Health care product production labeling machine detection system based on machine vision

Abstract

The invention discloses a detection system of a labeling machine for production of health care products based on machine vision, relates to the technical field of machine vision detection, and solves the technical problems that the repetition occurrence rate of similar abnormalities is high due to lack of intelligent analysis based on historical data, and the influence of detection environment and production process parameters on a detection result is not fully considered at the same time. According to the method, historical abnormal data is subjected to structural analysis, standardized abnormal reason classification and data intervals are generated, detection threshold setting is more scientific, abnormal data interval information is combined, multi-feature detection is conducted on a real-time labeling image through a deep learning algorithm, full coverage of geometric defects and semantic defects is achieved, and the detection accuracy is improved. Similar records are screened from historical data through multi-dimensional feature matching, an optimal adjustment scheme is automatically recommended in combination with an anomaly elimination rate mean value, detection environment and anomaly correlation analysis is introduced, the environment suitability of an adjustment strategy is improved through secondary screening, and the accuracy of overall detection is improved.

Description

Technical Field

[0001] The present invention relates to the technical field of machine vision detection, and in particular to a detection system for a health product production labeling machine based on machine vision. Background Art

[0002] Before health products leave the factory, an automatic labeling machine is required to affix trademarks to the cylindrical surface of the health products. Since the automatic labeling machine may miss labels during normal operation, the cylindrical surface of the health products needs to be inspected for labeling. If manual inspection is used, it will consume a lot of human resources and the inspection efficiency will be low.

[0003] According to the patent application with publication number CN106628475A, a machine vision-based mirror imaging bottle cap surface labeling detection system and method are disclosed. The system includes a conveyor belt, an action trigger system, a mirror imaging system, a waste rejection control system and a waste rejection system. The action trigger system includes a light trigger sensor and an ultrasonic ranging sensor; the mirror imaging system includes an area array camera, a camera bracket and a plane mirror; the waste rejection control system includes a host computer and a programmable logic controller; the waste rejection system includes a nozzle, a waste rejection box and an anti-fall cover plate.

[0004] However, when some existing labeling machine detection systems are in use, the classification of abnormal causes relies on manual experience, lacks intelligent analysis based on historical data, and the adjustment method is selected blindly, resulting in a high recurrence rate of similar abnormalities. The impact of the detection environment and production process parameters on the detection results is not fully considered, resulting in fluctuations in cross-scene detection accuracy and errors in labeling detection results. Summary of the Invention

[0005] In response to the shortcomings of the existing technology, the present invention provides a health product production labeling machine detection system based on machine vision, which solves the problems of lack of intelligent analysis based on historical data, resulting in a high recurrence rate of similar anomalies, and failing to fully consider the impact of the detection environment and production process parameters on the detection results.

[0006] To achieve the above objectives, the present invention is implemented through the following technical solutions: a health care product production labeling machine detection system based on machine vision, comprising:

[0007] The abnormality cause classification processing unit is used to obtain the historical detection data transmitted by the detection data acquisition unit, and classify the abnormality identification records therein into the same type according to the abnormality cause, and organize the abnormal data to generate abnormal data interval information, and transmit it to the labeling image analysis unit at the same time;

[0008] The labeling image analysis unit is used to match the real-time labeling image with the abnormal data interval information, generate normal or abnormal recognition results, analyze the abnormal recognition results, perform matching analysis based on the abnormal situation corresponding to the real-time labeling image, determine the cause of the abnormality, and generate product abnormality information or parameter recognition abnormality information;

[0009] The labeling exception processing unit is used to analyze the acquired parameter identification exception information, screen the exception identification records in the historical data based on the exception cause to obtain similar exception identification records, and perform secondary screening based on the abnormal data to obtain pre-selected records, and analyze the impact of the detection environment to generate records to be analyzed and pre-selected records;

[0010] Then the abnormality elimination rate of the records to be analyzed and the pre-selected records is calculated, and the average of the abnormality elimination rate corresponding to multiple time periods is calculated. At the same time, the detection adjustment information is generated based on the standard and transmitted to the detection information output unit.

