injection molded parts have high requirements for finished products, and the appearance of air bubbles will require the use of equipment after the injection molded parts are installed. However, the traditional manual detection method is extremely inefficient and has limited accuracy visible to the naked eye, making it impossible to accurately identify bubble defect points. Although with the development of image processing technology, there are currently some researches on injection molded parts on the market for bubble defects, but they are all limited to the analysis of the cause of bubble generation, and cannot find bubble defects on injection molded parts; plus In the image detection process of annular injection molded parts, since the reflective points of the annular injection molded part itself will have a certain impact on bubble recognition, it is more difficult to detect bubble defects in annular injection molded parts.

Therefore, there is an urgent need for a new detection strategy for injection molded parts on the market, which can accurately identify bubble defects on injection molded parts, especially for ring-shaped injection molded parts, and improve the detection success rate and accuracy of injection molded parts.

Contents of the invention

The invention provides a method and system for detecting injection molded parts based on machine vision, which can accurately identify bubble defects on injection molded parts, especially the recognition of bubble defects on ring-shaped injection molded parts, which can improve the detection success rate and accuracy of injection molded parts .

In order to solve the above-mentioned technical problems, an embodiment of the present invention provides a machine vision-based inspection method for injection molded parts to detect bubble defects on ring-shaped injection molded parts. The method includes:

The image acquisition of the annular injection molded part to be tested is carried out by the photographing device in a closed space to obtain the first captured image; the positions of the photographing device and the annular injection molded part to be tested are kept unchanged, and after a light source is placed in the closed space, the The annular injection molded part to be tested is subjected to secondary image acquisition to obtain a second acquired image;

Recognizing and marking the annular boundary features in the first collected image and the second collected image, respectively determining the ring shape of the annular injection molded part to be tested in the first collected image and the second collected image boundary;

After preprocessing the first collected image and the second collected image, input them into a pre-established highlight area model for identification, respectively mark and output the highlighted positions existing on the first collected image and the second collected image point;

Respectively determine the highlighted position points of the first captured image and the second captured image on the circular boundary, and determine the brightness value of each highlighted position point on the circular boundary;

According to the difference between the luminance values of the highlighted position points at the same position on the circular boundary between the first captured image and the second captured image, the highlighted position points whose difference value is greater than a preset threshold are used as influencing factors in the second captured image. - filtering in the collected image to obtain the filtered image;

The filtered image is input into a pre-established bubble shape recognition model for recognition, and the highlighted position points in the filtered image whose shape meets the shape of the bubble defect are marked and output as the bubble defect on the annular injection molded part to be tested.

As a preferred solution, the identifying and marking the annular boundary features in the first collected image and the second collected image determine the to-be In the step of measuring the annular boundary of the annular injection molded part, it specifically includes:

Carrying out grid processing on the first collected image and the second collected image respectively, and determining a reference point;

Establishing a three-dimensional coordinate system, using the reference point as the origin, respectively moving the first captured image and the second captured image into the three-dimensional coordinate system, and determining that each gridded point is in the three-dimensional coordinate system The coordinate position in;

Determining a connection line formed between a plurality of consecutive gridded points with the same chromaticity in the first captured image, and the chromaticity between the connection line and adjacent gridded points not on the connection line If the difference reaches the chromaticity threshold, the connection is determined as a ring boundary;

According to the circular boundary determined in the first captured image, move to the second captured image with the reference point as a reference in the three-dimensional coordinate system, and determine that the ring-shaped injection molded part to be tested is in the second captured image The circular boundary in the image.

As a preferred solution, the step of preprocessing the first collected image and the second collected image specifically includes:

performing grayscale processing on the first collected image and the second collected image to obtain corresponding grayscale images respectively;

In the three-dimensional coordinate system, taking the reference point as a midpoint, stretching the grayscale image horizontally by a certain multiple to obtain a stretched image;

Identifying the spot features existing in the stretched image, and filtering the spot features in the stretched image to obtain a filtered image;

According to the multiple of the horizontal stretching, after the filtered image is horizontally reduced, the preprocessed image is input to the pre-established highlight area model.

As a preferred solution, the step of establishing the highlighted region model includes:

Acquiring a training image, wherein the training image is obtained by capturing images of the training ring-shaped injection molded part by a shooting device after placing a light source in a closed space;

According to the chromaticity of the training image, mark the shape boundary of the highlight region in the training image, and determine the closest distance point between the center point of each highlight region and the ring boundary of the training annular injection molded part , associating the distance point with the corresponding highlighted region;

Establishing an initial highlight model through a machine learning algorithm, inputting associated training images into the initial highlight model for training, until the number of training times reaches a threshold, and generating a training highlight model;

Obtaining a test image, wherein the test image is obtained by capturing images of training annular injection molded parts in a closed space by a photographing device;

The test image is input into the training highlight model for testing, when the accuracy of the highlight position points in the test image marked by the training highlight model in the output image reaches a preset threshold , to generate a highlighted region model.

As a preferred solution, the step of identifying the spot features existing in the stretched image specifically includes:

Identifying the irregular graphics existing in the stretched image, and determining the irregular graphics existing in the stretched image;

Divide a multi-layer ring area for each irregular figure, and determine multiple test points in each layer of ring area, and at the same time, determine the chromaticity of each test point;

Calculate the average chromaticity of all test points in the circular area of each layer, and use the average chromaticity as the chromaticity value of the circular area;

When it is determined that in the same irregular figure, the chromaticity value on the outermost circular area decreases successively toward the innermost circular area, then it is determined that the irregular figure is a spot feature existing in the stretched image.

