Car part image super-resolution method based on spatial channel joint attention

Abstract

The application provides a super-resolution method for automobile part images based on spatial channel joint attention, relates to the field of computer vision, and first divides the image into high-frequency and low-frequency parts for processing; for high-frequency information, the image definition is improved by enhancing the detailed features; for low-frequency information, the low-frequency features are made more prominent by removing redundancy and enhancing the low-frequency features; then, the high-frequency and low-frequency features are fused, and the features are further enhanced through a spatial channel joint attention block; finally, the fused features are restored to a high-resolution space through an upsampling operation to obtain a super-resolution image; the method effectively improves the image details and structural information and significantly improves the quality of the super-resolution image.

Description

Technical Field

[0001] This invention belongs to the field of computer vision, and specifically relates to a super-resolution method for automotive component images based on spatial channel joint attention. Background Technology

[0002] With the rapid development of industrial automation and intelligent manufacturing, the inspection and quality control of automotive parts have become increasingly important. In order to improve production efficiency and ensure the quality of parts, the application of image recognition technology in parts inspection has received widespread attention. However, in practical applications, many automotive parts images often exhibit low resolution or noise interference due to factors such as shooting angle and lighting changes, resulting in poor performance of traditional image processing and recognition methods. Therefore, how to improve the quality of low-resolution images while ensuring detail and structural information has become a research hotspot in the field of image processing.

[0003] In recent years, spatial and channel attention mechanisms have been widely used in image processing tasks. By weighting the features of different regions and channels, they have effectively improved the representational power of images. However, how to reasonably integrate spatial and channel attention mechanisms in super-resolution tasks and give full play to their enhancement effect on high-frequency and low-frequency features remains a topic worthy of exploration. This invention proposes a super-resolution method for automotive parts images based on spatial and channel joint attention. By dividing low-resolution images into high-frequency and low-frequency features, processing them separately, and then fusing the features, further enhancement and upsampling reconstruction are performed on this basis. This can better improve the details and structural information of the image and achieve better super-resolution reconstruction results. Summary of the Invention

[0004] This invention proposes a super-resolution method for automotive component images based on spatial channel joint attention. The aim is to improve the super-resolution effect by constructing a spatial channel joint network, dividing low-resolution features into high-frequency and low-frequency features, processing them, and then enhancing them with spatial channel joint attention.

[0005] This invention integrates spatial attention and channel attention to provide a super-resolution method for automotive component images based on joint spatial and channel attention, comprising the following steps:

[0006] S1. Use an industrial camera to photograph automotive parts to obtain low-resolution images of the automotive parts.

[0007] S2. Construct a high-frequency feature enhancement module, which includes high-frequency feature extraction, 3x3 convolution, and ReLU activation function;

[0008] S3. Construct a low-frequency feature enhancement module, which includes low-frequency feature extraction, thresholding redundancy removal, and local adaptation.

[0009] S4. Construct a spatial-channel joint attention block, which consists of two branches, an upper branch and a lower branch. The upper branch is the improved spatial attention and the lower branch is the channel attention. Finally, the outputs of the upper and lower branches are multiplied with the original feature map by element-wise multiplication.

[0010] S5. Construct a feature processing block. The feature processing block goes through a high-frequency feature enhancement module and a low-frequency feature enhancement module. The outputs of the two modules are connected and then connected through a 1x1 convolution and a spatial channel joint attention block. Then, through a 1x1 convolution and pixel normalization, the output features are finally connected to the input jump connection.

[0011] S6. Construct a spatial channel joint attention network to integrate image features. The input is fed into a multi-level spatial channel joint attention network, which includes n feature processing blocks and a feature fusion layer.

[0012] S7. Construct an image reconstruction module, which performs convolution and upsampling on the fused features;

[0013] S8. Construct an image super-resolution model, including input, feature extraction, spatial channel joint network, image reconstruction, and image output;

[0014] S9. Input the images of automotive parts into the super-resolution image model, and obtain the super-resolution image after processing.

