An image enhancement method for ceramic packaging substrates based on a scale-state-space model

Abstract

This invention provides an image enhancement method for ceramic packaging substrates based on a scale-state-space model, belonging to the field of image processing technology. Targeting ceramic packaging substrates widely used in semiconductor and optoelectronic devices under low-light industrial environments, this invention establishes an enhancement method for ceramic packaging substrates under low-light conditions. This method is based on prior illumination processing and utilizes a multi-scale balanced state-space model as its core technology. The multi-scale balanced state-space model is constructed to enhance the image of the ceramic packaging substrate after illumination correction. While ensuring image enhancement, this method balances the image color space and reduces the impact of noise during image enhancement, further guaranteeing the imaging quality of the ceramic packaging substrate during darkroom inspection. It not only enhances the capture of details and textures but also significantly improves color and brightness stability in low-light environments.

Description

Technical Field

[0001] This invention belongs to the field of image processing technology, and particularly relates to an image enhancement method for ceramic packaging substrates based on a scale-state-space model. Background Technology

[0002] Integrated circuits, as a core industry of modern electronics, are widely used in information processing, communications, automotive electronics, defense, and other fields. With the rapid development of technology, the complexity and performance requirements of integrated circuits are constantly increasing, placing higher demands on the precision and reliability of manufacturing processes. In integrated circuit manufacturing, packaging technology plays a crucial role not only in protecting the chip but also in optimizing electrical performance and thermal management, thus becoming key to ensuring the overall performance of integrated circuits.

[0003] Choosing the right packaging substrate is crucial during the packaging process. Ceramic packaging substrates are renowned for their excellent thermal conductivity and mechanical strength, making them the preferred choice for high-performance and reliability applications. However, the production of ceramic substrates is often plagued by low yields due to defect-related issues. Therefore, defect detection is key to improving product quality.

[0004] Defect detection is typically performed in a darkroom environment using automated optical inspection (AOI) equipment to minimize external light interference and ensure the accuracy of the results. AOI systems employ high-resolution imaging technology to capture defects on and within the substrate. However, variations in substrate thickness during manufacturing pose challenges to the optical imaging process. This problem is particularly pronounced when inspecting thicker ceramic packaging substrates, leading to insufficient image brightness and affecting accurate defect identification. Numerous researchers have conducted related studies to address these issues.

[0005] Chinese patent application "CN202510425764.3 A Low-Light Image Enhancement Method Based on Wavelet-Fourier Transform" proposes a low-light image enhancement method based on wavelet-Fourier transform, which fully leverages the advantages of the two frequency domain transforms, solves the problems of insufficient illumination adjustment and poor detail preservation in existing methods, and improves the utilization efficiency of frequency domain information.

[0006] Chinese patent application "CN202211512888.8 A method for enhancing low-light images based on depthwise separable convolution" builds a self-calibrating low-light image enhancement model based on depthwise separable convolution, determines the model parameters and the loss function of the neural network, and optimizes the model performance to the best, thereby enhancing the low-light image.

[0007] Chinese patent application "CN202411679411.8 A Low-Light Image Enhancement Method Based on Retinex Model" proposes a low-light image enhancement method based on the Retinex model, which includes: acquiring the low-light image to be enhanced, inputting it into a trained low-light image enhancement model, and obtaining the enhanced image through a pre-illumination module and a fine reconstruction module.

[0008] Despite significant progress in existing low-light enhancement methods, shortcomings remain. For example, the aforementioned patent application only considers brightening low-light images, neglecting the impact of color space and noise on the enhanced image. This is because the algorithms in these methods primarily focus on overcoming the overall low brightness caused by insufficient lighting during low-light enhancement. Their core optimization goal is to improve the global or local brightness of the image to a visually acceptable or informationally discernible level. In this process, the fidelity constraints of color channels and considerations for noise suppression mechanisms are often relegated to a secondary position. Summary of the Invention

[0009] To address the shortcomings of existing technologies, this invention provides an image enhancement method for ceramic packaging substrates based on a scale-state-space model. This method is designed for ceramic packaging substrate images widely used in semiconductor, optoelectronic device, and other fields, acquired under low-light industrial environments. It balances the image color space and reduces the impact of noise during the image enhancement process while ensuring image enhancement.

