Visual inspection system and device for quality of plastic bottle

Abstract

The invention relates to the technical field of plastic visual inspection, and particularly discloses a plastic bottle quality visual inspection system and device, and the system comprises an identification unit which is used for automatically recording and storing a test sample defect position image of a plastic bottle through an industrial camera; the classification unit is used for determining a screening condition of a defect feature point threshold range based on a FAST algorithm, counting the number of regional feature points of defect position images of different test samples under the current threshold range condition for classification confirmation, and obtaining a defect template coefficient to judge the similarity of visual feature points; the extraction unit is used for performing clustering analysis and classification according to the feature points with the similar visual features, reading and generating a dictionary file, obtaining vector representation of the image, marking the vector representation, reading defect text information of the plastic bottle detection sample and outputting a defect classification result; and the detection and correction unit is used for performing visual detection on the plastic bottles close to the target position in real time according to the defect classification result input by the system and generating correction prompt information.

Description

Technical Field

[0001] The present invention relates to the technical field of plastic visual inspection, and particularly to a plastic bottle quality visual inspection system and device. Background Art

[0002] With the rapid development of computer and machine vision technologies, the society and individuals have higher and higher requirements for the quality of industrial products; especially in the automated inspection of plastic bottle defects using machine vision technology, it is more efficient, stable and can quickly identify product defects compared with manual inspection, achieving a significant increase in the production task volume of products.

[0003] In the existing plastic bottle quality visual inspection systems, there are problems of fuzziness in the detection range of defect information and unclear comparison; therefore, it is necessary to automatically classify the information processed by the image, so as to accurately match the quality information of plastic bottles during the automated inspection process; and further optimize the real-time analysis of plastic bottle defects to improve the visual inspection accuracy.

[0004] However, in the existing technologies, for the visual inspection system and related devices for plastic bottles, there is a lack of a unified problem feature determination procedure for common plastic bottle defect problems during the process of using visual technology to detect the quality of plastic bottles. Therefore, it is difficult to directly confirm the feature of various defect problems and unable to formulate a unified feature recognition range; moreover, it is also impossible to quickly identify and confirm defective plastic bottles during the visual inspection process, thereby affecting the yield of qualified products of plastic bottles in the factory. Summary of the Invention

[0005] The purpose of the present invention is to provide a plastic bottle quality visual inspection system and device, and solve the following technical problems:

[0006] How to quickly confirm the feature of various defect problems by limiting the feature recognition range of defective plastic bottles.

[0007] The purpose of the present invention can be achieved by the following technical solutions:

[0008] A plastic bottle quality visual inspection system includes:

[0009] An identification unit for automatically recording and saving the defect position image of the test sample of the plastic bottle through an industrial camera;

[0010] A classification unit for determining the screening conditions of the defect feature point threshold range based on the FAST algorithm, and classifying and confirming according to the number of regional feature points of different test sample defect position images under the current threshold range condition by using visual recognition technology, and obtaining the defect template coefficient to judge the similarity of visual feature points;

[0011] An extraction unit, which is used to perform clustering analysis according to feature points with similar visual features, classify and read to generate a dictionary file, obtain the vector representation of an image and mark it, read the defect text information of a plastic bottle detection sample, and output a defect classification result;

[0012] A detection and correction unit, which is used to perform visual detection on a plastic bottle approaching the target position in real time according to the defect classification result input by the system, and generate a correction prompt message.

[0013] Preferably, the method for determining the screening conditions of the defect feature point threshold range based on the FAST algorithm includes:

[0014] S1. Perform grayscale processing to determine the grayscale value contrast of the defect position image;

[0015] S2. Set a square area of the defect position image as the feature point center;

[0016] S3. Define that the dynamic local threshold is related to the maximum grayscale value, minimum grayscale value, and grayscale average value of this area; and the dynamic local threshold is in a proportional relationship with the grayscale value contrast, and obtain the proportionality coefficient.