As a preferred solution, the step of respectively determining the highlighted position points of the first captured image and the second captured image on the circular boundary, and determining the brightness value of each highlighted position point on the circular boundary, specifically includes :

Respectively determining the area range where each highlighted position point on the circular boundary of the first captured image and the second captured image is located;

A circumscribed circle is determined for the area range of each highlighted position point, and the brightness value corresponding to the position of the center of the circumscribed circle is used as the brightness value of the corresponding highlighted position point.

As a preferred solution, according to the difference between the brightness values of the highlighted position points at the same position on the circular boundary between the first acquired image and the second acquired image, the highlighted position points whose difference is greater than a preset threshold are taken as In the step of filtering the impact factor in the first collected image to obtain the filtered image, it specifically includes:

Taking the coordinate position of each highlighted position point in the three-dimensional coordinate system in the first collected image as a whole, defining it as the first coordinate position;

Taking the coordinate position of each highlighted position point in the second captured image in the three-dimensional coordinate system as a whole as a second coordinate position;

Taking the first coordinate position as a reference, moving the second coordinate position in a three-dimensional coordinate system as a whole until the second coordinate position coincides with the first coordinate position;

Determining the difference between the brightness values of the highlighted position points at the same position after overlapping, and filtering the highlighted position points whose difference value is greater than a preset threshold as an influencing factor in the first collected image to obtain a filtered image.

As a preferred solution, the step of establishing the bubble shape recognition model includes:

In the process of pre-establishing the bubble shape recognition model, the filtered image obtained after the execution of the above steps is obtained;

Marking the air bubbles existing in the filtered image by means of manual identification to determine the range of air bubble defects;

Identifying the gray value in each bubble defect range, and dividing the bubble defect range into two regions according to the change of the gray value in the bubble defect range;

Determine the circumscribed circles for the two areas where the same bubble defect range is located, and correlate the centers of the circumscribed circles corresponding to the two areas;

An initial bubble model is established through a machine learning algorithm, and the associated filtered images are input into the initial bubble model for training and testing, until the number of training and testing reaches a threshold, and a bubble shape recognition model is generated.

Correspondingly, another embodiment of the present invention also provides a machine vision-based injection molded part detection system for detecting bubble defects on annular injection molded parts. The system includes: an image acquisition module, a ring boundary module, a high A bright recognition module, a highlight determination module, an image filter module and a bubble recognition module;

The image acquisition module is used to acquire the image of the annular injection molded part to be tested in a closed space by a photographing device to obtain a first captured image; keeping the positions of the photographing device and the annular injection molded part to be tested unchanged, in the After the light source is placed in the closed space, a secondary image acquisition is performed on the annular injection molded part to be tested to obtain a second acquired image;

The circular boundary module is configured to identify and mark the circular boundary features in the first collected image and the second collected image, and determine the circular boundary features in the first collected image and the second collected image respectively. Describe the annular boundary of the annular injection molded part to be tested;

The highlight recognition module is configured to preprocess the first captured image and the second captured image and then input them into a pre-established highlight area model for recognition, respectively mark and output the first captured image and the second captured image 2. Acquire the highlighted position points existing on the image;

The highlight determination module is configured to respectively determine the highlight position points of the first captured image and the second captured image on the circular boundary, and determine the brightness value of each highlighted position point on the circular boundary;

The image filtering module is configured to, according to the difference between the brightness values of the highlighted position points at the same position on the circular boundary between the first captured image and the second captured image, select the highlighted position whose difference is greater than a preset threshold Points are used as influencing factors to filter in the first collected image to obtain a filtered image;

The bubble recognition module is used to input the filtered image into the pre-established bubble shape recognition model for recognition, mark and output the highlighted position points in the filtered image whose shape meets the shape of the bubble defect, as the annular injection molding to be tested Bubble defects on parts.

As a preferred solution, the annular boundary module is specifically configured to: perform grid processing on the first acquired image and the second acquired image respectively, and determine a reference point; establish a three-dimensional coordinate system, with the reference point as origin, respectively moving the first acquired image and the second acquired image into the three-dimensional coordinate system, and determining the coordinate positions of each gridded point in the three-dimensional coordinate system; Determine the connection line formed between multiple consecutive grid points with the same chromaticity in the image, and the chromaticity difference between the connection line and the adjacent grid points that are not on the connection line reaches the chromaticity Threshold, determining the connecting line as a circular boundary; according to the circular boundary determined in the first captured image, moving to the second captured image in the three-dimensional coordinate system with the reference point as a reference , determining the annular boundary of the annular injection molded part to be tested in the second collected image.

As a preferred solution, the highlight recognition module is used in the step of preprocessing the first collected image and the second collected image, which specifically includes: processing the first collected image and the second collected image performing grayscale processing to obtain corresponding grayscale images respectively; in the three-dimensional coordinate system, taking the reference point as the midpoint, stretching the grayscale image horizontally by a certain multiple to obtain a stretched image; Identifying the spot features existing in the stretched image, and filtering the spot features in the stretched image to obtain a filtered image; according to the multiple of the horizontal stretching, after horizontally reducing the filtered image, The preprocessed image is input to the pre-established highlight region model.

As a preferred solution, the step of establishing the highlighted area model includes: acquiring a training image, wherein the training image is obtained by capturing images of training annular injection molded parts by a shooting device after placing a light source in a closed space; The chromaticity of the training image, marking the shape boundary of the highlighted area in the training image, and determining the shortest distance point between the center point of each highlighted area and the annular boundary of the training annular injection molded part, Associating the distance point with the corresponding highlight area; establishing an initial highlight model through a machine learning algorithm, and inputting the associated training image into the initial highlight model for training until the number of training times reaches a threshold, Generate a training highlight model; acquire a test image, wherein the test image is obtained by capturing images of the training annular injection molded part in a closed space by a shooting device; input the test image into the training highlight model for testing , when in the output image the accuracy of marking the highlighted position points in the test image with the highlighted region by the training highlighted model reaches a preset threshold, a highlighted region model is generated.

