Chip package detection method, device, equipment and storage medium

Abstract

The present invention discloses a detection method for a chip package and related equipment, wherein the detection method for the chip package comprises: when the light-emitting chip is in a power-on state; collecting optical images of the inner surface of the peripheral wall of the bowl cup and the top surface of the bowl cup to obtain first image data and second image data respectively; when the light-emitting chip is in a power-off state; collecting optical images of the top surface of the bowl cup to obtain third image data; performing multi-level image processing on the first image data, the second image data and the third image data to obtain corresponding pre-processed images respectively; performing feature extraction on each pre-processed image to obtain corresponding feature atlases respectively; inputting each feature atlas into a preset chip detection model, and performing optical characteristic detection of the chip package according to each feature atlas through the chip detection model to obtain corresponding optical characteristic detection results. The technical solution of the present invention solves the technical problem of low accuracy and reliability of the detection method of traditional chip packages.

Description

Technical Field

[0001] The present invention relates to the technical field of chip package detection, and in particular to a chip package detection method, device, equipment and storage medium. Background Art

[0002] In the modern optoelectronic industry, chip packages include a cup and a light-emitting chip inside the cup. In the quality control of chip packages, it is crucial to ensure that the optical performance of the cup meets the design standards. The test results based on the optical performance of the cup can help manufacturers identify possible problems that may occur during the production process, such as material defects, manufacturing process errors, etc., so as to take measures to make improvements and improve product reliability and competitiveness.

[0003] Traditional detection methods are unable to eliminate various influencing conditions for detection. For example, traditional detection methods are carried out in an environment with external light. When judging the light-guiding effect of the inner surface of the bowl wall on the light of the light-emitting chip (such as light radiation, refraction, etc.), it may be affected by external light. When judging the absorption effect of the top surface of the bowl on external light, it may be affected by the light emitted by the light-emitting chip. These interference factors limit the accuracy and reliability of traditional methods. Summary of the invention

[0004] The main purpose of the present invention is to provide a chip package detection method, aiming to solve the technical problems of low accuracy and reliability of traditional methods.

[0005] To achieve the above object, the present invention provides a method for detecting a chip package, wherein the chip package includes a bowl and a light-emitting chip fixed in the bowl. The method for detecting the chip package includes:

[0006] S100, powering on the light emitting chip in an environment without visible light, so that the light emitting chip is in a powered-on state;

[0007] S200, collecting optical images of the inner surface of the peripheral wall of the bowl and the top surface of the bowl to obtain first image data and second image data respectively;

[0008] S300, powering off the light emitting chip in an environment with visible light, so that the light emitting chip is in a power-off state;

[0009] S400, collecting an optical image of the top surface of the bowl or cup to obtain third image data;

[0010] S500, performing multi-level image processing on the first image data, the second image data and the third image data to obtain corresponding pre-processed images respectively;

[0011] S600, extracting features from each of the pre-processed images to obtain corresponding feature atlases;

[0012] S700, inputting each of the feature atlases into a preset chip detection model, and performing optical property detection of the chip package according to each of the feature atlases through the chip detection model to obtain corresponding optical property detection results.

[0013] Optionally, the step S500 includes:

[0014] S501, analyzing the grayscale histograms of the first image data, the second image data, and the third image data one by one, and determining the noise level of each image data respectively;

[0015] S502, determining Gaussian blur parameters applicable to the first image data, the second image data, and the third image data respectively according to the noise level of each image data;

[0016] S503, constructing an adaptive Gaussian kernel according to the Gaussian blur parameters;

[0017] S504, applying the adaptive Gaussian kernel to perform Gaussian blur processing on the first image data, the second image data, and the third image data respectively, reducing image noise through convolution operation, and obtaining processed image data;

[0018] S505, performing quality assessment on the processed image data to determine the effect of Gaussian blur processing;

[0019] S506: If the effect of the Gaussian blur processing does not meet the preset standard, adjust the Gaussian kernel parameters, and repeat steps S503 to S505 until the preset standard is met to obtain the final processed image data.

[0020] Optionally, after step S506, the following steps are included:

[0021] S507, performing preliminary edge extraction on the final processed image data through a high-pass filter to obtain a preliminary edge intensity map;

[0022] S508, performing non-maximum suppression processing on the preliminary edge intensity map to obtain a refined edge intensity map;

[0023] S509, based on the refined edge strength map, using a dual threshold algorithm to perform edge detection, by setting high and low thresholds to distinguish strong edges from weak edges in the refined edge strength map, and by connecting strong edges with adjacent weak edges to determine final connected edges, to obtain a connected edge map;

[0024] S510, evaluating edge continuity and integrity of the connected edge graph;

[0025] S511, according to the evaluation results of edge continuity and integrity, optimizing the edges that are displayed as broken or discontinuous, including using an edge repair algorithm to process edge breaks and fragments, and enhancing edge continuity and clarity through interpolation and morphological operations, to generate an optimized edge map;

[0026] S512: Overlay the optimized edge map with the original image to obtain final edge-enhanced image data.

[0027] Optionally, after step S512, the following steps are included:

[0028] S513, selecting a target area from the final edge-enhanced image data;

[0029] S514, applying a gray level co-occurrence matrix to analyze the texture features of the target area, extracting corresponding descriptors, and obtaining a preliminary texture feature descriptor set, wherein the descriptors include contrast, homogeneity, energy, and correlation;

[0030] S515, performing dimensionality reduction processing on the preliminary texture feature descriptor set by using principal component analysis to obtain texture features after dimensionality reduction;

[0031] S516, constructing a texture feature model using the texture features after dimensionality reduction;

[0032] S517, applying the texture feature model to analyze the texture characteristics of the image data to obtain a comprehensive texture analysis result;

[0033] S518: Superimpose the texture analysis result with the original image data to obtain final texture analysis image data.

[0034] S518. After step S518, the following steps are performed:

[0035] S519, selecting an area with significant texture features from the final texture analysis image data as a spectral scanning area;

[0036] S520, applying spectral analysis technology to perform spectral scanning on the spectral scanning area, and collecting and recording reflectivity and transmittance data at each spectral wavelength;

[0037] S521, performing Fourier transform on the collected spectral data;

[0038] S522, classifying and pattern recognizing the Fourier transform results by using a support vector machine to obtain a spectrum analysis result;

[0039] S523, comparing the spectrum analysis result with a known material spectrum database to obtain a spectrum characteristic report;

[0040] S524, integrating the spectral characteristic report with the final texture analysis image data, displaying the association between spectral characteristics and texture characteristics through image and data superposition enhancement method, and obtaining comprehensive spectral and texture analysis image data.

