A method and system for detecting the microstructure of aluminum foam using deep learning

Abstract

The present invention discloses a method and system for detecting the microstructure of aluminum foam using deep learning, relating to the field of microstructure detection technology. The method comprises: acquiring an image of an aluminum foam material to be detected, wherein the aluminum foam material to be detected is a raw material for preparing a diffuser cover for an LED lamp; using a convolutional neural network to identify microscopic pore features in the aluminum foam material image and output microscopic pore data; collecting lighting property parameters of the LED lamp and design structural parameters of the diffuser cover for the LED lamp; and using deep learning to predict the astigmatism performance of the aluminum foam material to be detected based on the microscopic pore data, lighting property parameters, and design structural parameters, outputting a predicted astigmatism uniformity coefficient as the astigmatism performance detection result. The present invention solves the technical problem of poor accuracy in detecting the astigmatism performance of aluminum foam in the prior art.

Description

Technical Field

[0001] The present invention relates to the technical field of microstructure detection, and in particular to a foam aluminum microstructure detection method and system using deep learning. Background Art

[0002] The porous structure of aluminum foam can be used to diffuse light, particularly in lighting systems and some optical devices. By passing light through the pores of aluminum foam, the light can be dispersed in different directions, evenly illuminating the target area, thereby avoiding strong direct light. Due to its excellent light diffusion properties, aluminum foam can help LED lamps achieve uniform lighting without creating glaring hot spots. Aluminum foam diffusers are particularly widely used in environments with high lighting quality requirements, such as museums and offices.

[0003] The quality of aluminum foam directly affects the light scattering path and uniformity, thus determining the diffuser's performance. Existing technology cannot accurately measure the quality of aluminum foam under actual working conditions, leading to unstable diffuser performance. This technical issue urgently needs to be addressed. Summary of the Invention

[0004] The present application provides a method and system for detecting the microstructure of foamed aluminum using deep learning, which is used to solve the technical problem of insufficient accuracy and practicality in detecting the astigmatism performance of foamed aluminum in the existing technology.

[0005] In view of the above problems, the present application provides a method and system for detecting the microstructure of foam aluminum using deep learning.

[0006] In a first aspect, the present application provides a method for detecting the microstructure of aluminum foam using deep learning, the method comprising:

[0007] An image of the foamed aluminum material to be inspected is collected to obtain an image of the foamed aluminum material, wherein the foamed aluminum material to be inspected is a raw material for preparing a light diffuser cover of an LED lamp.

[0008] A convolutional neural network is used to identify microscopic pore features of the foam aluminum material image and output microscopic pore data.

[0009] Collect the lighting property parameters of LED lamps and the design structure parameters of LED lamp diffusers.

[0010] Based on the microscopic pore data, lighting property parameters and design structure parameters, deep learning is used to predict the astigmatism performance of the foam aluminum material to be tested, and the predicted astigmatism uniformity coefficient is output as the astigmatism performance test result.

[0011] In a second aspect, the present application provides a foam aluminum microstructure detection system using deep learning, comprising:

[0012] The material image acquisition module is used to acquire an image of the foam aluminum material to be detected and obtain an image of the foam aluminum material, wherein the foam aluminum material to be detected is a raw material for preparing a light diffuser cover of an LED lamp.

[0013] The feature recognition module is used to use a convolutional neural network to identify the microscopic pore features of the foam aluminum material image and output microscopic pore data.

[0014] The lamp parameter collection module is used to collect the lighting property parameters of LED lamps and the design structure parameters of the LED lamp diffuser.

[0015] The astigmatism performance detection module is used to predict the astigmatism performance of the foam aluminum material to be tested based on the microscopic pore data, lighting property parameters and design structure parameters using deep learning, and output the predicted astigmatism uniformity coefficient as the astigmatism performance detection result.

[0016] One or more technical solutions provided in this application have at least the following technical effects or advantages:

[0017] This application proposes a method and system for detecting the microstructure of aluminum foam using deep learning. By combining convolutional neural networks for identification and utilizing a deep learning prediction mechanism, the accuracy and efficiency of detecting the astigmatism performance of aluminum foam materials are significantly improved. Compared with traditional methods, the technical solution provided by this application overcomes the subjectivity and labor consumption of manual detection, realizes comprehensive detection from microstructure to macroscopic optical performance, breaks through the limitation of existing technologies that only focus on a single parameter, and comprehensively quantifies the astigmatism uniformity of aluminum foam under actual working conditions. This application achieves the technical effect of quickly and reliably detecting the astigmatism performance of aluminum foam without the need for destructive testing. BRIEF DESCRIPTION OF THE DRAWINGS

[0018] In order to more clearly illustrate the technical solutions in the embodiments of the present invention, the following briefly introduces the drawings required for use in the description of the embodiments. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without creative work.

