A packaging control method and system for LED chips

Abstract

This invention relates to the field of LED chip manufacturing technology and discloses a packaging control method and system for LED chips. The method includes: determining a target bracket image; determining the comprehensive dust value of the target bracket image based on analysis results; determining a predicted cleaning time for the LED bracket based on the comprehensive dust value and a dust encapsulation model; comparing the predicted cleaning time with a historical dataset; when it is determined that the predicted cleaning time needs correction, determining a time similarity based on the predicted cleaning time and the historical dataset; determining a first time correction coefficient or cleaning set based on the time similarity; reducing the historical dataset based on the cleaning set; determining a second time correction coefficient based on a suspected correction coefficient and a correction weight; and cleaning the LED bracket according to the adjusted predicted cleaning time. This invention avoids reliance on manual experience and dynamically adjusts according to actual conditions, ensuring packaging stability.

Description

Technical Field

[0001] This invention relates to the field of LED chip manufacturing technology, and more specifically, to a packaging control method and system for LED chips. Background Technology

[0002] With societal development, the number of street shops is increasing daily, posing a challenge to the production of LED chips for shop signs. During production, LED chips need to be encapsulated to prevent interference and damage to the electronic chips from the external environment, thus providing electrical connection and signal transmission functions. The LED bracket is the base for mounting the LED chips in the encapsulation process, and its surface cleanliness directly affects the encapsulation quality and luminous efficacy of the LED chips.

[0003] Currently, when using plasma instruments to clean LED brackets, the cleaning time is usually manually set based on the operator's experience. Due to the different skill levels and experience of different operators, the cleaning time is easily affected by the operator's skill level and judgment, and cannot be dynamically adjusted according to the actual situation. This leads to unstable cleaning quality, which in turn affects the precision of the packaging process, and ultimately results in insufficient production efficiency of LED chips.

[0004] Therefore, how to provide a packaging control method and system for LED chips is a technical problem that urgently needs to be solved by those skilled in the art. Summary of the Invention

[0005] In view of this, the present invention proposes a packaging control method and system for LED beads, aiming to solve the problem that the cleaning time is easily affected by the skill level and judgment of the operator, and cannot be dynamically adjusted according to the actual situation, resulting in unstable cleaning quality, which in turn affects the precision of the packaging process, and ultimately leads to insufficient production efficiency of LED beads.

[0006] In one aspect, the present invention proposes a packaging control method for LED chips, comprising:

[0007] An initial bracket image of the LED bracket is obtained and processed to determine a target bracket image. The target bracket image is analyzed, and the comprehensive dust value of the target bracket image is determined based on the analysis results. The cleaning prediction time of the LED bracket is determined based on the comprehensive dust value and the dust encapsulation model.

[0008] The cleaning prediction time is compared with the historical dataset, and the cleaning prediction time is adjusted based on the comparison results.

[0009] When it is determined that the cleaning prediction time needs to be corrected, the time similarity is determined based on the cleaning prediction time and the historical dataset, a first time correction coefficient or cleaning set is determined based on the time similarity, a suspected correction coefficient is determined based on the cleaning set, the historical dataset is reduced based on the cleaning set, the correction weight of the suspected correction coefficient is determined, and a second time correction coefficient is determined based on the suspected correction coefficient and the correction weight.

[0010] The cleaning prediction time is adjusted according to the first time correction coefficient or the second time correction coefficient, and the LED bracket is cleaned according to the adjusted cleaning prediction time.

[0011] Furthermore, when processing the initial stent image to determine the target stent image, the process includes:

[0012] The initial support image is preprocessed, including image denoising, color equalization, and normalization of pixel values.

[0013] The preprocessed initial stent image is substituted into the neural network model, and the image conformance value of the preprocessed initial stent image is output based on the neural network model.

[0014] When the image conformance value is greater than or equal to the preset image conformance value, the initial support image is determined as the target support image;

[0015] When the image conformance value is less than the preset image conformance value, the initial bracket image of the LED bracket is reacquired.

[0016] Furthermore, when analyzing the target support image and determining the overall dust value of the target support image based on the analysis results, the process includes:

[0017] Extract all the support pixels of the target support image and obtain the support pixel value corresponding to all the support pixels. Obtain the matching target support image corresponding to the target support image, extract all the matching support pixels of the matching target support image, and obtain the matching support pixel value corresponding to all the matching support pixels.

[0018] Compare the pixel values ​​of all brackets with the pixel values ​​of all adapted brackets, and classify all bracket pixels into the dust set based on the comparison results;

[0019] When the pixel value of the bracket is greater than the pixel value of the adapter bracket, the bracket pixel is assigned to the first dust set.

[0020] When the pixel value of the bracket is equal to the pixel value of the adapter bracket, the pixel of the bracket will not be divided.

[0021] When the pixel value of the bracket is less than the pixel value of the adapter bracket, the bracket pixel is assigned to the second dust set.

[0022] Count the first number of bracket pixels in the first dust set, count the second number of bracket pixels in the second dust set, obtain the first average bracket pixel value of the first dust set, and obtain the second average bracket pixel value of the second dust set.

[0023] The comprehensive dust value is determined based on the first quantity, the second quantity, the first average bracket pixel value, and the second average bracket pixel value.

[0024] Furthermore, when determining the predicted cleaning time for the LED bracket based on the comprehensive dust value and dust encapsulation model, the following steps are included:

[0025] Obtain a comprehensive dust value dataset and divide it into a training set and a test set. Use grid search to find the hyperparameters of the random forest model and build the random forest model.