[0011] As a further solution of the present invention, it also includes a detection data acquisition unit and a detection information output unit;

[0012] A detection data acquisition unit is used to acquire historical detection data and real-time labeling images of the labeling machine, and transmit the historical detection data to the abnormal cause classification processing unit and transmit the real-time labeling images to the labeling image analysis unit;

[0013] The detection information output unit is used to display the detection adjustment information to the corresponding operation management personnel.

[0014] As a further solution of the present invention, the specific manner in which the abnormal cause classification processing unit generates abnormal data interval information is as follows:

[0015] Extract all abnormality identification records from historical detection data, determine the abnormality cause corresponding to each record, classify the abnormality causes into categories of the same type, mark them as i, and i = 1, 2, ..., a, where a is the number of abnormality cause types, and generate abnormality cause classification information;

[0016] For each abnormal reason i, the corresponding abnormal data is sorted out to obtain the abnormal data interval, and the abnormal data interval sorting of all abnormal reasons is completed. The abnormal data interval information is generated and transmitted to the labeling image analysis unit.

[0017] As a further solution of the present invention, the specific manner in which the labeling image analysis unit generates product abnormality information or parameter identification abnormality information is:

[0018] Identify and detect real-time label image information. If it can be fully recognized, it is normal and a normal recognition result is generated, stored and transmitted to the operator. If it cannot be fully recognized, it is abnormal and an abnormal recognition result is generated;

[0019] The abnormal recognition results need to analyze the abnormal situation and match it with the abnormal data interval information to determine the cause. If the image itself is abnormal, product abnormality information is generated to the detection information output unit. If the image itself is not abnormal, parameter recognition abnormality information is generated to the labeling abnormality processing unit.

[0020] As a further solution of the present invention, the specific method of the labeling exception processing unit to screen and obtain similar exception identification records is:

[0021] Obtain real-time labeled images and their anomaly causes, filter out anomaly recognition records similar to the current anomaly cause from historical data, mark them as n and n=1, 2, …, m, where m is the number of similar records, and obtain the detection parameter adjustment methods corresponding to these records.

[0022] As a further solution of the present invention, the specific manner in which the labeling exception processing unit generates the records to be analyzed and the pre-selected records is as follows:

[0023] Obtain the anomaly parameters corresponding to the current anomaly cause and use them to filter similar records n to obtain pre-selected records o, where o = 1, 2, ..., p, where p is the number of pre-selected records. At the same time, obtain the current detection environment to determine whether it has an impact. If so, use the detection environment as a criterion to filter the pre-selected records o again to obtain the records to be analyzed. Then, analyze the anomaly elimination rate of the records to be analyzed.

[0024] If there is no impact, directly analyze the anomaly elimination rate of the pre-selected records.

[0025] As a further solution of the present invention, the specific manner in which the labeling exception processing unit generates the detection adjustment information is as follows:

[0026] In the time period T, the total number of tested products and the number of abnormal identifications after adjustment are counted, and the abnormality elimination rate is calculated. , calculate the multi-cycle anomaly elimination rate with T as the period, take the mean as the evaluation index, select the adjustment method with the largest mean to generate detection adjustment information and transmit it to the detection information output unit.