[0010] This invention provides an image enhancement method for ceramic packaging substrates based on a scale-state-space model, comprising the following steps:

[0011] Images of ceramic packaging substrates under different lighting conditions were collected, including normal lighting images and low lighting images. The ceramic packaging substrate images under different lighting conditions were preprocessed and a dataset was constructed. The dataset included preprocessed normal lighting images and preprocessed low lighting images. The dataset was divided into training set and test set, both of which included preprocessed normal lighting images and preprocessed low lighting images.

[0012] The low-light image after centralized preprocessing of the dataset is subjected to illumination prior processing to obtain the illumination-corrected image of the ceramic packaging substrate.

[0013] A multi-scale equilibrium state space model is constructed to enhance the image of the ceramic packaging substrate after illumination correction. The multi-scale equilibrium state space model is trained and evaluated using training and test sets to obtain a well-trained multi-scale equilibrium state space model.

[0014] The multi-scale equilibrium state space model includes a multi-scale feature extraction module, multiple equilibrium state space modules, and a scale fusion module. The multi-scale feature extraction module is used to extract features from the illumination-corrected ceramic packaging substrate image to obtain features at multiple different scales. Each scale feature corresponds to an input to an equilibrium state space module. The equilibrium state space module is used to perform noise reduction, enhancement, and balancing of the image color space on the input features to obtain enhanced features, which are then input to the scale fusion module. The scale fusion module is used to fuse the enhanced features output from multiple equilibrium state space modules to obtain the enhanced ceramic packaging substrate image.

[0015] The trained illumination equalization state space model is used to enhance the image of the ceramic packaging substrate under low illumination conditions, resulting in an enhanced image of the ceramic packaging substrate.

[0016] Furthermore, the illumination prior processing includes:

[0017] Calculate the illumination prior image based on the preprocessed low-light image;

[0018] The preprocessed low-light image and the prior lighting image are stitched together, and the stitched image is processed using a first 1×1 convolution and a depthwise separable convolution to obtain the first lighting feature;

[0019] The first illumination feature is reduced in channel dimension using a second 1×1 convolution to obtain the illumination map;

[0020] Integrating the illumination mapping and preprocessed low-light image yields an illumination-corrected image of the ceramic packaging substrate.

[0021] Furthermore, the multi-scale feature extraction module includes three branches: a downsampling branch, a basic scale branch, and an upsampling branch.

[0022] The image of the ceramic packaging substrate after illumination correction is input into the multi-scale feature extraction module, which obtains features of various scales through three branches, including low-scale features, basic scale features and high-scale features.

[0023] The process includes several branches: a downsampling branch with multiple 4×4 kernels and a stride of 2, which downsamples the image of the illuminated ceramic packaging substrate to obtain low-scale features; a base-scale branch with multiple 3×3 kernels and a stride of 2, which extracts features from the image of the illuminated ceramic packaging substrate to obtain base-scale features; and an upsampling branch with multiple 2×2 kernels and a stride of 2, which upsamples the image of the illuminated ceramic packaging substrate to obtain high-scale features.

[0024] Furthermore, the balanced state space module includes a balanced attention submodule and a feedforward neural network submodule;

[0025] The balanced attention submodule consists of two branches. The first branch includes the balanced attention mechanism and two-dimensional selective scanning. The second branch performs global average pooling on the features input to the balanced attention module, and after passing through the sigmoid function, it merges with the output of the first branch to obtain the output of the balanced attention submodule.

[0026] Furthermore, the feedforward neural network submodule includes two branches. The first branch includes layer normalization and a feedforward neural network. The second branch performs global average pooling on the output of the balanced attention submodule, and then merges it with the output of the first branch after passing through the sigmoid function.

[0027] The output of the equalization attention submodule is input into the feedforward neural network submodule to obtain the output of the feedforward neural network submodule, which is the enhanced feature output by the equalization state space module.

[0028] Furthermore, in the first branch of the equal attention submodule, the equal attention mechanism performs a scale transformation on the first illumination feature to obtain a second illumination feature with the same feature scale as the input equal state space module; the output of the equal attention mechanism is obtained based on the features of the input equal state space module and the second illumination feature; and a two-dimensional selective scan is performed on the output of the equal attention mechanism to obtain the output of the first branch.