[0019] Figure 1 A schematic flow chart of a method for detecting the microstructure of aluminum foam using deep learning provided in an embodiment of the present application.

[0020] Figure 2 A schematic structural diagram of a foam aluminum microstructure detection system using deep learning provided in an embodiment of the present application.

[0021] In the accompanying drawings, the components represented by the reference numerals are described as follows:

[0022] Material image acquisition module 100, feature recognition module 200, lamp parameter acquisition module 300, and astigmatism performance detection module 400. DETAILED DESCRIPTION

[0023] This application provides a method and system for detecting the microstructure of foam aluminum using deep learning, which is used to solve the technical problem of insufficient accuracy and practicality in the detection of the astigmatism performance of foam aluminum in the existing technology.

[0024] The following will be combined with the drawings in the embodiments of this application to clearly and completely describe the technical solutions in the embodiments of this application. Obviously, the embodiments described are only part of the embodiments of this application, not all of them. Based on the embodiments in this application, all other embodiments obtained by ordinary technicians in this field without making creative efforts are within the scope of protection of this application.

[0025] It should be noted that the terms "including" and "having" are intended to cover non-exclusive inclusions. For example, a process, method, system, product or server that includes a series of steps or units is not necessarily limited to those steps or units clearly listed, but may include other steps or modules that are not clearly listed or are inherent to these processes, methods, products or devices.

[0026] Example 1, as Figure 1 As shown, the present application provides a method for detecting the microstructure of aluminum foam using deep learning, wherein the method includes:

[0027] S10: Capturing an image of the aluminum foam material to be inspected to obtain an image of the aluminum foam material, wherein the aluminum foam material to be inspected is a raw material for preparing a light diffuser cover for an LED lamp.

[0028] In the production and use of LED lamps, the characteristics of the LED lamp diffuser have a significant impact on the performance of the LED lamp. However, when studying the performance of the lamp diffuser, the performance testing of the foam aluminum material used to make the LED lamp diffuser is imprecise and often requires destructive testing.

[0029] In the embodiment of the present application, micro-CT is used to capture images of the aluminum foam to be inspected, thereby obtaining images of the aluminum foam material, wherein the aluminum foam to be inspected is the material for preparing the diffuser cover of an LED lamp.

[0030] In order to finely quantify the microscopic characteristics of foam aluminum, high-precision 3D microscopic images of foam aluminum, the material used to make LED lamp diffusers, are collected. This provides high-quality original data support for subsequent identification. At the same time, due to the use of non-destructive testing, there is no need to destroy the material, which greatly saves costs.

[0031] S20: Using a convolutional neural network to identify microscopic pore features of the foam aluminum material image, and outputting microscopic pore data.

[0032] Existing aluminum foam inspection methods mostly rely on manual microscopic observation, which has limitations in quantifying nonlinear characteristics such as pore size distribution and pore wall roughness. Furthermore, manual identification is inefficient and subject to significant subjective errors. Traditional processing methods are unable to effectively capture characteristics such as the pore morphology coefficient, resulting in inaccurate identification results.

[0033] Step S20 in the method provided in the embodiment of the present application includes:

[0034] Pore ​​characteristic indicators for identifying the foam aluminum material are configured, wherein the pore characteristic indicators for identifying the foam aluminum material include pore size distribution, pore wall roughness, pore wall thickness, pore morphology coefficient, porosity, pore density and pore distribution uniformity coefficient.

[0035] Based on historical detection data of similar foam aluminum materials and with the pore feature identification index as a constraint, a sample material image set and a sample pore feature set are collected.

[0036] The sample material image set and the sample pore feature set are used to train a convolutional neural network until convergence to obtain a pore feature identifier.

[0037] The sample material image set and the sample pore feature set are used to train a convolutional neural network until convergence to obtain a pore feature identifier, including:

[0038] The sample material image set and the sample pore feature set are used as training data, divided into K parts, and K parts are selected with replacement from the K parts of the data set to construct a first sample data set. The K parts are selected iteratively K times to obtain K parts of the sample data set, where K is an integer greater than or equal to 5.

[0039] Taking the sample material image as input and the sample pore characteristics as supervision, the K sample data sets are used to perform supervised training and verification on the convolutional neural network respectively until the preset convergence conditions are met, and K pore feature recognition branches are obtained, which are combined to obtain a pore feature identifier.

[0040] The pore feature identifier is used to perform microscopic pore feature identification on the foam aluminum material image and output microscopic pore data.

[0041] The pore feature identifier is used to perform microscopic pore feature recognition on the foam aluminum material image and output microscopic pore data, including:

[0042] A real-time image quality coefficient of the foam aluminum material image is obtained based on image acquisition accuracy and real-time interference factor intensity evaluation.