[0015] Preferably, in step S2, a high-frequency feature enhancement module is constructed, characterized in that:

[0016] Input initial features of automotive component images Initial features of automotive component images Applied to a 7x7 Gaussian kernel and using the initial features Subtracting the features processed by the Gaussian kernel, using Gaussian kernel blurring to approximate the low-frequency components, and then subtracting the low-frequency components from the initial features yields the high-frequency components. , The code uses a 7x7 Gaussian kernel. High-frequency features are then processed through a 3x3 convolution and a ReLU activation function. The 3x3 convolution extracts the high-frequency features, and the ReLU activation function generates an attention map. Skip connections are then used to multiply the high-frequency features and the attention map pixel-by-pixel to further enhance the high-frequency region. To avoid over-enhancement, residual connections are used to combine the enhanced high-frequency features with the original high-frequency features. Finally, the combined features are processed through a 3x3 convolution and a ReLU activation function. This process can be represented as: , This represents a 3×3 convolution. Represents the ReLU activation function. This represents element-wise multiplication. This represents element-wise addition. This is the output of the high-frequency feature enhancement module.

[0017] Preferably, in step S2, low-frequency features are extracted by Gaussian kernel blurring and high-frequency features are obtained by difference operation, which effectively combines high-frequency features with attention map to improve the performance of detailed features; residual connection is used to avoid over-enhancement, preserve the original information, and improve the stability and training effect of the network; pixel-wise multiplication and convolution processing enhance the expressive power of high-frequency parts while maintaining the diversity and fineness of features.

[0018] Preferably, in step S3, a low-frequency feature enhancement module is constructed, characterized in that:

[0019] Input initial features of automotive component images Initial features of automotive component images Low-frequency features are extracted using a 3x3 convolutional layer. Then, by calculating the range of variation in the low-frequency region, redundancy is removed based on a pre-set threshold. The resulting features are then... , This indicates thresholding to remove redundancy. This indicates the preset threshold. The mathematical model is , ,in Input features The number of elements in the middle, Indicates the first The value of each element, , Input features The mean, Input features The standard deviation of the low-frequency features is used to remove redundancy. For low-frequency features, a difference operation is used to subtract the local mean of the low-frequency region to avoid loss of detail or over-enhancement. Then, an enhancement factor is multiplied to enhance the low-frequency features. Residual connections are then used to combine the enhanced low-frequency features with the original low-frequency features. Finally, a 3×3 convolution and ReLU activation function are applied. This process can be represented as: , This is the output of the low-frequency feature enhancement module. Represents the ReLU activation function. This is the local mean value in the low-frequency region. As an enhancing factor, The mathematical model is , , ,in, It is the mean of the low-frequency features after redundancy removal. Input features The number of elements in the middle, Indicates the first The value of each element, It is the standard deviation of the low-frequency features after redundancy removal. It is the standard deviation of the input features.

[0020] Preferably, in step S3, low-frequency features are extracted by 3×3 convolution, and redundancy is removed by thresholding, effectively reducing noise and unnecessary information; the combination of difference operation and enhancement factor ensures that the details of low-frequency features are preserved and enhanced, avoiding the risk of over-enhancement; residual connection helps to retain the original low-frequency information, improving the stability and convergence speed of the model; the subtraction operation of local mean further enhances the expressive power of low-frequency features and improves image quality.