[0029] Furthermore, the scale fusion module fuses the enhanced features output by the equilibrium state space model at different scales by unifying the feature scale and combining a staged fusion method to obtain an enhanced ceramic packaging substrate image.

[0030] The beneficial effects of adopting the above technical solution are as follows: The ceramic packaging substrate image enhancement method based on the scale state space model provided by the present invention establishes an enhancement method for ceramic packaging substrates under low illumination conditions based on illumination prior processing and multi-scale balanced state space model as the core technology. This method can enhance low illumination images under different illumination conditions, and balance the color space of the image while ensuring illumination intensity, reducing the noise impact in the low illumination enhancement process, and further ensuring the imaging quality of ceramic packaging substrates in the darkroom detection process. It not only enhances the capture of details and textures, but also significantly improves the color and brightness stability in low light environments. Attached Figure Description

[0031] Figure 1 Flowchart of the image enhancement method for ceramic packaging substrate based on a scale-state-space model provided in Embodiment 1 of the present invention;

[0032] Figure 2 A schematic diagram of the multi-scale equilibrium state-space model provided in Embodiment 1 of the present invention;

[0033] Figure 3 Image of a low-light ceramic packaging substrate to be enhanced, provided in Embodiment 1 of the present invention;

[0034] Figure 4 Histogram of a low-light ceramic packaging substrate image provided in Embodiment 1 of the present invention;

[0035] Figure 5 HSV color space visualization of a low-light ceramic packaging substrate image provided in Embodiment 1 of the present invention;

[0036] Figure 6 Image of the enhanced ceramic packaging substrate provided in Embodiment 1 of the present invention;

[0037] Figure 7 Histogram of the enhanced ceramic packaging substrate image provided in Embodiment 1 of the present invention;

[0038] Figure 8 The HSV color space visualization of the enhanced ceramic packaging substrate image provided in Embodiment 1 of the present invention. Detailed Implementation

[0039] The specific implementation methods of this application will be further described in detail below with reference to the accompanying drawings and embodiments.

[0040] Example 1:

[0041] An image enhancement method for ceramic packaging substrates based on a scale-state-space model, such as... Figure 1 As shown, it includes the following steps:

[0042] Step 1: Collect images of ceramic packaging substrates under different lighting conditions, including normal lighting images and low lighting images. Preprocess the ceramic packaging substrate images under different lighting conditions to construct a dataset, including preprocessed normal lighting images and preprocessed low lighting images. Divide the dataset into training set and test set, both of which include preprocessed normal lighting images and preprocessed low lighting images.

[0043] Step 2: Perform illumination prior processing on the pre-processed low-light images in the dataset to obtain illumination-corrected images of the ceramic packaging substrate.

[0044] This embodiment uses the Retinex algorithm to perform illumination prior processing on the preprocessed low-light images in the dataset. The Retinex algorithm aims to simulate human visual perception and enhance the visibility of images by emphasizing details and normalizing brightness.

[0045] Based on the preprocessed low-light image with 3 channels Calculate the illumination prior image with 1 channel. As shown in the formula below:

[0046] ;

[0047] in, This is an operation to calculate the average value of each pixel along the channel dimension. For low-light images height, For low-light images width, For all real numbers;

[0048] Preprocessed low-light images and lighting prior images The images are stitched together, and the stitched images are processed using a first 1×1 convolution and a depthwise separable convolution to obtain the first illumination features. ,in, For lighting characteristics The number of channels; in this embodiment, 1×1 convolution and depthwise separable convolution are used to effectively reduce computational complexity while preserving key spatial information, providing crucial local illumination information for subsequent image enhancement.

[0049] Use the second 1×1 convolution on the first illumination feature Perform channel dimensionality reduction to obtain the dimension as Lighting mapping ;

[0050] Lighting mapping and pre-processed low-light images Integrating the image yields the illumination-corrected image of the ceramic packaging substrate. ;

[0051] Step 3: Construct a multi-scale equilibrium state space model to enhance the image of the ceramic packaging substrate after illumination correction. Use the training set and test set to train and evaluate the multi-scale equilibrium state space model to obtain the trained multi-scale equilibrium state space model.