[0017] Preferably, the method for classifying and confirming according to the number of regional feature points of the defect position images of different test samples under the current threshold range condition by using visual recognition technology is:

[0018] Through the formula Calculate to obtain the template matching value E(i, j) of the feature point center p(i, j);

[0019] Wherein, H is the total number of grid rows, and h ∈ [1, H]; V is the total number of grid columns, and v ∈ [1, V]; i is the abscissa of the feature point center, j is the ordinate of the feature point center; k is the defect structure value; and s is the area of the defect region; l is the perimeter of the defect region edge; M is the defect coverage area map; T is the template;

[0020] Compare the template matching value E(i, j) with the preset standard threshold E thr for comparison:

[0021] If the matching value E(i, j) < E thr , it is determined that the matching degree of the feature point center of the current defect position image is low;

[0022] If the matching value E(i, j) ≥ E thr , it is determined that the matching degree of the feature point center of the current defect position image is high, and the feature point center of this defect position image is marked as the target area feature point center.

[0023] Preferably, the method for obtaining the defect template coefficient to judge the similarity of visual feature points is:

[0024] Statistically analyze the number of center points of feature points in the target area, the input proportionality coefficient, and the corresponding matching values for classification:

[0025] Through the formula Calculate the defect template coefficient Dtc;

[0026] Where m is the total number of center points of feature points in the target area, τ ∈ [1, m]; K τ Is the local threshold - overall gray - scale contrast proportionality coefficient of the τ - th center point of the feature points in the target area; E τ Is the matching value of the τ - th center point of the feature points in the target area; Is the average value of the matching values;

[0027] Compare the defect template coefficient Dtc with the standard threshold interval [D1, D2] of the preset features:

[0028] If Dtc ∈ [D1, D2], then it is judged that there is visual feature similarity;

[0029] If Then it is judged that there is no visual feature similarity.

[0030] Preferably, the extraction unit includes:

[0031] SS1. Use the K - means clustering algorithm to group feature points with similar visual features into one category to form "visual words" and generate a dictionary file;

[0032] SS2. According to the dictionary file, use a histogram to perform vector representation on each image;

[0033] SS3. Label the images with the obtained vector representation with the category label label and perform SVM training;

[0034] SS4. Obtain the SVM classification text, output the classification result, and end the algorithm.

[0035] Preferably, the detection and correction unit includes:

[0036] Determine different detection distance points of the plastic bottle and the detection device and mark them;

[0037] Real - time statistics of the visual images of the plastic bottle at each marked detection distance point;

[0038] Obtain the abnormal visual image area for defect matching. When the matching meets the threshold range, obtain the current detection distance point as the target position; otherwise, generate a correction prompt message.

[0039] Preferably, the system further includes:

[0040] A feedback unit, configured to give early warning prompts for unclassified defect problems obtained in real time, and send the warning information to the detection and correction unit.

[0041] A visual inspection device for plastic bottle quality, applied to a visual inspection system for plastic bottle quality. The device includes:

[0042] An identification unit, configured to automatically record and save images of the defect positions of test samples of plastic bottles through an industrial camera;

[0043] A classification unit, configured to determine screening conditions for the threshold range of defect feature points based on the FAST algorithm, and classify and confirm according to the number of regional feature points of defect position images of different test samples under the current threshold range conditions based on visual recognition technology, and obtain defect template coefficients to judge the similarity of visual feature points;

[0044] An extraction unit, configured to perform clustering analysis according to feature points with similar visual features, classify and read to generate a dictionary file, obtain the vector representation of the image and mark it, read the defect text information of the plastic bottle detection sample, and output the defect classification result;

[0045] A detection and correction unit, configured to perform visual inspection on plastic bottles approaching the target position in real time according to the defect classification result input by the system, and generate correction prompt information;

[0046] A feedback unit, configured to give early warning prompts for unclassified defect problems obtained in real time, and send the warning information to the detection and correction unit.