[0021] Preferably, in step S4, a spatial channel joint attention block is constructed, which includes:

[0022] S41, the spatial channel joint attention block is divided into an upper branch and a lower branch, inputting processed image features. The processed image features Divided into The process is divided into two parts, each processed by a separate branch. For the upper branch, the input features are... Two feature maps are obtained by performing max pooling and average pooling respectively. The max pooling layer can preserve the main features in the feature map and suppress noise and unimportant details. , This represents the max pooling operation. The average pooling layer reduces the size of the feature map by calculating the average value of local regions. , This represents an average pooling operation, and then the pooling results are concatenated. , The concatenation operation involves passing the concatenated features through a 3×3 convolutional layer, which helps capture spatial dependencies within local regions. A sigmoid activation function is then used to obtain a spatial attention map. To enhance the response in important regions, the output map is element-wise multiplied with the original feature map, and a 3×3 convolutional layer and ReLU activation function are used to further enhance the expressive power of local features. This process can be described as follows: , Represents the ReLU activation function. This represents the sigmoid activation function. This represents a 3×3 convolution. This represents element-wise multiplication;

[0023] S42. For the lower branch, input features The algorithm is further divided into two branches. The upper branch performs global average pooling, and then uses 1×1 convolution and ReLU activation function to obtain the attention for each channel. , This indicates a global average pooling operation. The next branch uses global average pooling and then reverses, swapping the preceding and following channels. This process can be described as follows: Then, a 1×1 convolution and ReLU activation function are used, followed by skip connections to the deflected result, and finally, the result is deflected again. This process can be described as follows: , This indicates a reverse operation, and finally, the outputs of the two branches are weighted and concatenated. This process can be described as follows: , This represents the weight, and its value is equal to 0.7. This represents element-wise addition;

[0024] S43. After enhancing the high-frequency and low-frequency features, the outputs of the two branches are then compared with the initial input features. Element-wise multiplication yields the final output of the spatial channel attention block. , This represents element-wise multiplication.

[0025] Preferably, in step S4, the advantage of the spatial channel joint attention block is that it can effectively fuse high-frequency and low-frequency features. By processing spatial and channel attention separately through the upper and lower branches, the expressive power of the feature map is improved. The upper branch enhances local spatial dependencies through max pooling and average pooling, while the lower branch enhances the relationship between channels through global pooling and deflection operations, thereby better capturing important feature information. The final element-wise multiplication operation ensures the effective fusion between different features, improving the performance and generalization ability of the model.

[0026] Preferably, in step S5, a feature processing block is constructed, characterized in that:

[0027] Feature processing blocks for input automotive component image features First, it passes through a high-frequency feature enhancement module and a low-frequency feature enhancement module. After enhancement, the high-frequency features... , This indicates a high-frequency feature enhancement module, which enhances low-frequency features. , This represents a low-frequency feature enhancement module, which then concatenates the enhanced high-frequency and low-frequency features and fuses them using a 1×1 convolution. The fused features... , Indicates a connection operation. This represents a 1×1 fusion convolution; subsequently, a spatial channel attention joint module is used to further refine the features, employing 1×1 convolution for feature transformation, while pixel normalization is added to improve model performance. Finally, long skip connections are used to enhance the model's residual learning ability, yielding the final output of the feature processing block. , Represents pixel normalization operation. This represents a spatial channel joint attention block.

[0028] Preferably, in step S5, the feature processing block effectively separates and enhances image information through high-frequency and low-frequency feature enhancement modules, improving the model's ability to process features of different frequencies. The fused features are further optimized through a joint module of 1×1 convolution and spatial channel attention, enabling more refined adjustment of feature representation and enhancing the model's ability to learn local and global information; pixel normalization and long skip connections help improve the model's stability and residual learning ability, ultimately achieving stronger feature representation and improved overall performance.

[0029] Preferably, in step S6, a spatial channel joint attention network is constructed, characterized in that:

[0030] For input features We concatenate the outputs of each feature processing block and then perform a concatenation operation on the outputs of each feature processing block. The concatenated results are then fused using a 1x1 convolution, followed by a 3x3 convolution to further refine the features. Finally, the original features are... The spatial channel attention network's final output is obtained by adding skip connections to the network output. , This represents the output of the nth spatial channel attention block.

[0031] Preferably, in step S6, by cascading the outputs of multiple spatial channel attention blocks, feature information at different levels is fully integrated, improving the model's ability to express multi-scale features; 1×1 convolutions are used for feature fusion, effectively reducing dimensionality while retaining important information, and 3×3 convolutions further refine the features, improving the expression of spatial details; skip connections are used to combine the original features... Directly incorporating low-level information preserves residual learning capabilities, enhancing model stability and convergence speed.