[0027] The present invention provides a machine vision-based detection system for health product production labeling machines. Compared with the existing technology, it has the following advantages:

[0028] The present invention conducts structured analysis on historical abnormal data to generate standardized abnormal cause classification and data intervals, making the detection threshold setting more scientific. In combination with abnormal data interval information, it uses a deep learning algorithm to perform multi-feature detection on real-time labeled images, achieving full coverage of "geometric defects + semantic defects". Similar records are screened from historical data through multi-dimensional feature matching, and the optimal adjustment plan is automatically recommended based on the mean abnormality elimination rate. The detection environment and abnormality correlation analysis are introduced to improve the environmental adaptability of the adjustment strategy through secondary screening, thereby improving the overall detection accuracy. BRIEF DESCRIPTION OF THE DRAWINGS

[0029] Figure 1 It is a diagram of the steps and methods of the present invention;

[0030] Figure 2 This is a flow chart for troubleshooting the low elimination rate problem of the present invention. DETAILED DESCRIPTION

[0031] The following will clearly and completely describe the technical solutions in the embodiments of the present invention in conjunction with the accompanying drawings. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without making creative efforts are within the scope of protection of the present invention.

[0032] Example 1

[0033] See also Figure 1 This application provides a health product production labeling machine detection system based on machine vision, including: a detection data acquisition unit, an abnormality cause classification processing unit, a labeling graphic analysis unit, a labeling abnormality processing unit and a detection information output unit. Figure 1 It can be seen from the shown content that the above functional units are electrically connected in a unidirectional manner.

[0034] The detection data acquisition unit is used to acquire the historical detection data and real-time labeling images of the labeling machine, and at the same time transmit the historical detection data to the abnormal cause classification processing unit and transmit the real-time labeling images to the labeling image analysis unit.

[0035] Abnormal cause classification processing unit, which is used to analyze the acquired historical detection data,

[0036] Perform preliminary cleaning and screening of acquired historical inspection data. Remove obvious errors, duplicate data, and information irrelevant to anomaly analysis to ensure the accuracy and validity of the data used in subsequent analysis. Comprehensively search pre-processed historical inspection data to accurately capture all anomaly identification records. Furthermore, for each anomaly identification record, trace the corresponding anomaly cause in detail. This can be obtained from a variety of sources, including but not limited to automatic annotation by the inspection system, manual recording by operators, and operation logs of relevant equipment.

[0037] Utilizing advanced classification algorithms and specialized domain knowledge, we meticulously categorize the acquired anomaly causes into similar categories. This classification process fully considers factors such as the anomaly cause's essential characteristics, its mechanism of occurrence, and its impact on labeling quality. After classification, we generate unique classification information for each anomaly cause and number it, denoted by i, where i = 1, 2, ..., a, and a represents the total number of anomaly causes.

[0038] For each type of abnormal cause i, the corresponding abnormal data is systematically organized and analyzed. Using statistical analysis methods such as calculating the mean, standard deviation, maximum, and minimum values, a reasonable range for this type of abnormal data is determined. This range serves as an important reference for determining whether new test data is abnormal. Similarly, all abnormal data ranges corresponding to the same type of abnormal cause i are organized one by one, ultimately generating comprehensive abnormal data range information.

[0039] Assume that the historical test data of a health product labeling machine covers 1,000 records. After data preprocessing, 200 abnormal identification records are screened out. After careful analysis, it is found that the abnormal causes corresponding to these abnormal identification records are mainly the following:

[0040] 1. Label pasting position deviation, 2. Label printing quality problems (such as blurred fonts, color deviation), 3. Label wrinkles, 4. Label missing, 5. Label does not match the bottle body;

[0041] The above abnormal reasons were classified and numbered. Label affixation position deviation (i=1): After analyzing all abnormal data caused by label affixation position deviation, it was found that the deviation was mainly concentrated between -3mm and +5mm (based on the ideal affixation position). Therefore, the range of this type of abnormal data was determined to be [-3mm, +5mm];

[0042] Label printing quality issues (i=2): For abnormal data with blurred fonts, statistics were collected using image clarity evaluation indicators (such as clarity scores), and it was found that scores below 70 were more common. For abnormal data with color deviation, analysis was performed using color similarity indicators (such as similarity percentage), and it was found that similarities below 85% were more common. After comprehensive consideration, the range of this type of abnormal data was determined to be: font clarity scores below 70 points, and color similarity below 85%;

[0043] Label wrinkles (i=3): Quantify the degree of label wrinkling, measuring the ratio of wrinkle area to total label area. Statistical analysis indicates that wrinkle area exceeding 5% is considered abnormal. Therefore, the abnormal data interval is defined as: wrinkle area greater than 5%.