[0041] S524, each of the feature atlases includes a first feature atlas corresponding to the first image data, a second feature atlas corresponding to the second image data, and a third feature atlas corresponding to the third image data;

[0042] The step S700 includes:

[0043] S701, inputting the first feature atlas, the second feature atlas, and the third feature atlas into a preset chip detection model, and calculating the attention weight vectors of the first feature atlas, the second feature atlas, and the third feature atlas respectively through the attention mechanism layer of the chip detection model;

[0044] S702, performing weighted fusion on the first feature atlas, the second feature atlas, and the third feature atlas according to the weight vector through a feature fusion layer in the chip detection model to obtain a comprehensive feature vector;

[0045] S703 , calculating an optical characteristic detection result of the chip package according to the comprehensive feature vector through the classification layer in the chip detection model, and outputting the optical characteristic detection result through the output layer in the chip detection model.

[0046] A second aspect of the present invention provides a detection device for a chip package, wherein the chip package comprises a bowl and a light-emitting chip fixed in the bowl, and the detection device for the chip package comprises:

[0047] A power-on module, used for powering on the light-emitting chip in an environment without visible light, so that the light-emitting chip is in a power-on state;

[0048] A first image acquisition module is used to acquire optical images of the inner surface of the peripheral wall of the bowl and the top surface of the bowl to obtain first image data and second image data respectively;

[0049] A power-off module, used to power off the light-emitting chip in an environment with visible light, so that the light-emitting chip is in a power-off state;

[0050] A second image acquisition module is used to acquire an optical image of the top surface of the bowl or cup to obtain third image data;

[0051] An image processing module, used for performing multi-level image processing on the first image data, the second image data and the third image data to obtain corresponding pre-processed images respectively;

[0052] A feature extraction module, used to extract features from each of the preprocessed images to obtain corresponding feature atlases;

[0053] The model input module is used to input each of the feature atlases into a preset chip detection model, and perform optical property detection of the chip package according to each of the feature atlases through the chip detection model to obtain corresponding optical property detection results.

[0054] The third aspect of the present invention provides a chip package detection device, comprising: a memory and at least one processor, wherein instructions are stored in the memory, and the memory and the at least one processor are interconnected via lines; the at least one processor calls the instructions in the memory so that the chip package detection device executes the steps of the above-mentioned chip package detection method.

[0055] A fourth aspect of the present invention provides a computer-readable storage medium, wherein the computer-readable storage medium stores instructions, which, when executed on a computer, enable the computer to execute the steps of the above-mentioned chip package detection method.

[0056] Compared with the prior art, the present invention has the following beneficial effects:

[0057] The detection method of the chip package includes activating the light-emitting chip by powering on in an environment without visible light, and collecting optical images of the inner surface of the bowl wall and the top surface of the bowl, so as to accurately evaluate the light-guiding effect of the bowl on the light of the light-emitting chip, such as the radiation and refraction of light. In addition, by disconnecting the power in an environment with visible light and collecting the optical image of the top surface, this method eliminates the interference of the light of the light-emitting chip on the light absorption performance detection of the top surface of the bowl, thereby more accurately evaluating its absorption effect on external light. This targeted detection environment setting not only ensures the accuracy of optical measurement, but also makes it possible to comprehensively evaluate the optical performance of the chip package under different operating conditions.

[0058] Through the implementation of this method, manufacturers can effectively identify and improve material defects and manufacturing process errors that may occur in the production process, and then optimize the design and production process of the product, which not only improves the optical performance and overall quality of the product, but also enhances the competitiveness of the product in the market. Overall, this detection method provides an efficient and reliable solution for the optoelectronic industry to ensure that chip packages can exhibit optimal performance in different environments and meet higher standards of application requirements. BRIEF DESCRIPTION OF THE DRAWINGS

[0059] Figure 1 A schematic diagram of an embodiment of a method for detecting a chip package in an embodiment of the present invention;

[0060] Figure 2 A schematic diagram of an embodiment of a chip package detection device according to an embodiment of the present invention;

[0061] Figure 3 FIG. 1 is a schematic diagram of an embodiment of a chip package detection device according to an embodiment of the present invention. DETAILED DESCRIPTION

[0062] In order to make the purpose, technical solution and advantages of the embodiments of the present invention clearer, the technical solution of the present invention will be clearly and completely described below in conjunction with the accompanying drawings. Obviously, the described embodiments are part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0063] The terms "including" and "having" and any variations thereof mentioned in the embodiments of the present invention are intended to cover non-exclusive inclusions. For example, a process, method, system, product or device end including a series of steps or units is not limited to the listed steps or units, but may optionally include other steps or units that are not listed, or may optionally include other steps or units that are inherent to these processes, methods, products or device ends.

[0064] The technical solution of the present invention is further described below with reference to the accompanying drawings and through specific implementation methods.

[0065] Combined with Figure 1 As shown, an embodiment of the present invention provides a method for detecting a chip package, wherein the chip package includes a bowl and a light-emitting chip fixed in the bowl. The method for detecting the chip package includes:

[0066] S100, powering on the light emitting chip in an environment without visible light, so that the light emitting chip is in a powered-on state;

[0067] This step ensures that in the subsequent image acquisition, the light-guiding effect and optical properties of the inner and top surfaces of the bowls and cups can be accurately recorded without any interference from external light sources. This controlled environmental condition can be achieved by using a special darkroom in which all incident light is blocked to ensure the purity of the experimental conditions.

[0068] S200, collecting optical images of the inner surface of the peripheral wall of the bowl and the top surface of the bowl to obtain first image data and second image data respectively;

[0069] High-precision imaging equipment is used to collect optical images of the inner surface of the bowl wall and the top surface of the bowl to obtain first image data and second image data respectively. This step is completed by photographic equipment set up in a dark room, such as using high-resolution cameras, which can capture high-quality images under extremely low light conditions. The key to this step is that through the light-guiding properties of the bowl, it is possible to observe how the light-emitting chip affects the light distribution and optical response of its surrounding structures when it is powered on, so as to evaluate its performance.

[0070] It should be noted that during the product production process, it is hoped that the top surface of the bowl will not affect the light propagation of the light-emitting chip. Therefore, detecting the image data of the top surface of the bowl is a crucial step for the detection result.

[0071] S300, powering off the light emitting chip in an environment with visible light, so that the light emitting chip is in a power-off state;

[0072] Specifically, the light-emitting chip was powered off and the light absorption properties of the top of the bowl were observed in a controlled light environment to ensure the diversity and completeness of the test.

[0073] S400, collecting an optical image of the top surface of the bowl or cup to obtain third image data;

[0074] In this step, since the light-emitting chip is powered off, any changes in light will directly reflect the optical properties of the bowl material itself, such as light absorption and scattering, which is crucial for evaluating the properties of the material.

[0075] S500, performing multi-level image processing on the first image data, the second image data and the third image data to obtain corresponding pre-processed images respectively;

[0076] This step can be done through image processing techniques to improve the image quality and prepare it for more in-depth analysis. This step is necessary because it helps reduce image noise and enhance key features in the image, providing a clear basis for feature extraction.

[0077] S600, extracting features from each of the pre-processed images to obtain corresponding feature atlases;

[0078] This step includes extracting key visual and optical features from the image, such as light intensity, color distribution, texture, etc. These features are compiled into feature atlases. Through these detailed feature data, the optical performance of the chip package can be more accurately described and analyzed.