[0032] Compared with the prior art, the present invention has the following technical effects:

[0033] The spatial channel joint attention network provided by this invention consists of n spatial channel joint attention blocks and a feature fusion module. The spatial channel joint attention blocks divide the image into high-frequency and low-frequency parts, which are processed separately. For high-frequency features, image clarity is improved by enhancing their detail features; for low-frequency features, redundancy is removed and low-frequency features are enhanced to make them more prominent. Then, the high-frequency and low-frequency features are fused, and subsequently, the features are further enhanced by the spatial channel joint attention blocks. Finally, an upsampling operation restores the fused features to the high-resolution space, resulting in a super-resolution image. This method effectively improves image details and structural information, significantly enhancing the quality of super-resolution images. Attached Figure Description

[0034] Figure 1 This is a flowchart of the super-resolution image of automotive parts proposed in this invention.

[0035] Figure 2 This is an example diagram of the high-frequency feature enhancement module proposed in this invention.

[0036] Figure 3 This is an example diagram of the low-frequency feature enhancement module proposed in this invention.

[0037] Figure 4 This is an example diagram of the spatial channel joint attention block proposed in this invention.

[0038] Figure 5 This is a diagram of the feature processing block structure proposed in this invention.

[0039] Figure 6 This is a diagram of the spatial channel joint attention network structure proposed in this invention. Detailed Implementation

[0040] The technical solutions in the embodiments of the present invention will now be clearly and completely described with reference to the accompanying drawings. Obviously, the described embodiments are only a part of the embodiments of the present invention, and not all of them. All other embodiments obtained by those skilled in the art based on the embodiments of the present invention without creative effort are within the scope of protection of the present invention.

[0041] This invention proposes a super-resolution method for automotive component images based on spatial channel joint attention. First, a spatial channel joint attention network is constructed, consisting of n feature processing blocks and a feature fusion module. The spatial channel joint attention blocks divide the image into high-frequency and low-frequency parts, processing them separately. For high-frequency features, their detail features are enhanced to improve image clarity; for low-frequency features, redundancy is removed and low-frequency features are enhanced to make them more prominent. Next, the high-frequency and low-frequency features are fused, and then further enhanced by the spatial channel joint attention blocks. Finally, an upsampling operation restores the fused features to the high-resolution space, obtaining the super-resolution image. This method effectively improves image details and structural information, significantly enhancing the quality of the super-resolution image.

[0042] Please see Figure 1 As shown, the super-resolution method for automotive component images based on spatial channel joint attention in this application embodiment is as follows:

[0043] S1. Use an industrial camera to photograph automotive parts to obtain low-resolution images of the automotive parts.

[0044] S2. Construct a high-frequency feature enhancement module, which includes high-frequency feature extraction, 3x3 convolution, and ReLU activation function.

[0045] Furthermore, such as Figure 2 As shown, the high-frequency feature enhancement module is constructed, and the specific steps are as follows.

[0046] Input initial features of automotive component images Initial features of automotive component images Applied to a 7x7 Gaussian kernel and using the initial features Subtracting the features processed by the Gaussian kernel, using Gaussian kernel blurring to approximate the low-frequency components, and then subtracting the low-frequency components from the initial features yields the high-frequency components. , The code uses a 7x7 Gaussian kernel. High-frequency features are then processed through a 3x3 convolution and a ReLU activation function. The 3x3 convolution extracts the high-frequency features, and the ReLU activation function generates an attention map. Skip connections are then used to multiply the high-frequency features and the attention map pixel-by-pixel to further enhance the high-frequency region. To avoid over-enhancement, residual connections are used to combine the enhanced high-frequency features with the original high-frequency features. Finally, the combined features are processed through a 3x3 convolution and a ReLU activation function. This process can be represented as: , This represents a 3×3 convolution. Represents the ReLU activation function. This represents element-wise multiplication. This represents element-wise addition. This is the output of the high-frequency feature enhancement module.

