LED packaging detection method and system

Abstract

The invention relates to an LED packaging detection method and system, and the method comprises the steps: firstly obtaining a light source image of a to-be-detected LED light source, and then carrying out the feature extraction and analysis of the light source image, thereby obtaining image feature information; meanwhile, electrical testing is carried out on the to-be-detected LED, and electrical characteristic information is obtained; carrying out a thermal shock test and a vibration shock test on the to-be-detected LED package to respectively obtain corresponding thermal characteristic information and mechanical characteristic information; and finally, inputting the image feature information, the optical feature information, the thermal feature information and the mechanical feature information into a package detection model trained by combining a plurality of machine learning classifiers to obtain an LED package quality detection report, and judging the package detection quality grade by the report according to a preset threshold value. The detection accuracy and reliability are improved by comprehensively evaluating the quality of LED packaging in multiple aspects such as optics, electricity, thermal and mechanical properties.

Description

Technical Field

[0001] The present invention belongs to the field of information technology, and particularly relates to an LED packaging detection method and system. Background Art

[0002] In recent years, LED technology has developed rapidly, and its applications have gradually expanded from traditional lighting fields to multiple fields such as display, automotive, medical, and agriculture. For example, in the display field, the demand for small-pitch LED displays is increasing day by day, posing higher requirements for the accuracy and reliability of LED packaging; in the automotive field, LED vehicle lights need to have characteristics such as high temperature resistance, high humidity resistance, and earthquake resistance to adapt to the complex automotive use environment. These diverse application scenarios have prompted the LED packaging industry to continuously improve detection technologies to ensure that products can meet the stringent requirements of different fields. Traditional detection means have many limitations. For example, manual detection has low efficiency, is easily affected by subjective factors, and it is difficult to detect internal micro-defects; a single detection method cannot comprehensively evaluate the quality of LED packaging in terms of optical, electrical, thermal, and mechanical properties, making it difficult to improve the accuracy and reliability of detection. Summary of the Invention

[0003] Based on this, in view of the above technical problems, it is necessary to provide an LED packaging detection method and system that can comprehensively evaluate the quality of LED packaging in terms of optical, electrical, thermal, and mechanical properties.

[0004] In a first aspect, the present application provides an LED packaging detection method, including:

[0005] Obtaining a light source image of an LED light source to be detected.

[0006] Performing feature extraction and analysis on the light source image to obtain image feature information.

[0007] Performing electrical tests on the LED to be detected to obtain electrical feature information for corresponding electrical performance analysis.

[0008] Performing a thermal shock test and a vibration shock test on the LED packaging to be detected to obtain corresponding thermal feature information and mechanical feature information.

[0009] Inputting the image feature information, optical feature information, thermal feature information, and mechanical feature information into a trained packaging detection model to obtain a quality detection report for the LED packaging; the quality detection report determines the packaging detection quality level according to a preset threshold; wherein, the packaging detection model is trained by combining multiple machine learning classifiers.

[0010] In one embodiment, performing feature extraction and analysis on the light source image to obtain image feature information includes:

[0011] Sharpen the light source image of the LED light source to be detected using the Laplace operator to obtain a light source enhanced image with enhanced image.

[0012] Based on the light source enhanced image, perform positioning using an image recognition model constructed by a convolutional neural network to obtain the target image area; use the principal component analysis algorithm to extract features from the target image area to obtain the spectral features of the light source image.

[0013] Use the Harris corner detection algorithm to extract features from the target image area to obtain the corner features of the light source image.

[0014] Use the Canny edge detection algorithm to extract features from the target image area to obtain the edge features of the light source image.

[0015] Use the Laplacian of Gaussian operator blob detection method to extract features from the target image area to obtain the blob features of the light source image.

[0016] Based on the spectral features, corner features, edge features, and blob features of the light source image, perform feature point marking to obtain image feature points.

[0017] Use the feature fusion formula to fuse the features of each image feature point to obtain image feature information.

[0018] Use the least absolute shrinkage and selection operator algorithm to assist in verifying the image feature information to obtain updated image feature information.

[0019] In one of the embodiments, use the feature fusion formula to fuse the features of each image feature point to obtain image feature information, including:

[0020] Use the following feature fusion formula to calculate and obtain image feature information:

[0021] Y = Re(BN(Conv(concat(x 1 ,x 2 ,x 3 ,x 4 ))))

[0022] θ = Sig(Re(GP(Y)))

[0023] Y c = atten(θ,x 1 )+x 2 +x 3 +x 4

[0024] where, x 1 ,x 2 ,x 3 and x 4respectively represent the spectral features, corner points, edges, and blob feature points in the image feature points, concat(*) represents the concatenation operation, Conv(*) represents the convolution operation, BN(*) represents the batch normalization process, Re(*) represents the activation function, Y represents the fused feature map, GP(*) represents the global mean pooling process, Sig(*) represents the activation function, θ represents the weight vector, atten(*) represents the weighted operation on the feature image, Y c represents the image feature information.

[0025] In one embodiment, the LED to be detected is electrically tested to obtain the electrical feature information for the corresponding electrical performance analysis, including:

[0026] Obtain the electrical data of the LED test points; the electrical data includes the forward voltage, forward current, reverse voltage, reverse current of the LED, and the electrical signal waveform of the LED; wherein, the electrical signal waveform includes the switching response time and the noise dynamic characteristics.

[0027] Construct a data set from the electrical data to obtain the electrical sample data to be processed.

[0028] Based on the electrical sample data to be processed, perform normalization and outlier removal operations to obtain an electrical data set.

[0029] Based on the electrical data set, use the Synthetic Minority Over-sampling Technique (SMOTE) to balance the data and obtain a balanced data set.

[0030] Process the data in the balanced data set using the Recursive Feature Elimination (RFE) method to obtain different electrical feature subsets.

[0031] Construct a feature vector group for different electrical feature subsets to obtain an electrical feature vector group.

[0032] Use the performance evaluation formula to calculate the electrical feature vector group to obtain an electrical performance value.