[0043] The ratio of the average historical image quality coefficient of the same type of foam aluminum material to the real-time image quality coefficient is set as the branch selection adjustment coefficient, which is multiplied by the initial number of selected branches and rounded to obtain the adaptive selection branch number Q, wherein the initial number of selected branches is 2, and Q is greater than or equal to 1 and less than or equal to K.

[0044] Among the K pore feature recognition branches of the pore feature identifier, Q pore feature recognition branches are randomly selected to perform microscopic pore feature recognition on the foam aluminum material image, and output microscopic pore data after feature mean fitting.

[0045] In the embodiment of the present application, the identification pore characteristic index of the foam aluminum material is configured, wherein the identification pore characteristic index includes pore size distribution, which characterizes the density of pore size; pore wall roughness, in micrometers, which characterizes the flatness of pore wall; pore wall thickness, in micrometers; pore morphology coefficient, which characterizes the regularity of pores. Regular pores are conducive to uniform light dispersion; porosity, in %, which characterizes the area ratio of pores in the whole material; pore density, in %, which characterizes the pore density; pore distribution uniformity coefficient, which characterizes the degree of uniformity of pore distribution. By configuring multiple parameters, the pore characteristics of the foam aluminum material can be comprehensively characterized, which is conducive to a comprehensive understanding of the performance of the foam aluminum.

[0046] Based on the historical detection data of similar foam aluminum materials, with the identification of pore characteristic indicators as a constraint, a sample material image set and a sample pore feature set were collected. The sample pore features in the sample pore feature set and the sample material image are in a one-to-one correspondence, and a set of sample pore features corresponds to one sample material image.

[0047] A sample material image set and a sample pore feature set are used as training data. The sample material image set and the sample pore feature set are divided into K equal parts to obtain K datasets. A cyclic sampling is performed within the K datasets. Specifically, one dataset is selected each time with replacement K times, and the K selected datasets are integrated to construct a first sample dataset. The sampling is iterated K times to obtain K sample datasets. Where K is an integer greater than or equal to 5. Multiple iterative sampling with replacement is performed to form multiple sample datasets, which facilitates model training.

[0048] A convolutional neural network is used to construct a pore feature recognition branch. For example, a four-layer convolutional neural network is employed, wherein the input layer is used to receive the material image, the convolution layer uses 64 3×3 convolution kernels, the pooling layer uses 2×2 maximum pooling, and the output layer uses multiple nodes to output various identified pore features such as pore size distribution, pore wall roughness, pore wall thickness, pore morphology coefficient, porosity, pore density, and pore distribution uniformity coefficient. The preset convergence condition is set to an accuracy rate of more than 85% for the various pore features identified. Using sample material images as input and sample pore features as supervision, the constructed convolutional neural network is supervised and validated using K sample data sets until the preset convergence condition is met, indicating that the convolutional neural network training is complete. Each pore recognition branch is trained using the same training method until convergence, resulting in K pore feature recognition branches. These K pore feature recognition branches are combined to form a pore feature identifier.

[0049] The real-time image quality coefficient of the aluminum foam material image is obtained based on image acquisition accuracy and real-time interference factor intensity assessment. For example, the standard image acquisition area is set to 512×512. When the actual acquired image area is smaller than the standard image acquisition area, the image acquisition accuracy is calculated as follows: actual acquired image area divided by standard image acquisition area. When the actual acquired image size is equal to or greater than the standard image acquisition area, the image acquisition accuracy is 1. For example, when the actual acquired image size is 256×256, the image acquisition accuracy is calculated as (256×256) divided by (512×512) = 0.25. A higher image acquisition accuracy indicates better image quality. Interference factor intensity can be assessed using the area of ​​the stain in the acquired image. Interference factor intensity is calculated as: stain area divided by actual acquired image area. For example, if the stain area is 655 pixels, and the actual acquired image area is 65536 pixels, the interference factor intensity is calculated as 655 divided by 65536 = 0.01. A higher interference factor intensity indicates worse image quality.

[0050] Real-time image quality coefficient = [image acquisition accuracy + (1 - interference factor intensity)] ÷ 2. For example, if the image acquisition accuracy is 0.25 and the interference factor intensity is 0.01, the real-time image quality coefficient = [0.25 + (1 - 0.01)] ÷ 2 = 0.62.