[0047] Advantages of the present invention:

[0048] (1) By setting a classification unit in the present invention to classify according to the number of regional feature points of defect position images corresponding to different test samples under the determined threshold range conditions, obtain defect template information confirmation, ensure the formulation of a unified feature recognition range, and optimize the marker points of defect problem features.

[0049] (2) The present invention also sets a detection and correction unit to perform visual inspection of plastic bottles in real time, identify and adjust the target position, realize an automatic correction process, perform visual inspection on plastic bottles approaching the target position in real time according to the defect template input by the system, generate correction prompt information, and realize a rapid recognition and confirmation process for defective plastic bottles, improving the yield of qualified products of plastic bottles in the factory.

[0050] Of course, it is not necessary for any product implementing the present invention to achieve all the above-described advantages simultaneously. Description of the Drawings

[0051] To more clearly illustrate the technical solutions of the embodiments of the present invention, the following will briefly introduce the accompanying drawings required for the description of the embodiments. Obviously, the accompanying drawings in the following description are only some embodiments of the present invention. For those of ordinary skill in the art, without creative efforts, other accompanying drawings can be obtained based on these drawings.

[0052] Figure 1 It is a unit diagram of a visual inspection system for the quality of plastic bottles according to the present invention;

[0053] Figure 2 It is a method step diagram of the screening conditions for determining the threshold range of defect feature points based on the FAST algorithm according to the present invention;

[0054] Figure 3 It is a step diagram of the extraction unit outputting the defect analysis result according to the present invention;

[0055] Figure 4 It is the regional feature points of the defect position image according to the present invention. Specific embodiments

[0056] 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 embodiments of the present invention, rather than all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those of ordinary skill in the art without creative efforts belong to the scope of protection of the present invention.

[0057] Please refer to Figure 1 As shown, the present invention is a visual inspection system for the quality of plastic bottles, including:

[0058] An identification unit for automatically recording and saving the defect position images of the test samples of plastic bottles through an industrial camera;

[0059] A classification unit for determining the screening conditions of the threshold range of defect feature points based on the FAST algorithm, and classifying and confirming according to the number of regional feature points of the defect position images of different test samples under the current threshold range conditions based on visual recognition technology, and obtaining the defect template coefficient to judge the similarity of visual feature points;

[0060] An extraction unit for performing clustering analysis according to the feature points with similar visual features, classifying and reading to generate a dictionary file, obtaining the vector representation of the image and marking it, reading the defect text information of the plastic bottle detection sample, and outputting the defect classification result;

[0061] A detection and correction unit for visually inspecting the plastic bottles approaching the target position in real time according to the defect classification result input by the system, and generating a correction prompt message.

[0062] In the above technical solution, an identification unit is set to automatically record the defect information in the plastic bottle test sample through an industrial camera, and the recorded content is the image of the defect position of the test sample; and after automatically recording the defect position image of the test sample through the identification unit, it is saved; so as to facilitate the identification and determination of defect features in the next step; a classification unit is set to further process the feature points. Specifically, first, the screening conditions for the threshold range of defect feature points are determined based on the FAST algorithm, and then, according to the visual recognition technology, the number of regional feature points in the defect position images of different test samples under the current threshold range conditions is counted for classification confirmation, and the defect template coefficient is obtained to judge the similarity of visual feature points; according to the limiting conditions of the threshold range, the regional feature points and the number of the defect position images corresponding to the test samples are determined; ensuring the formulation of a unified feature recognition range and optimizing the landmark points of defect problem features; a detection and correction unit is also set to perform visual detection of plastic bottles in real time, and through the adjustment of the target position, an automatic correction process is realized. Specifically, according to the defect classification result input by the system, visual detection is performed on the plastic bottles approaching the target position in real time, a correction prompt message is generated, and the correction prompt message is used to quickly identify and confirm the defective plastic bottles; through the limitation of the feature recognition range of defective plastic bottles above, the rapid confirmation of various defect problem features is realized, and the qualified product output rate of factory plastic bottles is improved.