[0033]

[0034] Among them, D i represents the electrical performance value of the tested LED, V a represents the forward voltage, I a is the forward current, R represents the standard resistance, V b represents the reverse voltage, I b represents the reverse current, μ represents the noise factor, μ max represents the maximum noise factor, T c represents the switching response time, T c,max represents the maximum switching response time, ω 1 、ω 2 、ω 3and ω 4 respectively represent the corresponding weight coefficients.

[0035] Based on a preset value, the electrical performance values are screened to obtain electrical characteristic information that meets the preset evaluation conditions.

[0036] In one embodiment, a thermal shock test and a vibration shock test are performed on the LED package to be detected, and the corresponding thermal characteristic information and mechanical characteristic information are obtained, including:

[0037] The LED package to be detected is placed in a high-temperature environment for a thermal shock experiment and a vibration shock test, and two-dimensional image information of the change in the micro-defect area of the LED package to be detected with thickness after the thermal shock experiment and the vibration shock test is obtained.

[0038] Calculations are performed based on the two-dimensional image information to obtain thermal shock parameters and vibration shock parameters.

[0039] Missing value processing is performed on the thermal shock parameters and vibration shock parameters to obtain the corresponding processed characteristic parameters.

[0040] A feature selection algorithm is used to encode each characteristic parameter to obtain a characteristic subset corresponding to each characteristic parameter.

[0041] According to the fitness function, the advantages and disadvantages of each characteristic subset are evaluated, and the optimal characteristic subsets corresponding to the thermal shock parameters and vibration shock parameters are the thermal characteristic information and the mechanical characteristic information respectively.

[0042] In one embodiment, calculations are performed based on the two-dimensional image information to obtain thermal shock parameters and vibration shock parameters, including:

[0043] The thermal shock parameters and vibration shock parameters are calculated using the following formula:

[0044]

[0045] v = k*(T c - t) 2

[0046] where E c represents the thermal shock parameter E 1 or the vibration shock parameter E 2 , T c represents the thickness T of the surface of the package body after experiencing the thermal shock or vibration shock test obtained by ultrasonic ranging 1 or T 2, dt represents the preset detection thickness. High-frequency ultrasonic waves are sent and collected at intervals of the preset detection thickness dt within a preset frequency range to obtain the area S of micro-defects in the ultrasonic image at this thickness. Data points (t, S) corresponding to the number of micro-defects at each thickness are recorded, and a two-dimensional image showing the variation of the micro-defect area S with the thickness t is generated. The corresponding functional relationship S = f c (t) is obtained through image fitting. v represents the expansion speed of micro-defects at each preset thickness, where k represents a preset influence factor.

[0047] In one embodiment, the image feature information, optical feature information, thermal feature information, and mechanical feature information are input into the trained package detection model to obtain a quality inspection report for the LED package, including:

[0048] Based on historical training data, a machine learning classifier is used for model training to obtain a detection analyzer; the machine learning classifier is at least one of a support vector machine, decision tree, neural network, and random forest.

[0049] Each detection analyzer is trained using the stochastic gradient descent method to obtain a pre-trained package detection model.

[0050] The AUC value of each pre-trained package detection model is calculated using a formula to obtain a trained package detection model; the trained package detection model is a package detection model whose performance meets preset conditions.

[0051] The image feature information, optical feature information, thermal feature information, and mechanical feature information are input into the trained package detection model to obtain the package detection result of the LED package.

[0052] The following formula is used to calculate the image feature information, optical feature information, thermal feature information, and mechanical feature information to obtain the comprehensive score of the package detection result:

[0053]

[0054] Among them, S represents the comprehensive score of the package detection result, n represents the number of feature information affecting the package detection result, w i represents the weight of the i-th feature information, S i represents the value of the i-th feature information, f i (S i ) represents a score conversion function for the i-th feature information, which converts the feature evaluation value into a score within the range of [0, 1].

[0055] The quality level is determined based on the preset threshold for the comprehensive score to obtain a quality inspection report for the quality discrimination of the LED package detection.

[0056] Second aspect, the present application also provides an LED package detection system, which includes:

[0057] An information acquisition module, configured to acquire a light source image of an LED light source to be detected; also configured to perform feature extraction and analysis on the light source image to obtain image feature information; also configured to perform electrical tests on the LED to be detected to obtain electrical feature information for analyzing its electrical performance; also configured to perform thermal shock tests and vibration shock tests on the LED package to be detected to obtain corresponding thermal feature information and mechanical feature information.

[0058] A report generation module, configured to input the image feature information, optical feature information, thermal feature information, and mechanical feature information into a trained package detection model to obtain a quality detection report of the LED package; the quality detection report determines the package detection quality level according to a preset threshold; wherein, the package detection model is trained by combining multiple machine learning classifiers.

[0059] Third aspect, the present application also provides a computer device, including a memory and a processor, the memory stores a computer program, and when the processor executes the computer program, the method described in the first aspect of the present application is implemented.

[0060] Fourth aspect, the present application also provides a computer-readable storage medium, on which a computer program is stored, and when the computer program is executed by a processor, the method described in the first aspect of the present application is implemented.

[0061] For the above-mentioned LED package detection method and system, first, a light source image of the LED light source to be detected is acquired, and then feature extraction and analysis are performed on it to obtain image feature information; at the same time, electrical tests are performed on the LED to be detected to obtain electrical feature information for analyzing its electrical performance; and thermal shock tests and vibration shock tests are carried out on the LED package to be detected to obtain corresponding thermal feature information and mechanical feature information respectively; finally, the image feature information, optical feature information, thermal feature information, and mechanical feature information are input into a trained package detection model that combines multiple machine learning classifiers to obtain a quality detection report of the LED package, and this report determines the package detection quality level according to a preset threshold. By comprehensively evaluating the quality of the LED package in multiple aspects such as optical, electrical, thermal, and mechanical properties, the accuracy and reliability of the detection are improved. Description of the Drawings

[0062] In order to more clearly illustrate the technical solutions in the embodiments of the present application or related technologies, the following will briefly introduce the drawings required for use in the description of the embodiments or related technologies. Obviously, the following drawings are only some embodiments of the present application. For those of ordinary skill in the art, other drawings can be obtained based on these drawings without creative efforts.