[0079] S700, inputting each of the feature atlases into a preset chip detection model, and performing optical property detection of the chip package according to each of the feature atlases through the chip detection model to obtain corresponding optical property detection results.

[0080] The model is based on machine learning algorithms to evaluate and predict the optical properties of chip packages. The results will be used to generate a detailed optical property inspection report. The success of this process depends on the accurate execution of all previous steps. The final output report provides manufacturers with important information about product performance and supports subsequent product improvements and quality control decisions.

[0081] For example, suppose the optical performance test of an LED package is being conducted. This method can be used to evaluate in detail the impact of the LED under different electrical states on the internal structure of the package and how these structures affect the propagation and distribution of light, which is crucial for designing high-efficiency and high-performance LED products.

[0082] The following is an example of an LED product being an example of a chip package.

[0083] Optionally, the step S500 includes:

[0084] S501, analyzing the grayscale histograms of the first image data, the second image data, and the third image data one by one, and determining the noise level of each image data respectively;

[0085] Since the surface quality of the bowls and cups directly affects the light output and distribution of the LEDs, ensuring the accuracy and quality of the image data is critical. In this step, each original RGB image is first converted into a grayscale image to simplify the data processing process and focus on analyzing the distribution of light intensity rather than color information. A histogram is then generated for each grayscale image, which is a statistical chart that shows the frequency distribution of pixels at different gray levels.

[0086] The analysis of grayscale histogram and noise level can be achieved by the following formula:

[0087] 1. Grayscale histogram:

[0088] .

[0089] in is the image at coordinates The gray value of and are the width and height of the image respectively.

[0090] 2. Noise variance estimation:

[0091] .

[0092] in is the average grayscale value of the image, calculated as , is the total number of pixels.

[0093] 3. Information entropy calculation:

[0094] .

[0095] in , represents the gray level Probability of occurrence.

[0096] Explanation of the meaning of parameters and operators: Indicates a specific gray level The number of times the pixel appears is used to measure the distribution of the gray level in the image; Represents the noise variance of the image, reflecting the degree of fluctuation of the grayscale value of the image. The greater the fluctuation, the higher the noise level is usually; Information entropy quantifies the information content of an image, and high entropy usually indicates that there is more noise in the image.

[0097] For example, suppose we are analyzing a bowl image of an LED chip package. Through the above calculation, if the noise variance is Higher, indicating that there is significant random noise in the image. In this case, high information entropy It may also indicate that the grayscale distribution in the image is relatively uniform, but is more affected by noise. Through these analyses, we can determine the necessary image preprocessing steps, such as adjusting the Gaussian blur parameters in subsequent steps to optimize the image quality and prepare for a more in-depth analysis of the bowl surface, which is critical for evaluating the optical performance of the LED.

[0098] S502, determining Gaussian blur parameters applicable to the first image data, the second image data, and the third image data respectively according to the noise level of each image data;

[0099] The appropriate Gaussian kernel size and standard deviation are selected according to the noise level to ensure that the random noise in the image is effectively smoothed without causing the loss of important features. A Gaussian kernel with calculated parameters is generated, which will be used for the convolution process of the image to achieve noise reduction.

[0100] For the setting of Gaussian blur parameters, the following calculation formula can be used:

[0101] 1. Adaptive Gaussian kernel standard deviation:

[0102] .

[0103] in It is a coefficient used to adjust the size of the standard deviation, which can be adjusted according to the actual denoising requirements.

[0104] 2. Gaussian kernel size:

[0105] .

[0106] 3. Gaussian function:

[0107] .

[0108] Used to generate Gaussian kernel, and Represents the position coordinates in the nucleus.

[0109] Explanation of the meaning of parameters and operators: is the standard deviation of the Gaussian kernel, which controls the degree of blur and is adjusted dynamically , which can provide the best denoising effect for different noise levels. It is the size of the Gaussian kernel, which determines the range of blur. A suitable kernel size can ensure effective noise suppression while retaining important image details. is the coordinate in the generated Gaussian kernel This function ensures that the image can smooth out noise and maintain edge clarity when convolving.

[0110] For example, considering that we are processing an image of a bowl of an LED chip package whose surface details have a great impact on the performance of the final product, assuming that the noise variance obtained from step S501 is is 0.03. Select one The value is 0.5, we calculate , and then, we determine the size of the Gaussian kernel is 5. Such parameter settings allow us to effectively reduce image noise while keeping the key optical features of the bowl and cup visible. By applying such a Gaussian kernel, we can optimize the quality of the image and provide clearer and more accurate image data for the next step of optical property analysis.

[0111] S503, constructing an adaptive Gaussian kernel according to the Gaussian blur parameters;

[0112] In this step, the previously determined and , generate a Gaussian kernel, which will be applied to the convolution of the image. Afterwards, to ensure that the image brightness remains unchanged after the kernel is applied, the Gaussian kernel needs to be normalized.

[0113] In order to accurately construct and apply the Gaussian kernel, this application uses the following formula in this step:

[0114] 1. Gaussian kernel calculation:

[0115] .

[0116] This is at every position in the nucleus Formula for calculating the value of the Gaussian function.

[0117] 2. Gaussian kernel normalization:

[0118] .

[0119] in is the kernel radius, which ensures that the total brightness of the image remains unchanged after the convolution operation.

[0120] Explanation of the meaning of parameters and operators: For location The unnormalized Gaussian kernel value at determines the smoothing weight of the point on the surrounding pixels; is the normalized Gaussian kernel value, ensuring that the kernel does not change the overall brightness of the image after being applied to it. is the standard deviation of the Gaussian distribution, which determines the degree of blur and the width of the kernel. is the sum of all values ​​in the kernel and is used to normalize the kernel.

[0121] For example, when processing the image of the LED chip package cup, suppose we have calculated , kernel size , we generate a A Gaussian kernel is used and the above formula is applied for normalization. Through the convolution processing of this kernel, the random noise in the image is effectively suppressed. At the same time, the key details on the surface of the bowl and cup, such as tiny scratches or dents, are still clearly visible. This processing not only optimizes the image quality, but also provides a solid foundation for subsequent feature extraction and optical performance evaluation, ensuring the accuracy and reliability of the performance evaluation of LED chip packages.

[0122] S504, applying the adaptive Gaussian kernel to perform Gaussian blur processing on the first image data, the second image data, and the third image data respectively, reducing image noise through convolution operation, and obtaining processed image data;

[0123] This step convolves the image with a normalized Gaussian kernel. This operation smoothes each pixel by weighting the surrounding pixels, thereby reducing noise. However, while the main goal is to reduce noise, it is also necessary to ensure that key edges and details in the image are not overly blurred in order to accurately assess the optical properties of the bowl.

[0124] In order to effectively implement image convolution and ensure the optimization of the results, this application adopts the convolution kernel application formula, which is:

[0125] .