[0063] As an embodiment of the present invention, please refer to Figure 2 As shown, the method for determining the screening conditions for the threshold range of defect feature points based on the FAST algorithm includes:

[0064] S1. Perform gray-scale processing to determine the gray-scale value contrast of the defect position image;

[0065] S2. Set a square area of the defect position image as the feature point center;

[0066] S3. Define that the dynamic local threshold is related to the maximum gray-scale value, the minimum gray-scale value, and the gray-scale average value of this area; and the dynamic local threshold is in a proportional relationship with the gray-scale value contrast, and obtain the proportional coefficient.

[0067] In the above technical solution, first, the defect position image is gray-scaled to determine the gray-scale value contrast of the pixels; then, the pixels of the selected defect position image are used as the center of the circle, and a square area with a fixed number of pixel ranges is determined. Specifically, the machine learning method is used for corner detection, and multiple groups of pixel interest points are trained to determine the defect position image interest points as the feature point center; finally, according to the feature center, the dynamic local threshold is defined to be related to the maximum gray-scale value, the minimum gray-scale value, and the gray-scale average value of this area, and is in a proportional relationship with the gray-scale value contrast of the square area of the image, and the proportional coefficient is obtained.

[0068] As an embodiment of the present invention, please refer toFigure 4 As shown in Figure 4 , the method for classifying and confirming by statistically counting the number of regional feature points in the defect position images of different test samples under the current threshold range according to visual recognition technology is as follows:

[0069] Through the formula calculate the template matching value E(i, j) of the feature point center p(i, j);

[0070] where H is the total number of grid rows, and h ∈ [1, H]; V is the total number of grid columns, and v ∈ [1, V]; i is the abscissa of the feature point center, j is the ordinate of the feature point center; k is the defect structure value; and s is the area of the defect region; l is the perimeter of the defect region edge; M is the defect coverage region map; T is the template;

[0071] Compare the template matching value E(i, j) of the feature point center (i, j) with the preset standard threshold E thr for comparison:

[0072] If the matching value E(i, j) < E thr , it is determined that the matching degree of the feature point center of the current defect position image is low;

[0073] If the matching value E(i, j) ≥ E thr , it is determined that the matching degree of the feature point center of the current defect position image is high, and the feature point center of the current defect position image is marked as the target region feature point center.

[0074] In the above technical solution, the template matching value is used to realize the automation of defect judgment. Specifically, by using visual recognition technology to statistically count the number of regional feature points in the defect position images of different test samples under the current threshold range, and classifying and calculating the regional feature points. The specific calculation formula is through the formula calculate the template matching value E(i, j) of the feature point center p(i, j); use the plane rectangular coordinate system to locate the feature points, and the matching result is represented by the correlation coefficient obtained by normalizing the feature point center coordinates through template matching, and realize the judgment of the matching degree of the feature point center of the defect position image; since different defect coverage region maps have different structural features, the template matching results of the coverage region maps with the determined structural values of the defect regions are used; that is, analyze the normalized difference between the defect coverage region map with the corresponding structural value index and the number of points of the template. The larger the obtained template matching value E(i, j), the better the matching degree. Generally, the maximum value of the template matching value can be set as the target value, that is, the corresponding preset standard value E thr , the closer to the preset standard value, the higher the matching degree; specifically, compare the template matching value E(i, j) of the feature point center p(i, j) with the preset standard threshold E thrCompare the sizes. When the matching value E(i,j) < E thr , it is determined that the matching degree of the center of the image feature points at the current defect position is low. Otherwise, the matching degree is high.