[0063] Figure 1 This is a flowchart of an LED packaging detection method provided by an embodiment of the present invention;

[0064] Figure 2 This is a flowchart of extracting and analyzing features of a light source image to obtain image feature information provided by an embodiment of the present invention;

[0065] Figure 3 This is a flowchart of performing thermal shock tests and vibration shock tests on the LED packaging to be detected to obtain corresponding thermal feature information and mechanical feature information provided by an embodiment of the present invention;

[0066] Figure 4 This is a flowchart of inputting image feature information, optical feature information, thermal feature information, and mechanical feature information into a trained packaging detection model to obtain a quality detection report of the LED packaging provided by an embodiment of the present invention;

[0067] Figure 5 This is a structural block diagram of an LED packaging detection system provided by an embodiment of the present invention. Detailed implementation manners

[0068] In order to make the objectives, technical solutions, and advantages of the present application clearer, the present application will be further described in detail below with reference to the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are only used to explain the present application and are not used to limit the present application.

[0069] First, the implementation environment of the embodiments of the present application will be described. Exemplarily, the implementation environment includes a detection device, data acquisition hardware, and a data processing device.

[0070] In the LED packaging detection method and system, the detection device transmits the detected analog signals to a data acquisition card through a specific interface. After the data acquisition card converts the analog signals into digital signals, it transmits them to an industrial computer. At the same time, the industrial computer is connected to the detection device to control the detection process to achieve automatic control and coordination of the entire detection system. As a data processing device, the industrial computer processes, analyzes, and stores the collected data to complete the data interaction and processing of the entire detection process.

[0071] The detection device is the front-end sensing component of the LED packaging detection system, which obtains various key information of the LED packaging. Optical detection devices, such as microscopes, are used for fine observation of microscopic structures, and spectrometers and integrating spheres can measure optical performance parameters. Electrical detection devices measure the basic and precise electrical parameters of LEDs through multimeters, semiconductor parameter testers, etc.

[0072] The data acquisition hardware includes a data acquisition card that receives analog signals from the detection equipment and converts the signals into digital signals. It controls the operation process of the detection equipment in an automated detection scenario and is connected to the data acquisition at the same time to ensure that the data acquisition can proceed orderly according to the preset detection process and accurately collect various types of detection data.

[0073] The data processing device includes an industrial computer that runs the detection system software, deeply processes and analyzes the digital signals transmitted by the data acquisition hardware, determines whether the LED package is qualified according to the preset algorithm, generates a detailed detection report and stores the data.

[0074] In combination with the above implementation environment, the application scenarios of the embodiments of the present application are described.

[0075] A method and system for detecting LED packages provided by an embodiment of the present application use an optical detection device such as a microscope to observe the microscopic structure of the LED package, and at the same time measure its optical performance parameters with the help of a spectral analyzer and an integrating sphere; measure basic and precise electrical parameters with a multimeter, a semiconductor parameter tester, etc. Then, the data acquisition card converts the analog signal output by the detection equipment into a digital signal and transmits it to the industrial computer for analysis and processing of the acquired data, and determines whether the LED package is qualified according to the preset standard, and generates a detailed detection report. Exemplarily, a method and system for detecting LED packages provided by an embodiment of the present application can be applied to at least one of the following scenarios including but not limited to the following scenarios.

[0076] First, the method and system for detecting LED packages are applied to the LED lighting product production line. On the LED lighting product production line, first, the spectral analyzer and the integrating sphere in the detection equipment measure the optical parameters of the just-packaged LED, such as luminous flux, color rendering index, etc., and the multimeter and the semiconductor parameter tester detect its electrical performance. The data acquisition card quickly collects the analog signal and converts it into a digital signal, and transmits it to the industrial computer. The computer analyzes and determines whether the LED meets the requirements according to the set standard parameters of the lighting product.

[0077] Second, the method and system for detecting LED packages are applied to the automotive lighting scenario. In this scenario, the aging test equipment in the detection equipment simulates the complex environment and long-term working state that the LED faces during the driving of the car. At the same time, the microscope is used to detect the changes in the microscopic structure of the LED before and after aging, such as whether the solder joints are loose, etc. The data acquisition card real-time collects the changes in various performance data during the aging process and transmits it to the industrial computer. The computer analyzes the data through a professional algorithm to evaluate the performance of the LED at different stages, predicts its lifespan in actual use in the car, and the LED that passes the strict detection is applied to the automotive lighting system.

[0078] With the pursuit of energy conservation, environmental protection and high-quality life by people, the market has higher and higher requirements for the performance indicators of LED products, such as luminous efficacy, lifespan, stability, etc. Consumers expect LED lamps to provide a more uniform and comfortable lighting effect and maintain good performance during long-term use. This requires more strict and precise detection methods during the LED packaging process to ensure product quality. At the same time, LED packaging materials are constantly updated. New types of encapsulation adhesives, brackets and other materials have better optical properties, thermal stability and weather resistance. However, the characteristics of these new materials also require corresponding detection technologies to evaluate and verify. For example, some encapsulation adhesives with high refractive index can improve the light extraction efficiency of LEDs, but may be more sensitive to temperature and humidity. Therefore, it is necessary to conduct tests under specific environmental conditions to determine their impact on the performance of LEDs. An LED packaging detection method and system provided by an embodiment of the present application can be exemplarily applied to experimental steps for pretreatment in the following environmental conditions including but not limited to:

[0079] a. Dehumidify the LED packaging materials. Dehumidify the in-factory materials at 150°C ± 5°C for 2 - 6H, and dehumidify other materials at 85°C ± 5°C for 10 - 14H.

[0080] b. Perform taping and pasting operations after dehumidification.

[0081] c. Place the LED packaging products in different environmental conditions for testing in sequence. First, in an environment of 85°C ± 5°C and 85%RH ± 5%, continuously cycle for 36 - 108H (a total of 9 cycles) in the way of lighting for 2 - 6H and extinguishing for 2 - 6H (energized) with white balance.

[0082] d. Then, store the products with the power on and the screen blacked out at 25°C ± 5°C and 95%RH ± 5%RH for 22 - 26H.

[0083] e. Finally, again in an environment of 85°C ± 5°C and 85%RH ± 5%, continuously cycle for 36 - 108H (a total of 9 cycles) in the way of lighting for 2 - 6H and extinguishing for 2 - 6H (energized) with white balance.