As a preferred solution, the highlight identification module is used in the step of identifying the spot features existing in the stretched image, which specifically includes: identifying irregular patterns existing in the stretched image, and determining the presence of Irregular figures in the stretched image; each irregular figure is divided into multi-layer ring regions, and a plurality of test points are determined in each layer of ring regions, and at the same time, the location of each test point is determined Chromaticity; Calculate the average chromaticity of all test points in the ring area of each layer, and use the average chromaticity as the chromaticity value of the ring area; when determining the same irregular figure, the outermost ring If the chromaticity value on the area decreases successively toward the innermost circular area, then it is determined that the irregular figure is the spot feature existing in the stretched image.

As a preferred solution, the highlight determination module is specifically configured to: respectively determine the area range where each highlight position point on the circular boundary of the first captured image and the second captured image is located; for each highlight position The circumscribed circle is determined by the area range of the point, and the brightness value corresponding to the position of the center of the circumscribed circle is used as the brightness value of the point corresponding to the highlighted position.

As a preferred solution, the image filtering module is specifically configured to: define the coordinate position of each highlighted position point in the first captured image in the three-dimensional coordinate system as a whole as the first coordinate position; define the second The coordinate position of each highlighted position point in the collected image in the three-dimensional coordinate system is defined as a second coordinate position as a whole; taking the first coordinate position as a reference, the second coordinate position is carried out in the three-dimensional coordinate system Move as a whole until the second coordinate position coincides with the first coordinate position; determine the difference between the brightness values of the highlighted position points at the same position after the coincidence, and use the highlighted position points whose difference is greater than the preset threshold as the influence Factors are filtered in the first acquired image to obtain a filtered image.

As a preferred solution, the step of establishing the bubble shape recognition model includes: in the process of pre-establishing the bubble shape recognition model, obtaining the filtered image obtained after the execution of the above steps; manually identifying the filtered image Mark the bubbles existing in the bubble to determine the bubble defect range; identify the gray value in each bubble defect range, and divide the bubble defect range into two regions according to the change of the gray value in the bubble defect range; Determine the circumscribed circles for the two areas where the same bubble defect range is located, and correlate the centers of the circumscribed circles corresponding to the two areas; establish an initial bubble model through a machine learning algorithm, and input the associated filtered image into the initial bubble model Perform training and testing in , until the number of training and testing reaches the threshold, a bubble shape recognition model is generated.

An embodiment of the present invention also provides a computer-readable storage medium, where the computer-readable storage medium includes a stored computer program; wherein, when running, the computer program controls the device where the computer-readable storage medium is located to execute the following steps: The machine vision-based inspection method for injection molded parts described in any one of the above.

An embodiment of the present invention also provides a terminal device, including a processor, a memory, and a computer program stored in the memory and configured to be executed by the processor. When the processor executes the computer program, the A machine vision-based inspection method for injection molded parts as described in any one of the above.

Compared with the prior art, the embodiments of the present invention have the following beneficial effects:

The technical scheme of the present invention processes two images of the same ring-shaped injection molded part to be tested, which are irradiated by different light sources, respectively, and utilizes the feature that the reflective points on the ring-shaped injection molded part can be enlarged after being irradiated by the light source. The brightness value of the position point filters the reflective point, which eliminates the influence noise of the reflective point on the ring-shaped injection molded part for the shape recognition of the bubble defect, and overcomes the inability of the existing technology to accurately detect the bubble defect on the injection molded part, especially the bubble defect on the ring-shaped injection molded part To solve the technical problems of identification, realize accurate identification of bubble defects on injection molded parts, especially the identification of bubble defects on annular injection molded parts, which can improve the detection success rate and accuracy of injection molded parts.

Description of drawings

Fig. 1: a flow chart of the steps of a machine vision-based injection molded part detection method provided by an embodiment of the present invention;

Fig. 2: a schematic structural diagram of a machine vision-based injection molding inspection system provided by an embodiment of the present invention;

FIG. 3 is a schematic structural diagram of an embodiment of a terminal device provided by an embodiment of the present invention.

Detailed ways

The following will clearly and completely describe the technical solutions in the embodiments of the present invention with reference to the accompanying drawings in the embodiments of the present invention. Obviously, the described embodiments are only some, not all, embodiments of the present invention. Based on the embodiments of the present invention, all other embodiments obtained by persons of ordinary skill in the art without creative efforts fall within the protection scope of the present invention.

Embodiment one

Please refer to FIG. 1 , which is a flow chart of steps of a machine vision-based inspection method for injection molded parts provided by an embodiment of the present invention. The method is used to detect bubble defects on annular injection molded parts, including steps 101 to 106, and each step is specifically as follows:

Step 101: Collecting images of the annular injection molded part to be tested in a closed space by a photographing device to obtain a first captured image; keeping the positions of the photographing device and the annular injection molded part to be tested unchanged, and placing a light source in the closed space Afterwards, a second image acquisition is performed on the annular injection molded part to be tested to obtain a second acquired image.

