Apparent color difference control method for mini LED printed circuit board processing

Abstract

The invention relates to the field of printed circuit board processing, and provides an apparent color difference control method for mini LED printed circuit board processing, which comprises the following steps of: firstly, providing a circuit board substrate comprising a dielectric layer and a copper layer, then covering dry film type solder resist ink on the surface of the substrate through a vacuum film pressing process, and firmly attaching the dry film type solder resist ink to the substrate through vacuum negative pressure adsorption; the method comprises the following steps of: pressing a film on a printed circuit board, applying pressure to perform leveling treatment to obtain a pressed circuit board, performing solder resist exposure and development on the pressed circuit board, and finally manufacturing the mini LED printed circuit board, in order to control the apparent chromatic aberration problem, detecting the printed circuit board by adopting a local area color distribution prototype feature aggregation analysis technology based on a feature domain, therefore, whether a chromatic aberration problem exists is judged. Therefore, the production quality of the mini LED printed circuit board can be effectively improved.

Description

Technical Field

[0001] This application relates to the field of printed circuit board processing, and more specifically, to a method for controlling apparent color difference in the processing of mini-LED printed circuit boards. Background Art

[0002] With the rapid development of mini-LED display technology towards high density and high precision, the apparent quality of printed circuit boards has a more significant impact on the terminal display effect. In the mini-LED packaging structure, it is necessary to accurately form tens of thousands of solder mask openings arrays on the circuit board surface to meet the bonding requirements of micron-level LED chips, which puts forward sub-micron-level precision control requirements for the thickness uniformity and surface flatness of the solder mask layer.

[0003] Traditional manufacturing processes use liquid solder mask ink to be coated on the circuit board surface by printing. Due to the inherent fluidity characteristics of liquid materials, wetting differences are likely to occur at the junction of the dielectric layer and the copper layer, resulting in uneven thickness distribution of the solder mask layer and a microscopically uneven surface morphology. This thickness gradient change will cause multiple refraction effects of incident light, visually presenting as regional color difference patches, seriously affecting the color uniformity of the display panel. More prominently, in the development process of the unique matrix-type solder mask opening structure of mini-LED (each opening contains 6 precision pads), due to the inconsistency of the underlying ink thickness, there will be differences in the etching rate of the opening sidewalls, which will further cause geometric deviations in the opening size. Such size anomalies not only directly cause visual color differences in spot offset, but may also lead to misalignment of micron-level pads and LED chips. In addition, the surface of the traditional liquid solder mask layer lacks a dense protective layer and is easily damaged by mechanical scratches in subsequent processes, forming surface defects that affect optical performance.

[0004] Therefore, an optimized solution for controlling apparent color difference in the processing of mini-LED printed circuit boards is desired. Summary of the Invention

[0005] To solve the above technical problems, this application is proposed. Embodiments of this application provide a method for controlling apparent color difference in the processing of mini-LED printed circuit boards.

[0006] According to one aspect of this application, there is provided a method for controlling apparent color difference in the processing of mini-LED printed circuit boards, which includes:

[0007] Circuit board substrate preparation step: providing a circuit board substrate, the circuit board substrate includes a dielectric layer and a copper layer stacked on the dielectric layer; Vacuum laminating step: laying a dry film type solder mask ink flat on the circuit board substrate, and using vacuum negative pressure adsorption to make the dry film type solder mask ink firmly adhere to the surface of the circuit board substrate, and performing leveling treatment on it by applying pressure to obtain a circuit board after vacuum laminating;

[0008] Post - processing steps: The circuit board after vacuum laminating is subjected to solder mask exposure and development to obtain a mini - LED printed circuit board; among them, the apparent color difference control method further includes an apparent color difference detection step: performing local area color distribution prototype feature aggregation analysis based on the feature domain on the mini - LED printed circuit board to determine whether there is a color difference problem with the mini - LED printed circuit board.

[0009] Compared with the prior art, the apparent color difference control method for mini - LED printed circuit board processing provided by the present application first provides a circuit board substrate including a dielectric layer and a copper layer, then covers the surface of the substrate with a dry - film solder mask ink through a vacuum laminating process, and uses vacuum negative pressure adsorption to make it firmly adhere, while applying pressure for leveling treatment to obtain the laminated circuit board. Then, the laminated circuit board is subjected to solder mask exposure and development, and finally a mini - LED printed circuit board is made. To control the apparent color difference problem, a local area color distribution prototype feature aggregation analysis technology based on the feature domain is used to detect the printed circuit board to determine whether there is a color difference problem. In this way, the production quality of the mini - LED printed circuit board can be effectively improved. Description of the Drawings

[0010] By describing the embodiments of the present application in more detail in conjunction with the drawings, the above - mentioned and other objects, features, and advantages of the present application will become more obvious. The drawings are used to provide a further understanding of the embodiments of the present application, and constitute a part of the specification. Together with the embodiments of the present application, they are used to explain the present application and do not constitute a limitation to the present application. In the drawings, the same reference numerals generally represent the same components or steps.

[0011] Figure 1 It is a flowchart of the apparent color difference control method for mini - LED printed circuit board processing according to an embodiment of the present application. Figure 2 It is a flowchart of the vacuum laminating step in the apparent color difference control method for mini - LED printed circuit board processing according to an embodiment of the present application.

[0012] Figure 3 It is a flowchart of the apparent color difference detection step in the apparent color difference control method for mini - LED printed circuit board processing according to an embodiment of the present application.

[0013] Figure 4 It is a flowchart of S43 in the apparent color difference control method for mini - LED printed circuit board processing according to an embodiment of the present application.

[0014] Figure 5 It is a flowchart of S44 in the apparent color difference control method for mini - LED printed circuit board processing according to an embodiment of the present application.

[0015] Figure 6 It is a flowchart of S44-2 in the apparent color difference control method for mini-LED printed circuit board processing according to an embodiment of the present application.

[0016] Figure 7 It is a flowchart of S45 in the apparent color difference control method for mini-LED printed circuit board processing according to an embodiment of the present application. Detailed implementation manners

[0017] Hereinafter, exemplary embodiments according to the present application will be described in detail with reference to the accompanying drawings. Obviously, the described embodiments are only a part of the embodiments of the present application, rather than all embodiments of the present application. It should be understood that the present application is not limited by the exemplary embodiments described herein.

[0018] With the development of mini-LED display technology towards high density and high precision, the surface quality of printed circuit boards is crucial for the terminal display effect. In mini-LED packaging, the circuit board needs to accurately form a large number of micron-level solder mask opening arrays to meet the bonding requirements of LED chips, which poses sub-micron-level precision requirements for the thickness uniformity and surface flatness of the solder mask layer.

[0019] Traditional processes use liquid solder mask ink to print and coat the surface of the circuit board. However, due to the fluidity of liquid materials, wetting differences are likely to occur at the junction of the dielectric layer and the copper layer, resulting in uneven thickness of the solder mask layer and surface micro-concavities and convexities. This thickness difference will cause an optical refraction effect, resulting in color difference patches and affecting the color uniformity of the display panel. In addition, in the development process of the mini-LED matrix solder mask opening structure (each opening contains 6 precision pads), the sidewall etching rate difference due to uneven ink thickness will lead to deviation of the opening size. This will not only cause visual color difference of spot offset, but also may lead to misalignment of the bonding between the pad and the LED chip. More importantly, the surface of the traditional solder mask layer lacks a dense protective layer and is easily mechanically damaged in subsequent processes, forming surface defects that affect optical performance.