[0084] f. When the experiment ends, record and analyze the defective conditions including color deficiency, dark and bright spots, and caterpillars.

[0085] In one embodiment, as Figure 1 shown, a method for detecting LED packaging is provided, which may include the following steps:

[0086] Step S101, obtain the light source image of the LED light source to be detected.

[0087] Typically, a high-resolution imaging device, such as a professional industrial camera, is used in combination with a specific optical lens to capture the LED light source to be detected under a stable lighting environment and fixed shooting parameters. During shooting, it is necessary to ensure that the LED light source is in a normal lit state to obtain the image information when it emits light. The light source images collected in this way can cover initial information such as the light-emitting area, brightness distribution, and color uniformity of the LED.

[0088] Step S102: Extract and analyze the features of the light source image to obtain image feature information.

[0089] After obtaining the light source image of the LED light source to be detected, first perform preprocessing operations on the image such as noise reduction and enhancement. Then, extract features from multiple dimensions. For example, use edge detection algorithms to extract the contour features of the LED light source image to determine whether its shape is regular and whether there are abnormalities such as missing corners and burrs; analyze the texture features of the image through the gray-level co-occurrence matrix to evaluate the uniformity of LED light emission; use color space conversion and statistical methods to obtain color features and detect whether there are color deviations. Comprehensively analyze the extracted features, and use classification and clustering algorithms in machine learning or deep learning to compare with the standard LED light source image feature library, and finally obtain image feature information including detailed situations in multiple aspects such as shape, texture, and color.

[0090] Step S103: Conduct electrical tests on the LED to be detected to obtain electrical feature information for corresponding electrical performance analysis.

[0091] During the test, professional electrical measurement instruments are used to simulate various electrical conditions of the LED in actual operation. First, perform a forward voltage test. Under a specified forward current, measure the voltage across the LED. The magnitude of the forward voltage can directly reflect the performance and quality of the LED chip. If the forward voltage deviates from the standard range, the chip has defects or poor performance. Then, perform a reverse leakage current test. Under a reverse bias voltage, detect the magnitude of the leakage current of the LED. Excessive reverse leakage current affects the stability and service life of the LED. In addition, also test the relationship between the luminous efficiency and current of the LED. By changing the input current, measure the corresponding luminous flux, analyze its luminous efficiency curve, and judge the luminous performance of the LED under different current drives. Organize and analyze the test data to obtain electrical feature information such as forward voltage values, reverse leakage current magnitudes, and luminous efficiency curves, and accurately evaluate the electrical performance of the LED and determine whether it meets the quality standards.

[0092] Step S104: Conduct thermal shock tests and vibration shock tests on the LED package to be detected to obtain corresponding thermal feature information and mechanical feature information.

[0093] Specifically, in the thermal shock test, a professional thermal shock test chamber is used to rapidly switch the LED package sample between high-temperature and low-temperature environments. For example, it quickly switches from a high temperature of 125°C to a low temperature of -40°C and repeats this cycle multiple times to simulate the severe temperature change scenarios that the LED may encounter during actual use. During this process, a high-precision temperature sensor is used to monitor the temperature changes at key parts inside the LED package in real time, and a thermal imager is used to record the surface temperature distribution. Through analysis, thermal characteristics information such as thermal resistance and thermal diffusivity can be obtained to evaluate the heat dissipation performance and thermal stability of the LED package and to determine whether problems such as material delamination and chip damage occur during temperature changes.

[0094] In the vibration and shock test, the LED package is fixed on a vibration table, and vibration and shock tests are carried out according to preset parameters such as vibration frequency, amplitude, and shock acceleration. By simulating the jolts during transportation, mechanical vibrations, and accidental shocks in the use environment, an acceleration sensor is used to monitor the acceleration changes during vibration, and a strain gauge is used to measure the strain of the package structure. After the test, a comprehensive inspection of the LED package is carried out to observe whether there are mechanical damages such as pin fractures, chip displacements, and package body cracks, and the monitoring data is analyzed to obtain mechanical characteristics information such as structural strength, vibration resistance, and fatigue life.

[0095] Step S105: Input the image feature information, optical feature information, thermal feature information, and mechanical feature information into the trained package detection model to obtain a quality inspection report for the LED package; the quality inspection report determines the quality level of the package inspection according to a preset threshold; among them, the package detection model is trained by combining multiple machine learning classifiers.

[0096] After completing the multi-faceted inspection of the LED light source and obtaining the image feature information, optical feature information, thermal feature information, and mechanical feature information, the data is input into a specially trained package detection model. This package detection model is trained by integrating multiple machine learning classifiers to generate a quality inspection report for the LED package. The report determines the quality level of the package inspection according to a preset threshold, and the threshold is determined based on multiple factors such as a large amount of historical data, industry standards, and actual production requirements. For example, if the image features show that the LED light source has obvious spot non-uniformity and the thermal feature information indicates that the thermal resistance exceeds the normal range, combined with the model analysis results, when the feature parameters exceed the corresponding preset threshold, the report determines that the quality level of the LED package is low and there may be problems affecting its performance and reliability; on the contrary, if all the feature information is within the threshold range, the report gives a higher quality level evaluation.

[0097] The above LED package detection method first obtains the light source image of the LED light source to be detected, and then extracts and analyzes its features to obtain image feature information. At the same time, electrical tests are performed on the LED to be detected to obtain electrical feature information for analyzing its electrical performance. In addition, thermal shock tests and vibration shock tests are carried out on the LED package to be detected, and the corresponding thermal feature information and mechanical feature information are obtained respectively. Finally, the image feature information, optical feature information, thermal feature information and mechanical feature information are input into the package detection model trained by combining multiple machine learning classifiers to obtain the quality detection report of the LED package. This report determines the quality level of the package detection according to the preset threshold. By comprehensively evaluating the quality of the LED package in terms of optical, electrical, thermal and mechanical performance, the accuracy and reliability of the detection are improved.

[0098] In one embodiment, as Figure 2 shown, the feature extraction and analysis of the light source image to obtain image feature information may include the following steps:

[0099] Step S201, perform sharpening processing on the light source image of the LED light source to be detected using the Laplacian operator to obtain a light source enhanced image with enhanced image.