[0126] Explanation of the meaning of parameters and operators: For the original image at position The pixel value of . After Gaussian blur processing, at position The pixel value of . is the normalized Gaussian kernel at the offset The weight value at , used to smooth the image. is the radius of the nucleus, The double summation indicates that all positions in the kernel are accumulated to achieve the convolution operation.

[0127] To make it easier to understand, let's take an example. Suppose we are processing an image of a bowl of an LED chip package with slight surface defects. By applying the above convolution formula, we can reduce the random noise in the image while keeping key surface details such as scratches and tiny dents visible. This processed image is crucial for evaluating the quality of the bowl and its impact on the LED light output. For example, if there are subtle defects on the surface of the bowl that are not masked by noise, these defects will be clearly visible in the image after Gaussian blur processing, so that their potential impact on light propagation can be further analyzed to ensure the optimal performance of LED equipment.

[0128] S505, performing quality assessment on the processed image data to determine the effect of Gaussian blur processing;

[0129] First, we need to evaluate the effectiveness of Gaussian blur processing by comparing the image data before and after processing. This includes observing the reduction in noise levels and the degree of retention of key details such as the edges and surface features of bowls and cups. To this end, image quality assessment indicators such as Signal-to-Noise Ratio (SNR) and Edge Preservation Index (EPI) can be used to quantitatively evaluate the effect of blur processing.

[0130] The signal-to-noise ratio is a commonly used indicator to measure image quality. It compares the intensity of the image signal with the intensity of the background noise. It is calculated by analyzing the brightness value of the image and its standard deviation. The higher the signal-to-noise ratio, the lower the noise level of the image and the better the image quality. The edge preservation index is evaluated by comparing the clarity of the edges in the image before and after processing. This indicator ensures that important geometric features in the image are not over-blurred during the denoising process. Through these indicators, we can get an objective evaluation of the Gaussian blur effect, so as to decide whether it is necessary to adjust the blur parameters or take other image processing measures.

[0131] For example, when processing an image showing a bowl of an LED chip package, we first applied a designed Gaussian kernel for blurring, and then calculated the signal-to-noise ratio and edge preservation index of the processed image. If the signal-to-noise ratio is significantly improved, while the edge preservation index indicates that key surface details such as micro-seam and chip boundaries are still clearly visible, it means that the Gaussian blurring process has achieved the desired effect, that is, effectively reducing image noise without damaging the important visual information of the bowl. This method ensures that subsequent steps can be performed at the best image quality, providing a solid foundation for further analysis.

[0132] S506: If the effect of the Gaussian blur processing does not meet the preset standard, adjust the Gaussian kernel parameters, and repeat steps S503 to S505 until the preset standard is met, thereby obtaining the final processed image data.

[0133] The purpose of this step is to ensure that the Gaussian blur processing achieves the best effect, that is, to minimize noise while maintaining key image details. The implementation of this step involves continuous adjustment of Gaussian kernel parameters and repeated convolution processing and quality assessment until the image data meets the preset quality standards. In specific operations, once it is found that the initial Gaussian blur processing fails to achieve the expected effect, such as the image signal-to-noise ratio is low or the edge preservation index fails to reach the predetermined threshold, it is necessary to adjust the standard deviation or kernel size of the Gaussian kernel.

[0134] The adjustment process begins with analyzing the current processing results and evaluating whether image quality indicators such as signal-to-noise ratio and edge preservation index meet the standards. If not, the Gaussian kernel parameters are adjusted according to the evaluation results. Usually, the standard deviation and kernel size are modified to find a better balance between noise suppression and detail retention. After the adjustment, a new Gaussian kernel is generated again and applied to the image for convolution processing. The quality of the processed image is then evaluated again. This iterative process may need to be repeated many times, and the kernel parameters are carefully adjusted each time to gradually approach the optimal image processing effect.

[0135] For example, suppose that after the initial processing of the image of the LED chip package cup, it is found that although the image noise has been reduced, important edges, such as the tiny textures on the inner wall of the cup, have begun to become blurred. In this case, the standard deviation of the Gaussian kernel is reduced to try to retain more details. After adjustment, the Gaussian blur process is reapplied and the image quality is evaluated again. If the edge clarity has improved and the noise level is still within an acceptable range, continue to use this parameter setting; otherwise, further fine-tune the Gaussian kernel parameters. This repeated optimization ensures that the final image has both low noise and retains the necessary surface details of the cup, thus meeting the high standards required for the optical performance evaluation of the chip package.

[0136] Further, after step S506, the following steps are included:

[0137] S507, performing preliminary edge extraction on the final processed image data through a high-pass filter to obtain a preliminary edge intensity map;

[0138] This step involves using a high-pass filter to perform preliminary edge extraction on the Gaussian blurred image in order to obtain a basic map that expresses the edge strength of the image. In the bowl image analysis of LED chip packages, edge extraction is a key step in evaluating the bowl shape and any surface defects, because the clarity of the edge directly affects the measurement and evaluation of optical performance.

[0139] In this step, the formula .

[0140] Explanation of the meaning of parameters and operators: It is the response function of the high-pass filter in the frequency domain, which is used to emphasize the high-frequency information in the image, that is, the edge part. It is the coordinate in the frequency domain, which is used to determine the position of each point in the frequency domain. The cut-off frequency is used to control the frequency range affected by the filter. Higher values ​​will cause more high-frequency components to be emphasized, enhancing the edge effect.

[0141] When processing an image of a LED chip package cup, suppose we want to detect the edge clarity of the cup to evaluate its manufacturing quality. Using the above high-pass filter formula, set the cutoff frequency A filtering process is performed, which will emphasize the high-frequency parts of the image, namely the edges of the bowls and cups, so that these edges are clearly displayed in the preliminary edge intensity map. This step is crucial because it provides the necessary basic information for the subsequent edge refinement processing, ensuring the accuracy and efficiency of edge extraction. Through this method, the structural edges and possible defects of the bowls and cups can be effectively distinguished, providing key visual data for further optical property analysis.

[0142] S508, performing non-maximum suppression processing on the preliminary edge intensity map to obtain a refined edge intensity map;

[0143] This step involves performing non-maximum suppression on the preliminary edge intensity map. The purpose of this process is to further refine the edge, that is, to retain only the local maximum edge response in the obtained edge map, thereby eliminating those non-edge responses and enhancing the performance of the real edge.

[0144] Specifically, the non-maximum suppression algorithm is used for processing, which is as follows:

[0145] .

[0146] in is the initial edge strength map at position The pixel value of The pixel value in the edge intensity map after non-maximum suppression processing, It means that in the gradient direction of the current pixel, the edge strength of this pixel is the largest, which ensures that the point is part of the actual edge.

[0147] Similarly, for ease of understanding, suppose we are analyzing an image of the surface of a bowl of an LED chip package, which shows the edge of the bowl and possible manufacturing defects. After applying non-maximum suppression, only the most obvious edges along the edge gradient direction will be retained, which can more clearly distinguish between real physical edges and background noise or other non-edge structures. For example, if the edge of the bowl appears as a bright line in the image, non-maximum suppression will ensure that these lines are refined to a single pixel level in their width, enhancing their identifiability in subsequent analysis. In this way, we can accurately identify and evaluate the optical properties of the bowl surface, such as reflection and refraction properties, which is crucial to ensuring the light efficiency and consistency of the LED.