[0051] The ratio of the mean historical image quality coefficient to the real-time image quality coefficient for similar aluminum foam materials is used as the branch selection adjustment coefficient. The branch selection adjustment coefficient is calculated as follows: real-time image quality coefficient ÷ mean historical image quality coefficient. For example, if the real-time image quality coefficient is 0.62 and the mean historical image quality coefficient is 0.7, the branch selection adjustment coefficient is 0.62 ÷ 0.7 = 0.89. This is multiplied by the number of initially selected branches and rounded to the nearest integer to obtain the number of adaptively selected branches, Q. The initial number of selected branches is 2, and Q is greater than or equal to 1 and less than or equal to K. For example, if the branch selection adjustment coefficient is 0.89, Q = 0.89 × 2 = 1.78, which is rounded to 2. If the calculated result is less than or equal to 2, Q = 2; if the calculated result is greater than or equal to K, Q = K.

[0052] Within the K pore feature recognition branches of the pore feature identifier, Q pore feature recognition branches are randomly selected to perform microscopic pore feature recognition on the aluminum foam material image. After feature mean fitting, the microscopic pore data is output. Quantifiable data such as pore wall roughness, pore wall thickness, pore morphology coefficient, porosity, pore density, and pore distribution uniformity coefficient are output as arithmetic mean values, while less quantifiable data such as pore size distribution are output as fitted average values.

[0053] A convolutional neural network automatically identifies microscopic pore features, overcoming the inaccuracy of manual identification. Leveraging the nonlinear modeling capabilities of deep learning, complex metrics such as pore distribution are precisely quantified. The resulting microscopic pore data is comprehensive and objective, providing a highly reliable foundation for predicting astigmatism performance. By selecting the appropriate number of branches based on real-time image quality, the system ensures accurate pore feature recognition while avoiding unnecessary computational resources and improving recognition efficiency.

[0054] S30: Collecting lighting property parameters of the LED lamp and design structure parameters of the LED lamp diffuser.

[0055] Current performance evaluations of aluminum foams consider only the structural parameters of the material itself, ignoring the coupled effects of lighting properties and diffuser design on optical performance. This single-dimensional analysis can lead to significant deviations from actual luminaire operating conditions, failing to reflect the material's dynamic scattering behavior in real-world conditions and resulting in inaccurate evaluations of diffuse light uniformity.

[0056] Collect the lighting attribute parameters of the LED lamp and the design structure parameters of the LED lamp diffuser, wherein the lighting attribute parameters include light source brightness, beam angle and spectral distribution, and the design structure parameters include diffuser shape, diffuser size and diffuser thickness.

[0057] In the embodiment of the present application, the production parameter design of the LED lamp is checked, the lighting property parameters of the LED lamp and the design structure parameters of the LED lamp diffuser are collected.

[0058] Among them, the lighting attribute parameters include light source brightness, which is measured in lumens; beam angle, which is measured in degrees; and spectral distribution, which is represented by color temperature, such as 4500K natural white light.

[0059] The design structural parameters include the shape of the diffuser, such as round or square; the size of the diffuser, such as the diameter of a round diffuser and the side length of a square diffuser, in centimeters; and the thickness of the diffuser, in millimeters.

[0060] The lighting attribute parameters and diffuser design structure parameters are collected and incorporated into the analysis framework, providing diverse data input for performance analysis under real working conditions.

[0061] S40: Based on the microscopic pore data, lighting property parameters and design structure parameters, deep learning is used to predict the astigmatism performance of the foam aluminum material to be tested, and a predicted astigmatism uniformity coefficient is output as the astigmatism performance test result.

[0062] Existing technologies cannot integrate the microstructure of aluminum foam and the macroscopic design parameters of the diffuser for analysis. Instead, they only correlate single indicators such as porosity with the diffuser performance through static experiments. This results in the test results being unable to respond to dynamic factors such as light source changes and structural changes.

[0063] Step S40 in the method provided in the embodiment of the present application includes:

[0064] Based on the historical lighting monitoring data of LED lamps, a sample micropore data set and a sample design structure parameter set of the LED lamp diffuser cover, as well as a sample lighting attribute parameter set of the LED lamp are collected. In addition, a sample astigmatism uniformity coefficient of the LED lamp is collected to obtain a sample astigmatism uniformity coefficient set.

[0065] The sample light uniformity coefficient of LED lamps is collected, including:

[0066] Acquire multiple regional brightness images of the lighting area of ​​the LED lamp at multiple historical monitoring time points, and perform image grayscale processing to obtain multiple regional grayscale images.

[0067] Grayscale value standard deviations are calculated for the multiple regional grayscale images respectively to obtain multiple image grayscale standard deviations, and then an average of the image grayscale standard deviations is calculated.

[0068] The ratio of the standard deviation of the grayscale standard deviations of the multiple images to the mean of the grayscale standard deviations of the images is set as the illumination fluctuation coefficient, and the illumination uniformity stability coefficient is obtained by subtracting the illumination fluctuation coefficient from 1.