[0100] Step S202, perform localization on the light source enhanced image using the image recognition model constructed by the convolutional neural network to obtain the target image area; use the principal component analysis algorithm to extract features from the target image area to obtain the spectral features of the light source image.

[0101] Step S203, use the Harris corner detection algorithm to extract features from the target image area to obtain the corner features of the light source image.

[0102] Step S204, use the Canny edge detection algorithm to extract features from the target image area to obtain the edge features of the light source image.

[0103] Step S205, use the Laplacian of Gaussian operator spot detection method to extract features from the target image area to obtain the spot features of the light source image.

[0104] Step S206, perform feature point marking based on the spectral features, corner features, edge features and spot features of the light source image to obtain image feature points.

[0105] Step S207, use the feature fusion formula to fuse the features of each image feature point to obtain image feature information.

[0106] Step S208, use the least absolute shrinkage and selection operator algorithm to assist in verifying the image feature information to obtain updated image feature information.

[0107] During the detection process of the LED light source, first, the Laplace operator is used to sharpen the light source image of the LED light source to be detected, and a light source enhancement image with enhanced image is obtained. Then, based on the light source enhancement image, an image recognition model constructed by a convolutional neural network is used for positioning to determine the target image area. Subsequently, the principal component analysis algorithm is used to extract features from this area to obtain the spectral features of the light source image. Then, the Harris corner detection algorithm (Harris & Stephens corner detection algorithm), Canny edge detection algorithm (Canny edge detection algorithm), and Laplacian of Gaussian operator spot detection method are respectively used to extract features from the target image area, and the corner features, edge features, and spot features of the light source image are obtained in sequence. After that, feature point marking is performed according to the spectral features, corner features, edge features, and spot features of the light source image, and then image feature points are obtained. The feature fusion formula is used to fuse each image feature point to obtain image feature information. Finally, the least absolute shrinkage and selection operator algorithm is used to assist in verifying the image feature information to obtain updated image feature information, thereby completing the comprehensive extraction and optimization of the LED light source image features.

[0108] In one embodiment, using the feature fusion formula to perform feature fusion on each image feature point to obtain image feature information may include the following steps:

[0109] Using the following feature fusion formula, calculate to obtain image feature information:

[0110] Y = Re(BN(Conv(concat(x 1 ,x 2 ,x 3 ,x 4 ))))

[0111] θ = Sig(Re(GP(Y)))

[0112] Y c = atten(θ,x 1 )+x 2 +x 3 +x 4

[0113] Among them, x 1 , x 2 , x 3 and x 4They respectively represent the spectral features, corner points, edges, and spot feature points in the image feature points. concat(*) represents the concatenation operation, Conv(*) represents the convolution operation, BN(*) represents the batch normalization process, Re(*) represents the activation function, Y represents the fused feature map, GP(*) represents the global average pooling process, Sig(*) represents the activation function, θ represents the weight vector, atten(*) represents the weighted operation on the feature image, and Y c represents the image feature information.

[0114] Feature fusion is performed on each image feature point using the feature fusion formula to obtain the image feature information, and calculations are carried out using a specific feature fusion formula. Different types of feature points such as spectral features, corner points, edges, and spots are combined through the concatenation operation, and then through the convolution operation, batch normalization process, activation using the activation function, and finally through the global average pooling process, the fused feature map is obtained. Among them, the feature image is weighted through the weight vector, and finally through the operation of the activation function, the image feature information is obtained.

[0115] In one embodiment, electrical tests are performed on the LED to be detected to obtain the electrical feature information corresponding to the electrical performance analysis, which may include the following steps:

[0116] Step S301, obtaining the electrical data of the LED test point; the electrical data includes the forward voltage, forward current, reverse voltage, reverse current of the LED, and the electrical signal waveform of the LED; among them, the electrical signal waveform includes the switching response time and the noise dynamic characteristics.

[0117] Step S302, constructing a data set for the electrical data to obtain the electrical sample data to be processed.

[0118] Step S303, performing normalization and outlier removal operations based on the electrical sample data to be processed to obtain the electrical data set.

[0119] Step S304, balancing the data based on the electrical data set using the synthetic minority over-sampling technique to obtain the balanced data set.

[0120] Step S305, processing the data in the balanced data set using the recursive feature elimination method to obtain different electrical feature subsets.

[0121] Step S306, constructing a feature vector group for different electrical feature subsets to obtain the electrical feature vector group.

[0122] Step S307, calculating the electrical feature vector group using the performance evaluation formula to obtain the electrical performance value.

[0123]

[0124] Among them, D i represents the electrical performance value of the tested LED, V a represents the forward voltage, I a is the forward current, R represents the standard resistance, V b represents the reverse voltage, I b represents the reverse current, μ represents the noise factor, μ max represents the maximum noise factor, T c represents the switching response time, T c,max represents the maximum switching response time, ω 1 、ω 2 、ω 3 and ω 4 respectively represent the corresponding weight coefficients.

[0125] Step S308, screen the electrical performance values based on preset values to obtain electrical characteristic information that meets the preset evaluation conditions.

[0126] First, obtain the electrical data of the LED test points, construct a data set for this electrical data, so as to obtain the electrical sample data to be processed; then perform normalization and outlier removal operations based on the electrical sample data to be processed to obtain an electrical data set; use the synthetic minority over-sampling technique to perform data balancing processing on this electrical data set, and then obtain a balanced data set; then process the data in the balanced data set using the recursive feature elimination method to obtain different electrical feature subsets; then construct a feature vector group for different electrical feature subsets to obtain an electrical feature vector group; calculate the electrical feature vector group using the performance evaluation formula, and finally screen the calculated electrical performance values based on preset values to obtain electrical characteristic information that meets the preset evaluation conditions.

[0127] The above embodiments of electrically testing the LED to be detected and obtaining electrical characteristic information improve the accuracy and effectiveness of the evaluation of the electrical performance of the LED, are conducive to realizing the refined quality control of LED products, optimizing the production process, reducing the defective rate, and thus promoting the improvement of the performance and quality of LED products.