[0026] The random forest model is iteratively trained based on the training set, and the test set is substituted into the iteratively trained random forest model for verification.

[0027] When the verification value reaches the verification threshold, the iteratively trained random forest model is determined as the dust encapsulation model, and the cleaning prediction time is obtained based on the comprehensive dust value.

[0028] Furthermore, when comparing the predicted cleaning time with historical datasets and determining whether to correct the predicted cleaning time based on the comparison results, the process includes:

[0029] The historical dataset includes several historical cleaning prediction times, several historical time correction coefficients, several historical comprehensive dust values, and the average of historical qualified data. Each historical comprehensive dust value corresponds to a historical cleaning prediction time and a historical time correction coefficient.

[0030] The cleaning prediction time is compared with the average of the historical qualified data, and the cleaning prediction time is adjusted based on the comparison results.

[0031] If the predicted cleaning time is less than the average of the historical qualified data, it is determined that the predicted cleaning time should be corrected; otherwise, the predicted cleaning time is not corrected, and the LED bracket is cleaned according to the predicted cleaning time.

[0032] Furthermore, when determining the time similarity based on the cleaning prediction time and the historical dataset, and determining the first time correction coefficient or cleaning set based on the time similarity, the process includes:

[0033] The time similarity is determined by comparing the predicted cleaning time with each historical prediction cleaning time in the historical dataset.

[0034] The time similarity is compared with a preset time similarity threshold, and the first time correction coefficient or the cleaning set is determined based on the comparison result.

[0035] When there is data in the historical dataset with a time similarity greater than or equal to the time similarity threshold, the first time correction coefficient is determined based on the historical dataset; otherwise, the cleaning set is determined based on the historical dataset.

[0036] When determining the first time correction coefficient based on the historical dataset, if there is only one data in the historical dataset that is greater than or equal to the time similarity threshold, the historical time correction coefficient corresponding to the data shall be used as the first time correction coefficient.

[0037] If the data in the historical dataset that is greater than or equal to the time similarity threshold is not unique, the historical time correction coefficient corresponding to the data with the highest time similarity will be used as the first time correction coefficient.

[0038] If the data in the historical dataset that is greater than or equal to the time similarity threshold is not unique, and there are several data with the maximum time similarity, the average of the historical time correction coefficients corresponding to each data with the maximum time similarity is used as the first time correction coefficient.

[0039] Furthermore, when determining the suspected correction coefficient based on the cleaning set, the process includes:

[0040] The comprehensive dust value, the cleaning prediction time, and the historical dataset are used as the dataset to be clustered. The historical time correction coefficient corresponding to each data in the dataset to be clustered is extracted, the expected number of clusters k is determined to be 3, and the parameters of the Gaussian distribution are initialized.

[0041] Based on the probability that each data point in the dataset to be clustered belongs to each Gaussian distribution, a responsibility value is obtained. Based on the responsibility value, the dataset corresponding to the comprehensive dust value and the cleaning prediction time is obtained, and the dataset is used as the cleaning set.

[0042] Obtain the corrected mean of the historical time correction coefficients in the cleaning set, and determine the corrected mean as the suspected correction coefficient.

[0043] Furthermore, when reducing the historical dataset based on the cleaned set, determining the correction weight of the suspected correction coefficient, and determining the second time correction coefficient based on the suspected correction coefficient and the correction weight, the process includes:

[0044] Remove data from the historical dataset that is identical to the cleaned set, record the number of data removed, and count the number of data in the historical dataset that was not removed.

[0045] Obtain the difference between the number of deletions and the number of data, and obtain the absolute value of the difference. The ratio of the absolute value of the difference to the number of deletions is recorded as the deviation dispersion.

[0046] The result of a power operation with the natural constant as the base and the deviation dispersion as the exponent is used as the correction weight of the suspected correction coefficient, and the second time correction coefficient is the product of the suspected correction coefficient and the correction weight.

[0047] Furthermore, when adjusting the cleaning prediction time according to the first time correction coefficient or the second time correction coefficient, the following steps are included:

[0048] The first time correction coefficient is directly proportional to the cleaning prediction time, and the second time correction coefficient is also directly proportional to the cleaning prediction time.

[0049] Compared with existing technologies, the advantages of this invention are as follows: By processing and analyzing the initial image of the LED bracket, the cleaning prediction time is determined based on the comprehensive dust value and dust encapsulation model. This avoids the risk of mismatch between the cleaning time and the actual contamination situation caused by differences in human experience, thus improving the reliability and stability of the cleaning prediction time. The cleaning prediction time is compared with historical datasets to dynamically determine whether correction is needed. Historical experience is used to verify the accuracy of the cleaning prediction time, providing direction for precise correction. When correction is required, a first time correction coefficient or cleaning set is determined based on time similarity, making full use of historical datasets and flexibly correcting the cleaning prediction time according to different situations, ensuring that the corrected cleaning prediction time closely matches the current actual contamination situation of the LED bracket. By determining a suspected correction coefficient from the cleaning set and then determining the correction weight based on the cleaning set, a second time correction coefficient is obtained. This comprehensively considers multiple factors, reducing encapsulation quality problems caused by improper cleaning time, ensuring the stability of cleaning quality, and thus improving the production efficiency of LED chips.

[0050] On the other hand, this application also provides a packaging control system for LED chips, for applying the above-described packaging control method for LED chips, including:

[0051] The acquisition module is configured to acquire an initial bracket image of the LED bracket, process the initial bracket image to determine a target bracket image, analyze the target bracket image, determine the comprehensive dust value of the target bracket image based on the analysis results, and determine the predicted cleaning time of the LED bracket based on the comprehensive dust value and the dust encapsulation model.