Specifically, since the image of the injection molded part will produce reflective points during the process of identifying defects, and the problem of generating reflective points in the ring-shaped injection molded part will be more serious due to the radian itself. When the image of the injection molded part produces reflective points, some of the reflective points are often misidentified as bubble defects during the recognition process, or the recognition of bubble defects is affected due to the excessive coverage of the reflective points. Based on the above reasons, in the process of inspecting injection molded parts, especially in the process of inspecting annular injection molded parts, it is especially necessary to eliminate the influence of reflective points on the image of the injection molded part itself. This step needs to collect two images, one image is based on the collection of annular injection molded parts in a closed space, and the other image is an image collected under the condition of an external light source. Through research, it is found that after adding a light source to the original image, the reflective points on the image of the injection molded part will be enlarged, which is more conducive to our identification and removal of reflective points.

Step 102: Identify and mark the annular boundary features in the first collected image and the second collected image, and determine the annular injection mold to be tested in the first collected image and the second collected image respectively. The circular boundary of the piece.

In this embodiment, the step 102 specifically includes: step 1021, performing grid processing on the first collected image and the second collected image respectively, and determining a reference point; step 1022, establishing a three-dimensional coordinate system, Taking the reference point as the origin, respectively moving the first captured image and the second captured image into the three-dimensional coordinate system, and determining the coordinate positions of each gridded point in the three-dimensional coordinate system; Step 1023, determining a connection line formed between a plurality of consecutive gridded points with the same chromaticity in the first captured image, and the distance between the connection line and adjacent gridded points that are not on the connection line If the difference between the chromaticity reaches the chromaticity threshold, the connecting line is determined as a circular boundary; step 1024, according to the circular boundary determined in the first captured image, in the three-dimensional coordinate system with the reference point As a reference, move to the second captured image, and determine the annular boundary of the annular injection molded part to be tested in the second captured image.

Specifically, through the above step 101, the reflection point of the injection molded part can be enlarged. In order to further identify the impact of the annular boundary on the reflective point, specifically for the annular injection molded part, that is, to identify the reflective point on the annular injection molded part that is more difficult to break through, we need to first determine the annular boundary on the annular injection molded part. In the specific recognition process, in order to eliminate the impact of external light sources on the recognition of the ring boundary (for example, if the exposure time is too long, the ring boundary will be blurred), in this step, we use the first and second collected images to compare Align at the point position. Then, according to the color difference formed by the circular boundary and the surrounding grid area on the non-boundary, the complete circular boundary can be identified to prevent the inability to fully identify the reflective points that appear on the circular boundary in the subsequent identification process.

Step 103, preprocessing the first collected image and the second collected image and inputting them into a pre-established highlight area model for identification, respectively marking and outputting the Highlight the location point.

In this embodiment, the step 103 is used in the step of preprocessing the first collected image and the second collected image, which specifically includes: Step 1031, processing the first collected image and the second collected image 2. Gather images and perform grayscale processing to obtain corresponding grayscale images respectively; step 1032, in the three-dimensional coordinate system, take the reference point as the midpoint, and stretch the grayscale image horizontally by a certain multiple , to obtain a stretched image; step 1033, identify the spot features existing in the stretched image, and filter the spot features in the stretched image to obtain a filtered image; step 1034, according to the horizontally stretched After the filtered image is reduced horizontally, the preprocessed image is input to the pre-established highlight area model.

Wherein, in another aspect of this embodiment, the step 1033 is used in the step of identifying the spot features existing in the stretched image, which specifically includes: identifying irregular patterns existing in the stretched image Identify and determine the irregular graphics that exist in the stretched image; divide each irregular graphic into a multi-layer ring area, and determine a plurality of test points in each layer of ring area, and at the same time, determine The chromaticity of each test point; calculate the average chromaticity of all test points in the circular area of each layer, and use the average chromaticity as the chromaticity value of the circular area; when determining the same irregular figure, If the chromaticity values on the outermost ring area decrease successively toward the innermost ring area, then it is determined that the irregular pattern is a spot feature existing in the stretched image.

Specifically, in order to determine the reflective points on the annular boundary in the subsequent steps (basically all the reflective points are concentrated on the annular boundary, because during the light source irradiation process, the annular interface will reflect the light source to form reflection. Therefore, on the annular injection molded part The reflective points are basically concentrated on the circular boundary), we need to use the model to identify the reflective points on the injection molded parts. In order to more accurately identify the highlighted position point (that is, the suspected reflective point, which will be judged later), the image needs to be preprocessed first. In the preprocessing process, in order to reduce noise, it is necessary to remove other influencing factors in the image area other than the annular injection molded part (for example, noise and other light spots). At this time, after the image is stretched, the light spot can be stretched into an irregular image, and the light spot noise in the image can be judged and filtered by using the characteristic that the chromaticity of the light spot in the image decreases from the outer layer to the inside.

In this embodiment, the step of establishing the highlighted area model includes: acquiring a training image, wherein the training image is obtained by capturing images of the training annular injection molded part by a shooting device after placing a light source in a closed space ; According to the chromaticity of the training image, mark the shape boundary of the highlight region in the training image, and determine the shortest distance between the center point of each highlight region and the ring boundary of the training annular injection molded part point, associate the distance point with the corresponding highlight area; establish an initial highlight model through a machine learning algorithm, and input the associated training images into the initial highlight model for training until the number of training times reaches the threshold Afterwards, generate a training highlight model; obtain a test image, wherein the test image is obtained by capturing images of the training annular injection molded part in a closed space by a shooting device; input the test image into the training highlight model A test is performed, and when the accuracy of marking the highlight position points in the test image by the training highlight model in the output image reaches a preset threshold, a highlight area model is generated.