[0047] S3. Construct a low-frequency feature enhancement module, which includes low-frequency feature extraction, thresholding redundancy removal, and local adaptation.

[0048] Furthermore, such as Figure 3 As shown, the low-frequency feature enhancement module is constructed, and the specific steps are as follows.

[0049] Input initial features of automotive component images Initial features of automotive component images Low-frequency features are extracted using a 3x3 convolutional layer. Then, by calculating the range of variation in the low-frequency region, redundancy is removed based on a pre-set threshold. The resulting features are then... , This indicates thresholding to remove redundancy. This indicates the preset threshold. The mathematical model is , ,in Input features The number of elements in the middle, Indicates the first The value of each element, , Input features The mean, Input features The standard deviation of the low-frequency features is used to remove redundancy. For low-frequency features, a difference operation is used to subtract the local mean of the low-frequency region to avoid loss of detail or over-enhancement. Then, an enhancement factor is multiplied to enhance the low-frequency features. Residual connections are then used to combine the enhanced low-frequency features with the original low-frequency features. Finally, a 3×3 convolution and ReLU activation function are applied. This process can be represented as: , This is the output of the low-frequency feature enhancement module. Represents the ReLU activation function. This is the local mean value in the low-frequency region. As an enhancing factor, The mathematical model is , , ,in, It is the mean of the low-frequency features after redundancy removal. Input features The number of elements in the middle, Indicates the first The value of each element, It is the standard deviation of the low-frequency features after redundancy removal. It is the standard deviation of the input features.

[0050] S4. Construct a spatial-channel joint attention block, which consists of two branches: an upper branch and a lower branch. The upper and lower branches are the improved spatial attention and channel attention, respectively. Finally, the outputs of the upper and lower branches are multiplied with the original feature map through element-wise multiplication.

[0051] Furthermore, such as Figure 4 As shown, the specific steps for constructing a joint attention block for spatial channels are as follows.

[0052] S41, the spatial channel joint attention block is divided into an upper branch and a lower branch, inputting processed image features. The processed image features Divided into The process is divided into two parts, each processed by a separate branch. For the upper branch, the input features are... Two feature maps are obtained by performing max pooling and average pooling respectively. The max pooling layer can preserve the main features in the feature map and suppress noise and unimportant details. , This represents the max pooling operation. The average pooling layer reduces the size of the feature map by calculating the average value of local regions. , This represents an average pooling operation, and then the pooling results are concatenated. , The concatenation operation involves passing the concatenated features through a 3×3 convolutional layer, which helps capture spatial dependencies within local regions. A sigmoid activation function is then used to obtain a spatial attention map. To enhance the response in important regions, the output map is element-wise multiplied with the original feature map, and a 3×3 convolutional layer and ReLU activation function are used to further enhance the expressive power of local features. This process can be described as follows: , Represents the ReLU activation function. This represents the sigmoid activation function. This represents a 3×3 convolution. This represents element-wise multiplication;

[0053] S42. For the lower branch, input features The algorithm is further divided into two branches. The upper branch performs global average pooling, and then uses 1×1 convolution and ReLU activation function to obtain the attention for each channel. , This indicates a global average pooling operation. The next branch uses global average pooling and then reverses, swapping the preceding and following channels. This process can be described as follows: Then, a 1×1 convolution and ReLU activation function are used, followed by skip connections to the deflected result, and finally, the result is deflected again. This process can be described as follows: , This indicates a reverse operation, and finally, the outputs of the two branches are weighted and concatenated. This process can be described as follows: , This represents the weight, and its value is equal to 0.7. This represents element-wise addition;

[0054] S43. After enhancing the high-frequency and low-frequency features, the outputs of the two branches are then compared with the initial input features. Element-wise multiplication yields the final output of the spatial channel attention block. , This represents element-wise multiplication.