[0052] The judgment module is configured to compare the cleaning prediction time with the historical dataset and determine whether to correct the cleaning prediction time based on the comparison result.

[0053] The processing module is configured to, when it is determined that the cleaning prediction time needs to be corrected, determine the time similarity based on the cleaning prediction time and the historical dataset, determine a first time correction coefficient or cleaning set based on the time similarity, determine a suspected correction coefficient based on the cleaning set, delete the historical dataset based on the cleaning set, determine the correction weight of the suspected correction coefficient, and determine a second time correction coefficient based on the suspected correction coefficient and the correction weight.

[0054] The cleaning module is configured to adjust the cleaning prediction time according to the first time correction coefficient or the second time correction coefficient, and to clean the LED bracket according to the adjusted cleaning prediction time.

[0055] It is understandable that the above-mentioned packaging control method and system for LED beads have the same beneficial effects, and will not be elaborated further here. Attached Figure Description

[0056] Various other advantages and benefits will become apparent to those skilled in the art upon reading the following detailed description of preferred embodiments. The accompanying drawings are for illustrative purposes only and are not intended to limit the invention. Furthermore, the same reference numerals denote the same parts throughout the drawings. In the drawings:

[0057] Figure 1 A flowchart illustrating a packaging control method for LED beads provided in an embodiment of the present invention;

[0058] Figure 2 This is a functional block diagram of a packaging control system for LED beads provided in an embodiment of the present invention. Detailed Implementation

[0059] Exemplary embodiments of the present disclosure will now be described in more detail with reference to the accompanying drawings. While exemplary embodiments of the present disclosure are shown in the drawings, it should be understood that the present disclosure may be implemented in various forms and should not be limited to the embodiments set forth herein. Rather, these embodiments are provided to enable a more thorough understanding of the present disclosure and to fully convey the scope of the disclosure to those skilled in the art. It should be noted that, unless otherwise specified, embodiments and features in the embodiments of the present invention can be combined with each other. The present invention will now be described in detail with reference to the accompanying drawings and embodiments.

[0060] See Figure 1 As shown in some embodiments of this application, this embodiment provides a packaging control method for LED chips, including:

[0061] S100: Obtain the initial bracket image of the LED bracket, process the initial bracket image, determine the target bracket image, analyze the target bracket image, determine the comprehensive dust value of the target bracket image based on the analysis results, and determine the predicted cleaning time of the LED bracket based on the comprehensive dust value and the dust encapsulation model.

[0062] S200: Compare the cleaning prediction time with the historical dataset, and determine whether to correct the cleaning prediction time based on the comparison results.

[0063] S300: When it is determined that the cleaning prediction time needs to be corrected, the time similarity is determined based on the cleaning prediction time and the historical dataset. The first time correction coefficient or cleaning set is determined based on the time similarity. The suspected correction coefficient is determined based on the cleaning set. The historical dataset is reduced based on the cleaning set. The correction weight of the suspected correction coefficient is determined. The second time correction coefficient is determined based on the suspected correction coefficient and the correction weight.

[0064] S400: Adjust the cleaning prediction time according to the first time correction coefficient or the second time correction coefficient, and clean the LED bracket according to the adjusted cleaning prediction time.

[0065] Specifically, initial images of the LED bracket are acquired using acquisition devices such as infrared cameras. These initial images may contain interference factors such as noise and ambient brightness, affecting subsequent judgments. Therefore, the initial images are processed to determine the target image, avoiding subjective errors caused by manual judgment. The target image is then analyzed to quantitatively calculate the comprehensive dust value on the LED bracket. This comprehensive dust value reflects the actual contamination level on the LED bracket surface. Using a trained dust encapsulation model, a predicted cleaning time is determined based on the comprehensive dust value obtained from image analysis. Predicting the required cleaning time based on the actual contamination level using a big data model ensures consistency between the predicted cleaning time and the comprehensive dust value, thereby improving encapsulation quality. Using a dust encapsulation model for prediction cannot guarantee the accuracy and reliability of the cleaning prediction time. Historical datasets cover various data from past cleaning processes. By comparing the cleaning prediction time with historical datasets, it is possible to accurately determine whether the cleaning prediction time conforms to the actual situation in the past. The cleaning prediction time can be continuously optimized during the cleaning process, ensuring that the cleaning prediction time meets the actual cleaning needs while avoiding encapsulation quality problems caused by over- or under-cleaning. Compared with manual experience, the comparison of big data models and historical datasets improves the accuracy and reliability of the cleaning prediction time, avoiding the risk of unstable cleaning quality due to differences in operator skills and experience. When it is necessary to correct the cleaning prediction time, the time similarity between the cleaning prediction time and the historical dataset is calculated to measure the closeness between the cleaning prediction time and the historical dataset. Based on the time similarity, a first time correction coefficient or cleaning set is determined. For the cleaning set, suspected correction coefficients are screened, and based on this, the historical dataset is screened to determine the dispersion between data to improve the data relevance, thereby obtaining the correction weight of the suspected correction coefficients, and finally determining the second time correction coefficient. The dynamic correction mechanism based on historical data ensures that the adjustment of the cleaning prediction time conforms to the actual situation of the LED bracket. Through multi-dimensional analysis and data screening, the stability of cleaning quality is further guaranteed, and problems with encapsulation quality caused by improper cleaning time are avoided.