Specifically, the key points of this solution also exist in the link of establishing the highlighted area model. Because the function of the highlight area model is used to accurately identify the highlight position points, when we train this model, we need to guide the position of the highlight area in the training image and the associated features on the circular boundary. The continuous circular boundary of the model learns the position of the highlighted area of the image, and the finally generated highlighted area model can recognize the input image and generate the marked highlighted position point.

Step 104, respectively determine the highlighted position points on the circular boundary of the first captured image and the second captured image, and determine the brightness value of each highlighted position point on the circular boundary.

In this embodiment, the step 104 specifically includes: step 1041, respectively determining the area range of each highlighted position point on the circular boundary of the first captured image and the second captured image; step 1042, for The area range of each highlighted position point determines a circumscribed circle, and the brightness value corresponding to the position of the center of the circumscribed circle is used as the brightness value of the corresponding highlighted position point.

Specifically, the highlighted position points are determined by means of circumscribed circles, which actually uses the center of the circumscribed circle as the brightness value of the corresponding highlighted position points, which can make the brightness value assignment of each highlighted position point more accurate, so that the following One step to distinguish which of these highlighted points are the real reflective points.

Step 105, according to the difference between the brightness values of the highlighted position points at the same position on the circular boundary between the first acquired image and the second acquired image, use the highlighted position points whose difference is greater than a preset threshold as the influencing factors in the filtering in the first collected image to obtain a filtered image.

In this embodiment, the step 105 specifically includes: step 1051, taking the coordinate position of each highlighted position point in the first captured image in the three-dimensional coordinate system as a whole, and defining it as the first coordinate position; step 1052 , define the coordinate position of each highlighted position point in the three-dimensional coordinate system in the second captured image as a whole as a second coordinate position; step 1053, take the first coordinate position as a reference, and define the first coordinate position The two coordinate positions move as a whole in the three-dimensional coordinate system until the second coordinate position coincides with the first coordinate position; step 1054, determine the difference in brightness values of the highlighted position points at the same position after the coincidence, and convert the difference The highlighted position points whose values are greater than the preset threshold are used as influencing factors and filtered in the first collected image to obtain a filtered image.

Specifically, by using the characteristic that the reflective points are enlarged after adding a light source, by calculating the difference between the corresponding brightness values in the two captured images at the same position, it can be determined which of these highlighted positions are the real reflective points, and then filtered After leaving a noise-free image (remove the reflective points). In order to make the data more accurate, we will also consider the step of determining the "same position" to avoid image movement during multiple acquisitions and processing. We use the mobile equivalence relationship of the three-dimensional coordinate system to convert the first acquisition The image is moved as a whole until it coincides with the second collected image, and then the position on the same coordinate point can be regarded as the "same position".

Step 106, input the filtered image into the pre-established bubble shape recognition model for recognition, mark and output the highlighted position points in the filtered image whose shape satisfies the shape of the bubble defect, as the bubble defect on the annular injection molded part to be tested .

In this embodiment, the step of establishing the bubble shape recognition model includes: in the process of pre-establishing the bubble shape recognition model, obtaining the filtered image obtained after the execution of the above steps; Filter the bubbles in the image to mark and determine the bubble defect range; identify the gray value of each bubble defect range, and divide the bubble defect range into two regions according to the change of the gray value in the bubble defect range ; Determine the circumscribed circles for the two areas where the same bubble defect range is located, and correlate the centers of the circumscribed circles corresponding to the two areas; establish an initial bubble model through a machine learning algorithm, and input the associated filtered image to the initial Train and test in the bubble model, until the number of training and testing reaches the threshold, a bubble shape recognition model is generated.

Specifically, through the above steps 101 to 105, we have obtained a noise-free image, and now we only need to input the noise-free image (ie, the filtered image) into the model for identifying the shape of the bubble defect for recognition. It can be understood that in the process of recognizing the bubble shape recognition model, we pre-operated steps 101 to 105, and then used the "filtered image" output in step 105 as the training image of the bubble shape recognition model. When the model training is completed Finally, in the process of retrieving the annular injection molded parts that need to be tested, there is no need to re-establish and train the bubble shape recognition model, and it can be used directly. In the process of building the bubble shape recognition model, the bubble defects in the image can be marked very accurately first by artificial means. Considering that in practical applications, the performance of bubble defects in the image will be due to the appearance of the shadow part, so that the bubbles will present two obviously layered areas on the injection molded part, and through the correlation of the circumscribed centers of the two areas, the model passes through After training, the bubble defects in the filtered image can be identified according to the shape of the bubble itself and the relationship between the two layered areas corresponding to the circumcenter of the bubbles, so as to accurately identify the bubble defects on the annular injection molded parts.

The technical scheme of the present invention processes two images of the same ring-shaped injection molded part to be tested, which are irradiated by different light sources, respectively, and utilizes the feature that the reflective points on the ring-shaped injection molded part can be enlarged after being irradiated by the light source. The brightness value of the position point filters the reflective point, which eliminates the influence noise of the reflective point on the ring-shaped injection molded part for the shape recognition of the bubble defect, and overcomes the inability of the existing technology to accurately detect the bubble defect on the injection molded part, especially the bubble defect on the ring-shaped injection molded part To solve the technical problems of identification, realize accurate identification of bubble defects on injection molded parts, especially the identification of bubble defects on annular injection molded parts, which can improve the detection success rate and accuracy of injection molded parts.

Embodiment two

Please refer to FIG. 2 , which is a schematic structural diagram of a machine vision-based injection molding inspection system provided by another embodiment of the present invention. The system is used to detect bubble defects on annular injection molded parts, including: an image acquisition module, a ring boundary module, a highlight recognition module, a highlight determination module, an image filter module and a bubble recognition module.