[0020] Based on this, the present application proposes an apparent color difference control method for mini-LED printed circuit board processing. Figure 1 It is a flowchart of the apparent color difference control method for mini-LED printed circuit board processing according to an embodiment of the present application. As Figure 1As shown, the apparent color difference control method for mini-LED printed circuit board processing according to an embodiment of the present application includes: a circuit board substrate preparation step: providing a circuit board substrate, the circuit board substrate including a dielectric layer and a copper layer stacked on the dielectric layer; a vacuum laminating step: laying a dry film type solder resist ink flat on the circuit board substrate, and using vacuum negative pressure adsorption to firmly paste the dry film type solder resist ink on the surface of the circuit board substrate, and performing leveling treatment on it by applying pressure to obtain a circuit board after vacuum lamination; a post-treatment step: performing solder resist exposure and development on the circuit board after vacuum lamination to obtain a mini-LED printed circuit board; an apparent color difference detection step: performing local area color distribution prototype feature aggregation analysis based on a feature domain on the mini-LED printed circuit board to determine whether there is a color difference problem with the mini-LED printed circuit board.

[0021] In the circuit board substrate preparation step, a circuit board substrate is provided, the circuit board substrate including a dielectric layer and a copper layer stacked on the dielectric layer. It should be understood that the dielectric layer, as an insulating layer, undertakes the functions of supporting the copper layer and achieving electrical insulation, while the copper layer constitutes the conductive circuit part of the circuit board, and the two together form the basic electrical structure of the circuit board. In the manufacture of mini-LED printed circuit boards, the solder resist layer needs to cover the surfaces of the copper layer and the dielectric layer to form a precise solder mask opening array. In traditional processes, the liquid solder resist ink is prone to problems such as uneven thickness and surface unevenness at the junction of the dielectric layer and the copper layer due to wetting differences. However, in this technical solution, by using a dry film type solder resist ink and combining with a vacuum laminating process, it is necessary to be based on a circuit board substrate with a complete structure and a surface state meeting the requirements to ensure that the dry film type solder resist ink can be evenly laid flat on the surfaces of the copper layer and the dielectric layer, providing a stable base for subsequent use of vacuum negative pressure adsorption to achieve firm adhesion between the dry film and the substrate surface and leveling treatment, thereby avoiding problems such as inconsistent solder resist layer thickness and poor surface flatness caused by substrate structure defects, and ensuring that the solder mask openings with consistent sizes are formed in the subsequent exposure and development processes of the solder resist layer, and further improving the apparent color difference of the mini-LED printed circuit board.

[0022] In the vacuum laminating step, a dry film type solder resist ink is laid flat on the circuit board substrate, and using vacuum negative pressure adsorption to firmly paste the dry film type solder resist ink on the surface of the circuit board substrate, and performing leveling treatment on it by applying pressure to obtain a circuit board after vacuum lamination. Specifically, Figure 2 is a flowchart of the vacuum laminating step in the apparent color difference control method for mini-LED printed circuit board processing according to an embodiment of the present application. As Figure 2As shown, the vacuum laminating step includes: S21, evenly covering the surface of the printed circuit board substrate with dry film solder mask ink; S22, using vacuum negative pressure adsorption to ensure that the dry film solder mask ink can be firmly and tightly adhered to the surface of the printed circuit board substrate; S23, after vacuum negative pressure adsorption, performing leveling treatment on it by applying pressure to obtain the printed circuit board after vacuum lamination.

[0023] In step S21, the surface of the printed circuit board substrate is evenly covered with dry film solder mask ink. It should be understood that by evenly covering the surface of the printed circuit board substrate with dry film solder mask ink, it can be ensured that the solder mask material can adhere to the substrate surface evenly and tightly, providing initial conditions for subsequent vacuum negative pressure adsorption and leveling treatment. This operation can avoid the problem of uneven thickness caused by the fluidity of traditional liquid solder mask ink, and at the same time reduce the surface unevenness caused by uneven ink distribution. That is, by pre - evenly covering the dry film solder mask ink, the risk of local poor adhesion or bubble residue that may occur in the subsequent vacuum lamination process can be effectively reduced, thereby improving the overall uniformity and surface flatness of the solder mask layer.

[0024] In step S22, using vacuum negative pressure adsorption to ensure that the dry film solder mask ink can be firmly and tightly adhered to the surface of the printed circuit board substrate. It should be understood that using vacuum negative pressure adsorption to firmly and tightly adhere the dry film solder mask ink to the surface of the printed circuit board substrate can eliminate the air gap between the dry film and the substrate, ensuring complete bubble - free adhesion between the two. That is, this operation forces the dry film material to evenly extend and tightly adhere to the substrate surface through the negative pressure effect in a vacuum environment, including areas prone to poor adhesion such as the junction between the dielectric layer and the copper layer. This tightly adhered state can avoid the problem of uneven thickness caused by the fluidity of materials in traditional printing processes, and at the same time prevent the dry film from peeling or local deformation that may occur in subsequent processes. The stable bonding interface formed by vacuum negative pressure adsorption provides a uniform force - bearing basis for subsequent leveling treatment, ensuring that the surface of the solder mask layer meets the required flatness requirements, thus meeting the strict standards for the solder mask opening size accuracy and optical consistency of miniLED products.

[0025] In step S23, after vacuum negative pressure adsorption, performing leveling treatment on it by applying pressure to obtain the printed circuit board after vacuum lamination. It should be understood that performing leveling treatment by applying pressure after vacuum negative pressure adsorption can eliminate the possible microscopic unevenness between the dry film solder mask ink and the substrate surface, ensuring that the solder mask layer has a uniform thickness distribution and an ideal surface flatness. Specifically, this operation further compacts the tightly adhered dry film material through mechanical pressure, eliminating the local minor undulations that may remain after vacuum adsorption, especially for the transition area between the dielectric layer and the copper layer. This pressure leveling treatment can correct the ink thickness fluctuation caused by the microscopic surface topography difference of the substrate, forming a solder mask surface with consistent optical properties.

[0026] In the post - processing step, the circuit board after vacuum laminating is subjected to solder mask exposure and development to obtain a mini - LED printed circuit board. It should be understood that performing solder mask exposure and development on the circuit board after vacuum laminating can accurately transfer the design pattern to the planarized solder mask layer through a photochemical reaction, forming a solder mask opening array that meets the requirements of mini - LED packaging. Specifically, during the exposure process, ultraviolet light is used to irradiate the selective areas of the dry - film type solder mask ink to cause a cross - linking and curing reaction, while the development process removes the ink material in the unexposed areas through chemical dissolution, thereby forming an accurate opening pattern on the solder mask layer. This step can ensure the size and position accuracy of each solder mask opening. Especially for the matrix - type opening structure containing 6 precision pads, its processing accuracy directly affects the alignment of subsequent LED chip bonding. That is, through the precise control of the exposure and development processes, micron - level graphic processing is achieved on the basis of a solder mask layer with good flatness, and finally a finished printed circuit board that meets the requirements of high - density mini - LED displays is obtained.

[0027] This application, by adopting a dry - film type solder mask ink combined with a vacuum laminating process and using a three - stage vacuum press to press the dry - film type ink, can effectively address the apparent color difference problem caused by the traditional liquid solder mask process. The solid - state characteristics of the dry - film material fundamentally avoid the wetting differences caused by the fluidity of liquid ink. Through vacuum negative pressure adsorption, a gap - free fit is formed between the solder mask layer and the substrate surface, eliminating the thickness mutation region at the junction of the dielectric layer and the copper layer, thereby suppressing the optical refraction differences caused by the change in the ink thickness gradient and significantly improving the regional color difference patches. The applied leveling pressure further optimizes the surface topography of the solder mask layer to reach sub - micron - level flatness. This not only ensures the spatial consistency of the etching rate during the subsequent development process, making the etching profiles of the side walls of the matrix - type solder mask openings vertically symmetric, but also avoids the problem of solder pad bonding misalignment caused by the deviation of the opening size. In addition, the inherently dense structure of the dry - film solder mask layer forms a continuous protective layer after film formation, and its scratch - resistant performance is significantly improved compared with the traditional liquid curing layer, which can resist mechanical damage during subsequent processing operations, the risk of foreign matter adhesion and scratching during the operation process, and maintain the integrity of the optical performance of the circuit board surface. This solution synchronously solves the three major technical bottlenecks of thickness uniformity, surface flatness, and mechanical protection during the film - forming stage of the solder mask layer through material form innovation and process collaborative optimization, meeting the extreme requirements of mini - LED for the micro - area optical characteristics and structural accuracy of the circuit board.