[0148] S509, based on the refined edge strength map, using a dual threshold algorithm to perform edge detection, by setting high and low thresholds to distinguish strong edges from weak edges in the refined edge strength map, and by connecting strong edges with adjacent weak edges to determine final connected edges, to obtain a connected edge map;

[0149] The purpose of this step is to distinguish strong edges from weak edges in the image, thereby ensuring that only truly important edges are ultimately determined as connected edges.

[0150] The specific algorithm formula is:

[0151] 1. Adaptive threshold setting formula:

[0152] .

[0153] 2. Edge detection and connection formula:

[0154] .

[0155] Specifically, It is an adaptive threshold used at each pixel location, dynamically calculated based on local image statistics. and Pixels The mean intensity and standard deviation within its neighborhood, It is an adjustment factor used to control the sensitivity of the threshold to adapt to different image regions and edge characteristics. : Indicates the position The decision variable whether the pixel is marked as an edge, is the pixel value in the edge intensity map after non-maximum suppression processing.

[0156] Similarly, for ease of understanding, imagine that we are processing an image of a bowl showing an LED chip package, which contains subtle edges due to tiny defects in the manufacturing process. Through the above-mentioned adaptive dual threshold method, we can dynamically adjust the threshold in each local area based on the actual image content. If in a certain area, the average pixel intensity is high and the standard deviation is small, it indicates that the area may be more uniform, so the threshold will be increased accordingly to avoid mistakenly marking noise as an edge. This method not only improves the accuracy of edge detection, but also ensures that the integrity and continuity of the real edge in the image are better maintained, thereby providing more accurate basic data for subsequent optical performance analysis.

[0157] S510, evaluating edge continuity and integrity of the connected edge graph;

[0158] During this process, each identified edge is carefully inspected to confirm that their representation in the image is both continuous and complete, which is critical for identifying and evaluating tiny defects and manufacturing defects in the bowl of the LED chip package.

[0159] The evaluation process first involves checking the visual continuity of each edge, which means we need to observe whether the edges in the image appear broken or have gaps. If the edge is found to be suddenly interrupted in a certain area, it may indicate that there is excessive filtering in the image processing process or that the edge detection algorithm fails to fully capture all the information of the edge. For these problems, we will re-examine the previous processing steps, including the application of non-maximum suppression and Gaussian blur, to determine whether the parameters of these processing steps need to be adjusted.

[0160] Next, a completeness assessment is performed to check whether the complete representation of the edge in the image can truly reflect the physical boundaries of the bowl or cup. This includes analyzing the shape and length of the edge to ensure that they can accurately map the geometric structure of the bowl or cup. If it is found that the edge image deviates from the actual physical structure, it may be necessary to adjust the image acquisition environment or image processing algorithm, such as changing the lighting conditions or adjusting the image enhancement parameters, to better reveal the true shape of the bowl or cup.

[0161] For example, suppose that when processing an image of a bowl of an LED chip package, we noticed that the edge appeared discontinuous in a critical area of ​​the bowl, which could affect the accuracy of subsequent optical performance test results. In this case, we might re-execute the edge detection step, use more detailed parameter settings, or introduce additional image processing steps such as edge connection algorithms to ensure that all important edges are accurately and completely captured. Such adjustments not only help us more accurately evaluate the manufacturing quality of the bowl, but also ensure that the performance evaluation of the final product is based on the most accurate image data.

[0162] S511, according to the evaluation results of edge continuity and integrity, optimizing the edges that are displayed as broken or discontinuous, including using an edge repair algorithm to process edge breaks and fragments, and enhancing edge continuity and clarity through interpolation and morphological operations, to generate an optimized edge map;

[0163] Accurate edge mapping of the bowl is fundamental to analyzing light distribution and reflection efficiency, so any incomplete or incorrect edge mapping may lead to inaccurate performance test results.

[0164] In order to achieve edge optimization, we use image processing technology, combined with computational geometry and image morphology operations to enhance the continuity and clarity of the edges. The following are two key calculation steps:

[0165] .

[0166] in It is the edge image after non-maximum suppression and double threshold processing. is the optimized edge image. Indicate point Neighborhood of Calculate the Euclidean distance between two points, It is the set distance threshold used to determine whether two edge points can be considered continuous.

[0167] .

[0168] in Represents a morphological closing operation, which is used to enhance the edge integrity in an image by filling the breaks in the edge by dilating first and then corroding. is the edge image to be processed, is a structural element, and Represent the morphological dilation and erosion operations respectively.

[0169] When processing images of LED chip package bowls, we found that some edges were broken due to manufacturing defects. Using the edge connection and repair formulas described above, we are able to automatically connect these broken edges if their distance is less than a preset threshold. In addition, applying morphological closing operations further smooths the edges and fills in tiny gaps left by the initial processing. This optimized edge image is clearer and more continuous, providing a more accurate visual basis for the next step of optical performance evaluation. Through these complex and creative calculation steps, we ensure high accuracy and efficiency of image analysis, which effectively supports the quality control and performance testing of bowls and cups.

[0170] S512: Overlay the optimized edge map with the original image to obtain final edge-enhanced image data.

[0171] In this process, first, we will overlay the two images through a simple but effective image processing technique. Typically, this involves overlaying the edge-optimized image on the original image with a certain degree of transparency. The purpose of this is to maintain the detail information of the original image while highlighting the edges that have been identified and enhanced, making them more visually obvious and easier to recognize.

[0172] The method of overlay processing usually involves adjusting the transparency of the two image layers or using image blending techniques, such as merging images by weighted averaging. Here, the optimized edge map is usually overlaid with the original image with higher contrast and obvious colors to ensure that the visualization effect of the edge is maximized without excessively interfering with other important information of the original image.

[0173] For example, when analyzing an image of an LED chip package cup with subtle defects, these defects may only be slightly visible at the edges. By overlaying the optimized edge map with the original image, these subtle defects can be made more visually obvious, making it easier to detect and evaluate these potential problems. In addition, this overlay image also facilitates the intuitive display of the treatment effect in presentations or technical reports, helping technicians or researchers better understand the optical properties of the cup and its possible manufacturing defects.

[0174] Furthermore, after step S512, the following steps are included:

[0175] S513, selecting a target area from the final edge-enhanced image data;

[0176] Selecting target areas from the final edge-enhanced image data refers to those areas that show potential defects or significant features. The purpose of selecting these areas is to ensure that subsequent analysis focuses on the most critical parts, thereby improving the efficiency and effectiveness of the analysis. In actual operation, this step involves image segmentation technology to automatically or manually mark the areas of interest in the image.