The image acquisition module is used to acquire the image of the annular injection molded part to be tested in a closed space by a photographing device to obtain a first captured image; keeping the positions of the photographing device and the annular injection molded part to be tested unchanged, in the After the light source is placed in the closed space, a secondary image acquisition is performed on the annular injection molded part to be tested to obtain a second acquired image.

The circular boundary module is configured to identify and mark the circular boundary features in the first collected image and the second collected image, and determine the circular boundary features in the first collected image and the second collected image respectively. Describe the annular boundary of the annular injection molded part to be tested.

In this embodiment, the annular boundary module is specifically configured to: perform grid processing on the first acquired image and the second acquired image respectively, and determine a reference point; establish a three-dimensional coordinate system, and use the reference point point as the origin, respectively move the first captured image and the second captured image into the three-dimensional coordinate system, and determine the coordinate positions of each gridded point in the three-dimensional coordinate system; A connection line formed between a plurality of consecutive grid points with the same chromaticity is determined in an acquired image, and the chromaticity difference between the connection line and adjacent grid points not on the connection line reaches If the chromaticity threshold is used, the connecting line is determined as a circular boundary; according to the circular boundary determined in the first captured image, move to the second captured image with the reference point as a reference in the three-dimensional coordinate system In the image, the annular boundary of the annular injection molded part to be tested in the second collected image is determined.

The highlight recognition module is configured to preprocess the first captured image and the second captured image and then input them into a pre-established highlight area model for recognition, respectively mark and output the first captured image and the second captured image 2. Acquire the highlighted position points existing on the image.

In this embodiment, the highlight recognition module is used in the step of preprocessing the first captured image and the second captured image, specifically including: Collecting images and performing grayscale processing to obtain corresponding grayscale images respectively; in the three-dimensional coordinate system, taking the reference point as the midpoint, stretching the grayscale image horizontally by a certain multiple to obtain the stretched Image; identifying the spot features existing in the stretched image, filtering the spot features in the stretched image to obtain a filtered image; horizontally shrinking the filtered image according to the multiple of the horizontal stretch After that, the preprocessed image is input to the pre-established highlight region model.

Wherein, in another aspect of this embodiment, the step of identifying the spot features existing in the stretched image by the highlight recognition module specifically includes: identifying irregularities in the stretched image Graphics are identified to determine the irregular graphics that exist in the stretched image; each irregular graphic is divided into a multi-layer ring area, and a plurality of test points are determined in each layer of the ring area, and at the same time, Determine the chromaticity where each test point is located; calculate the average chromaticity of all test points in the circular area of each layer, and use the average chromaticity as the chromaticity value of the circular area; when determining the same irregular figure , the chromaticity value on the outermost ring area decreases successively toward the innermost ring area, then it is determined that the irregular figure is a spot feature existing in the stretched image.

In this embodiment, the step of establishing the highlighted area model includes: acquiring a training image, wherein the training image is obtained by capturing images of the training annular injection molded part by a shooting device after placing a light source in a closed space ; According to the chromaticity of the training image, mark the shape boundary of the highlight region in the training image, and determine the shortest distance between the center point of each highlight region and the ring boundary of the training annular injection molded part point, associate the distance point with the corresponding highlight area; establish an initial highlight model through a machine learning algorithm, and input the associated training images into the initial highlight model for training until the number of training times reaches the threshold Afterwards, generate a training highlight model; obtain a test image, wherein the test image is obtained by capturing images of the training annular injection molded part in a closed space by a shooting device; input the test image into the training highlight model A test is performed, and when the accuracy of marking the highlight position points in the test image by the training highlight model in the output image reaches a preset threshold, a highlight area model is generated.

The highlight determination module is configured to respectively determine the highlight position points of the first captured image and the second captured image on the circular boundary, and determine the brightness value of each highlight position point on the circular boundary.

In this embodiment, the highlight determination module is specifically configured to: respectively determine the area range where each highlight position point on the circular boundary of the first captured image and the second captured image is located; The circumscribed circle is determined by the area range of the bright position point, and the brightness value corresponding to the position of the center of the circumscribed circle is used as the brightness value corresponding to the highlighted position point.

The image filtering module is configured to, according to the difference between the brightness values of the highlighted position points at the same position on the circular boundary between the first captured image and the second captured image, select the highlighted position whose difference is greater than a preset threshold Points are used as influencing factors to filter in the first collected image to obtain a filtered image.

In this embodiment, the image filtering module is specifically configured to: define the coordinate positions of each highlighted position point in the first captured image in the three-dimensional coordinate system as a whole as a first coordinate position; The coordinate position of each highlighted position point in the second captured image in the three-dimensional coordinate system is defined as a second coordinate position as a whole; taking the first coordinate position as a reference, the second coordinate position is placed in the three-dimensional coordinate system Carry out overall movement in the center until the second coordinate position coincides with the first coordinate position; determine the difference between the brightness values of the highlighted position points at the same position after the coincidence, and select the highlighted position points whose difference value is greater than the preset threshold Filtering in the first collected image as an impact factor to obtain a filtered image.

The bubble recognition module is used to input the filtered image into the pre-established bubble shape recognition model for recognition, mark and output the highlighted position points in the filtered image whose shape meets the shape of the bubble defect, as the annular injection molding to be tested Bubble defects on parts.