[0066] Understandably, the derived first or second time correction coefficient is applied to the cleaning prediction time to determine the adjusted cleaning prediction time, and the LED bracket is cleaned accordingly. Through analysis and adjustment, the adjusted cleaning prediction time avoids both over-cleaning leading to resource waste and under-cleaning affecting packaging quality. This helps improve the precision of the packaging process, reduces the defect rate of LED chips, and thus improves the production efficiency of LED chips.

[0067] In some embodiments of this application, the process of processing the initial support image to determine the target support image includes: preprocessing the initial support image, the preprocessing including image denoising, color equalization and normalization of pixel values, substituting the preprocessed initial support image into a neural network model, outputting the image conformance value of the preprocessed initial support image based on the neural network model, determining the initial support image as the target support image when the image conformance value is greater than or equal to a preset image conformance value, and re-acquiring the initial support image of the LED support when the image conformance value is less than the preset image conformance value.

[0068] Specifically, image denoising removes random noise introduced during the initial acquisition of the support image. This noise interferes with subsequent dust analysis of the LED support, leading to deviations in the calculation of the overall dust value. After denoising, image details are preserved, while color equalization ensures a more uniform color distribution, preventing localized overexposure or underexposure due to uneven lighting, which could affect the assessment of the LED support's surface condition. Pixel normalization unifies pixel values ​​from images acquired by different devices within a specific range, eliminating data bias caused by device differences. To ensure image data consistency, a neural network model is pre-trained. This model includes multiple convolutional layers, pooling layers, and fully connected layers. Convolutional layers extract deep features, pooling layers reduce feature dimensionality, and fully connected layers perform final classification. With a preset image conformance value of 0.8, the neural network model learns features to distinguish image conformance and outputs the probability of conformance or non-conformity using a softmax function, ensuring that the target support image accurately reflects the true contamination status of the LED support.

[0069] In some embodiments of this application, when analyzing a target stent image and determining the comprehensive dust value of the target stent image based on the analysis results, the process includes: extracting all stent pixels from the target stent image and obtaining the stent pixel values ​​corresponding to all stent pixels; obtaining an adapted target stent image corresponding to the target stent image; extracting all adapted stent pixels from the adapted target stent image and obtaining the adapted stent pixel values ​​corresponding to all adapted stent pixels; comparing all stent pixel values ​​with all adapted stent pixel values; classifying all stent pixels into a dust set based on the comparison results; classifying a stent pixel into a first dust set when the stent pixel value is greater than the adapted stent pixel value; not classifying a stent pixel when the stent pixel value is equal to the adapted stent pixel value; classifying a stent pixel into a second dust set when the stent pixel value is less than the adapted stent pixel value; counting a first number of stent pixels in the first dust set; counting a second number of stent pixels in the second dust set; obtaining a first average stent pixel value for the first dust set; obtaining a second average stent pixel value for the second dust set; and determining the comprehensive dust value based on the first number, the second number, the first average stent pixel value, and the second average stent pixel value.

[0070] Specifically, the overall dust value is derived from the following formula:

[0071]

[0072] Where L represents the overall dust value, A1 represents the first average bracket pixel value, p represents the first quantity, Xi represents the bracket pixel value of the i-th bracket pixel in the first dust set, Yi represents the adapted bracket pixel value corresponding to the i-th bracket pixel in the first dust set, A2 represents the second average bracket pixel value, q represents the second quantity, Nj represents the adapted bracket pixel value corresponding to the j-th bracket pixel in the second dust set, and Mj represents the bracket pixel value of the j-th bracket pixel in the second dust set.

[0073] Specifically, the target bracket image is obtained based on a dust-free LED bracket. The bracket pixels and the adapter bracket pixels correspond one-to-one, ensuring a one-to-one correspondence between the bracket pixel values ​​and the adapter bracket pixel values. The dust set includes a first dust set and a second dust set. Although the target bracket image has undergone image denoising and color equalization, the actual lighting conditions during shooting can cause variations in pixel values ​​in different areas of the target bracket image. When dust is in a darker area and the background is in a brighter area, even if the dust's reflective properties would normally result in a larger bracket pixel value under normal lighting, the background lighting will affect the bracket pixel value, leading to a smaller value. Conversely, when dust is under strong light and the background lighting is weak, even if the dust itself is dark, its bracket pixel value may be larger. Therefore, dividing the bracket pixels that do not equal the adapter bracket pixel value lays the foundation for calculating the overall dust value and ensures the accuracy of the overall dust value calculation.

[0074] In some embodiments of this application, when determining the predicted cleaning time of an LED bracket based on the comprehensive dust value and the dust encapsulation model, the process includes: obtaining a comprehensive dust value dataset and dividing the comprehensive dust value dataset into a training set and a test set; using grid search to find the hyperparameters of a random forest model; establishing a random forest model; iteratively training the random forest model based on the training set; substituting the test set into the iteratively trained random forest model and verifying it; when the verification value reaches the verification threshold, determining the iteratively trained random forest model as the dust encapsulation model; and obtaining the predicted cleaning time based on the comprehensive dust value.