[0069] The sample astigmatism uniformity coefficient is obtained by evaluating the mean value of the image grayscale standard deviation and the illumination uniformity stability coefficient, wherein the sample astigmatism uniformity coefficient is negatively correlated with the mean value of the image grayscale standard deviation and positively correlated with the illumination uniformity stability coefficient.

[0070] The sample micropore dataset, sample design structure parameter set and sample illumination property parameter set are used as input, and the sample astigmatism uniformity coefficient set is used as output. The deep learning model is trained until convergence to obtain an astigmatism performance predictor.

[0071] The microscopic pore data, lighting property parameters and design structure parameters are input into the astigmatism performance predictor to predict and obtain the predicted astigmatism uniformity coefficient of the foam aluminum material to be tested.

[0072] In the embodiment of the present application, based on the historical lighting monitoring data of LED lamps, a sample microscopic pore data set and a sample design structure parameter set of the LED lamp diffuser cover are collected according to the method provided in the previous steps of the embodiment of the present application.

[0073] Collect a sample set of lighting attribute parameters for the LED lamps. Obtain multiple regional brightness images of the LED lamp's lighting area at multiple historical monitoring time points. For example, set the historical monitoring time point to acquire a brightness image of the lighting area every three hours. Use the average method, which directly takes the average of the image's R, B, and G components as the image grayscale: grayscale = (R + B + G) ÷ 3. Based on the grayscale, perform grayscale processing on the image to obtain multiple regional grayscale images.

[0074] Grayscale standard deviations of multiple regional grayscale images are calculated respectively to obtain multiple image grayscale standard deviations, and the mean of the image grayscale standard deviations is calculated.

[0075] The ratio of the standard deviation of the grayscale standard deviations of multiple images to the mean of their grayscale standard deviations is defined as the illumination fluctuation coefficient. The illumination fluctuation coefficient = standard deviation of the image grayscale standard deviations / mean of the image grayscale standard deviations. For example, if the standard deviation of the grayscale standard deviations of multiple images is 1.5 and the mean of the image grayscale standard deviations is 12, then the illumination fluctuation coefficient = 1.5 ÷ 12 = 0.125. A smaller illumination fluctuation coefficient indicates closer brightness distributions between different images. Subtract the illumination fluctuation coefficient from 1 to obtain the illumination uniformity and stability coefficient. For an illumination fluctuation coefficient of 0.125, the illumination uniformity and stability coefficient = 1 - illumination fluctuation coefficient = 1 - 0.125 = 0.875. A larger illumination uniformity and stability coefficient indicates more uniform and stable lighting.

[0076] The sample astigmatism uniformity coefficient is calculated based on the mean image grayscale standard deviation and the illumination uniformity stability coefficient. The sample astigmatism uniformity coefficient is negatively correlated with the mean image grayscale standard deviation and positively correlated with the illumination uniformity stability coefficient. For example, the sample astigmatism uniformity coefficient = (1 ÷ mean sample image grayscale standard deviation) × grayscale weight + sample illumination uniformity stability coefficient × illumination weight, where grayscale weight + illumination weight = 1. For an image grayscale standard deviation mean of 12, a grayscale weight of 0.4, an illumination uniformity stability coefficient of 0.875, and an illumination weight of 0.6, the sample astigmatism uniformity coefficient = (1 ÷ 12) × 0.4 + 0.875 × 0.6 = 0.558. A larger sample astigmatism uniformity coefficient indicates better astigmatism uniformity, which is more beneficial for the daily use of aluminum foam diffusers.

[0077] A deep learning-based predictor for astigmatism performance was constructed. The input layer employed 13 neurons to receive seven types of micropore data, three design structure parameters, and three lighting property parameters. The first hidden layer employed 64 neurons activated using the ReLU function, while the second hidden layer employed 32 neurons activated using the ReLU function. The output layer employed one neuron, linearly activated to output the astigmatism uniformity coefficient. The loss function employed the mean squared error (MSE). The deep learning model was trained under supervision until convergence, using a sample micropore dataset, a sample design structure parameter set, and a sample lighting property parameter set as input and a sample astigmatism uniformity coefficient set as output. The model was considered trained successfully if the error in the output astigmatism uniformity coefficient for the input micropore data, design structure parameters, and lighting property parameters was within ±0.05, resulting in a predictor for astigmatism performance.

[0078] The microscopic pore data, lighting property parameters and design structure parameters are input into the astigmatism performance predictor to obtain the predicted astigmatism uniformity coefficient of the foam aluminum material to be tested.