[0044] Missing label (i=4): This exception is relatively clear, meaning that the absence of a label is detected as an exception. Therefore, the abnormal data interval is: the number of labels is 0.

[0045] Label mismatch (i=5): Analyzing the compatibility of label size and shape with the bottle reveals that label size deviations exceeding ±2mm or significant shape differences (measured by the shape matching index, with a matching degree below 80%) are considered abnormal. Therefore, the data interval for this type of abnormality is determined to be: label size deviations exceeding ±2mm or shape matching below 80%.

[0046] The label image analysis unit analyzes the real-time label image based on the acquired abnormal data interval information and uses advanced image recognition algorithms (such as deep learning-based convolutional neural networks) to identify and detect the pre-processed real-time label image. It focuses on key information on the label, such as text, patterns, barcodes, QR codes, etc., to determine whether this information can be completely and accurately recognized;

[0047] If all key information in the image can be fully identified and the characteristic values ​​of each information are within the normal range specified by the abnormal data interval information, the real-time labeled image is determined to be normal and a detailed normal recognition result is generated. The normal recognition result should include the basic information of the image (such as shooting time, labeled product number, etc.) and the details of the identified label information;

[0048] If the key information in the image cannot be fully identified, or the characteristic values ​​of some information exceed the range set by the abnormal data interval information, the real-time labeled image is judged to be abnormal and an abnormal recognition result is generated. In addition to containing the basic information of the image, the abnormal recognition result should also clearly indicate the information content that cannot be identified or the characteristic values ​​that are out of range;

[0049] For the normal recognition results generated, although there is no abnormality in the image itself, they still need to be processed in a standardized manner. The recognition results are stored safely and reliably for subsequent query and statistical analysis;

[0050] After the anomaly recognition results are generated, it is necessary to conduct in-depth mining and analysis of the anomalies in the real-time labeled image. The obtained anomalies are accurately matched with the abnormal data interval information to determine the specific cause of the anomaly;

[0051] If the analysis determines that the cause of the abnormality is a problem with the real-time labeling image itself, such as label printing errors, excessive deviation in the pasting position, or label damage, detailed product abnormality information will be generated. The product abnormality information should include a screenshot of the abnormal image, a specific abnormality description (such as "label text is blurred" or "label offset is 5mm", etc.), the time when the abnormality occurred, and the corresponding product number, etc. The product abnormality information will be transmitted to the detection information output unit in a timely manner;

[0052] If the cause of the anomaly is not a problem with the real-time labeling image itself, but rather due to factors such as improper detection parameter settings, equipment failure, or environmental interference, a parameter recognition anomaly message is generated. This message should include an analysis of the parameters or factors that may have caused the anomaly (e.g., "abnormal lighting intensity" or "inaccurate camera focus"), the time the anomaly occurred, and relevant equipment information. This message is then promptly transmitted to the labeling anomaly handling unit.

[0053] The detection information output unit is used to display the acquired product abnormality information to the corresponding operator.

[0054] Example 2

[0055] As the second embodiment of the present invention, it is implemented on the basis of the first embodiment, and differs from the first embodiment in the following aspects:

[0056] The labeling exception processing unit is used to analyze the acquired parameter identification exception information. Upon receiving the parameter identification exception information, the integrity and accuracy of the data must be ensured. In addition to obtaining the real-time labeling image and the corresponding exception cause, further information related to the exception needs to be collected, such as the current equipment operating status (such as labeling speed and pressure), raw material batch information, etc. At the same time, historical data is comprehensively organized and classified to establish an efficient exception identification record database to facilitate subsequent query and screening;

[0057] Using the current anomaly cause as the core criterion, extensively screen anomaly identification records in historical data to identify all similar anomaly identification records. To improve screening accuracy, a multi-dimensional matching method can be used, considering not only the textual description of the anomaly cause but also key parameter characteristics of the anomaly (such as the location and severity of the anomaly). The similar anomaly identification records screened out are labeled n, where n = 1, 2, ..., m, and m represents the number of similar anomaly identification records.