[0075] As an implementation manner of the present invention, the method for obtaining the defect template coefficient to judge the similarity of visual feature points is as follows:

[0076] Count the number of centers of feature points in the target area, the input ratio coefficient, and the corresponding matching values for classification analysis:

[0077] Through the formula Calculate to obtain the defect template coefficient Dtc;

[0078] Among them, m is the total number of centers of feature points in the target area, τ ∈ [1, m]; K τ is the local threshold - overall gray - scale contrast ratio coefficient of the τ - th center of the target area feature point; E τ is the matching value of the τ - th center of the target area feature point; is the average value of the matching values;

[0079] Compare the defect template coefficient Dtc with the standard threshold interval [D1, D2] of the preset feature:

[0080] If Dtc ∈ [D1, D2], it is determined that there is visual feature similarity;

[0081] If it is determined that there is no visual feature similarity.

[0082] In the above - mentioned technical solution, the method for obtaining the defect template coefficient is based on the change of the matching value, and classification analysis is carried out according to the number of centers of feature points in the statistical target area and the pre - input ratio coefficient of the current system. The calculation is carried out through the formula Calculate to obtain the defect template coefficient Dtc; analyze by using the product of the difference change of the ratio coefficient corresponding to the number of feature points in all target areas and the matching values of different feature points The ratio coefficient K τ is the proportional relationship between the dynamic local threshold and the gray - scale value contrast. It reflects that the more obvious the gray - scale value (the greater the defect probability), the larger the set dynamic local threshold, and the larger the ratio coefficient. Finally, the result of the calculated defect template coefficient Dtc changes significantly. Judge the similarity of visual feature points through the result of the defect template coefficient Dtc. When the defect template coefficient Dtc is compared with the standard threshold interval [D1, D2] of the preset feature, if it falls within the preset interval range, the visual feature similarity obtained by the current classification unit is obvious; it is determined that there is visual feature similarity; otherwise, a judgment that the visual feature similarity is not obvious is obtained, so there is no visual feature similarity.

[0083] As an implementation manner of the present invention, please refer to Figure 3 As shown, the extraction unit includes:

[0084] SS1. Using the K-means clustering algorithm to group feature points with similar visual features into one category, forming individual "visual words" and generating a dictionary file;

[0085] SS2. According to the dictionary file, using a histogram to perform vector representation on each image;

[0086] SS3. Labeling the images with vector representation with the labels label of their respective categories and performing SVM training;

[0087] SS4. Obtaining the SVM classification text, outputting the classification result, and ending the algorithm.

[0088] In the above technical solution, feature points with detailed visual features are statistically extracted. The specific steps are as follows: First, using the K-means clustering algorithm to group feature points with similar visual features into one category, forming individual "visual words" and generating a dictionary file; then, performing vector representation on each image according to the dictionary file, and then classifying the images with vector representation to obtain labels and performing machine training. The training method is SVM training, that is, support vector machine training, to ensure that the output result of the trained text is more accurate; finally, obtaining the SVM classification text, outputting the classification result, and ending the algorithm.

[0089] As an implementation manner of the present invention, the detection and correction unit includes:

[0090] Determining different detection distance sites between the plastic bottle and the detection device and marking them;

[0091] Real-time statistically collecting the visual images of the plastic bottle at each marked detection distance site;

[0092] Obtaining the abnormal visual image area for defect matching. When the matching meets the threshold range, obtaining the current detection distance site as the target position; otherwise, generating a correction prompt message.

[0093] In the above technical solution, the detection and correction unit determines the correction process of the plastic bottle vision during the process of identifying and matching the defects of the plastic bottle that needs to be monitored in real time. It is mainly by determining different detection distance sites between the plastic bottle and the detection device and marking them; then statistically collecting the visual images of the detection positions of the plastic bottle at each marked site in real time, obtaining the abnormal visual image area for defect matching. When the image area of the current detection distance site meets the matching threshold range, obtaining the current detection distance site as the target position; otherwise, generating a correction prompt message; and performing an adaptive movement adjustment of the detection distance point position.