[0075] Specifically, the comprehensive dust value dataset contains key data such as various pollution levels of LED brackets, environmental dust concentration, and cleaning times under different conditions. This dataset is divided into training and testing sets. 50%-90% of the data is used as the training set, and the remainder as the testing set. This ensures that both the training and testing sets contain diverse data to improve the model's generalization ability. The training set is used for the random forest model, while the testing subset is used to validate the performance of the trained model. A grid search is used to exhaustively search for hyperparameters in the parameter space to build the random forest model. The random forest model includes parameters such as the number of trees and the maximum tree depth, aiming to capture complex relationships in the data. During iterative training of the random forest model using data from the training set, the model attempts to learn patterns and relationships in the data to improve its prediction or classification capabilities. After iterative training, the model is validated using data from the testing set. Validation metrics include accuracy and F1 score. Through continuous iterative training of the random forest model, a stable output of the predicted cleaning time can be achieved. When the validation value reaches the validation threshold, the iteratively trained random forest model is determined as a dust encapsulation model. The validation threshold is dynamically set according to the validation metrics, for example, the accuracy is set to 80% and the F1 score is set to 0.7.

[0076] Understandably, the training process for the neural network model is consistent with that for the random forest model. The training of the neural network model and its application in image processing are repetitive and redundant, therefore not described in detail. Iteratively training the random forest model allows for the determination of a dust encapsulation model. Substituting the comprehensive dust values ​​into this model yields the predicted cleaning time, thus improving the stability and reliability of the cleaning process.

[0077] In some embodiments of this application, when comparing the predicted cleaning time with a historical dataset and determining whether to correct the predicted cleaning time based on the comparison result, the following steps are taken: the historical dataset includes several historical predicted cleaning times, several historical time correction coefficients, several historical comprehensive dust values, and the average of historical qualified data. Each historical comprehensive dust value corresponds to a historical predicted cleaning time and a historical time correction coefficient. The predicted cleaning time is compared with the average of historical qualified data, and a determination is made based on the comparison result whether to correct the predicted cleaning time. When the predicted cleaning time is less than the average of historical qualified data, it is determined that the predicted cleaning time should be corrected; otherwise, the predicted cleaning time is not corrected, and the LED bracket is cleaned using the predicted cleaning time.

[0078] Specifically, the historical average of qualified data is set according to cleaning requirements and used as a judgment indicator. Its specific value can be adjusted during actual cleaning. By introducing the historical average of qualified data as a reference benchmark, the accuracy and automation of cleaning prediction time correction are improved. When the cleaning prediction time is less than the historical average of qualified data, a correction is made. This is because an excessively short cleaning prediction time may result in incomplete cleaning of the LED bracket, and residual dust may affect the packaging quality and luminous efficacy of the LED chips. By correcting the cleaning prediction time, interference and errors from human judgment are reduced, improving adaptability to different actual contamination levels of LED brackets, thereby effectively ensuring cleaning quality and improving the efficiency of LED chip production.

[0079] In some embodiments of this application, when determining time similarity based on the cleaning prediction time and historical dataset, and determining a first time correction coefficient or cleaning set based on the time similarity, the process includes: comparing the cleaning prediction time with each historical cleaning prediction time in the historical dataset to determine time similarity; comparing the time similarity with a pre-set time similarity threshold; determining a first time correction coefficient or cleaning set based on the comparison result; when there is data in the historical dataset with a time similarity greater than or equal to the time similarity threshold, determining the first time correction coefficient based on the historical dataset; otherwise, determining the cleaning set based on the historical dataset; when determining the first time correction coefficient based on the historical dataset, if the data in the historical dataset with a time similarity greater than or equal to the time similarity threshold is unique, the historical time correction coefficient corresponding to the data is used as the first time correction coefficient; if the data in the historical dataset with a time similarity greater than or equal to the time similarity threshold is not unique, the historical time correction coefficient corresponding to the data with the maximum time similarity is used as the first time correction coefficient; if the data in the historical dataset with a time similarity greater than or equal to the time similarity threshold is not unique, and there are several data with the maximum time similarity, the average of the historical time correction coefficients corresponding to each data with the maximum time similarity is used as the first time correction coefficient.

[0080] Specifically, time similarity is derived from the following formula:

[0081]

[0082] Where Q represents time similarity, Lh represents the normalized result of the h-th historical comprehensive dust value in the historical dataset, LA represents the normalized result of the comprehensive dust value, Th represents the normalized result of the historical cleaning prediction time corresponding to the h-th historical comprehensive dust value, and TA represents the normalized result of the cleaning prediction time.

[0083] Specifically, by normalizing the data to facilitate calculations under the same scale, comparisons can be made between data under different cleaning conditions. A time similarity threshold is used to determine the degree of matching between the current cleaning conditions and historical successful conditions. When historical data with high time similarity is found, this data can be directly used to determine the first-time correction coefficient, thus ensuring the reliability and consistency of the cleaning process. If only one data point in the historical dataset has a time similarity greater than or equal to the threshold, its corresponding historical time correction coefficient can be directly used as the first-time correction coefficient, avoiding complex calculations and judgments and improving the accuracy of the correction process. If multiple data points in the historical dataset have a time similarity greater than or equal to the threshold, different approaches are taken: if only one data point has the highest time similarity, its corresponding historical time correction coefficient is used as the first-time correction coefficient; if several data points have the highest time similarity, the average of their corresponding historical time correction coefficients is used as the first-time correction coefficient. By fully utilizing historical data in the historical dataset and finding the most suitable first-time correction coefficient among multiple similarity cases, the accuracy of the correction is ensured.

[0084] Understandably, when the current cleaning conditions do not perfectly match the historical data, the cleaning set is determined based on the historical dataset to further analyze the relationship between the historical data, laying the foundation for determining the second time correction coefficient.