In this embodiment, the step of establishing the bubble shape recognition model includes: in the process of pre-establishing the bubble shape recognition model, obtaining the filtered image obtained after the execution of the above steps; Filter the bubbles in the image to mark and determine the bubble defect range; identify the gray value of each bubble defect range, and divide the bubble defect range into two regions according to the change of the gray value in the bubble defect range ; Determine the circumscribed circles for the two areas where the same bubble defect range is located, and correlate the centers of the circumscribed circles corresponding to the two areas; establish an initial bubble model through a machine learning algorithm, and input the associated filtered image to the initial Train and test in the bubble model, until the number of training and testing reaches the threshold, a bubble shape recognition model is generated.

Embodiment Three

An embodiment of the present invention also provides a computer-readable storage medium, the computer-readable storage medium includes a stored computer program; wherein, when running, the computer program controls the device where the computer-readable storage medium is located to execute the above-mentioned The machine vision-based inspection method for injection molded parts described in any one of the embodiments.

Embodiment four

Please refer to FIG. 3 , which is a schematic structural diagram of an embodiment of a terminal device provided by an embodiment of the present invention. The terminal device includes a processor, a memory, and a program stored in the memory and configured to be executed by the processor. A computer program, when the processor executes the computer program, it implements the machine vision-based injection molded part inspection method described in any one of the above embodiments.

Preferably, the computer program can be divided into one or more modules/units (such as computer programs, computer programs), and the one or more modules/units are stored in the memory and executed by the processor Execute to complete the present invention. The one or more modules/units may be a series of computer program instruction segments capable of accomplishing specific functions, and the instruction segments are used to describe the execution process of the computer program in the terminal device.

The processor may be a central processing unit (Central Processing Unit, CPU), or other general-purpose processors, digital signal processors (Digital Signal Processor, DSP), application specific integrated circuits (Application Specific Integrated Circuit, ASIC), off-the-shelf Programmable gate array (Field-Programmable Gate Array, FPGA) or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components, etc., the general-purpose processor can be a microprocessor, or the processor can be any A conventional processor, the processor is the control center of the terminal equipment, and uses various interfaces and lines to connect various parts of the terminal equipment.

The memory mainly includes a program storage area and a data storage area, wherein the program storage area can store an operating system, an application program required by at least one function, etc., and the data storage area can store related data, etc. In addition, the memory can be a high-speed random access memory, or a non-volatile memory, such as a plug-in hard disk, a smart memory card (SmartMedia Card, SMC), a secure digital (Secure Digital, SD) card and a flash memory card ( Flash Card), etc., or the memory may also be other volatile solid-state memory devices.

It should be noted that the above-mentioned terminal device may include, but not limited to, a processor and a memory. Those skilled in the art can understand that the above-mentioned terminal device is only an example and does not constitute a limitation on the terminal device, and may include more or less parts, or a combination of certain parts, or different parts.

The specific embodiments described above have further described the purpose, technical solutions and beneficial effects of the present invention in detail. It should be understood that the above descriptions are only specific embodiments of the present invention and are not intended to limit the protection scope of the present invention. . In particular, for those skilled in the art, any modifications, equivalent replacements, improvements, etc. made within the spirit and principles of the present invention shall be included within the protection scope of the present invention.

Claims (10)

1. A machine vision based injection molding inspection method for detecting bubble defects on a ring-shaped injection molding, the method comprising:

acquiring an image of the annular injection molding piece to be detected in a closed space through shooting equipment to obtain a first acquired image; keeping the positions of the shooting equipment and the annular injection molding piece to be detected unchanged, and after a light source is put into the closed space, carrying out secondary image acquisition on the annular injection molding piece to be detected to obtain a second acquired image;

identifying and marking the annular boundary characteristics in the first collected image and the second collected image, and determining the annular boundary of the annular injection molding piece to be detected in the first collected image and the second collected image respectively;

preprocessing the first collected image and the second collected image, inputting the preprocessed first collected image and the preprocessed second collected image into a pre-established highlight area model for recognition, and respectively marking and outputting highlight position points existing on the first collected image and the second collected image;

respectively determining highlight position points of the first collected image and the second collected image on an annular boundary, and determining the brightness value of each highlight position point on the annular boundary;

filtering highlight position points with difference values larger than a preset threshold value in the first acquired image as influence factors according to the difference between the brightness values of the highlight position points at the same position on the annular boundary of the first acquired image and the second acquired image to obtain a filtered image;

and inputting the filtering image into a pre-established bubble shape recognition model for recognition, marking and outputting highlight position points with shapes meeting the shapes of bubble defects in the filtering image, and taking the highlight position points as the bubble defects on the annular injection molding piece to be detected.

2. The machine-vision-based injection molding inspection method of claim 1, wherein said step of identifying and marking the annular boundary features in the first captured image and the second captured image, and determining the annular boundary of the annular injection molding under test in the first captured image and the second captured image, respectively, comprises:

respectively carrying out gridding processing on the first collected image and the second collected image, and determining a reference point;

establishing a three-dimensional coordinate system, respectively moving the first collected image and the second collected image to the three-dimensional coordinate system by taking the reference point as an origin, and determining the coordinate position of each gridding point in the three-dimensional coordinate system;

determining a connecting line formed among a plurality of continuous gridding points with the same chromaticity in the first collected image, and determining the connecting line as an annular boundary when the chromaticity difference value between the connecting line and the adjacent gridding points which are not on the connecting line reaches a chromaticity threshold value;

and moving the reference point in the three-dimensional coordinate system to the second acquired image by taking the reference point as a reference according to the determined annular boundary in the first acquired image, and determining the annular boundary of the annular injection molding piece to be detected in the second acquired image.