[0094] As an implementation of the present invention, the system further includes:

[0095] A feedback unit for giving early warning prompts for unclassified defect problems obtained in real time and sending the warning information to the detection and correction unit.

[0096] A visual inspection device for the quality of plastic bottles, which is applied to a visual inspection system for the quality of plastic bottles, and includes:

[0097] An identification unit for automatically recording and saving images of the defect positions of test samples of plastic bottles through an industrial camera;

[0098] A classification unit for determining the screening conditions of the threshold range of defect feature points based on the FAST algorithm, and classifying and confirming according to the number of regional feature points of different test sample defect position images under the current threshold range condition by using visual recognition technology, and obtaining the defect template coefficient to judge the similarity of visual feature points;

[0099] An extraction unit for performing clustering analysis according to feature points with similar visual features, classifying and reading to generate a dictionary file, obtaining the vector representation of the image and marking it, reading the defect text information of the plastic bottle detection sample, and outputting the defect classification result;

[0100] A detection and correction unit for visually inspecting plastic bottles approaching the target position in real time according to the defect classification result input by the system and generating correction prompt information;

[0101] A feedback unit for giving early warning prompts for unclassified defect problems obtained in real time and sending the warning information to the detection and correction unit.

[0102] Each embodiment in this specification is described in a progressive manner. The same or similar parts among the embodiments can be referred to each other, and the differences between each embodiment and other embodiments are emphasized. In particular, for the embodiments of the device, equipment, and non-volatile computer storage medium, since they are basically similar to the method embodiments, the description is relatively simple, and the relevant parts can be referred to the partial description of the method embodiments.

[0103] The above describes specific embodiments of this specification. Other embodiments are within the scope of the attached documents. In some cases, the actions or steps recorded in this application can be executed in a different order from that in the embodiments and still achieve the desired results. Additionally, the processes depicted in the drawings do not necessarily require the specific order or sequential order shown to achieve the desired results. In certain embodiments, multitasking and parallel processing are also possible or may be advantageous.

[0104] The above content is only an example and illustration of the concept of the present invention. Those skilled in the art of this technology can make various modifications, supplements, or use similar methods of substitution to the specific embodiments described. As long as they do not deviate from the concept of the invention or exceed the scope defined by this application, they should all fall within the protection scope of the present invention.

Claims

1. A visual inspection system for the quality of plastic bottles, characterized in that, Including: An identification unit for automatically recording and saving images of the defect positions of test samples of plastic bottles through an industrial camera; A classification unit for determining the screening conditions of the threshold range of defect feature points based on the FAST algorithm, and classifying and confirming according to the number of regional feature points of the defect position images of different test samples under the current threshold range condition by using visual recognition technology, and obtaining the defect template coefficient to judge the similarity of visual feature points; An extraction unit for performing clustering analysis on the feature points with similar visual features, classifying and reading to generate a dictionary file, obtaining the vector representation of the image and marking it, reading the defect text information of the plastic bottle detection sample, and outputting the defect classification result; A detection and correction unit for visually detecting the plastic bottles approaching the target position in real time according to the defect classification result input by the system, and generating a correction prompt message.

2. The visual inspection system for the quality of plastic bottles according to claim 1, wherein The method for determining the screening conditions of the threshold range of defect feature points based on the FAST algorithm includes: S1. Performing graying processing to determine the gray value contrast of the defect position image; S2. Setting a square area of the defect position image as the center of the feature points; S3. Defining that the dynamic local threshold is related to the maximum gray value, minimum gray value and average gray value of the area; and the dynamic local threshold is in a proportional relationship with the gray value contrast to obtain the proportional coefficient.