[0052] Multi-scale equilibrium state-space model, such as Figure 2 As shown, it includes a multi-scale feature extraction module, multiple equilibrium state space modules, and a scale fusion module;

[0053] The multi-scale feature extraction module is used to extract features from the illumination-corrected ceramic packaging substrate image to obtain features at multiple different scales. Each scale feature corresponds to an input equalization state space module. The equalization state space module is used to perform noise reduction, enhancement, and balancing of the image color space on the input features to obtain enhanced features, which are then input into the scale fusion module. The scale fusion module is used to fuse the enhanced features output from multiple equalization state space modules to obtain the enhanced ceramic packaging substrate image.

[0054] The multi-scale extraction module includes three branches: a downsampling branch, a base-scale branch, and an upsampling branch; it extracts the illumination-corrected image of the ceramic packaging substrate. The input multi-scale feature extraction module, after passing through three branches, yields features of various scales, including low-scale features, basic-scale features, and high-scale features.

[0055] The downsampling branch includes multiple downsampling convolutions with a kernel size of 4×4 and a stride of 2; the downsampling branch processes the image of the ceramic packaging substrate after illumination correction. Downsampling is performed to obtain low-scale features. The basic scale branch includes multiple convolutions with a kernel size of 3×3 and a stride of 2; the basic scale branch applies to the image of the illumination-corrected ceramic packaging substrate. Feature extraction is performed to obtain basic scale features. The upsampling branch includes multiple upsampling convolutions with a kernel size of 2×2 and a stride of 2; the upsampling branch applies the image of the ceramic packaging substrate after illumination correction. Upsampling is performed to obtain high-scale features. ;

[0056] Features at different scales are input into the corresponding equalization state space module for noise reduction, enhancement, and balancing of the image color space, resulting in enhanced features at different scales. The equalization state space module includes an equalization attention submodule and a feedforward neural network submodule.

[0057] The balanced attention submodule consists of two branches. The first branch includes the balanced attention mechanism and two-dimensional selective scanning. The second branch performs global average pooling on the features input to the balanced attention module, and after passing through the sigmoid function, it is fused with the output of the first branch to obtain the output of the balanced attention submodule.

[0058] In the first branch, the balanced attention mechanism affects the first illumination feature. Perform a scaling transformation to obtain the features of the input equalization state space module. Second lighting features of the same scale Among them, the characteristics of the input equalization state-space module Including low-scale features Basic scale characteristics and high-scale features , Indicates low scale, Indicates the basic scale, Indicates high scale, second illumination feature Including second low-scale lighting features Second basic scale features Second high-scale features ;

[0059] Features based on input equalization state space module The output of the equal attention mechanism is obtained from the second illumination feature. Output of the balanced attention mechanism A two-dimensional selective scan is performed to obtain the output of the first branch. The second branch performs global average pooling on the features input to the balanced attention module, and after passing through the sigmoid function, it is fused with the output of the first branch to obtain the output of the balanced attention submodule. As shown in the formula below:

[0060] ;

[0061] ;

[0062] in, To balance attention mechanisms, It is a two-dimensional selective scan. For global average pooling, The softmax activation function is used. For the sigmoid function, , and These are the learnable weight parameter sets, "This is an element-wise multiplication operation." The dimension of the learnable weight parameter set K; the output of the balanced attention submodule. Including the low-scale output of the equal attention submodule The basic scale output of the balanced attention submodule and the high-scale output of the balanced attention submodule ;

[0063] The Balanced Attention Mechanism submodule, based on the core design principles of the Balanced Attention Mechanism, achieves a dynamic balance between enhancement effect and preservation of original feature information during image enhancement by effectively integrating the interaction between multi-scale features and illumination features. The Balanced Attention Mechanism not only introduces cross-scale information fusion at the feature level but also explicitly models the impact of illumination conditions on the local and global feature representation of the image, thus avoiding the problems of detail loss or over-enhancement during the enhancement process. Furthermore, the Balanced Attention Mechanism is combined with 2D selective scanning to construct a hierarchical feature interaction framework. 2D selective scanning systematically captures long-range dependencies and local contextual information through structured scanning of the image space, thereby enhancing the model's ability to model complex spatial relationships. This integrated strategy can simultaneously capture contextual associations and spatial interaction information in the feature space, improving semantic consistency and structural integrity in image enhancement tasks. Overall, the Balanced Attention Mechanism submodule maintains computational efficiency while strengthening robustness and adaptability under complex lighting and structural conditions.