[0177] S514, applying a gray level co-occurrence matrix to analyze the texture features of the target area, extracting corresponding descriptors, and obtaining a preliminary texture feature descriptor set, wherein the descriptors include contrast, homogeneity, energy, and correlation;

[0178] The gray-level co-occurrence matrix is ​​a statistical method used to quantify the spatial relationship between pixels in an image, thereby extracting texture descriptors such as contrast, homogeneity, energy, and correlation. These descriptors provide information about the roughness, uniformity, and pattern repeatability of the image texture, and are important data for understanding the surface characteristics of bowls and cups. By calculating these descriptors, we can obtain a preliminary quantitative understanding of the surface texture conditions of bowls and cups, which is crucial for identifying possible problems in the manufacturing process.

[0179] S515, performing dimensionality reduction processing on the preliminary texture feature descriptor set by using principal component analysis to obtain texture features after dimensionality reduction;

[0180] Principal component analysis (PCA) was used to reduce the dimensionality of the preliminary texture feature descriptor set obtained from the gray-level co-occurrence matrix. PCA reduces the complexity of the dataset by extracting the main variables and patterns in the data while retaining the most important information, which makes subsequent model building more efficient and accurate. In this step, we simplified the texture data through PCA transformation and obtained a set of reduced-dimensional features that characterize key texture characteristics.

[0181] S516, constructing a texture feature model using the texture features after dimensionality reduction;

[0182] These reduced texture features are used to build a texture feature model. This model is based on a machine learning algorithm that can predict the optical properties of bowls and cups or identify potential defects based on texture features. The training of the model involves supervised learning, in which known sample data is used to train the algorithm so that it can recognize similar patterns in unknown data.

[0183] S517, applying the texture feature model to analyze the texture characteristics of the image data to obtain a comprehensive texture analysis result;

[0184] S518: Superimpose the texture analysis result with the original image data to obtain final texture analysis image data.

[0185] This overlay image not only provides a visual analysis result, but also facilitates further interpretation and reporting. For example, when evaluating LED chip packages, this method can reveal texture anomalies caused by material defects or improper processing, providing a scientific basis for improving production processes and enhancing product quality.

[0186] It is worth noting that texture analysis can help identify defects that may occur during the production process, such as scratches, cracks, or uneven material deposition, which may cause problems in the product's service life and performance. By detecting these problems early, manufacturers can adjust the manufacturing process in time, reduce scrap rates and costs, and improve production efficiency.

[0187] Furthermore, after step S518, the following steps are included:

[0188] S519, selecting an area with significant texture features from the final texture analysis image data as a spectral scanning area;

[0189] This often includes areas that display irregular or unconventional textures that could indicate potential defects or material inhomogeneities.

[0190] S520, applying spectral analysis technology to perform spectral scanning on the spectral scanning area, and collecting and recording reflectivity and transmittance data at each spectral wavelength;

[0191] This data is then used to analyze the optical response of the material. Spectral data is collected by shining different wavelengths of light and recording the amount of light reflected and transmitted through the surface of the bowl or cup.

[0192] S521. Perform Fourier transform on the collected spectral data. Fourier transform is performed to convert time or space samples into the frequency domain. Fourier transform is used to analyze the presence and intensity of each frequency component, which helps to reveal periodic patterns and important frequency components hidden in the spectral data. Through this transformation, we can more clearly see how material properties affect the scattering and absorption of light.

[0193] S522. Combine the support vector machine to classify and recognize the Fourier transform results to obtain the spectral analysis results; Support vector machine (SVM) is a powerful machine learning algorithm to classify and recognize the results of Fourier transform. SVM distinguishes data points of different categories by constructing one or more classification boundaries, so that samples can be effectively classified according to spectral characteristics to determine the type or state of the bowl and cup material, such as distinguishing between normal and defective bowl and cup areas.

[0194] S523. Compare the spectral analysis results with a known material spectrum database to obtain a spectral characteristics report. This step is accomplished by matching the spectral data with reference spectra in the database, so that the specific type of material can be quickly identified or deviations from conventional spectral patterns can be detected, which is very useful for ensuring product quality and consistency.

[0195] S524, integrating the spectral characteristic report with the final texture analysis image data, displaying the association between spectral characteristics and texture characteristics through image and data superposition enhancement method, and obtaining comprehensive spectral and texture analysis image data.

[0196] This integration process is completed through image and data overlay enhancement methods, which not only improves the visual representation of the data, but also makes it easier for analysts to understand the correlation between spectral characteristics and textural characteristics. This comprehensive analysis method provides a comprehensive view of how the physical and chemical properties of the bowl interact to affect the overall performance of the chip package.

[0197] When evaluating a newly developed LED package cup, these steps allow us to analyze in detail which areas of the material may have unstable performance due to improper temperature control during processing. By integrating the data from spectral and texture analysis, the R&D team can intuitively see which improvement measures are most effective, thereby guiding future design optimization.

[0198] Step S600 begins by taking the pre-processed images obtained through multi-level image processing in step S500 as input. These pre-processed images have been processed such as Gaussian blur to reduce noise and highlight important visual features in the image. In S600, we convert these image data into feature atlases, which includes using a series of image processing algorithms to extract key information in the image, such as edge, texture, color and shape features.

[0199] The core technology to achieve this step involves applying a variety of image analysis techniques, such as edge detection algorithms to identify obvious boundaries in images, texture analysis tools to evaluate the roughness and detail of the surface of objects, and color analysis technology to capture the color distribution and intensity in chip packages. Each technology generates a feature map, which will be used as part of the feature set for subsequent model training and analysis.

[0200] For example, if we are dealing with an image of an LED chip package, the feature extraction in step S600 may include identifying tiny cracks or gaps that may be caused by improper manufacturing, which will be marked during the edge detection process. Similarly, texture analysis may reveal whether the coating inside the bowl is uniform, or whether there are abnormal texture areas caused by uneven material distribution. In addition, color analysis can help identify color changes caused by different materials, which is critical for detecting different compounds or materials that may be used in the packaging process.

[0201] Each extracted feature is converted into numerical data to form a feature vector, which constitutes a feature atlas. These feature atlases are then input into the chip inspection model in step S700, and the model uses these data to evaluate the optical and structural properties of the chip package.

[0202] Further, each of the feature atlases includes a first feature atlas corresponding to the first image data, a second feature atlas corresponding to the second image data, and a third feature atlas corresponding to the third image data;

[0203] The step S700 includes:

[0204] S701, inputting the first feature atlas, the second feature atlas, and the third feature atlas into a preset chip detection model, and calculating the attention weight vectors of the first feature atlas, the second feature atlas, and the third feature atlas respectively through the attention mechanism layer of the chip detection model;

[0205] In this step, the attention mechanism pays special attention to the feature areas that are most critical for detecting the optical properties of the chip package, such as the optical texture or reflective properties of the inner wall of a bowl or cup.

[0206] In the specific implementation, the first, second and third feature atlases are standardized and input into the detection model, using an attention mechanism based on the Transformer architecture, which evaluates the importance of each feature in optical property detection by calculating the dot product of the query, key and value.