[0028] In the apparent color difference detection step, local area color distribution prototype feature aggregation analysis based on the feature domain is performed on the miniLED printed circuit board to determine whether there is a color difference problem with the miniLED printed circuit board. It should be understood that the processing technology of the MiniLED printed circuit board is complex, covering multiple key steps such as substrate preparation, vacuum laminating, solder mask exposure, and development. Even with an optimized process plan, due to factors such as subtle differences in material properties, equipment precision limitations, and process parameter fluctuations, the finished product may still have problems with apparent color differences. Therefore, by implementing efficient apparent color difference detection, products with color difference problems can be identified in a timely manner, preventing defective circuit boards from flowing into subsequent processes or the final market, ensuring product quality, and improving the yield rate and market competitiveness. However, the regional color differences caused by uneven thickness of the liquid solder mask layer often exhibit microscopic scale characteristics (such as chromaticity shifts corresponding to nanoscale optical path differences), and manual visual inspection or conventional optical detection equipment is limited by the human eye resolution limit and equipment sampling precision, making it difficult to capture such sub-micron scale optical anomalies. Even more severe is the complex surface structure formed by tens of thousands of solder mask openings arrays (each containing 6 micron-sized pads) on the miniLED circuit board, which makes it extremely easy for the overall chromaticity analysis used in traditional detection methods to mask local abnormal areas. Currently, existing technologies usually rely on simple color difference algorithms based on fixed threshold determination and cannot dynamically adapt to the natural chromaticity distribution differences caused by process fluctuations in different batches of products, resulting in a high false positive rate. The existence of such detection blind spots makes it impossible to effectively identify problems such as refractive color differences caused by thickness gradient changes in the solder mask layer and geometric optical distortions caused by opening etching deviations during the manufacturing process, ultimately resulting in visible color patches or bright and dark stripes on the terminal display panel. Based on this, the technical concept of this application is to first preprocess the surface image of the solder mask layer collected by the AOI device with high resolution to optimize the recognizability of microscopic features, and then divide the entire board image into several independent detection areas according to the geometric distribution of the solder mask openings arrays. For each microarea, a multi-dimensional color feature vector containing chromaticity and brightness information is extracted. Through self-learning enhanced prototype feature extraction, a color distribution prototype model of the current batch of products is dynamically constructed, and the spatial distance difference between the feature vectors of each area and the prototype model is calculated. Finally, a quantitative index reflecting the color consistency of the entire board is generated. This intelligent comparison mechanism based on a dynamic reference can not only eliminate the interference of natural chromaticity distribution caused by process parameter fluctuations, but also capture local anomalies that are difficult to detect by traditional methods, thereby intercepting macroscopic display defects caused by refractive color differences or structural distortions at the microscopic scale.

[0029] Figure 3 It is a flowchart of the apparent color difference detection step in the apparent color difference control method for miniLED printed circuit board processing according to an embodiment of the present application. As Figure 3As shown, the apparent color difference detection steps include: S41, feeding the miniLED printed circuit board into the AOI device to collect the surface image of the PCB solder mask layer by using the AOI device; S42, performing image preprocessing on the surface image of the PCB solder mask layer to obtain the preprocessed surface image of the PCB solder mask layer; S43, performing region division and color feature extraction on the preprocessed surface image of the PCB solder mask layer to obtain a set of color feature coding vectors of the surface region of the PCB solder mask layer; S44, performing self-learning optimized color distribution prototype feature extraction on the set of color feature coding vectors of the surface region of the PCB solder mask layer to obtain the surface color distribution modeling vector of the PCB solder mask layer; S45, determining whether there is a color difference problem with the miniLED printed circuit board based on the difference representation between the set of color feature coding vectors of the surface region of the PCB solder mask layer and the surface color distribution modeling vector of the PCB solder mask layer.

[0030] In step S41, the miniLED printed circuit board is fed into the AOI device to collect the surface image of the PCB solder mask layer by using the AOI device. It should be understood that in the surface image of the PCB solder mask layer collected by the AOI device, it mainly contains the optical reflection characteristic information of the solder mask layer surface, specifically manifested as the chromaticity value distribution, brightness change and microscopic texture characteristics of different regions. Specifically, this image records the optical response difference caused by the thickness change of the solder mask material at the junction of the dielectric layer and the copper layer, and at the same time captures the geometric shape of the solder mask opening array and the sharpness of its edges. Through high-resolution imaging, the image can reflect the subtle chromaticity shift caused by the nanoscale optical path difference, as well as the local reflectivity change caused by the uneven thickness or surface flatness difference of the solder mask layer. In short, by obtaining the surface image of the PCB solder mask layer, it can provide the original data basis for subsequent image processing and analysis, enabling the apparent color difference at the microscopic scale to be accurately identified by the model.

[0031] In step S42, image preprocessing is performed on the surface image of the PCB solder mask layer to obtain the surface image of the preprocessed PCB solder mask layer. Accordingly, considering that the microscopic color difference on the surface of the solder mask layer is often caused by differences in ink thickness, etching deviation, or mechanical damage, its optical characteristics may be weakened or masked in the images collected by the original AOI device due to uneven ambient lighting, equipment noise interference, or surface reflection effects. For example, the chromaticity shift corresponding to the nanoscale optical path difference caused by the thickness gradient change of the liquid solder mask layer may not be effectively resolved in the original image due to pixel blurring or local overexposure. In addition, the complex solder mask opening array on the circuit board surface will generate artifacts due to geometric structure shadows or edge diffraction effects, further interfering with the extraction of true color features. If the original image is directly subjected to region division and feature analysis, it is very easy to misjudge local abnormal regions as normal due to insufficient image signal-to-noise ratio or loss of details, or to amplify normal process fluctuations as defects. Therefore, in this application, image preprocessing is performed on the surface image of the PCB solder mask layer to obtain the surface image of the preprocessed PCB solder mask layer. Specifically, in an example of this application, it specifically includes operations such as multi-scale filtering denoising, adaptive contrast enhancement, and non-uniform illumination correction. The denoising process uses a hybrid filtering algorithm based on wavelet transform to suppress high-frequency noise while retaining microscopic details; contrast enhancement is achieved through local histogram equalization, focusing on strengthening the chromaticity gradient region formed by the difference in ink thickness on the surface of the solder mask layer; illumination correction uses background modeling technology to eliminate the influence of uneven spatial distribution of the device light source on color features. These processes work together to clearly present the microscopic color difference features on the surface of the solder mask layer (such as the chromaticity gradient corresponding to refractive chromatic aberration and the edge color deviation caused by etching distortion), while eliminating artifacts introduced by environmental interference or device inherent characteristics. In this way, microscopic features in the image, such as the subtle color difference on the surface of the solder mask layer, can be highlighted, which helps to accurately extract color features and discover potential color difference problems in the subsequent process.

[0032] In step S43, region division and color feature extraction are performed on the surface image of the preprocessed PCB solder mask layer to obtain a set of PCB solder mask layer surface region color feature coding vectors. Specifically, Figure 4 FIG. is a flowchart of S43 in the apparent color difference control method for miniLED printed circuit board processing according to an embodiment of the present application. As Figure 4 shown, step S43 includes: S431, performing region division on the surface image of the preprocessed PCB solder mask layer to obtain a set of surface region images of the PCB solder mask layer; S432, respectively extracting the color features of each surface region image in the set of surface region images of the PCB solder mask layer to obtain a set of PCB solder mask layer surface region color feature coding vectors.