[0128] In one of the embodiments, as Figure 3 shown, perform a thermal shock test and a vibration shock test on the LED package to be detected to obtain corresponding thermal characteristic information and mechanical characteristic information, which may include the following steps:

[0129] Step S401, place the LED package to be detected in a high-temperature environment for a thermal shock experiment and a vibration shock test to obtain two-dimensional image information of the change in the area of the micro-defects of the LED package to be detected with the thickness after the thermal shock experiment and the vibration shock test.

[0130] Specifically, high-frequency ultrasonic waves are transmitted and collected at a preset frequency range based on a preset detection thickness to obtain an ultrasonic image at this thickness; defects are extracted from the ultrasonic image to obtain the area of micro-defects. Based on the data points formed by the number of micro-defects corresponding to each thickness, two-dimensional image information of the change of the micro-defect area with the thickness is generated.

[0131] Step S402, calculate based on the two-dimensional image information to obtain the thermal shock parameter and the vibration shock parameter.

[0132] Step S403, perform missing value processing on the thermal shock parameter and the vibration shock parameter to obtain the corresponding processed characteristic parameters.

[0133] Step S404, use a feature selection algorithm to encode each characteristic parameter to obtain a feature subset corresponding to each characteristic parameter.

[0134] Step S405, evaluate the quality of each feature subset according to the fitness function, and obtain that the optimal feature subsets corresponding to the thermal shock parameter and the vibration shock parameter are the thermal characteristic information and the mechanical characteristic information respectively.

[0135] First, place the LED package to be detected in a high-temperature environment and conduct thermal shock experiments and vibration shock tests simultaneously, and then obtain two-dimensional image information of the change of the micro-defect area of the LED package to be detected with the thickness after the thermal shock experiment and the vibration shock test; then calculate based on this two-dimensional image information to obtain the thermal shock parameter and the vibration shock parameter; then perform missing value processing on the thermal shock parameter and the vibration shock parameter to obtain the corresponding processed characteristic parameters; use a feature selection algorithm to encode each characteristic parameter to obtain a feature subset corresponding to each characteristic parameter; finally, evaluate the quality of each feature subset according to the fitness function to determine the optimal feature subsets corresponding to the thermal shock parameter and the vibration shock parameter respectively, and use them as the thermal characteristic information and the mechanical characteristic information.

[0136] The thermal characteristic information and the mechanical characteristic information can accurately reflect the key characteristics of the package in terms of thermal and mechanical properties, help to accurately evaluate the reliability and stability of the LED package, thereby improving the overall performance and service life of the product, reducing the failure risk caused by thermal shock and vibration shock, and ensuring the normal use of the LED product in different environments.

[0137] In one of the embodiments, calculating based on the two-dimensional image information to obtain the thermal shock parameter and the vibration shock parameter includes:

[0138] Use the following formula to calculate the thermal shock parameter and the vibration shock parameter:

[0139]

[0140] v=k*(Tc -t) 2

[0141] Among them, E c represents the thermal shock parameter E 1 or the vibration shock parameter E 2 , T c represents the thickness T of the surface of the package body obtained by ultrasonic ranging after undergoing thermal shock or vibration shock tests 1 or T 2 , dt represents the preset detection thickness. High-frequency ultrasonic waves are sent and collected at intervals of the preset detection thickness dt within a preset frequency range to obtain the area S of micro-defects in the ultrasonic image at this thickness. Data points (t, S) corresponding to the number of micro-defects at each thickness are recorded, and a two-dimensional image of the change in the area S of micro-defects with the thickness t is generated. The corresponding functional relationship S = f c (t) is obtained through image fitting. v represents the expansion speed of micro-defects at preset thicknesses, where k represents a preset influence factor.

[0142] First, various information such as the thickness of the package body surface obtained from ultrasonic ranging, the preset detection thickness, the area of micro-defects at different thicknesses in the ultrasonic image, and the functional relationship obtained through image fitting are organically integrated by using a specific formula. The complex two-dimensional image information can be quantified into concise thermal shock parameters and vibration shock parameters, providing specific quantitative indicators for evaluating the performance of LED packages after thermal shock and vibration shock tests, making it possible to compare the performance between different package bodies. This calculation method takes into account the expansion speed of micro-defects at each thickness, helps to deeply analyze the defect evolution of the package body during thermal shock and vibration shock processes, and can predict in advance the failure possibility of the package body in extreme environments. It not only helps to improve the accuracy of detection but also provides data support for improving the package material and structure design, thereby enhancing the quality and reliability of LED package products, reducing the failure risk caused by thermal shock and vibration shock, and improving the durability and stability of the product under different environmental conditions.

[0143] In one embodiment, as Figure 4 shown, inputting image feature information, optical feature information, thermal feature information, and mechanical feature information into a trained package detection model to obtain a quality detection report for the LED package may include the following steps:

[0144] Step S501, training a model using a machine learning classifier based on historical training data to obtain a detection analyzer; the machine learning classifier is at least one of a support vector machine, a decision tree, a neural network, and a random forest.

[0145] Step S502: Train each detection analyzer using the stochastic gradient descent method to obtain a pre-trained packaged detection model.

[0146] Step S503: Calculate the AUC value of each pre-trained packaged detection model using a formula to obtain a trained packaged detection model; the trained packaged detection model is a packaged detection model whose performance meets the preset conditions.

[0147] Step S504: Input the image feature information, optical feature information, thermal feature information, and mechanical feature information into the trained packaged detection model to obtain the packaged detection result of the LED package.

[0148] Step S505: Calculate the comprehensive score of the packaged detection result using the following formula for the image feature information, optical feature information, thermal feature information, and mechanical feature information:

[0149]

[0150] where S represents the comprehensive score of the packaged detection result, n represents the number of feature information affecting the packaged detection result, w i represents the weight of the i-th feature information, S i represents the value of the i-th feature information, f i (s i ) represents the score conversion function for the i-th feature information, which converts the feature evaluation value into a score within the range of [0, 1].

[0151] Step S506: Determine the quality level based on the preset threshold for the comprehensive score to obtain a quality inspection report for the quality discrimination of the LED package detection.