3. The machine-vision-based injection molded part inspection method of claim 2, wherein the step of preprocessing the first captured image and the second captured image comprises:

graying the first collected image and the second collected image to respectively obtain corresponding grayscale images;

in the three-dimensional coordinate system, the reference point is used as a midpoint, and the gray level image is transversely stretched by a certain multiple to obtain a stretched image;

identifying light spot features existing in the stretched image, and filtering the light spot features in the stretched image to obtain a filtered image;

and according to the transverse stretching multiple, transversely reducing the filtered image to obtain a preprocessed image, and inputting the preprocessed image into a pre-established highlight area model.

4. The machine-vision-based injection molded part inspection method of claim 3, wherein the step of establishing the highlight region model comprises:

acquiring a training image, wherein the training image is obtained by shooting an image acquisition of a training annular injection molding piece by shooting equipment after a light source is put in a closed space;

according to the chromaticity of the training image, marking the shape boundary of the highlight generation area in the training image, respectively determining the center point of each highlight generation area and the nearest distance point on the annular boundary of the training annular injection molding part, and associating the distance points with the corresponding highlight generation areas;

establishing an initial highlight model through a machine learning algorithm, inputting the associated training images into the initial highlight model for training until the training times reach a threshold value, and generating a training highlight model;

acquiring a test image, wherein the test image is obtained by acquiring an image of a training annular injection molding piece in a closed space through shooting equipment;

and inputting the test image into the training highlight model for testing, and generating a highlight area model when the accuracy of the training highlight model for marking highlight position points with highlight areas in the test image in the output image reaches a preset threshold value.

5. The machine-vision-based injection molding inspection method of claim 3, wherein the step of identifying the spot features present in the stretched image comprises:

identifying irregular figures existing in the stretched image, and determining the irregular figures existing in the stretched image;

dividing a plurality of layers of circular ring areas in each irregular graph respectively, determining a plurality of test points in each layer of circular ring area, and simultaneously determining the chromaticity of each test point;

calculating the average chroma of all test points in each layer of ring area, and taking the average chroma as the chroma value of the ring area;

and when the chromatic values on the ring area at the outermost layer in the same irregular figure are determined to be sequentially decreased towards the ring area at the innermost layer, determining that the irregular figure is the spot feature existing in the stretched image.

6. The machine-vision-based injection molded part inspection method of claim 1, wherein said steps of determining highlight site points of said first captured image and said second captured image, respectively, on an annular boundary, and determining a brightness value of each highlight site point on the annular boundary, specifically comprise:

respectively determining the area range of each highlight position point on the annular boundary of the first collected image and the second collected image;

and determining an circumscribed circle aiming at the area range of each highlight position point, and taking the corresponding brightness value at the position of the circle center of the circumscribed circle as the brightness value of the corresponding highlight position point.

7. The machine-vision-based injection molding detection method of claim 6, wherein the step of filtering the highlight position points with the difference value larger than a preset threshold value in the first captured image as influence factors according to the difference between the brightness values of the highlight position points at the same position on the annular boundary of the first captured image and the second captured image to obtain a filtered image specifically comprises:

defining the coordinate position of each highlight position point in the first collected image in a three-dimensional coordinate system as a whole to be a first coordinate position;

defining the coordinate position of each highlight position point in the second collected image in the three-dimensional coordinate system as a whole to be a second coordinate position;

taking the first coordinate position as a reference, and integrally moving the second coordinate position in a three-dimensional coordinate system until the second coordinate position is superposed with the first coordinate position;

and determining the difference of the brightness values of the highlight position points at the same position after superposition, and filtering the highlight position points with the difference value larger than a preset threshold value in the first collected image as influence factors to obtain a filtered image.

8. A machine vision-based injection molding inspection system for detecting bubble defects on a ring-shaped injection molded part, the system comprising: the device comprises an image acquisition module, an annular boundary module, a highlight identification module, a highlight determination module, an image filtering module and a bubble identification module;

the image acquisition module is used for acquiring an image of the annular injection molding piece to be detected in a closed space through shooting equipment to obtain a first acquired image; keeping the positions of the shooting equipment and the annular injection molding piece to be detected unchanged, and after a light source is put into the closed space, carrying out secondary image acquisition on the annular injection molding piece to be detected to obtain a second acquired image;

the annular boundary module is used for identifying and marking annular boundary characteristics in the first collected image and the second collected image, and determining an annular boundary of the annular injection molding piece to be detected in the first collected image and the second collected image respectively;

the highlight identification module is used for preprocessing the first collected image and the second collected image, inputting the preprocessed first collected image and the preprocessed second collected image into a preset highlight area model for identification, and respectively marking and outputting highlight position points existing on the first collected image and the second collected image;

the highlight determining module is used for respectively determining highlight position points of the first acquired image and the second acquired image on the annular boundary and determining the brightness value of each highlight position point on the annular boundary;

the image filtering module is used for filtering the highlight position points with the difference value larger than a preset threshold value in the first collected image as influence factors according to the difference between the brightness values of the highlight position points at the same position on the annular boundary of the first collected image and the second collected image to obtain a filtered image;

and the bubble identification module is used for inputting the filtering image into a pre-established bubble shape identification model for identification, marking and outputting highlight position points with shapes meeting the shapes of bubble defects in the filtering image, and the highlight position points are used as the bubble defects on the annular injection molding part to be detected.

9. A computer-readable storage medium, characterized in that the computer-readable storage medium comprises a stored computer program; wherein the computer program, when executed, controls an apparatus on which the computer-readable storage medium resides to perform the machine-vision-based injection molding inspection method of any one of claims 1-7.

10. A terminal device comprising a processor, a memory, and a computer program stored in the memory and configured to be executed by the processor, the processor when executing the computer program implementing the machine-vision based injection molding detection method of any one of claims 1-7.

Images