[0033] In step S431, the preprocessed surface image of the PCB solder mask layer is divided into regions to obtain a set of surface region images of the PCB solder mask layer. Correspondingly, considering that the solder mask opening array (each containing 6 micron-level pads) distributed on the surface of the solder mask layer forms highly dense and complex microscopic geometric features, such structures are likely to cause local anomalies in optical detection to be masked by global chromaticity analysis. For example, the nanoscale chromaticity shift caused by uneven ink thickness or the edge color difference caused by opening etching deviation may only exist in specific opening units or the transition regions between adjacent openings. However, the overall image analysis used in traditional detection methods will weaken or even ignore such local defects due to global data averaging. In addition, the natural chromaticity distribution differences generated by process fluctuations on the surface of the solder mask layer often show regional characteristics. If feature extraction is directly performed on the entire image, the accurate positioning of abnormal signals will be interfered by the chromaticity superposition effect in different regions, resulting in a decrease in detection sensitivity and accuracy. Based on this, in the technical solution of this application, the preprocessed surface image of the PCB solder mask layer is divided into regions to obtain a set of surface region images of the PCB solder mask layer. That is, in this step, the preprocessed high-resolution surface image of the solder mask layer is divided into several independent detection regions according to the geometric distribution law of the solder mask opening array, and each region corresponds to a single or a group of opening structures, ensuring the spatial matching between the detection unit and the physical structure. During the division process, an edge detection algorithm based on morphological operations is used to accurately locate the opening boundaries, and the adaptive grid generation technology is combined to dynamically adjust the region division granularity, which not only ensures the integrity of the opening features (such as the layout integrity of 6 pads) in each region but also avoids feature fragmentation caused by over-segmentation. Through this operation, the originally complex entire board surface is deconstructed into multiple independent detection units, providing a structured data basis for subsequent local feature extraction.

[0034] In step S432, the color features of each surface region image of the PCB solder mask layer in the set of surface region images of the PCB solder mask layer are respectively extracted to obtain a set of surface region color feature coding vectors of the PCB solder mask layer. Specifically, in the embodiment of this application, step S432 includes: using a ResNet-based color extractor to respectively extract the color features of each surface region image of the PCB solder mask layer in the set of surface region images of the PCB solder mask layer to obtain a set of surface region color feature coding vectors of the PCB solder mask layer.

[0035] Accordingly, considering that after the surface image of the solder mask layer of the miniLED printed circuit board is divided into regions, each region may contain different color information details. Due to factors such as subtle differences in material properties, equipment precision limitations, and process parameter fluctuations, the color characteristics of these regions will be different, and these differences may be the key factors leading to apparent color differences. Therefore, in this application, the color characteristics of each PCB solder mask layer surface region image in the set of PCB solder mask layer surface region images are extracted respectively to obtain a set of PCB solder mask layer surface region color feature encoding vectors. In particular, in an example of this application, a ResNet-based color extractor is used to extract the color characteristics of each PCB solder mask layer surface region image in the set of PCB solder mask layer surface region images respectively to obtain a set of PCB solder mask layer surface region color feature encoding vectors. Specifically, the ResNet-based color extractor has a deep convolutional neural network structure, which can automatically learn complex features in the image and accurately extract color information at the microscopic level. Through its multi-scale feature learning ability of the deep convolutional neural network, the chromaticity distribution characteristics of local regions are analyzed. The residual connection structure of ResNet effectively alleviates the problem of gradient disappearance in the training of deep networks, enabling it to separate low-frequency chromaticity gradients and high-frequency chromaticity mutation features from high-resolution regional images. Specifically, the network aggregates color responses under different receptive fields step by step through multiple residual blocks. For example, shallow convolutional kernels capture local chromaticity mutations (such as the edges of color patches caused by uneven ink thickness), and deep convolutional kernels extract the global chromaticity distribution trend (such as the natural chromaticity reference shift within a batch). The finally output encoding vector not only contains the dominant statistical features (such as chromaticity mean, variance) in the LAB color space, but also implicitly encodes the non-linear relationship between optical path difference and material optical properties at the microscopic scale, forming a highly discriminative color representation.

[0036] The following is a detailed elaboration of a specific implementation process of "using a ResNet-based color extractor to extract the color characteristics of each PCB solder mask layer surface region image in the set of PCB solder mask layer surface region images respectively to obtain a set of PCB solder mask layer surface region color feature encoding vectors":

[0037] First, perform preprocessing and normalization on the regional images. After the regional division of the surface image of the PCB solder mask layer, the obtained regional images may have different resolutions and color space characteristics, and they need to be uniformly normalized to meet the input requirements of the ResNet network. Since the ResNet network is usually optimized for input images of a specific size during design, the resolution of each regional image needs to be adjusted to a fixed size, such as 224×224 pixels. During the resolution adjustment process, the bilinear interpolation algorithm is used to scale the image. This algorithm calculates the weighted average of the gray values of the neighboring pixels around the target pixel, which can effectively reduce the jagged edges or color distortion phenomena that occur during image scaling and retain the color detail information of the original image to the greatest extent. After the resolution adjustment is completed, the image needs to be normalized. The pixel values are normalized from the original range of [0,255] to the interval of [0,1]. The specific operations include calculating the mean and standard deviation of the training dataset, and then subtracting the corresponding mean from the R, G, and B channel values of each pixel and dividing by the standard deviation to eliminate the brightness and contrast deviations caused by differences in shooting lighting conditions and device parameters between different images, ensuring the distribution consistency of the input data and providing standardized input data for subsequent feature extraction. Next, construct a color feature extraction model based on ResNet. Considering the excellent performance of the ResNet network in image feature extraction tasks, especially its residual connection structure can effectively alleviate the gradient vanishing problem in deep neural networks, thus supporting the network to learn more complex feature representations. Select the pre-trained ResNet-50 as the basic architecture. This model has been pre-trained on large-scale datasets such as ImageNet and has learned rich underlying visual features of images, including edges, textures, color distributions, etc. These features have important transfer learning value for the extraction of the surface color features of the PCB solder mask layer. During the model construction process, remove the top fully connected layer used for image classification in the original network because the target task is feature extraction rather than classification, and retain all network structures from the input layer to the last convolutional layer, including convolutional layers, pooling layers, and residual connection modules. These network levels from the bottom to the middle can effectively extract the basic color features in the image and their spatial combination patterns, while the deep network is responsible for capturing the global color distribution trend and complex feature mappings, thus forming a multi-scale color feature representation.