[0207] The specific calculation formula is:

[0208] .

[0209] Explanation of the meaning of parameters and operators:

[0210] : Query, key, and value are converted from the feature atlas and represent the encoding of the original features from different perspectives and processing methods.

[0211] : The dimension of the key vector, used to normalize the result of the dot product to prevent excessively large internal product values ​​from affecting the gradient of the softmax function.

[0212] Softmax: Normalization function that ensures that the sum of the elements of the output weight vector is 1, representing the probability distribution.

[0213] In the quality control of LED chip packages, the optical properties of different regions, such as reflectivity and refractive index, are crucial to the overall performance. By applying the above-mentioned attention mechanism, the model can automatically identify the feature areas that have the greatest impact on the optical performance, such as areas where defects may exist or areas where the optical properties are abnormal, and then apply higher weights to these areas, thereby paying more attention to these key features in subsequent analysis. This method not only improves the accuracy of detection, but also provides strong technical support for quality assurance in the manufacturing process.

[0214] S702, performing weighted fusion on the first feature atlas, the second feature atlas, and the third feature atlas according to the weight vector through a feature fusion layer in the chip detection model to obtain a comprehensive feature vector;

[0215] This step applies the corresponding attention weight to each feature map set. This process is achieved through element-by-element multiplication (Hadamard product). During the feature fusion process, the weighted feature maps are merged into a single comprehensive feature vector through weighted summation. This vector will contain all the key optical property information.

[0216] Specifically, the following algorithm formula is used:

[0217] .

[0218] in, is the weight vector obtained from the attention mechanism layer, are their respective eigenvectors, Represents element-wise multiplication.

[0219] Suppose that when inspecting a high-performance LED chip package with a complex internal structure, different images capture the optical properties from the top and inside. The first feature atlas may emphasize the reflective properties of the top coating, while the second and third feature atlases emphasize the internal light scattering and the internal material uniformity, respectively. By applying the weighted fusion method in S702, it can be ensured that the model does not only focus on a single perspective, but integrates the information of all key perspectives, thereby providing a comprehensive and accurate view in the final optical property analysis.

[0220] S703 , calculating an optical characteristic detection result of the chip package according to the comprehensive feature vector through the classification layer in the chip detection model, and outputting the optical characteristic detection result through the output layer in the chip detection model.

[0221] The purpose of this step is to use the model to determine the optical properties of the chip package, identify possible defects or quality issues, and ultimately determine the overall quality level of the chip.

[0222] Specific implementation steps:

[0223] 1. Feature vector input: Input the comprehensive feature vector to the classification layer.

[0224] 2. Classification processing: Use one or more fully connected layers in a deep neural network, which are configured with activation functions such as ReLU or Sigmoid to process the feature vector and produce classification output.

[0225] 3. Result output: The output layer converts the classification results into specific test results, such as pass / fail, or specific defect types.

[0226] The specific algorithm formula is:

[0227] .

[0228] in, is the weight matrix, is the comprehensive feature vector obtained from the feature fusion layer, is the bias term, It is an activation function, such as Sigmoid, used to make the final classification decision.

[0229] In step S703, if the comprehensive feature vector indicates that the lens bowl has uneven polishing or scratches, the classification layer will accurately map these problems to specific defect types by analyzing these features and combining the trained weights and bias parameters. For example, the model may identify optical scattering problems caused by improper polishing. Such output not only helps the quality control team solve the problem, but also guides improvement measures in the production process.

[0230] The above describes the detection method of the chip package in the embodiment of the present invention. The following describes the detection device of the chip package in the embodiment of the present invention. Figure 2 In one embodiment of the present invention, a detection device for a chip package includes:

[0231] A power-on module 201, used to power on the light-emitting chip in an environment without visible light, so that the light-emitting chip is in a power-on state;

[0232] A first image acquisition module 202 is used to acquire optical images of the inner surface of the peripheral wall of the bowl and the top surface of the bowl to obtain first image data and second image data respectively;

[0233] A power-off module 203, used to perform power-off processing on the light-emitting chip in an environment with visible light, so that the light-emitting chip is in a power-off state;

[0234] The second image acquisition module 204 is used to acquire an optical image of the top surface of the bowl or cup to obtain third image data;

[0235] An image processing module 205 is used to perform multi-level image processing on the first image data, the second image data and the third image data to obtain corresponding pre-processed images respectively;

[0236] A feature extraction module 206 is used to extract features from each of the pre-processed images to obtain corresponding feature atlases;

[0237] The model input module 207 is used to input each of the feature atlases into a preset chip detection model, and perform optical property detection of the chip package according to each of the feature atlases through the chip detection model to obtain corresponding optical property detection results.

[0238] above Figure 2 The detection device of the chip package in the embodiment of the present invention is described in detail from the perspective of modular functional entities. The detection device of the chip package in the embodiment of the present invention is described in detail from the perspective of hardware processing.

[0239] Figure 3 3 is a schematic diagram of the structure of a chip package detection device provided by an embodiment of the present invention. The chip package detection device 300 may have relatively large differences due to different configurations or performances, and may include one or more processors (central processing units, CPU) 310 (for example, one or more processors) and a memory 320, and one or more storage media 330 (for example, one or more mass storage device terminals) storing application programs 333 or data 332. Among them, the memory 320 and the storage medium 330 can be short-term storage or permanent storage. The program stored in the storage medium 330 may include one or more modules (not shown in the figure), and each module may include a series of instruction operations in the chip package detection device 300. Furthermore, the processor 310 can be configured to communicate with the storage medium 330, and execute a series of instruction operations in the storage medium 330 on the chip package detection device 300 to implement the steps of the above-mentioned chip package detection method.

[0240] The chip package detection device 300 may also include one or more power supplies 340, one or more wired or wireless network interfaces 350, one or more input and output interfaces 360, and / or one or more operating systems 331, such as Windows Serve, Mac OS X, Unix, Linux, FreeBSD, etc. It will be appreciated by those skilled in the art that Figure 3 The structure of the detection equipment for the chip package shown does not constitute a limitation on the computing resource calling equipment of the intelligent camera cluster provided by the present invention, and may include more or fewer components than shown in the figure, or a combination of certain components, or a different arrangement of components.

[0241] The present invention also provides a computer-readable storage medium, which may be a non-volatile computer-readable storage medium or a volatile computer-readable storage medium. Instructions are stored in the computer-readable storage medium. When the instructions are executed on a computer, the computer executes the steps of the chip package detection method.

[0242] Those skilled in the art can clearly understand that, for the convenience and brevity of description, the specific working process of the above-described system, device, or unit can refer to the corresponding process in the aforementioned method embodiment and will not be repeated here.

[0243] If the integrated unit is implemented in the form of a software functional unit and sold or used as an independent product, it can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of the present invention is essentially or the part that contributes to the prior art or the whole or part of the technical solution can be embodied in the form of a software product. The computer software product is stored in a storage medium, including several instructions to enable a computer device (which can be a personal computer, server, or network device, etc.) to perform all or part of the steps of the method described in each embodiment of the present invention. The aforementioned storage medium includes: U disk, mobile hard disk, read-only memory (ROM), random access memory (RAM), disk or optical disk and other media that can store program code.