[0064] The output of the equal attention submodule The input is a feedforward neural network submodule, which consists of two branches. The first branch includes layer normalization and the feedforward neural network, and the second branch is the output of the balanced attention submodule. Global average pooling is performed, and after passing through the sigmoid function, it is fused with the output of the first branch to obtain the output of the feedforward neural network submodule, which is the enhanced feature of the output of the equalization state space module.

[0065] Output enhancement features of the equalization state-space module As shown in the formula below:

[0066] ;

[0067] in, For layer normalization, For feedforward networks, enhance features Including low-scale enhancement features Basic scale enhancement features and high-scale enhancement features ;

[0068] The scale fusion module fuses the enhanced features output from the equilibrium state-space model at different scales by unifying the feature scale and combining a staged fusion method to obtain the enhanced image of the ceramic packaging substrate. As shown in the formula below:

[0069] ;

[0070] in, For upsampling operation, This is a downsampling operation;

[0071] The problems of noise and color space imbalance can be fundamentally attributed to the limited expressive power of image feature space and insufficient response and integration of multi-scale contextual information. A limited feature space often makes it difficult for models to effectively distinguish between signal and noise, and also prevents them from fully encoding complex color distribution relationships, thus easily introducing artifacts or causing color distortion during enhancement. To systematically address this core challenge, this embodiment proposes and constructs a multi-scale balanced state space model. The core innovation of this model lies in its parallel multi-branch structure, which simultaneously extracts and fuses image features at multiple scales. This design ensures that the model can simultaneously capture subtle local texture details and macroscopic global structural information, thereby enabling robust analysis and modeling within a broader and richer feature space.

[0072] Specifically, the multi-scale feature extraction network provides a more discriminative input foundation for the subsequent equalization state space module. The equalization state space module, operating on this enhanced foundation, efficiently integrates features from different scales through its internal mechanisms and state propagation, and dynamically adjusts its attention weights to noise components and color channels. This process achieves two key improvements: firstly, by leveraging the redundancy and contextual information inherent in the expanded feature set, noise can be identified and suppressed more accurately, significantly reducing its interference with the final image quality; secondly, modeling the long-range dependencies of cross-scale color features helps correct color shifts introduced by illumination, sensors, or processing links, thereby partially restoring the natural balance and visual realism of image colors.

[0073] Step 3: Use the trained illumination equalization state space model to enhance the image of the ceramic packaging substrate under low illumination conditions to obtain the enhanced ceramic packaging substrate image.

[0074] This embodiment proposes a low-light enhancement method for ceramic packaged substrate images based on a multi-scale state space model. A low-light enhancement system for ceramic packaged substrate images is constructed by fusing illumination priors, a multi-scale equalization state space model, and an equalization attention mechanism. Illumination priors provide initial illumination intensity enhancement for low-light images, ensuring brightness priors. The multi-scale equalization state space model not only enhances the capture of details and textures but also significantly improves color and brightness stability in low-light environments. By extracting features across multiple scales, analysis is performed in a broader feature space, providing a stronger foundation for the equalization state space model. The equalization state space model significantly reduces the impact of noise on image quality and restores the natural color balance by expanding the feature set. The equalization attention submodule ensures a balance between image enhancement and feature information preservation through the interaction between scale features and illumination features. This not only enhances image brightness but also maintains image information entropy, preventing overexposure and color space imbalance.

[0075] Compared with existing methods, the technical solution proposed in this embodiment establishes an enhancement framework for ceramic packaging substrates under low-light conditions, based on an illumination prior technology framework and with a multi-scale equilibrium state-space model and equilibrium attention mechanism as core technologies. This embodiment conducts image enhancement experiments on low-light ceramic packaging substrate images, as shown in the image to be enhanced. Figure 3 As shown, the histogram of the low-light ceramic packaging substrate image is as follows: Figure 4 As shown, the HSV color space visualization of the low-light ceramic packaging substrate image is as follows: Figure 5 As shown, the enhanced image of the ceramic packaging substrate is as follows: Figure 6 As shown, the histogram of the enhanced ceramic packaging substrate image is as follows: Figure 7 As shown, the HSV color space visualization of the enhanced ceramic package substrate image is as follows: Figure 8 As shown. The ceramic packaging substrate image enhancement method based on the scale-state-space model provided in this embodiment can enhance low-light images under different lighting conditions, and balance the color space of the image while ensuring the light intensity, reducing the noise impact in the low-light image enhancement process, and further ensuring the imaging quality of the ceramic packaging substrate detection process under darkroom conditions.