[0038] In the feature extraction stage, the normalized regional images are sequentially input into the constructed ResNet network for forward propagation calculation. The image first passes through shallow convolution layers, such as the 7×7 convolution layer and the 3×3 maximum pooling layer in the first convolution block. These layers extract basic color features such as edges, corners, and monochrome blocks in the image through convolution kernels of different sizes, such as identifying the color block area with clear boundaries formed by the difference in ink thickness on the surface of the solder mask. As the network level deepens, the middle convolution layer, such as the residual block of the bottleneck structure, begins to integrate the color information of the local area. Through the parallel processing of multiple convolution kernels, the spatial contextual relationship between color blocks is captured, such as the gradient transition of adjacent color blocks or the pattern of specific color combination, forming a color feature representation with local semantics. Deep convolution layers, such as the last few residual blocks, are responsible for processing the color distribution in the global range, integrating the color information of the whole image through a larger receptive field, and identifying large-area color gradient areas or specific color aggregation patterns, such as the overall chromaticity shift trend caused by process fluctuations on the surface of the solder mask. The unique residual connection structure of ResNet allows the network to directly learn the residual mapping between input and output, avoiding the problem of reduced feature learning ability caused by gradient attenuation in deep networks, so that the network can effectively capture the subtle color changes on the surface of the solder mask layer caused by nanometer-level optical path difference, such as the weak chromaticity difference at the junction of the dielectric layer and the copper layer. In the generation of feature encoding vectors, a global average pooling layer is added after the last convolutional layer of the ResNet network to convert the feature map of the spatial dimension into a one-dimensional feature vector. The feature vector usually has 2048 dimensions, and each dimension corresponds to a color-related feature learned by the network, such as the reflectivity difference at a specific wavelength, the intensity ratio between RGB color channels, and the texture entropy value of the color distribution. These features are not direct color space values ​​(such as Lab or HSV color values), but abstract feature representations formed after multiple layers of nonlinear transformations, which can more sensitively reflect the microscopic color differences on the surface of the solder mask layer. For each input region image, after layer-by-layer feature extraction by the network, a unique feature vector is finally output in the global average pooling layer. This vector integrates the color information of all pixels in the image and their spatial distribution characteristics to form a high-level abstract encoding of the color features of the region.

[0039] After completing the feature extraction of all regional images, the generated feature vectors need to be integrated to construct a set of color feature encoding vectors of the surface area of ​​the PCB solder mask layer. According to the spatial position order when the area is divided, the feature vectors corresponding to each regional image are stored in the data structure in sequence to ensure that the feature vectors correspond to the position information of the original regional image one by one. Each vector in the set not only contains the color feature information of the corresponding area, but also implies its spatial position attribute in the entire board, providing a structured data basis for the subsequent color distribution prototype feature extraction.

[0040] In particular, during the actual application process, attention should be paid to the training and optimization of the model. Although a pre-trained model is used for initialization, due to the difference in the color feature distribution between the PCB solder mask layer images and natural images, it is necessary to fine-tune the model using the PCB image dataset in actual production. The specific operations include freezing the parameters of the shallow network and only training the middle and deep networks to retain the basic feature extraction ability learned by the pre-trained model while adapting to the color feature distribution in a specific domain. Data augmentation techniques such as random flipping, rotation, and brightness adjustment can be used to expand the training dataset and improve the generalization ability of the model to ensure the stability of PCB image feature extraction under different batches and different process conditions. In addition, the configuration of the hardware environment has an important impact on the implementation effect. Since the ResNet network involves a large number of convolutional operations and parameters, a high-performance GPU (such as a graphics processing unit) needs to be used for accelerated computing to meet the efficiency requirements of real-time or batch processing.

[0041] In step S44, self-learning optimization of the color distribution prototype feature extraction is performed on the set of color feature encoding vectors of the PCB solder mask layer surface area to obtain the PCB solder mask layer surface color distribution modeling vector. Specifically, Figure 5 FIG. is a flowchart of S44 in the apparent color difference control method for miniLED printed circuit board processing according to an embodiment of the present application. As Figure 5 shown, step S44 includes: S44-1, performing linear clustering analysis on the set of color feature encoding vectors of the PCB solder mask layer surface area to obtain the initial linear clustering center encoding vector of the PCB solder mask layer surface area color feature; S44-2, performing clustering compensation component analysis on the set of color feature encoding vectors of the PCB solder mask layer surface area to obtain the linear clustering compensation component encoding vector of the PCB solder mask layer surface area color feature; S44-3, fusing the linear clustering compensation component encoding vector of the PCB solder mask layer surface area color feature and the initial linear clustering center encoding vector of the PCB solder mask layer surface area color feature to obtain the PCB solder mask layer surface color distribution modeling vector.

[0042] It should be understood that the complexity of the color characteristics of the solder mask surface stems from the nonlinear coupling of process fluctuations and microscopic defects. Traditional methods use static thresholds or fixed color difference models for judgment, which cannot adapt to the changes in natural chromaticity distribution caused by differences in material properties, equipment parameter drift or environmental interference in different batches of products. For example, the color shift caused by the thickness gradient of the liquid solder mask and the change in scattering characteristics caused by the window etching deviation may present a highly nonlinearly correlated characteristic distribution in the color space, and simple linear clustering or global statistics are difficult to capture such complex patterns. More importantly, the color difference signals of local micro-areas in tens of thousands of solder mask window arrays (such as nanometer-level optical path differences caused by excessive ink thickness in a single window) are easily masked by global averaging analysis, resulting in missed defects. The existing technology lacks the ability to decouple linear and nonlinear components in the feature space, and cannot build a benchmark model that dynamically adapts to process fluctuations, and it is difficult to distinguish normal chromaticity fluctuations from real defect characteristics. Therefore, in order to capture the distribution pattern of color on the entire solder mask surface, the present application obtains the PCB solder mask surface color distribution modeling vector by performing self-learning optimization on the set of color feature coding vectors of the PCB solder mask surface area.

[0043] Specifically, firstly, the initial color distribution prototype (characterizing the main chromaticity trend within the batch) is extracted based on linear clustering to construct the linear skeleton of the global color distribution; then, the nonlinear association between the features of each region and the global prototype is mined through deep collaborative coding to generate a deep implicit vector reflecting the details of local color difference; then, the feature compensation increment is calculated to correct the local deviation of the linear prototype, for example, for the area with abnormal etching at the edge of the window or the sudden change of ink thickness, the local color difference sensitivity is enhanced through nonlinear compensation; finally, the linear prototype and the compensation component are fused to generate a dynamic color distribution modeling vector. This vector not only retains the macroscopic chromaticity benchmark of the batch product, but also embeds the chromaticity elasticity range allowed by the process fluctuation through the compensation mechanism, and at the same time strengthens the ability to capture nonlinear defect features. In this way, the color distribution modeling vector of the surface of the PCB solder mask layer is obtained in order to establish a model that can accurately describe the color distribution of the surface of the solder mask layer. This model can be used as a benchmark for subsequent judgment of apparent color difference. By comparing the actual measured color data with the model, it can quickly and accurately detect whether there is color difference and the degree of color difference.

[0044] Specifically, in the embodiment of the present application, step S44-1 includes: performing a linear cluster analysis on a set of color feature coding vectors of the surface area of ​​the PCB solder mask layer to obtain an initial linear cluster center coding vector of the color feature of the surface area of ​​the PCB solder mask layer, which can be expressed as follows:

[0045] X={x1,x2,...,x i ,...,x n}

[0046]

[0047] Among them, X is a set of color feature coding vectors of the surface area of the PCB solder mask layer, and x1, x2, x i and x n are respectively the 1st, 2nd, ith, and nth color feature coding vectors of the surface area of the PCB solder mask layer in the set of color feature coding vectors of the surface area of the PCB solder mask layer. n is the number of vectors in X, and x c is the initial linear clustering center coding vector of the color features of the surface area of the PCB solder mask layer.

[0048] It should be understood that the surface color distribution of the solder mask layer is affected by process fluctuations such as material differences and equipment drift, and traditional static thresholds cannot adapt to chromaticity non-linear offsets. By performing unsupervised grouping on the set of color feature coding vectors of the surface area of the PCB solder mask layer through linear clustering, the dominant chromaticity trend within the batch can be extracted as the global skeleton. This processing assumes that although the coupling of microscopic defects and process fluctuations leads to a complex distribution, there is still a low-rank structure that can be linearly approximated at the macroscopic level, such as the chromaticity principal component caused by the thickness gradient of the liquid solder mask. That is, by using the algorithm to iteratively optimize the within-cluster distance, the initial linear clustering center coding vector of the color features of the surface area of the PCB solder mask layer is extracted, and the high-dimensional data is projected onto the linear subspace to explicitly decouple the linear principal component (such as the batch chromaticity reference) from the subsequent non-linear details that need to be compensated (such as local etching deviation). The generated linear clustering center coding vector serves as a global anchor point, which can provide a structured reference for subsequent non-linear correction and avoid the information redundancy of directly processing the full amount of high-dimensional data.