[0152] In this embodiment, first, a machine learning classifier (including at least one of a support vector machine, decision tree, neural network, and random forest) is used for model training based on historical training data to obtain a detection analyzer; then, the stochastic gradient descent method is used to train each detection analyzer to obtain a pre-trained packaged detection model; next, the AUC value of each pre-trained packaged detection model is calculated using a formula, and a trained packaged detection model whose performance meets the preset conditions is selected; the image feature information, optical feature information, thermal feature information, and mechanical feature information are input into the trained packaged detection model to obtain the packaged detection result of the LED package; then, the above multi-faceted feature information is calculated using a formula to obtain the comprehensive score of the packaged detection result; finally, the quality level is determined based on the preset threshold for the comprehensive score to generate a quality inspection report for the quality discrimination of the LED package detection.

[0153] In one of the embodiments, as Figure 5As shown in the figure, the present application also provides an LED package detection system, which may include:

[0154] An information acquisition module 601, configured to acquire a light source image of an LED light source to be detected; further configured to perform feature extraction and analysis on the light source image to obtain image feature information; further configured to perform electrical tests on the LED to be detected to obtain electrical feature information for corresponding electrical performance analysis; further configured to perform thermal shock tests and vibration shock tests on the LED package to be detected to obtain corresponding thermal feature information and mechanical feature information.

[0155] A report generation module 602, configured to input the image feature information, optical feature information, thermal feature information, and mechanical feature information into a trained package detection model to obtain a quality detection report of the LED package; the quality detection report determines the package detection quality level according to a preset threshold; wherein, the package detection model is trained by combining multiple machine learning classifiers.

[0156] For the above-mentioned LED package detection method and system, the information acquisition module acquires a light source image of the LED light source to be detected, performs feature extraction and analysis on the light source image to obtain image feature information, and simultaneously performs electrical tests on the LED to be detected to obtain electrical feature information for electrical performance analysis, and performs thermal shock tests and vibration shock tests on the LED package to be detected to obtain corresponding thermal feature information and mechanical feature information. The report generation module inputs the image feature information, optical feature information, thermal feature information, and mechanical feature information obtained by the information acquisition module into the trained package detection model to obtain a quality detection report of the LED package. This quality detection report determines the package detection quality level according to a preset threshold. The package detection model is trained by combining multiple machine learning classifiers. This system comprehensively evaluates the quality of the LED package in multiple aspects such as optical, electrical, thermal, and mechanical performance, improving the accuracy and reliability of the detection.

[0157] It should be understood that although each step in the flowcharts involved in the above-described embodiments is displayed in sequence according to the indication of the arrows, these steps do not necessarily need to be executed in the order indicated by the arrows. Unless there is a clear indication in this article, the execution of these steps has no strict order limit, and these steps can be executed in other orders. Moreover, at least a part of the steps in the flowcharts involved in the above-described embodiments may include multiple steps or multiple stages. These steps or stages do not necessarily need to be executed at the same time, but can be executed at different times. The execution order of these steps or stages does not necessarily need to be sequential, but can be executed alternately or alternately with at least a part of other steps or steps or stages in other steps.

[0158] In one embodiment, a computer device is provided, including a memory and a processor. The memory stores a computer program, and when the processor executes the computer program, the steps of an LED packaging detection method as described above are implemented.

[0159] In one embodiment, a computer-readable storage medium is provided, on which a computer program is stored. When the computer program is executed by a processor, the steps in the above method embodiments are implemented.

[0160] For the device embodiments, since they basically correspond to the method embodiments, the relevant parts can be referred to the partial descriptions of the method embodiments. The device embodiments described above are only illustrative. The components described as separate components may or may not be physically separated, and the components shown as units may or may not be physical units, that is, they may be located in one place or distributed to multiple network units. Some or all of the modules can be selected according to actual needs to achieve the purpose of the present disclosure solution. A person of ordinary skill in the art can understand and implement it without creative efforts.

[0161] The above embodiments only represent several implementation manners of the embodiments of the present application. The description is relatively specific and detailed, but it should not be construed as a limitation on the patent scope of the embodiments of the application. It should be noted that for those of ordinary skill in the art, without departing from the concept of the embodiments of the present application, several modifications and improvements can still be made, and these all belong to the protection scope of the embodiments of the present application.

Claims

1. A method for detecting LED packaging, characterized in that: The method comprises: Acquire a light source image of the LED light source to be detected; Extracting and analyzing features of the light source image to obtain image feature information; Conduct electrical tests on the LED to be tested to obtain electrical characteristic information corresponding to electrical performance analysis; Perform thermal shock test and vibration shock test on the LED package to be tested to obtain corresponding thermal characteristic information and mechanical characteristic information; The image feature information, optical feature information, thermal feature information and mechanical feature information are input into a trained packaging inspection model to obtain a quality inspection report for LED packaging; the quality inspection report determines the packaging inspection quality level according to a preset threshold; wherein the packaging inspection model is obtained by training in combination with multiple machine learning classifiers.

2. The method according to claim 1, characterized in that The extracting and analyzing the features of the light source image to obtain image feature information includes: The light source image of the LED light source to be detected is sharpened using a Laplace operator to obtain an image-enhanced light source enhanced image; Based on the light source enhanced image, an image recognition model constructed by a convolutional neural network is used to locate the target image area; a principal component analysis algorithm is used to extract features of the target image area to obtain spectral features of the light source image; Using the Harris corner detection algorithm to extract features from the target image area, and obtain corner features of the light source image; Using the Canny edge detection algorithm to extract features from the target image area, and obtain edge features of the light source image; Using the Laplace Gaussian operator spot detection method to extract features of the target image area, to obtain spot features of the light source image; Marking feature points based on the spectral features, corner features, edge features and spot features of the light source image to obtain image feature points; Using a feature fusion formula to perform feature fusion on each of the image feature points to obtain image feature information; The image feature information is auxiliary verified using a minimum absolute shrinkage and selection operator algorithm to obtain updated image feature information.