[0079] A deep learning-based model of microscopic pore data, lighting properties, and structural parameters is constructed to dynamically predict the astigmatism uniformity coefficient. By exploiting the nonlinear interactions between complex parameters through deep learning, the effectiveness of predictions in real-world conditions is significantly improved, ultimately yielding intuitive and quantifiable astigmatism performance indicators.

[0080] Example 2, as Figure 2 As shown, based on the same inventive concept as the method for detecting the microstructure of aluminum foam using deep learning provided in Example 1, an embodiment of the present invention further provides a system for detecting the microstructure of aluminum foam using deep learning, comprising:

[0081] The material image acquisition module 100 is used to acquire an image of the foam aluminum material to be inspected, and obtain an image of the foam aluminum material, wherein the foam aluminum material to be inspected is a raw material for preparing a light diffuser cover of an LED lamp.

[0082] The feature recognition module 200 is used to use a convolutional neural network to identify the microscopic pore features of the foam aluminum material image and output microscopic pore data.

[0083] The lamp parameter collection module 300 is used to collect the lighting property parameters of the LED lamp and the design structure parameters of the LED lamp diffuser.

[0084] The astigmatism performance detection module 400 is used to predict the astigmatism performance of the foam aluminum material to be tested based on the microscopic pore data, lighting property parameters and design structure parameters using deep learning, and output the predicted astigmatism uniformity coefficient as the astigmatism performance detection result.

[0085] In one embodiment, the feature recognition module 200 is further configured to:

[0086] Pore ​​characteristic indicators for identifying the foam aluminum material are configured, wherein the pore characteristic indicators for identifying the foam aluminum material include pore size distribution, pore wall roughness, pore wall thickness, pore morphology coefficient, porosity, pore density and pore distribution uniformity coefficient.

[0087] Based on historical detection data of similar foam aluminum materials and with the pore feature identification index as a constraint, a sample material image set and a sample pore feature set are collected.

[0088] The sample material image set and the sample pore feature set are used to train a convolutional neural network until convergence to obtain a pore feature identifier.

[0089] The sample material image set and the sample pore feature set are used to train a convolutional neural network until convergence to obtain a pore feature identifier, including:

[0090] The sample material image set and the sample pore feature set are used as training data, divided into K parts, and K parts are selected with replacement from the K parts of the data set to construct a first sample data set. The K parts are selected iteratively K times to obtain K parts of the sample data set, where K is an integer greater than or equal to 5.

[0091] Taking the sample material image as input and the sample pore characteristics as supervision, the K sample data sets are used to perform supervised training and verification on the convolutional neural network respectively until the preset convergence conditions are met, and K pore feature recognition branches are obtained, which are combined to obtain a pore feature identifier.

[0092] The pore feature identifier is used to perform microscopic pore feature identification on the foam aluminum material image and output microscopic pore data.

[0093] The pore feature identifier is used to perform microscopic pore feature recognition on the foam aluminum material image and output microscopic pore data, including:

[0094] A real-time image quality coefficient of the foam aluminum material image is obtained based on image acquisition accuracy and real-time interference factor intensity evaluation.

[0095] The ratio of the average historical image quality coefficient of the same type of foam aluminum material to the real-time image quality coefficient is set as the branch selection adjustment coefficient, which is multiplied by the initial number of selected branches and rounded to obtain the adaptive selection branch number Q, wherein the initial number of selected branches is 2, and Q is greater than or equal to 1 and less than or equal to K.

[0096] Among the K pore feature recognition branches of the pore feature identifier, Q pore feature recognition branches are randomly selected to perform microscopic pore feature recognition on the foam aluminum material image, and output microscopic pore data after feature mean fitting.

[0097] In one embodiment, the lamp parameter collection module 300 is further configured to:

[0098] Collect the lighting attribute parameters of the LED lamp and the design structure parameters of the LED lamp diffuser, wherein the lighting attribute parameters include light source brightness, beam angle and spectral distribution, and the design structure parameters include diffuser shape, diffuser size and diffuser thickness.

[0099] In one embodiment, the astigmatism performance detection module 400 is further configured to:

[0100] Based on the historical lighting monitoring data of LED lamps, a sample micropore data set and a sample design structure parameter set of the LED lamp diffuser cover, as well as a sample lighting attribute parameter set of the LED lamp are collected. In addition, a sample astigmatism uniformity coefficient of the LED lamp is collected to obtain a sample astigmatism uniformity coefficient set.

[0101] The sample light uniformity coefficient of LED lamps is collected, including:

[0102] Acquire multiple regional brightness images of the lighting area of ​​the LED lamp at multiple historical monitoring time points, and perform image grayscale processing to obtain multiple regional grayscale images.

[0103] Grayscale value standard deviations are calculated for the multiple regional grayscale images respectively to obtain multiple image grayscale standard deviations, and then an average of the image grayscale standard deviations is calculated.