[0076] Example 2:

[0077] This embodiment proposes an electronic device, including: one or more processors, and a memory, wherein the memory is used to store instructions, and when the instructions are executed by the one or more processors, the one or more processors execute the ceramic packaging substrate image enhancement method based on the scale state space model.

[0078] The electronic device may be a mobile phone, computer, or tablet computer, etc., and includes a memory and a processor. The memory stores a computer program, which, when executed by the processor, implements the ceramic package substrate image enhancement method based on the scale-state-space model as described in the embodiments. It is understood that the electronic device may also include input / output (I / O) interfaces and communication components.

[0079] The processor is used to execute all or part of the steps in the ceramic packaging substrate image enhancement method based on the scale-state-space model as described in the above embodiments. The memory is used to store various types of data, which may include, for example, instructions for any application or method in the electronic device, as well as application-related data.

[0080] The processor can be implemented as an Application Specific Integrated Circuit (ASIC), Digital Signal Processor (DSP), Programmable Logic Device (PLD), Field Programmable Gate Array (FPGA), controller, microcontroller, microprocessor, or other electronic components, and is used to execute the ceramic package substrate image enhancement method based on the scale state space model described in the above embodiments.

[0081] Example 3:

[0082] This embodiment proposes a computer-readable storage medium that stores executable instructions. When these instructions are executed, if they are implemented as software functional units and sold or used as independent products, they can be stored in a computer-readable storage medium.

[0083] The computer software product is stored in a storage medium and includes several instructions to cause a computer device (which may be a personal computer, a server, or a network device, etc.) to execute all or part of the steps of the ceramic packaging substrate image enhancement method based on the scale state space model described in the various embodiments of this application.

[0084] The aforementioned storage media include: flash memory, hard disks, multimedia cards, card-type memory (e.g., SD (Secure Digital Memory Card) or DX (Memory Data Register, MDR) memory), random access memory (RAM), static random-access memory (SRAM), read-only memory (ROM), electrically erasable programmable read-only memory (EEPROM), programmable read-only memory (PROM), magnetic storage, disks, optical discs, servers, APP (Application) application stores, and other media capable of storing program verification codes. These media store computer programs, which, when executed by a processor, can implement the various steps of the ceramic package substrate image enhancement method based on the scale-state-space model described above.

[0085] Example 4:

[0086] This embodiment proposes a computer program product, including a computer program or instructions, which, when executed by a processor, implements the aforementioned image enhancement method for ceramic packaging substrates based on a scale-state-space model.

[0087] Based on this understanding, the technical solution of this application, in essence, or the part that contributes to the prior art, or part of the technical solution, can be embodied in the form of a computer program product.

[0088] The various embodiments in this application are described in a progressive manner. The same or similar parts between the various embodiments can be referred to each other. Each embodiment focuses on describing the differences from other embodiments.

[0089] The scope of protection of this application is not limited to the embodiments described above. Obviously, those skilled in the art can make various modifications and variations to this disclosure without departing from the scope and spirit of this disclosure. If such modifications and variations fall within the scope of this disclosure and its equivalents, then the intent of this disclosure also includes these modifications and variations.