[0058] From each similar anomaly identification record n, the corresponding adjustment method is accurately obtained. The adjustment method here not only includes the adjustment of parameters during the detection process, but also covers equipment maintenance operations, process adjustments, etc. These adjustment methods are recorded and classified in detail to provide comprehensive data support for subsequent analysis, and the specific abnormal parameters corresponding to the current abnormal cause are obtained, such as the specific value of label offset and the degree of color deviation. Using these abnormal parameters as strict standards, similar records n are re-screened to obtain pre-selected records o, where o = 1, 2, ..., p, and p represents the number of pre-selected records;

[0059] Comprehensively obtain the detection environment information corresponding to the current abnormality cause, including environmental factors such as temperature, humidity, and light intensity. By establishing a correlation model between environmental factors and abnormality occurrence, determine whether there is an impact of the detection environment;

[0060] If there is an impact from the test environment, the pre-selected records o are screened again using the test environment as a refined criterion to obtain the records to be analyzed. This ensures that the screened records have a high degree of similarity with the current anomaly in terms of environmental conditions, thereby improving the reliability of the analysis results.

[0061] If there is no influence of the detection environment, the abnormality elimination rate corresponding to the pre-selected records is directly analyzed;

[0062] Take time T as a standard inspection cycle and obtain the total number of products inspected within that cycle. At the same time, accurately record the number of abnormalities identified after adjustment. Calculate the abnormality elimination rate for a single cycle using the following formula: , taking time T as a fixed period, continuously calculate the corresponding anomaly elimination rate within multiple time periods. By averaging these anomaly elimination rates, the overall anomaly elimination rate mean is obtained;

[0063] Using the mean anomaly elimination rate as the core evaluation metric, a comprehensive assessment is conducted on the corresponding adjustment methods for all records to be analyzed (or preselected records). The adjustment method with the highest mean anomaly elimination rate is selected as the optimal solution. Based on this optimal adjustment method, detailed test adjustment information is generated, including specific adjustment parameters, operation steps, and expected results.

[0064] A detection information output unit is used to display the acquired detection adjustment information to the corresponding operator.

[0065] Embodiment 3, as the third embodiment of the present invention, focuses on combining the implementation processes of embodiment 1 and embodiment 2.

[0066] Some of the data in the above formulas are calculated based on their numerical values ​​and are not substituted into parameter units for calculation. At the same time, the contents not described in detail in this specification belong to the existing technology known to those skilled in the art.

[0067] The above embodiments are only used to illustrate the technical method of the present invention and are not intended to limit the present invention. Although the present invention has been described in detail with reference to the preferred embodiments, those skilled in the art should understand that the technical method of the present invention may be modified or replaced by equivalents without departing from the spirit and scope of the technical method of the present invention.

Claims

1. The health care product production labeling machine detection system based on machine vision is characterized by: include: The abnormality cause classification processing unit is used to obtain the historical detection data transmitted by the detection data acquisition unit, and classify the abnormality identification records therein into the same type according to the abnormality cause, and organize the abnormal data to generate abnormal data interval information, and transmit it to the labeling image analysis unit at the same time; The labeling image analysis unit is used to match the real-time labeling image with the abnormal data interval information, generate normal or abnormal recognition results, analyze the abnormal recognition results, perform matching analysis based on the abnormal situation corresponding to the real-time labeling image, determine the cause of the abnormality, and generate product abnormality information or parameter recognition abnormality information; The labeling exception processing unit is used to analyze the acquired parameter identification exception information, screen the exception identification records in the historical data based on the exception cause to obtain similar exception identification records, and perform secondary screening based on the abnormal data to obtain pre-selected records, and analyze the impact of the detection environment to generate records to be analyzed and pre-selected records; Then the abnormality elimination rate of the records to be analyzed and the pre-selected records is calculated, and the average of the abnormality elimination rate corresponding to multiple time periods is calculated. At the same time, the detection adjustment information is generated based on the standard and transmitted to the detection information output unit.