[0104] The ratio of the standard deviation of the grayscale standard deviations of the multiple images to the mean of the grayscale standard deviations of the images is set as the illumination fluctuation coefficient, and the illumination uniformity stability coefficient is obtained by subtracting the illumination fluctuation coefficient from 1.

[0105] The sample astigmatism uniformity coefficient is obtained by evaluating the mean value of the image grayscale standard deviation and the illumination uniformity stability coefficient, wherein the sample astigmatism uniformity coefficient is negatively correlated with the mean value of the image grayscale standard deviation and positively correlated with the illumination uniformity stability coefficient.

[0106] The sample micropore dataset, sample design structure parameter set and sample illumination property parameter set are used as input, and the sample astigmatism uniformity coefficient set is used as output. The deep learning model is trained until convergence to obtain an astigmatism performance predictor.

[0107] The microscopic pore data, lighting property parameters and design structure parameters are input into the astigmatism performance predictor to predict and obtain the predicted astigmatism uniformity coefficient of the foam aluminum material to be tested.

[0108] In summary, the embodiments of the present application have at least the following technical effects:

[0109] This application proposes a method and system for detecting the microstructure of aluminum foam using deep learning. By combining a convolutional neural network to accurately identify microscopic pore features, analyze lighting properties and diffuser structural parameters, and dynamically optimized deep learning prediction mechanisms, the accuracy and efficiency of microstructure detection and diffuser performance evaluation of aluminum foam materials are significantly improved. By introducing an adaptive branch selection mechanism based on real-time image quality, the impact of acquisition interference on feature recognition is suppressed, effectively improving model computational efficiency, reducing computing power consumption, and enhancing model robustness, making detection results more accurate and reliable.

[0110] Compared to traditional methods, the technical solution provided by this application overcomes the subjectivity and labor consumption of manual testing, enabling comprehensive testing from microstructure to macroscopic optical properties. This overcomes the limitations of existing technologies that focus solely on a single parameter and comprehensively quantifies the astigmatism uniformity of aluminum foam under actual operating conditions. This application achieves the technical effect of quickly and reliably testing the astigmatism properties of aluminum foam without the need for destructive testing.

[0111] It should be noted that the order in which the embodiments of the present application are presented is for illustrative purposes only and does not necessarily represent the superiority or inferiority of the embodiments. Furthermore, the foregoing descriptions of specific embodiments of this specification are provided. Furthermore, the processes depicted in the accompanying drawings do not necessarily require the specific order or sequential sequence shown to achieve the desired results. In certain embodiments, multitasking and parallel processing are also possible or may be advantageous.

[0112] The above description is only a preferred embodiment of the present application and is not intended to limit the present application. Any modifications, equivalent replacements, improvements, etc. made within the spirit and principles of the present application should be included in the scope of protection of the present application.

[0113] This specification and drawings are merely illustrative of the present application and are intended to cover any and all modifications, variations, combinations, or equivalents within the scope of this application. Obviously, those skilled in the art may make various modifications and variations to this application without departing from the scope of this application. Thus, this application is intended to include such modifications and variations as fall within the scope of this application and its equivalents.

Claims

1. A method for detecting the microstructure of aluminum foam using deep learning, characterized in that: Methods include: Capturing an image of the aluminum foam material to be inspected, and obtaining an image of the aluminum foam material, wherein the aluminum foam material to be inspected is a raw material for preparing a light diffuser cover for an LED lamp; Using a convolutional neural network to identify microscopic pore features of the aluminum foam material image and output microscopic pore data; Collect the lighting property parameters of LED lamps and the design structure parameters of LED lamp diffusers; Based on the microscopic pore data, lighting property parameters, and design structure parameters, deep learning is used to predict the astigmatism performance of the foam aluminum material to be tested, and a predicted astigmatism uniformity coefficient is output as the astigmatism performance test result; The method of using a convolutional neural network to identify the microscopic pore features of the foam aluminum material image and outputting microscopic pore data includes: Configuring identification pore characteristic indicators of the foam aluminum material, wherein the identification pore characteristic indicators include pore size distribution, pore wall roughness, pore wall thickness, pore morphology coefficient, porosity, pore density and pore distribution uniformity coefficient; Based on historical inspection data of similar foam aluminum materials, and with the pore feature identification index as a constraint, a sample material image set and a sample pore feature set are collected; Using the sample material image set and the sample pore feature set, training a convolutional neural network until convergence to obtain a pore feature identifier; Using the pore feature identifier, performing microscopic pore feature identification on the foam aluminum material image and outputting microscopic pore data; According to the microscopic pore data, lighting property parameters and design structure parameters, deep learning is used to predict the astigmatism performance of the foam aluminum material to be tested, and the predicted astigmatism uniformity coefficient is output, including: Based on historical lighting monitoring data of LED lamps, sample micropore data sets and sample design structure parameter sets of LED lamp diffusers, as well as sample lighting attribute parameter sets of LED lamps, are collected. Sample astigmatism uniformity coefficients of LED lamps are also collected to obtain a sample astigmatism uniformity coefficient set. Using the sample micropore dataset, the sample design structure parameter set, and the sample illumination property parameter set as input, and using the sample astigmatism uniformity coefficient set as output, training a deep learning model until convergence to obtain an astigmatism performance predictor; The microscopic pore data, lighting property parameters and design structure parameters are input into the astigmatism performance predictor to predict and obtain the predicted astigmatism uniformity coefficient of the foam aluminum material to be tested.