Claims

1. A method for image enhancement of ceramic packaging substrates based on a scale-state-space model, characterized in that, Includes the following steps: Images of ceramic packaging substrates under different lighting conditions were acquired, including normal lighting images and low lighting images. The ceramic packaging substrate images under different lighting conditions were preprocessed and a dataset was constructed. The dataset includes preprocessed normal lighting images and preprocessed low lighting images. The dataset is divided into a training set and a test set, both of which include preprocessed normal lighting images and preprocessed low lighting images. The low-light image after centralized preprocessing of the dataset is subjected to illumination prior processing to obtain the illumination-corrected image of the ceramic packaging substrate. A multi-scale equilibrium state space model is constructed to enhance the image of the ceramic packaging substrate after illumination correction. The multi-scale equilibrium state space model is trained and evaluated using training and test sets to obtain a well-trained multi-scale equilibrium state space model. The multi-scale equilibrium state space model includes a multi-scale feature extraction module, multiple equilibrium state space modules, and a scale fusion module. The multi-scale feature extraction module is used to extract features from the illumination-corrected ceramic packaging substrate image to obtain features of various different scales. Each scale feature corresponds to an input to an equilibrium state space module. The equilibrium state space module is used to perform noise reduction, enhancement, and balancing of the image color space on the input features to obtain enhanced features, which are then input to the scale fusion module. The scale fusion module is used to fuse the enhanced features output by multiple equal state space modules to obtain an enhanced image of the ceramic packaging substrate. The trained multi-scale equilibrium state-space model is used to enhance the image of the ceramic packaging substrate under low illumination conditions, resulting in an enhanced image of the ceramic packaging substrate.

2. The image enhancement method for ceramic packaging substrates based on a scale-state-space model according to claim 1, characterized in that, The illumination prior processing includes: Calculate the illumination prior image based on the preprocessed low-light image; The preprocessed low-light image and the prior lighting image are stitched together, and the stitched image is processed using a first 1×1 convolution and a depthwise separable convolution to obtain the first lighting feature; The first illumination feature is reduced in channel dimension using a second 1×1 convolution to obtain the illumination map; Integrating the illumination mapping and preprocessed low-light image yields an illumination-corrected image of the ceramic packaging substrate.

3. The image enhancement method for ceramic packaging substrates based on a scale-state-space model according to claim 1, characterized in that, The multi-scale feature extraction module includes three branches: a downsampling branch, a basic scale branch, and an upsampling branch. The image of the ceramic packaging substrate after illumination correction is input into the multi-scale feature extraction module, which obtains features of various scales through three branches, including low-scale features, basic scale features and high-scale features. The process includes several branches: a downsampling branch with multiple 4×4 kernels and a stride of 2, which downsamples the image of the illuminated ceramic packaging substrate to obtain low-scale features; a base-scale branch with multiple 3×3 kernels and a stride of 2, which extracts features from the image of the illuminated ceramic packaging substrate to obtain base-scale features; and an upsampling branch with multiple 2×2 kernels and a stride of 2, which upsamples the image of the illuminated ceramic packaging substrate to obtain high-scale features.

4. The image enhancement method for ceramic packaging substrates based on a scale-state-space model according to claim 1, characterized in that, The balanced state space module includes a balanced attention submodule and a feedforward neural network submodule; The balanced attention submodule consists of two branches. The first branch includes the balanced attention mechanism and two-dimensional selective scanning. The second branch performs global average pooling on the features input to the balanced attention module, and after passing through the sigmoid function, it merges with the output of the first branch to obtain the output of the balanced attention submodule.

5. The image enhancement method for ceramic packaging substrates based on a scale-state-space model according to claim 4, characterized in that, The feedforward neural network submodule includes two branches. The first branch includes layer normalization and the feedforward neural network. The second branch performs global average pooling on the output of the balanced attention submodule and then merges it with the output of the first branch after passing through the sigmoid function. The output of the equalization attention submodule is input into the feedforward neural network submodule to obtain the output of the feedforward neural network submodule, which is the enhanced feature output by the equalization state space module.

6. The image enhancement method for ceramic packaging substrates based on a scale-state-space model according to claim 4, characterized in that, In the first branch of the equal attention submodule, the equal attention mechanism performs a scale transformation on the first illumination feature to obtain a second illumination feature with the same feature scale as the input equal state space module; the output of the equal attention mechanism is obtained based on the features of the input equal state space module and the second illumination feature. A two-dimensional selective scan of the output of the balanced attention mechanism is performed to obtain the output of the first branch.

7. The image enhancement method for ceramic packaging substrates based on a scale-state-space model according to claim 1, characterized in that, The scale fusion module fuses the enhanced features output by the equilibrium state space module at different scales by unifying the feature scale and combining a staged fusion method to obtain an enhanced image of the ceramic packaging substrate.

Images