[0244] As described above, the above embodiments are only used to illustrate the technical solutions of the present invention, rather than to limit the same. Although the present invention has been described in detail with reference to the aforementioned embodiments, those skilled in the art should understand that the technical solutions described in the aforementioned embodiments may still be modified, or some of the technical features thereof may be replaced by equivalents. However, these modifications or replacements do not deviate the essence of the corresponding technical solutions from the spirit and scope of the technical solutions of the embodiments of the present invention.

Claims

1. A method for detecting a chip package, characterized in that: The chip package includes a bowl and a light-emitting chip fixed in the bowl. The detection method of the chip package includes: S100, powering on the light emitting chip in an environment without visible light, so that the light emitting chip is in a powered-on state; S200, collecting optical images of the inner surface of the peripheral wall of the bowl and the top surface of the bowl to obtain first image data and second image data respectively; S300, powering off the light emitting chip in an environment with visible light, so that the light emitting chip is in a power-off state; S400, collecting an optical image of the top surface of the bowl or cup to obtain third image data; S500, performing multi-level image processing on the first image data, the second image data and the third image data to obtain corresponding pre-processed images respectively; The step S500 includes: S501, analyzing the grayscale histograms of the first image data, the second image data, and the third image data one by one, and determining the noise level of each image data respectively; S502, determining Gaussian blur parameters applicable to the first image data, the second image data, and the third image data respectively according to the noise level of each image data; S503, constructing an adaptive Gaussian kernel according to the Gaussian blur parameters; S504, applying the adaptive Gaussian kernel to perform Gaussian blur processing on the first image data, the second image data, and the third image data respectively, reducing image noise through convolution operation, and obtaining processed image data; S505, performing quality assessment on the processed image data to determine the effect of Gaussian blur processing; S506: If the effect of the Gaussian blur processing does not meet the preset standard, adjust the Gaussian kernel parameters, and repeat steps S503 to S505 until the preset standard is met, thereby obtaining the final processed image data; S507, performing preliminary edge extraction on the final processed image data through a high-pass filter to obtain a preliminary edge intensity map; S508, performing non-maximum suppression processing on the preliminary edge intensity map to obtain a refined edge intensity map; S509, based on the refined edge strength map, using a dual threshold algorithm to perform edge detection, by setting high and low thresholds to distinguish strong edges from weak edges in the refined edge strength map, and by connecting strong edges with adjacent weak edges to determine final connected edges, to obtain a connected edge map; S510, evaluating edge continuity and integrity of the connected edge graph; S511, according to the evaluation results of edge continuity and integrity, optimizing the edges that are displayed as broken or discontinuous, including using an edge repair algorithm to process edge breaks and fragments, and enhancing edge continuity and clarity through interpolation and morphological operations, to generate an optimized edge map; S512, superimposing the optimized edge map with the original image to obtain final edge-enhanced image data; S513, selecting a target area from the final edge-enhanced image data; S514, applying a gray level co-occurrence matrix to analyze the texture features of the target area, extracting corresponding descriptors, and obtaining a preliminary texture feature descriptor set, wherein the descriptors include contrast, homogeneity, energy, and correlation; S515, performing dimensionality reduction processing on the preliminary texture feature descriptor set by using principal component analysis to obtain texture features after dimensionality reduction; S516, constructing a texture feature model using the texture features after dimensionality reduction; S517, applying the texture feature model to analyze the texture characteristics of the image data to obtain a comprehensive texture analysis result; S518, superimposing the texture analysis result with the original image data to obtain final texture analysis image data; S519, selecting an area with significant texture features from the final texture analysis image data as a spectral scanning area; S520, applying spectral analysis technology to perform spectral scanning on the spectral scanning area, and collecting and recording reflectivity and transmittance data at each spectral wavelength; S521, performing Fourier transform on the collected spectral data; S522, classifying and pattern recognizing the Fourier transform results by using a support vector machine to obtain a spectrum analysis result; S523, comparing the spectrum analysis result with a known material spectrum database to obtain a spectrum characteristic report; S524, integrating the spectral characteristic report with the final texture analysis image data, displaying the association between spectral characteristics and texture characteristics through an image and data superposition enhancement method, and obtaining comprehensive spectral and texture analysis image data; S600, extracting features from each of the pre-processed images to obtain corresponding feature atlases; S700, inputting each of the feature atlases into a preset chip detection model, and performing optical property detection of the chip package according to each of the feature atlases through the chip detection model to obtain corresponding optical property detection results; Each of the feature atlases includes a first feature atlas corresponding to the first image data, a second feature atlas corresponding to the second image data, and a third feature atlas corresponding to the third image data; the step S700 includes: S701, inputting the first feature atlas, the second feature atlas, and the third feature atlas into a preset chip detection model, and calculating the attention weight vectors of the first feature atlas, the second feature atlas, and the third feature atlas respectively through the attention mechanism layer of the chip detection model; S702, performing weighted fusion on the first feature atlas, the second feature atlas, and the third feature atlas according to the weight vector through a feature fusion layer in the chip detection model to obtain a comprehensive feature vector; S703 , calculating an optical characteristic detection result of the chip package according to the comprehensive feature vector through the classification layer in the chip detection model, and outputting the optical characteristic detection result through the output layer in the chip detection model.

2. A chip package detection device, characterized in that: The chip package detection device adopts the chip package detection method according to claim 1, the chip package comprises a bowl and a light-emitting chip fixed in the bowl, and the chip package detection device comprises: A power-on module, used for powering on the light-emitting chip in an environment without visible light, so that the light-emitting chip is in a power-on state; A first image acquisition module is used to acquire optical images of the inner surface of the peripheral wall of the bowl and the top surface of the bowl to obtain first image data and second image data respectively; A power-off module, used to power off the light-emitting chip in an environment with visible light, so that the light-emitting chip is in a power-off state; A second image acquisition module is used to acquire an optical image of the top surface of the bowl or cup to obtain third image data; An image processing module, used for performing multi-level image processing on the first image data, the second image data and the third image data to obtain corresponding pre-processed images respectively; A feature extraction module, used to extract features from each of the preprocessed images to obtain corresponding feature atlases; The model input module is used to input each of the feature atlases into a preset chip detection model, and perform optical property detection of the chip package according to each of the feature atlases through the chip detection model to obtain corresponding optical property detection results.

3. A chip package detection device, characterized in that: The chip package detection device comprises: a memory and at least one processor, wherein the memory stores instructions; The at least one processor calls the instructions in the memory to enable the chip package detection device to perform the steps of the chip package detection method as claimed in any one of claims 1.

4. A computer-readable storage medium having instructions stored thereon, characterized in that: When the instructions are executed by a processor, the steps of the chip package detection method according to any one of claims 1 are implemented.

Images