[0049] Specifically, Figure 6 is a flowchart of S44-2 in the apparent color difference control method for miniLED printed circuit board processing according to an embodiment of the present application. As Figure 6As shown in step S44-2, it includes: S44-21, constructing the PCB solder mask layer surface area color feature depth collaborative implicit coding vectors between each PCB solder mask layer surface area color feature coding vector in the set of PCB solder mask layer surface area color feature coding vectors and the PCB solder mask layer surface area color feature initial linear clustering center coding vector; S44-22, based on the PCB solder mask layer surface area color feature depth collaborative implicit coding vectors, calculating the feature clustering compensation increment operators of each PCB solder mask layer surface area color feature coding vector in the set of PCB solder mask layer surface area color feature coding vectors relative to the PCB solder mask layer surface area color feature initial linear clustering center coding vector to obtain the set of PCB solder mask layer surface area color feature clustering compensation increment operators; S44-23, based on the set of PCB solder mask layer surface area color feature clustering compensation increment operators, calculating the PCB solder mask layer surface area color feature linear clustering compensation component coding vectors of the set of PCB solder mask layer surface area color feature coding vectors. More specifically, in the embodiment of the present application, step S44-21 includes: constructing the PCB solder mask layer surface area color feature depth collaborative implicit coding vectors between each PCB solder mask layer surface area color feature coding vector in the set of PCB solder mask layer surface area color feature coding vectors and the PCB solder mask layer surface area color feature initial linear clustering center coding vector, which can be expressed by the following formula:

[0050] r i = Sigmoid{W i [concat(x i ; x c ) + b i}

[0051] Wherein, concat(·;·) is the concatenation operation, W i is the i-th learnable collaborative weight matrix among multiple learnable collaborative weight matrices, b i is the i-th collaborative bias vector among multiple collaborative bias vectors, Sigmoid is the activation function, and r i is the PCB solder mask layer surface area color feature depth collaborative implicit coding vector between x i and x c .

[0052] It should be understood that microscopic defects such as excessive ink thickness in a single window appear as nonlinear deviations from the global linear skeleton in color space, and a deep model is needed to capture the complex association between local features and cluster centers. For example, changes in scattering characteristics caused by abnormal etching at the edge of the window may only appear in a specific context. That is, a deep neural network is used to interactively model the color feature encoding vector of the surface area of ​​the PCB solder mask layer and the initial linear cluster center encoding vector of the color feature of the surface area of ​​the PCB solder mask layer, and learn the nonlinear synergistic relationship between the two to mine the context-sensitive features of local color difference relative to the global benchmark, and generate a deep synergistic implicit encoding vector of the color feature of the surface area of ​​the PCB solder mask layer. This step can enhance the ability to characterize nonlinear patterns such as sudden changes in ink thickness, thereby compensating for the smoothing effect of linear clustering on local details.

[0053] More specifically, in the embodiment of the present application, step S44-22 includes: performing nonlinear activation processing based on the sigmoid function on the initial linear cluster center encoding vector of the color feature of the surface area of ​​the PCB solder mask layer to obtain the initial linear cluster center activation encoding vector of the color feature of the surface area of ​​the PCB solder mask layer. The process can be expressed by the formula:

[0054] v c =Sigmoid(x c )

[0055] Among them, v c It is the initial linear cluster center activation encoding vector of the color feature of the surface area of ​​the PCB solder mask layer;

[0056] The initial linear cluster center activation coding vector of the PCB solder mask surface area color feature and the PCB solder mask surface area color feature deep collaborative implicit coding vector corresponding to the PCB solder mask surface area color feature coding vector are compensated based on the eigenvalue granularity to obtain the PCB solder mask surface area color feature clustering intermediate compensation variables. The process can be expressed by the formula:

[0057]

[0058] Among them, r ik Yes i The kth eigenvalue in , log2 is the logarithmic function value with base 2, v ck Yes c The kth eigenvalue in, D is r i and v c The length of the vector, and r i and v c The same length, λ i is x i The corresponding PCB solder mask surface area color feature clustering intermediate compensation variables;

[0059] Perform compensation collaborative correction based on fractal resonance coupling on the intermediate compensation variable of the color feature clustering of the PCB solder mask layer surface area to obtain the intermediate compensation correction variable of the color feature clustering of the PCB solder mask layer surface area. This process can be expressed by the formula:

[0060] Σ i = <x i , r i >

[0061]

[0062] where <,> represents the vector inner product, Σ i is the fractal coupling covariate of the color feature of the PCB solder mask layer surface area corresponding to x i , σ(x i - r i ) 2 represents the variance of the vector (x i - r i ), L is the length of the vector (x i - r i ), ∈ i is the fractal scattering perturbation factor of the color feature of the PCB solder mask layer surface area corresponding to x i , Γ i is the relaxation i kinetic coefficient of the color feature of the PCB solder mask layer surface area corresponding to x

[0063] λ′ i is the intermediate compensation correction variable of the color feature clustering of the PCB solder mask layer surface area corresponding to x i ;

[0064] Perform normalization processing based on softmax on the intermediate compensation correction variable of the color feature clustering of the PCB solder mask layer surface area to obtain the clustering compensation increment operator of the color feature of the PCB solder mask layer surface area corresponding to the color feature coding vector. This process can be expressed by the formula:

[0065] ε i = softmax(λ′ i )

[0066] where softmax is the normalization function, and ε i is the clustering compensation increment operator of the color feature of the PCB solder mask layer surface area corresponding to x i .

[0067] Specifically, in the calculation process of the clustering compensation increment operator of the color feature of the PCB solder mask layer surface area, due to the color feature coding vector x of the PCB solder mask layer surface areai On the basis of linear clustering, a non-linear incremental expansion of linear clustering is superimposed, resulting in topological instability of the clustering space generated within the feature space. To correct this problem, first, through the color feature coding vector x of the surface area of the PCB solder mask layer i and the depth co-implication coding vector r of the color feature of the surface area of the PCB solder mask layer i The inner product <x i , r i > is used to construct the fractal coupling co-variable Σ of the color feature of the surface area of the PCB solder mask layer i , and the relaxation kinetic coefficient Γ of the color feature of the surface area of the PCB solder mask layer, which characterizes the energy dissipation in the fractal space, is calculated based on the following formula i :

[0068]

[0069] where σ(x i - r i ) 2 represents the variance of the vector (x i - r i ), L is the length of the feature vector, ∈ i is the fractal scattering perturbation factor of the color feature of the surface area of the PCB solder mask layer related to the fractal coupling co-variable Σ of the color feature of the surface area of the PCB solder mask layer i , and Γ i its physical meaning characterizes the scattering energy dissipation effect caused by the topological characteristics of fractal geometry in the non-integer dimensional space. Finally, through the resonant coupling relationship λ′ i = λ i × e Γ i×L -1 / 2 , the resonant coupling strength of the energy distribution in the fractal space is enhanced, thereby establishing a non-linear incremental collaborative correction mechanism, and significantly improving the convergence stability and generalization performance of the clustering compensation incremental operator of the color feature of the surface area of the PCB solder mask layer in the non-Euclidean space.

[0070] It should be understood that linear clustering is prone to masking local anomalies (such as nanoscale optical path difference signals) due to global averaging, and personalized compensation needs to be generated based on deep collaborative coding. For example, the ink thickness fluctuations in different regions of the same batch need to be differentially corrected. That is, for each color feature coding vector of the PCB solder mask layer surface area, the offset relative to the initial linear clustering center coding vector of the PCB solder mask layer surface area color feature is calculated using its PCB solder mask layer surface area color feature deep collaborative implicit coding vector, and a clustering compensation increment operator for the PCB solder mask layer surface area color feature is generated. This operator is essentially a non-linear adjustment coefficient. For example, it enhances the amplitude of the color difference signal in the open window etching area or elastically suppresses the chromaticity offset within the normal process fluctuation range. In this way, the local sensitivity can be adaptively improved, enabling the model to distinguish real defects from allowable fluctuations.