3. A visual inspection system for the quality of plastic bottles according to claim 2, characterized in that, The method for classifying and confirming according to the number of regional feature points of the defect position images of different test samples under the current threshold range condition by using visual recognition technology is: Obtained through the formula Calculate the template matching value E(i,j) of the center p(i,j) of the feature point; Among them, H is the total number of rows of the grid, and h ∈ [1, H]; V is the total number of columns of the grid, and v ∈ [1, V]; i is the abscissa of the center of the feature point, j is the ordinate of the center of the feature point; k is the defect structure value; s is the area of the defect region; l is the perimeter of the edge of the defect region; M is the defect coverage area map; T is the template; Compare the template matching value E(i,j) with the preset standard threshold E thr for comparison: If the matching value E(i, j) < E thr , it is determined that the central matching degree of the image feature points at the current defect position is low; If the matching value E(i,j) ≥ E thr , it is determined that the matching degree of the center of the image feature point at the current defect position is high, and the center of the image feature point at the current defect position is marked as the center of the target area feature point.

4. The quality visual inspection system for plastic bottles according to claim 3, characterized in that, The method for obtaining the defect template coefficient to judge the similarity of visual feature points is: Statistically analyzing the number of centers of target region feature points, the input proportional coefficient and the corresponding matching values for classification: Obtained through the formula Calculate to obtain the defect template coefficient Dtc; where m is the total number of the centers of the feature points in the target area, τ ∈ [1, m]; K τ is the local threshold - overall gray - scale contrast ratio coefficient of the center of the τ - th feature point in the target area; E τ is the matching value of the center of the τ - th feature point in the target area; is the average value of the matching values; Comparing the defect template coefficient Dtc with the standard threshold interval [D1, D2] of the preset feature: If Dtc ∈ [D1, D2], it is determined that there is visual feature similarity; If it is determined that there is no visual feature similarity.

5. A visual inspection system for the quality of plastic bottles according to claim 1, wherein, The extraction unit includes: SS1. Using the K-means clustering algorithm to classify the feature points with similar visual features into one category to form "visual words" one by one, and generating a dictionary file; SS2. According to the dictionary file, using a histogram to perform vector representation on each image; SS3. Labeling the image with the obtained vector representation with the label of the belonging category and performing SVM training; SS4. Obtaining the SVM classification text, outputting the classification result and ending the algorithm.

6. The visual inspection system for the quality of plastic bottles according to claim 1, characterized in that, The detection and correction unit includes: Determining and marking different detection distance positions of the plastic bottle and the detection device; Statistically analyzing the visual images of the plastic bottle at each marked detection distance position in real time; Obtaining the abnormal visual image area for defect matching, and when the matching meets the threshold range, obtaining the current detection distance position as the target position; otherwise, generating a correction prompt message.

7. A visual inspection system for the quality of plastic bottles according to claim 1, characterized in that, The system further includes: A feedback unit for giving an early warning prompt for the unclassified defect problems obtained in real time, and sending the warning information to the detection and correction unit.

8. A visual inspection device for the quality of plastic bottles, which is applied to the visual inspection system for the quality of plastic bottles described in any one of claims 1-7, and is characterized in that, The device includes: An identification unit for automatically recording and saving images of the defect positions of test samples of plastic bottles through an industrial camera; A classification unit is used to determine the screening conditions for the threshold range of defect feature points based on the FAST algorithm, and statistically count the number of regional feature points in the defect location images of different test samples under the current threshold range conditions according to visual recognition technology for classification confirmation, and obtain the defect template coefficient to judge the similarity of visual feature points; An extraction unit is used to perform clustering analysis on feature points with similar visual features, classify and read to generate a dictionary file, obtain the vector representation of the image and mark it, read the defect text information of the plastic bottle detection sample, and output the defect classification result; A detection and correction unit is used to perform visual detection on plastic bottles approaching the target position in real time according to the defect classification result input by the system, and generate a correction prompt message; A feedback unit is used to give a warning prompt for the unclassified defect problems obtained in real time, and send the warning information to the detection and correction unit.

Images