[0085] In some embodiments of this application, when determining the suspected correction coefficient based on the cleaning set, the process includes: using the comprehensive dust value, cleaning prediction time, and historical dataset as the dataset to be clustered; extracting the historical time correction coefficient corresponding to each data point in the dataset to be clustered; determining the expected number of clusters k as 3; initializing the parameters of the Gaussian distribution; obtaining the responsibility value based on the probability that each data point in the dataset to be clustered belongs to each Gaussian distribution; obtaining the dataset corresponding to the comprehensive dust value and cleaning prediction time based on the responsibility value; using the dataset as the cleaning set; obtaining the corrected mean of the historical time correction coefficients in the cleaning set; and determining the corrected mean as the suspected correction coefficient.

[0086] In some embodiments of this application, when reducing the historical dataset based on the cleaning set, determining the correction weight of the suspected correction coefficient, and determining the second time correction coefficient based on the suspected correction coefficient and the correction weight, the process includes: reducing the data in the historical dataset that is the same as the cleaning set, recording the number of reductions, counting the number of data in the historical dataset that has not been reduced, obtaining the difference between the number of reductions and the number of data, obtaining the absolute value of the difference, recording the ratio of the absolute value of the difference to the number of reductions as the deviation dispersion, using the natural constant as the base and the deviation dispersion as the exponent for the power operation result as the correction weight of the suspected correction coefficient, and the second time correction coefficient being the product of the suspected correction coefficient and the correction weight.

[0087] Specifically, if historical datasets have not yet been effectively accumulated, statistics are compiled based on industry cleaning standards for LED chip packaging processes and the cleaning time for successfully completed cleaning operations. Clustering algorithms are used to analyze historical data, identifying the cleaning set most closely matching current cleaning conditions. This improves the accuracy of determining potential correction coefficients, reduces reliance on human experience and judgment, minimizes human error during the cleaning process, and enhances overall automation and reliability. The corrected mean of historical time correction coefficients within the cleaning set is used as the potential correction coefficient. This effectively integrates similar historical data after clustering, eliminating the influence of individual data points and resulting in a stable potential correction coefficient that represents the characteristics of the cleaning set.

[0088] Understandably, during the clustering process, the cleaning set utilizes historical data most closely related to the current cleaning conditions. Therefore, some un-clustered historical data still exists in the historical dataset. While these un-clustered historical data are less representative than the data in the cleaning set, they are still part of the historical dataset and can reflect the overall data distribution characteristics and trends. This avoids focusing solely on the cleaning set while ignoring the impact of the overall data, making the calculation of correction weights more comprehensive and objective. Bias dispersion measures the representativeness of the cleaning set to the historical dataset. A larger number of un-clustered data indicates fewer deletions, suggesting limited representativeness of the cleaning set, resulting in a larger bias dispersion and thus a larger correction weight. Conversely, fewer un-clustered data indicates better representativeness of the cleaning set, resulting in a smaller bias dispersion and thus a smaller correction weight. By comprehensively utilizing a large amount of historical data, rich reference information is provided for determining the suspected correction coefficients and correction weights. Learning and optimizing from historical experience improves the efficiency of the cleaning process.

[0089] In some embodiments of this application, when adjusting the cleaning prediction time according to a first time correction coefficient or a second time correction coefficient, the following are included: the first time correction coefficient is proportional to the cleaning prediction time, and the second time correction coefficient is proportional to the cleaning prediction time.

[0090] Specifically, assuming the first time correction coefficient or the first time correction coefficient is R, and the cleaning prediction time is T, the adjusted cleaning prediction time is determined to be R*T. By establishing a direct proportional relationship between the first time correction coefficient, the second time correction coefficient, and the cleaning prediction time, precise control of the cleaning prediction time is achieved. The initial bracket image of the LED bracket is acquired in real time to determine the cleaning prediction time, and the first time correction coefficient or the second time correction coefficient is obtained based on the historical dataset. The cleaning prediction time is then corrected according to the first time correction coefficient or the second time correction coefficient, which improves the reliability of the cleaning process, reduces the defect rate of LED beads, and thus improves the production efficiency of LED beads.

[0091] In summary, the beneficial effects of this invention are as follows: By processing and analyzing the initial image of the LED bracket, the predicted cleaning time is determined based on the comprehensive dust value and dust encapsulation model. This avoids the risk of mismatch between the cleaning time and the actual contamination situation caused by differences in human experience, thus improving the reliability and stability of the predicted cleaning time. The predicted cleaning time is compared with historical datasets to dynamically determine whether correction is needed. Historical experience is used to verify the accuracy of the predicted cleaning time, providing direction for precise correction. When correction is required, a first time correction coefficient or cleaning set is determined based on time similarity, making full use of historical datasets and flexibly correcting the predicted cleaning time according to different situations, ensuring that the corrected predicted cleaning time closely matches the actual contamination situation of the current LED bracket. By determining a potential correction coefficient from the cleaning set and then determining the correction weight based on the cleaning set, a second time correction coefficient is obtained. This comprehensively considers multiple factors, reducing encapsulation quality problems caused by improper cleaning time, ensuring the stability of cleaning quality, and thus improving the production efficiency of LED chips.

[0092] In another preferred embodiment based on the above embodiments, see [reference] Figure 2 As shown, this embodiment provides a packaging control system for LED chips, used to apply the above-described packaging control method for LED chips, including:

[0093] The acquisition module is configured to acquire an initial bracket image of the LED bracket, process the initial bracket image, determine the target bracket image, analyze the target bracket image, determine the comprehensive dust value of the target bracket image based on the analysis results, and determine the predicted cleaning time of the LED bracket based on the comprehensive dust value and the dust encapsulation model.