[0071] More specifically, in the embodiment of the present application, step S44-23 includes: calculating the linear clustering compensation component coding vector of the PCB solder mask layer surface area color feature for the set of PCB solder mask layer surface area color feature coding vectors based on the set of clustering compensation increment operators for the PCB solder mask layer surface area color feature, which can be expressed by the following formula:

[0072]

[0073] where x b is the linear clustering compensation component coding vector of the PCB solder mask layer surface area color feature.

[0074] It should be understood that individual increment operators need to remove random noise and extract common compensation patterns, such as systematic chromaticity offsets caused by equipment parameter drift in multi-batch data. That is, by globally aggregating the set of clustering compensation increment operators for the PCB solder mask layer surface area color feature, the linear clustering compensation component coding vector of the PCB solder mask layer surface area color feature is extracted. For example, for the overall chromaticity distribution offset caused by environmental interference between batches, the aggregated component coding vector can represent the direction and intensity of this offset. In this way, the scattered local correction information can be elevated to global compensation features, eliminating the risk of overfitting while retaining the elastic range of process fluctuations.

[0075] Specifically, in the embodiment of the present application, step S44-3 includes: fusing the linear clustering compensation component coding vector of the PCB solder mask layer surface area color feature and the initial linear clustering center coding vector of the PCB solder mask layer surface area color feature to obtain a modeling vector for the PCB solder mask layer surface color distribution, which can be expressed by the following formula:

[0076] v f =α·x b +β·x c

[0077] where α and β are the weighted hyperparameters corresponding to x b and x c respectively, and v f is the modeling vector of the surface color distribution of the PCB solder mask layer. It should be understood that the traditional method lacks the decoupling ability of linear and non-linear components, resulting in the inability to dynamically adapt to process fluctuations. For example, it is necessary to simultaneously retain the batch main chromaticity trend (linear) and the local defect sensitivity (non-linear). Through the gated fusion mechanism, the linear clustering compensation component coding vector of the surface area color characteristics of the PCB solder mask layer and the initial linear clustering center coding vector of the surface area color characteristics of the PCB solder mask layer can be effectively combined. For example, a learnable weight is used to balance the macro chromaticity reference and the micro compensation intensity, so that the generated modeling vector of the surface color distribution of the PCB solder mask layer contains both the main chromaticity gradient information of the solder mask layer thickness and the discriminant features of the window etching abnormality. That is, the model can achieve the balance of robustness and sensitivity of defect detection through the dynamic fusion mechanism, meeting the adaptive requirements of cross-batch process fluctuations. In step S45, based on the difference representation between the set of the surface area color feature coding vectors of the PCB solder mask layer and the modeling vector of the surface color distribution of the PCB solder mask layer, it is determined whether there is a color difference problem in the miniLED printed circuit board. Specifically, Figure 7 FIG. is a flowchart of S45 in the apparent color difference control method for miniLED printed circuit board processing according to an embodiment of the present application. As Figure 7 shown, step S45 includes: S451, calculating the color difference values between each surface area color feature coding vector in the set of the surface area color feature coding vectors of the PCB solder mask layer and the modeling vector of the surface color distribution of the PCB solder mask layer to obtain a surface color consistency coding vector of the PCB solder mask layer composed of multiple color difference values as the difference representation; S452, based on the surface color consistency coding vector of the PCB solder mask layer, determining whether there is a color difference problem in the miniLED printed circuit board.

[0078] In step S451, color difference values between each PCB solder mask layer surface area color feature coding vector in the set of PCB solder mask layer surface area color feature coding vectors and the PCB solder mask layer surface color distribution modeling vector are calculated to obtain a PCB solder mask layer surface color consistency coding vector composed of multiple color difference values as a difference representation. It should be understood that the PCB solder mask layer surface color distribution modeling vector represents the overall characteristics and rules of the entire solder mask layer surface color distribution. The color feature coding vectors of each area reflect local color information. Calculating the difference value between the two can measure the deviation degree of the color feature of each local area from the overall color distribution, so as to understand the color performance of each local area under the framework of the overall color distribution. Because in a miniLED printed circuit board, even if the overall color distribution meets certain standards, color difference problems may exist in local areas, and these local color differences may affect the final display effect, so it is necessary to evaluate the difference between the local and the whole. The obtained PCB solder mask layer surface color consistency coding vector can be used as an important basis for judging whether there is a color difference problem in the miniLED printed circuit board. In step S452, based on the PCB solder mask layer surface color consistency coding vector, it is determined whether there is a color difference problem in the miniLED printed circuit board. Specifically, in the embodiment of the present application, step S452 includes: inputting the PCB solder mask layer surface color consistency coding vector into a color difference discriminator based on a classifier to determine whether there is a color difference problem in the miniLED printed circuit board. Correspondingly, the PCB solder mask layer surface color consistency coding vector obtained through the previous steps is a set of quantified data, which reflects the difference between the color features of each area and the overall color distribution model. However, these data themselves cannot intuitively describe whether there is a color difference problem in the circuit board and need further processing and analysis. A classifier is a trained model that can perform pattern recognition and classification according to the input data features. In the color difference judgment of a miniLED printed circuit board, a color difference discriminator based on a classifier can learn a large amount of color consistency coding vector data of circuit boards with known color difference problems, so as to establish accurate classification rules. Compared with manual judgment or simple threshold comparison, the classifier can more comprehensively and accurately consider various information in the color consistency coding vector, avoid the limitations of subjective factors and simple rules, and improve the accuracy and reliability of judgment. For this reason, after determining whether there is a color difference problem in the circuit board, it can provide clear guidance for the production process. If it is determined that there is a color difference problem, production personnel can further analyze to find the areas with relatively large differences, so as to trace back to the possible problem links in the production process, such as the coating of solder mask ink, exposure and development and other process steps, and make targeted adjustments and optimizations to reduce or eliminate the color difference problem and improve production efficiency and product quality.

[0079] In summary, an apparent color difference control method for mini-LED printed circuit board processing based on the embodiments of the present application is elucidated. First, a circuit board substrate including a dielectric layer and a copper layer is provided. Subsequently, a dry film type solder resist ink is covered on the surface of the substrate through a vacuum laminating process, and it is firmly adhered by using vacuum negative pressure adsorption. At the same time, pressure is applied for leveling treatment to obtain the laminated circuit board. Then, the laminated circuit board is subjected to solder resist exposure and development, and finally a mini-LED printed circuit board is made. To control the problem of apparent color difference, a local area color distribution prototype feature aggregation analysis technology based on the feature domain is used to detect the printed circuit board to determine whether there is a color difference problem. In this way, the production quality of the mini-LED printed circuit board can be effectively improved.

Claims

1. A method for controlling apparent color difference in the processing of mini-LED printed circuit boards, characterized in that, Including: Circuit board substrate preparation step: providing a circuit board substrate, the circuit board substrate including a dielectric layer and a copper layer stacked on the dielectric layer; Vacuum laminating step: laying a dry film type solder mask ink flat on the circuit board substrate, and using vacuum negative pressure adsorption to make the dry film type solder mask ink firmly adhere to the surface of the circuit board substrate, and performing leveling treatment on it by applying pressure to obtain a circuit board after vacuum laminating; Post-treatment step: performing solder mask exposure and development on the circuit board after vacuum laminating to obtain a miniLED printed circuit board; Wherein, the apparent color difference control method further includes an apparent color difference detection step: performing local area color distribution prototype feature aggregation analysis based on a feature domain on the miniLED printed circuit board to determine whether there is a color difference problem with the miniLED printed circuit board.