2. The machine vision-based health care product production labeling machine detection system according to claim 1 is characterized in that: It also includes a detection data acquisition unit and a detection information output unit; A detection data acquisition unit is used to acquire historical detection data and real-time labeling images of the labeling machine, and transmit the historical detection data to the abnormal cause classification processing unit and transmit the real-time labeling images to the labeling image analysis unit; The detection information output unit is used to display the detection adjustment information to the corresponding operation management personnel.

3. The machine vision-based health care product production labeling machine detection system according to claim 1 is characterized in that: The specific method for the abnormal cause classification processing unit to generate abnormal data interval information is: Extract all abnormality identification records from historical detection data, determine the abnormality cause corresponding to each record, classify the abnormality causes into categories of the same type, mark them as i, and i = 1, 2, ..., a, where a is the number of abnormality cause types, and generate abnormality cause classification information; For each abnormal reason i, the corresponding abnormal data is sorted out to obtain the abnormal data interval, and the abnormal data interval sorting of all abnormal reasons is completed. The abnormal data interval information is generated and transmitted to the labeling image analysis unit.

4. The machine vision-based health care product production labeling machine detection system according to claim 1 is characterized in that: The specific method for the labeling image analysis unit to generate product abnormality information or parameter identification abnormality information is: Identify and detect real-time label image information. If it can be fully recognized, it is normal and a normal recognition result is generated, stored and transmitted to the operator. If it cannot be fully recognized, it is abnormal and an abnormal recognition result is generated; The abnormal recognition results need to analyze the abnormal situation and match it with the abnormal data interval information to determine the cause. If the image itself is abnormal, product abnormality information is generated to the detection information output unit. If the image itself is not abnormal, parameter recognition abnormality information is generated to the labeling abnormality processing unit.

5. The machine vision-based health care product production labeling machine detection system according to claim 1 is characterized in that: The specific method of the labeling exception processing unit to screen and obtain similar exception identification records is as follows: Obtain real-time labeled images and their anomaly causes, filter out anomaly recognition records similar to the current anomaly cause from historical data, mark them as n and n=1, 2, …, m, where m is the number of similar records, and obtain the detection parameter adjustment methods corresponding to these records.

6. The machine vision-based health care product production labeling machine detection system according to claim 1 is characterized in that: The specific method for the labeling exception processing unit to generate the records to be analyzed and the pre-selected records is: Obtain the anomaly parameters corresponding to the current anomaly cause and use them to filter similar records n to obtain pre-selected records o, where o = 1, 2, ..., p, where p is the number of pre-selected records. At the same time, obtain the current detection environment to determine whether it has an impact. If so, use the detection environment as a criterion to filter the pre-selected records o again to obtain the records to be analyzed. Then, analyze the anomaly elimination rate of the records to be analyzed. If there is no impact, directly analyze the anomaly elimination rate of the pre-selected records.

7. The machine vision-based health care product production labeling machine detection system according to claim 1 is characterized in that: The specific method for the labeling exception processing unit to generate detection adjustment information is as follows: In the time period T, the total number of tested products and the number of abnormal identifications after adjustment are counted, and the abnormality elimination rate is calculated. , calculate the multi-cycle anomaly elimination rate with T as the period, take the mean as the evaluation index, select the adjustment method with the largest mean to generate detection adjustment information and transmit it to the detection information output unit.

Images