[0094] The judgment module is configured to compare the cleaning prediction time with the historical dataset and determine whether to correct the cleaning prediction time based on the comparison results.

[0095] The processing module is configured to, when it is determined that the cleaning prediction time needs to be corrected, determine the time similarity based on the cleaning prediction time and the historical dataset, determine the first time correction coefficient or cleaning set based on the time similarity, determine the suspected correction coefficient based on the cleaning set, reduce the historical dataset based on the cleaning set, determine the correction weight of the suspected correction coefficient, and determine the second time correction coefficient based on the suspected correction coefficient and the correction weight.

[0096] The cleaning module is configured to adjust the cleaning prediction time according to a first time correction coefficient or a second time correction coefficient, and to clean the LED bracket according to the adjusted cleaning prediction time.

[0097] Those skilled in the art will understand that embodiments of this application can be provided as methods, systems, or computer program goods. Therefore, this application can take the form of a completely hardware embodiment, a completely software embodiment, or an embodiment combining software and hardware aspects. Furthermore, this application can take the form of a computer program goods embodied on one or more computer-usable storage media (including, but not limited to, disk storage, CD-ROM, optical storage, etc.) containing computer-usable program code.

[0098] This application is described with reference to flowchart illustrations and / or block diagrams of methods, apparatus (systems), and computer program goods according to embodiments of this application. It will be understood that each block of the flowchart illustrations and / or block diagrams, and combinations of blocks in the flowchart illustrations and / or block diagrams, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, special-purpose computer, embedded processor, or other programmable data processing apparatus to produce a machine, such that the instructions, which execute via the processor of the computer or other programmable data processing apparatus, generate instructions for implementing the flowchart... Figure 1 One or more processes and / or boxes Figure 1 A device that provides the functions specified in one or more boxes.

[0099] These computer program instructions may also be stored in a computer-readable storage device that can direct a computer or other programmable data processing device to operate in a particular manner, such that the instructions stored in the computer-readable storage device produce an article of manufacture including instruction means, which are implemented in a process Figure 1 One or more processes and / or boxes Figure 1 The function specified in one or more boxes.

[0100] These computer program instructions may also be loaded onto a computer or other programmable data processing equipment to cause a series of operational steps to be performed on the computer or other programmable equipment to produce a computer-implemented process, thereby providing instructions that execute on the computer or other programmable equipment for implementing the process. Figure 1 One or more processes and / or boxes Figure 1 The steps of the function specified in one or more boxes.

[0101] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and not to limit it. Although the present invention has been described in detail with reference to the above embodiments, those skilled in the art should understand that modifications or equivalent substitutions can still be made to the specific implementation of the present invention. Any modifications or equivalent substitutions that do not depart from the spirit and scope of the present invention should be covered within the scope of protection of the claims of the present invention.

Claims

1. A package control method for an LED lamp bead, characterized in that, The method comprises the following steps: acquiring an initial support image of an LED support, processing the initial support image to determine a target support image, analyzing the target support image, determining a comprehensive dust value of the target support image based on the analysis result, and determining a cleaning prediction time of the LED support according to the comprehensive dust value and a dust packaging model; comparing the cleaning prediction time with a historical data set, and determining whether to correct the cleaning prediction time according to the comparison result; when it is determined to correct the cleaning prediction time, determining a time similarity according to the cleaning prediction time and the historical data set, determining a first time correction coefficient or a cleaning set according to the time similarity, determining a suspected correction coefficient according to the cleaning set, deleting the historical data set based on the cleaning set, determining a correction weight of the suspected correction coefficient, and determining a second time correction coefficient according to the suspected correction coefficient and the correction weight; adjusting the cleaning prediction time according to the first time correction coefficient or the second time correction coefficient, and cleaning the LED support according to the adjusted cleaning prediction time; when comparing the cleaning prediction time with the historical data set and determining whether to correct the cleaning prediction time according to the comparison result, the method comprises the following steps: the historical data set comprises a plurality of historical cleaning prediction times, a plurality of historical time correction coefficients, a plurality of historical comprehensive dust values, and a historical qualified data mean value, and each historical comprehensive dust value corresponds to a historical cleaning prediction time and a historical time correction coefficient; comparing the cleaning prediction time with the historical qualified data mean value, and determining whether to correct the cleaning prediction time according to the comparison result; when the cleaning prediction time is less than the historical qualified data mean value, it is determined to correct the cleaning prediction time, otherwise, the cleaning prediction time is not corrected, and the LED support is cleaned according to the cleaning prediction time.

2. The encapsulation control method for LED lamp beads according to claim 1, wherein, when processing the initial support image to determine the target support image, the method comprises the following steps: preprocessing the initial support image, wherein the preprocessing comprises image denoising, color equalization, and normalized pixel value; inputting the preprocessed initial support image into a neural network model, and outputting an image compliance value of the preprocessed initial support image based on the neural network model; when the image compliance value is greater than or equal to a preset image compliance value, the initial support image is determined as the target support image; when the image compliance value is less than the preset image compliance value, the initial support image of the LED support is reacquired.