2. The apparent color difference control method for miniLED printed circuit board processing according to claim 1, wherein The vacuum laminating step includes: Laying the dry film type solder mask ink flat on the surface of the circuit board substrate; Using vacuum negative pressure adsorption to ensure that the dry film type solder mask ink can firmly and tightly adhere to the surface of the circuit board substrate; After vacuum negative pressure adsorption, performing leveling treatment on it by applying pressure to obtain the circuit board after vacuum laminating.

3. The apparent color difference control method for miniLED printed circuit board processing according to claim 1, wherein, The apparent color difference detection step includes: Feeding the miniLED printed circuit board into an AOI device to collect an image of the surface of the PCB solder mask layer by using the AOI device; Performing image preprocessing on the image of the surface of the PCB solder mask layer to obtain a preprocessed image of the surface of the PCB solder mask layer; Performing region division and color feature extraction on the preprocessed image of the surface of the PCB solder mask layer to obtain a set of color feature coding vectors for the surface regions of the PCB solder mask layer; Performing self-learning optimized color distribution prototype feature extraction on the set of color feature coding vectors for the surface regions of the PCB solder mask layer to obtain a color distribution modeling vector for the surface of the PCB solder mask layer; Based on the difference representation between the set of color feature coding vectors for the surface regions of the PCB solder mask layer and the color distribution modeling vector for the surface of the PCB solder mask layer, determining whether there is a color difference problem with the miniLED printed circuit board.

4. The apparent color difference control method for miniLED printed circuit board processing according to claim 3, wherein Performing region division and color feature extraction on the preprocessed image of the surface of the PCB solder mask layer to obtain a set of color feature coding vectors for the surface regions of the PCB solder mask layer, including: Performing region division on the preprocessed image of the surface of the PCB solder mask layer to obtain a set of images of the surface regions of the PCB solder mask layer; Respectively extracting the color features of each image of the surface region of the PCB solder mask layer in the set of images of the surface regions of the PCB solder mask layer to obtain the set of color feature coding vectors for the surface regions of the PCB solder mask layer.

5. The method for controlling the apparent color difference in the processing of mini-LED printed circuit boards according to claim 4, wherein, Respectively extracting the color features of each image of the surface region of the PCB solder mask layer in the set of images of the surface regions of the PCB solder mask layer to obtain the set of color feature coding vectors for the surface regions of the PCB solder mask layer, including: using a ResNet-based color extractor to respectively extract the color features of each image of the surface region of the PCB solder mask layer in the set of images of the surface regions of the PCB solder mask layer to obtain the set of color feature coding vectors for the surface regions of the PCB solder mask layer.

6. The apparent color difference control method for miniLED printed circuit board processing according to claim 3, wherein, Performing self-learning optimization of color distribution prototype feature extraction on the set of color feature encoding vectors of the surface area of the PCB solder mask layer to obtain the PCB solder mask layer surface color distribution modeling vector, including: Performing linear clustering analysis on the set of color feature encoding vectors of the surface area of the PCB solder mask layer to obtain the initial linear clustering center encoding vector of the color features of the surface area of the PCB solder mask layer; Performing clustering compensation component analysis on the set of color feature encoding vectors of the surface area of the PCB solder mask layer to obtain the linear clustering compensation component encoding vector of the color features of the surface area of the PCB solder mask layer; Fusing the linear clustering compensation component encoding vector of the color features of the surface area of the PCB solder mask layer and the initial linear clustering center encoding vector of the color features of the surface area of the PCB solder mask layer to obtain the PCB solder mask layer surface color distribution modeling vector.

7. The apparent color difference control method for miniLED printed circuit board processing according to claim 6, wherein, Performing clustering compensation component analysis on the set of color feature encoding vectors of the surface area of the PCB solder mask layer to obtain the linear clustering compensation component encoding vector of the color features of the surface area of the PCB solder mask layer, including: Constructing the deep collaborative implicit encoding vector of the color features of the surface area of the PCB solder mask layer between each color feature encoding vector in the set of color feature encoding vectors of the surface area of the PCB solder mask layer and the initial linear clustering center encoding vector of the color features of the surface area of the PCB solder mask layer; Based on the deep collaborative implicit encoding vector of the color features of the surface area of the PCB solder mask layer, calculating the feature clustering compensation increment operator of each color feature encoding vector in the set of color feature encoding vectors of the surface area of the PCB solder mask layer relative to the initial linear clustering center encoding vector of the color features of the surface area of the PCB solder mask layer to obtain the set of feature clustering compensation increment operators of the color features of the surface area of the PCB solder mask layer; Based on the set of feature clustering compensation increment operators of the color features of the surface area of the PCB solder mask layer, calculating the linear clustering compensation component encoding vector of the color features of the surface area of the set of color feature encoding vectors of the PCB solder mask layer.

8. The method for controlling the apparent color difference in the processing of mini-LED printed circuit boards according to claim 7, wherein Based on the deep collaborative implicit encoding vector of the color features of the surface area of the PCB solder mask layer, calculating the feature clustering compensation increment operator of each color feature encoding vector in the set of color feature encoding vectors of the surface area of the PCB solder mask layer relative to the initial linear clustering center encoding vector of the color features of the surface area of the PCB solder mask layer to obtain the set of feature clustering compensation increment operators of the color features of the surface area of the PCB solder mask layer, including: Performing non-linear activation processing on the initial linear clustering center encoding vector of the color features of the surface area of the PCB solder mask layer based on the sigmoid function to obtain the initial linear clustering center activation encoding vector of the color features of the surface area of the PCB solder mask layer; Perform compensation calculation based on eigenvalue granularity on the initial linear clustering center activation coding vector of the color feature of the PCB solder mask layer surface area and the depth collaborative implicit coding vector of the color feature of the PCB solder mask layer surface area corresponding to the color feature coding vector of the PCB solder mask layer surface area to obtain the intermediate compensation variable of the color feature clustering of the PCB solder mask layer surface area; Perform compensation collaborative correction based on fractal resonance coupling on the intermediate compensation variable of the color feature clustering of the PCB solder mask layer surface area to obtain the intermediate compensation correction variable of the color feature clustering of the PCB solder mask layer surface area; Perform normalization processing based on softmax on the intermediate compensation correction variable of the color feature clustering of the PCB solder mask layer surface area to obtain the clustering compensation increment operator of the color feature of the PCB solder mask layer surface area corresponding to the color feature coding vector of the PCB solder mask layer surface area.

9. The method for controlling the apparent color difference in the processing of mini-LED printed circuit boards according to claim 8, wherein Based on the difference representation between the set of color feature coding vectors of the PCB solder mask layer surface area and the color distribution modeling vector of the PCB solder mask layer surface color, determine whether there is a color difference problem in the miniLED printed circuit board, including: Calculate the color difference values between each color feature coding vector of the PCB solder mask layer surface area in the set of color feature coding vectors of the PCB solder mask layer surface area and the color distribution modeling vector of the PCB solder mask layer surface color to obtain a color consistency coding vector of the PCB solder mask layer surface color composed of multiple color difference values as the difference representation; Based on the color consistency coding vector of the PCB solder mask layer surface color, determine whether there is a color difference problem in the miniLED printed circuit board.

10. The method for controlling the apparent color difference in the processing of mini-LED printed circuit boards according to claim 9, characterized in that, Based on the color consistency coding vector of the PCB solder mask layer surface color, determine whether there is a color difference problem in the miniLED printed circuit board, including: input the color consistency coding vector of the PCB solder mask layer surface color into a color difference discriminator based on a classifier to determine whether there is a color difference problem in the miniLED printed circuit board.

Images