2. The method for detecting the microstructure of aluminum foam using deep learning according to claim 1, characterized in that: The sample material image set and the sample pore feature set are used to train a convolutional neural network until convergence to obtain a pore feature identifier, including: Using the sample material image set and the sample pore feature set as training data, dividing them into K equal parts, selecting K times with replacement from the K data sets to construct a first sample data set, and iterating the selection K times to obtain K sample data sets, where K is an integer greater than or equal to 5; Taking the sample material image as input and the sample pore characteristics as supervision, the K sample data sets are used to perform supervised training and verification on the convolutional neural network respectively until the preset convergence conditions are met, and K pore feature recognition branches are obtained, which are combined to obtain a pore feature identifier.

3. The method for detecting the microstructure of aluminum foam using deep learning according to claim 2, characterized in that: Using the pore feature identifier, performing microscopic pore feature identification on the foam aluminum material image and outputting microscopic pore data, including: Obtaining a real-time image quality coefficient of the foam aluminum material image based on image acquisition accuracy and real-time interference factor intensity assessment; The ratio of the average historical image quality coefficient of the same type of foam aluminum material to the real-time image quality coefficient is set as the branch selection adjustment coefficient, which is multiplied by the number of initially selected branches and rounded to obtain the number of adaptive selection branches Q, wherein the number of initially selected branches is 2, and Q is greater than or equal to 1 and less than or equal to K; Among the K pore feature recognition branches of the pore feature identifier, Q pore feature recognition branches are randomly selected to perform microscopic pore feature recognition on the foam aluminum material image, and output microscopic pore data after feature mean fitting.

4. The method for detecting the microstructure of aluminum foam using deep learning according to claim 1, characterized in that: Collect the lighting attribute parameters of the LED lamp and the design structure parameters of the LED lamp diffuser, wherein the lighting attribute parameters include light source brightness, beam angle and spectral distribution, and the design structure parameters include diffuser shape, diffuser size and diffuser thickness.

5. The method for detecting the microstructure of aluminum foam using deep learning according to claim 1, characterized in that: Collect sample light uniformity coefficient of LED lamps, including: Acquire multiple regional brightness images of the lighting area of ​​the LED lamp at multiple historical monitoring time points, and perform image grayscale processing to obtain multiple regional grayscale images; Calculating grayscale value standard deviations of the plurality of regional grayscale images respectively to obtain a plurality of image grayscale standard deviations, and calculating a mean of the image grayscale standard deviations; The ratio of the standard deviation of the grayscale standard deviations of the plurality of images to the mean of the grayscale standard deviations of the images is set as an illumination fluctuation coefficient, and the illumination fluctuation coefficient is subtracted from 1 to obtain an illumination uniformity stability coefficient; The sample astigmatism uniformity coefficient is obtained by evaluating the mean value of the image grayscale standard deviation and the illumination uniformity stability coefficient, wherein the sample astigmatism uniformity coefficient is negatively correlated with the mean value of the image grayscale standard deviation and positively correlated with the illumination uniformity stability coefficient.

6. A foam aluminum microstructure detection system using deep learning, characterized in that: A system for implementing a method for detecting the microstructure of aluminum foam using deep learning according to any one of claims 1 to 5, comprising: A material image acquisition module is used to acquire an image of the foam aluminum material to be inspected, wherein the foam aluminum material to be inspected is a raw material for preparing a light diffuser cover of an LED lamp; A feature recognition module is used to identify microscopic pore features of the foam aluminum material image using a convolutional neural network and output microscopic pore data; The lamp parameter collection module is used to collect the lighting attribute parameters of LED lamps and the design structure parameters of the LED lamp diffuser; The astigmatism performance detection module is used to predict the astigmatism performance of the foam aluminum material to be tested based on the microscopic pore data, lighting property parameters and design structure parameters using deep learning, and output the predicted astigmatism uniformity coefficient as the astigmatism performance detection result.

Images