3. The encapsulation control method for LED lamp beads according to claim 2, wherein, when analyzing the target support image based on the analysis result to determine the comprehensive dust value of the target support image, the method comprises the following steps: extracting all support pixel points of the target support image, acquiring corresponding support pixel values of all support pixel points, acquiring an adaptive target support image corresponding to the target support image, extracting all adaptive support pixel points of the adaptive target support image, and acquiring corresponding adaptive support pixel values of all adaptive support pixel points; Comparing all support pixels and all adaptive support pixels, and dividing all support pixels into a dust set according to the comparison result; When the support pixel value is greater than the adaptive support pixel value, the support pixel is divided into a first dust set; When the support pixel value is equal to the adaptive support pixel value, the support pixel is not divided; When the support pixel value is less than the adaptive support pixel value, the support pixel is divided into a second dust set; Statistically, the first number of support pixels in the first dust set is counted, the second number of support pixels in the second dust set is counted, the first average support pixel value of the first dust set is obtained, and the second average support pixel value of the second dust set is obtained; Determine the comprehensive dust value based on the first number, the second number, the first average support pixel value and the second average support pixel value.

4. The encapsulation control method for LED lamp beads according to claim 3, characterized in that, In determining the cleaning prediction time of the LED support according to the comprehensive dust value and the dust packaging model, comprising: Obtain a comprehensive dust value data set, and divide the comprehensive dust value data set into a training set and a test set, find the hyperparameters of the random forest model by grid search, and establish a random forest model; According to the training set, the random forest model is iteratively trained, the test set is substituted into the iteratively trained random forest model and verified; When the verification value reaches the verification threshold, the iteratively trained random forest model is determined as the dust packaging model, and the cleaning prediction time is obtained according to the comprehensive dust value.

5. The encapsulation control method for LED lamp beads of claim 1, wherein, In determining the time similarity according to the cleaning prediction time and the historical data set, and determining the first time correction coefficient or the cleaning set according to the time similarity, comprising: Comparing the cleaning prediction time with each historical cleaning prediction time in the historical data set to determine the time similarity; Comparing the time similarity with the pre-set time similarity threshold, and determining the first time correction coefficient or the cleaning set according to the comparison result; When there is data with time similarity greater than or equal to the time similarity threshold in the historical data set, the first time correction coefficient is determined according to the historical data set, otherwise, the cleaning set is determined according to the historical data set; When the first time correction coefficient is determined according to the historical data set, if the data with time similarity greater than or equal to the time similarity threshold is unique, the historical time correction coefficient corresponding to the data is taken as the first time correction coefficient; If the data with time similarity greater than or equal to the time similarity threshold is not unique, the historical time correction coefficient corresponding to the data with the maximum time similarity is taken as the first time correction coefficient; If the data with time similarity greater than or equal to the time similarity threshold is not unique, and the data with the maximum time similarity is several, the average of the historical time correction coefficients corresponding to each data with the maximum time similarity is taken as the first time correction coefficient.

6. The packaging control method for LED lamp beads according to claim 5, wherein, In determining the suspected correction coefficient according to the cleaning set, comprising: The comprehensive dust value, the cleaning prediction time and the historical data set are taken as a data set to be clustered, a historical time correction coefficient corresponding to each data in the data set to be clustered is extracted, an expected cluster number k is determined as 3, and parameters of a Gaussian distribution are initialized; Responsibility values are obtained according to probabilities of each data in the data set to be clustered belonging to each Gaussian distribution, a data set corresponding to the comprehensive dust value and the cleaning prediction time is obtained according to the responsibility values, and the data set is taken as the cleaning set; A correction mean value of the historical time correction coefficient in the cleaning set is obtained, and the correction mean value is determined as the suspected correction coefficient.

7. The packaging control method for LED lamp beads according to claim 6, wherein, When the historical data set is pruned based on the cleaning set, a correction weight of the suspected correction coefficient is determined, and a second time correction coefficient is determined according to the suspected correction coefficient and the correction weight, the method comprises: Data in the historical data set that is the same as the cleaning set is pruned, a pruned number is recorded, and a data number of data in the historical data set that is not pruned is counted; A number difference value of the pruned number and the data number is obtained, an absolute value of the number difference value is obtained, and a ratio of the absolute value to the pruned number is taken as a deviation dispersion; A power operation result of a natural constant as a base number and the deviation dispersion as an index is taken as the correction weight of the suspected correction coefficient, and the second time correction coefficient is a product value of the suspected correction coefficient and the correction weight.

8. The packaging control method for LED lamp beads according to claim 7, wherein, When the cleaning prediction time is adjusted according to the first time correction coefficient or the second time correction coefficient, the first time correction coefficient and the cleaning prediction time are in a proportional relationship, and the second time correction coefficient and the cleaning prediction time are in a proportional relationship. The method comprises:

9. A package control system for LED lamp beads, for applying the package control method for LED lamp beads as claimed in any one of claims 1-8, characterized in that, The collection module is configured to obtain an initial support image of an LED support, process the initial support image, determine a target support image, analyze the target support image, determine a comprehensive dust value of the target support image based on an analysis result, determine a cleaning prediction time of the LED support according to the comprehensive dust value and a dust packaging model; The judgment module is configured to compare the cleaning prediction time with a historical data set, and determine whether to correct the cleaning prediction time according to a comparison result; The processing module is configured to, when it is determined to correct the cleaning prediction time, determine a time similarity according to the cleaning prediction time and the historical data set, determine a first time correction coefficient or a cleaning set according to the time similarity, determine a suspected correction coefficient according to the cleaning set, prune the historical data set based on the cleaning set, determine a correction weight of the suspected correction coefficient, and determine a second time correction coefficient according to the suspected correction coefficient and the correction weight; The cleaning module is configured to adjust the cleaning prediction time according to the first time correction coefficient or the second time correction coefficient, and clean the LED support according to the adjusted cleaning prediction time. ​

Images