3. The method according to claim 2, characterized in that The step of using a feature fusion formula to fuse the feature points of each image to obtain image feature information includes: The image feature information is calculated using the following feature fusion formula: Y=Re(BN(Conv(concat(x1,x2,x3,x4)))) θ=Sig(Re(GP(Y))) Y c =atten(θ,x1)+x2+x3+x4 Among them, x1, x2, x3 and x4 represent the spectral features, corner points, edge and spot feature points in the image feature points respectively, concat(*) represents the splicing operation, Conv(*) represents the convolution operation, BN(*) represents the batch normalization processing, Re(*) represents the activation function, Y represents the fusion feature map, GP(*) represents the global mean pooling processing, Sig(*) represents the activation function, θ represents the weight vector, atten(*) represents the weighted operation on the feature image, and Y c Represents image feature information.

4. The method according to claim 1, characterized in that The electrical test is performed on the LED to be tested to obtain electrical characteristic information corresponding to the electrical performance analysis, including: Acquire electrical data of the LED test point; the electrical data includes LED forward voltage, forward current, reverse voltage, reverse current and LED electrical signal waveform; wherein the electrical signal waveform includes switch response time and noise dynamic characteristics; Constructing a data set for the electrical data to obtain electrical sample data to be processed; Performing normalization and outlier removal operations on the electrical sample data to be processed to obtain an electrical data set; balancing the data using a synthetic minority oversampling technique based on the electrical data set to obtain a balanced data set; Processing the data in the balanced data set by a recursive feature elimination method to obtain different electrical feature subsets; Constructing a feature vector group for different subsets of the electrical features to obtain an electrical feature vector group; The electrical characteristic vector group is calculated using the performance evaluation formula to obtain the electrical performance value: Among them, D i Indicates the electrical performance value of the tested LED, V a Represents the forward voltage, I a is the forward current, R is the standard resistance, V b Reverse voltage, I b represents the reverse current, μ represents the noise factor, μ max represents the maximum noise factor, T c Indicates the switch response time, T c,max represents the maximum switch response time, ω1, ω2, ω3 and ω4 represent the corresponding weight coefficients respectively; The electrical performance values ​​are screened based on preset values ​​to obtain electrical characteristic information that meets preset evaluation conditions.

5. The method according to claim 1, characterized in that The thermal shock test and the vibration shock test are performed on the LED package to be tested to obtain corresponding thermal characteristic information and mechanical characteristic information, including: Placing the LED package to be inspected in a high temperature environment to perform a thermal shock test and a vibration shock test, and obtaining two-dimensional image information of the change of the area of ​​micro defects of the LED package to be inspected with thickness after the thermal shock test and the vibration shock test; Perform calculation based on the two-dimensional image information to obtain thermal shock parameters and vibration shock parameters; Process the missing values ​​of thermal shock parameters and vibration shock parameters to obtain the corresponding processed characteristic parameters; Using a feature selection algorithm to encode each of the feature parameters to obtain a feature subset corresponding to each of the feature parameters; The quality of each feature subset is evaluated according to the fitness function, and the optimal feature subsets corresponding to the thermal shock parameters and the vibration shock parameters are obtained as thermal feature information and mechanical feature information respectively.

6. The method according to claim 5, characterized in that The step of calculating based on the two-dimensional image information to obtain thermal shock parameters and vibration shock parameters includes: The thermal shock parameter and vibration shock parameter are calculated using the following formula: v=k*(T c -t) 2 Among them, E c Indicates thermal shock parameter E1 or vibration shock parameter E2, T c It indicates that the thickness T1 or T2 of the package surface after the thermal shock or vibration shock test is obtained by ultrasonic ranging, dt indicates the preset detection thickness, high-frequency ultrasonic waves are sent and collected in a preset frequency range every preset detection thickness dt, and the small defect area S of the ultrasonic image at the thickness is obtained, and the data points (t, S) formed by the number of small defects corresponding to each thickness are recorded, and a two-dimensional image of the small defect area S changing with the thickness t is generated, and the corresponding function relationship S=f is obtained by image fitting. c (t), v represents the expansion speed of tiny defects at each preset thickness, and k represents the preset influencing factor.

7. The method according to claim 1, characterized in that The image feature information, optical feature information, thermal feature information and mechanical feature information are input into a trained package inspection model to obtain a quality inspection report of LED package, including: Based on historical training data, a machine learning classifier is used to perform model training to obtain a detection analyzer; the machine learning classifier is at least one of a support vector machine, a decision tree, a neural network, and a random forest; Using the stochastic gradient descent method to train each of the detection analyzers to obtain a pre-trained packaging detection model; The AUC value of each pre-trained packaging detection model is calculated using a formula to obtain a trained packaging detection model; the trained packaging detection model is a packaging detection model whose performance meets preset conditions; Inputting the image feature information, optical feature information, thermal feature information and mechanical feature information into the trained packaging inspection model to obtain a packaging inspection result of the LED package; The image feature information, optical feature information, thermal feature information and mechanical feature information are calculated using the following formula to obtain a comprehensive score of the packaging test result: Among them, S represents the comprehensive score of the packaging test result, n represents the number of feature information that affects the packaging test result, and w i represents the weight of the i-th feature information, S i represents the i-th feature information value, f i (S i ) represents the score conversion function for the i-th feature information, which converts the feature evaluation value into a score in the interval [0,1]; The quality grade of the comprehensive score is determined according to a preset threshold value to obtain a quality inspection report for determining the quality of LED packaging inspection.

8. An LED packaging detection system, characterized in that: The system comprises: An information acquisition module is used to acquire a light source image of the LED light source to be detected; it is also used to extract and analyze the features of the light source image to obtain image feature information; it is also used to perform electrical tests on the LED to be detected to obtain electrical feature information corresponding to electrical performance analysis; it is also used to perform thermal shock tests and vibration shock tests on the LED package to be detected to obtain corresponding thermal feature information and mechanical feature information; A report generation module is used to input the image feature information, optical feature information, thermal feature information and mechanical feature information into a trained packaging inspection model to obtain a quality inspection report for LED packaging; the quality inspection report determines the packaging inspection quality level according to a preset threshold; wherein the packaging inspection model is trained by combining multiple machine learning classifiers.

9. A computer device comprising a memory and a processor, wherein the memory stores a computer program, wherein: When the processor executes the computer program, the steps of the method according to any one of claims 1 to 7 are implemented.

10. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by a processor, the steps of the method according to any one of claims 1 to 7 are implemented.

Images