[0055] S5. Construct a feature processing block. The feature processing block goes through a high-frequency feature enhancement module and a low-frequency feature enhancement module. The outputs of the two modules are connected and then connected through a 1x1 convolution and a spatial channel joint attention block. Then, through a 1x1 convolution and pixel normalization, the output features are finally connected to the input in a jump connection.

[0056] Furthermore, such as Figure 5 As shown, the feature processing block is constructed, and the specific steps are as follows.

[0057] Feature processing blocks for input automotive component image features First, it passes through a high-frequency feature enhancement module and a low-frequency feature enhancement module. After enhancement, the high-frequency features... , This indicates a high-frequency feature enhancement module, which enhances low-frequency features. , This represents a low-frequency feature enhancement module, which then concatenates the enhanced high-frequency and low-frequency features and fuses them using a 1×1 convolution. The fused features... , Indicates a connection operation. This represents a 1×1 fusion convolution; subsequently, a spatial channel attention joint module is used to further refine the features, employing 1×1 convolution for feature transformation, while pixel normalization is added to improve model performance. Finally, long skip connections are used to enhance the model's residual learning ability, yielding the final output of the feature processing block. , Represents pixel normalization operation. This represents a spatial channel joint attention block.

[0058] S6. Construct a spatial channel joint attention network to integrate image features. The input spatial channel joint attention network consists of n feature processing blocks and a feature fusion layer.

[0059] S7. Construct an image reconstruction module, which performs convolution and upsampling on the fused features.

[0060] S8. Construct an image super-resolution model, including input, feature extraction, spatial channel joint attention network, image reconstruction, and image output.

[0061] S9. Input the images of automotive parts into the super-resolution image model, and obtain the super-resolution image after processing.

[0062] Furthermore, such as Figure 6 As shown, an image super-resolution model is constructed. After capturing low-resolution images of automotive parts using an industrial camera, the images are input into a feature extraction module to obtain image features. Image features The input spatial channel joint attention network concatenates the outputs of each feature processing block and performs a concatenation operation on the outputs of each feature processing block. The concatenation results are then fused using a 1x1 convolution, followed by a 3x3 convolution to further refine the features. Finally, the original features are combined. The spatial channel attention network's final output is obtained by adding skip connections to the network output. , This represents the output of the nth spatial channel attention block. Finally, the output of the spatial channel attention network is passed through the image reconstruction module to obtain the super-resolution image. This process can be described as follows: , This indicates upsampling.

[0063] The above are merely preferred embodiments of the present invention. It should be noted that those skilled in the art can make various modifications and improvements without departing from the inventive concept of the present invention, and these modifications and improvements all fall within the protection scope of the present invention.

Claims

1. A super-resolution method for automotive component images based on spatial channel joint attention, characterized in that, Includes the following steps: S1. Use an industrial camera to photograph automotive parts to obtain low-resolution images of the automotive parts. S2. Construct a high-frequency feature enhancement module, which includes high-frequency feature extraction, 3x3 convolution, and ReLU activation function; S3. Construct a low-frequency feature enhancement module, which includes low-frequency feature extraction, thresholding redundancy removal, and local adaptation. S4. Construct a spatial-channel joint attention block, which consists of two branches, an upper branch and a lower branch. The upper branch is the improved spatial attention and the lower branch is the channel attention. Finally, the outputs of the upper and lower branches are multiplied with the original feature map by element-wise multiplication. S41, the spatial channel joint attention block is divided into an upper branch and a lower branch, and the input processed image features F p The processed image features F p Divided into The process is divided into two parts, each processed by a separate branch. For the upper branch, the input features are... Two feature maps are obtained by performing max pooling and average pooling respectively. The max pooling layer can preserve the main features in the feature map and suppress noise and unimportant details. MaxPool() represents the max pooling operation. Average pooling layers reduce the size of the feature map by calculating the average value of local regions. AvgPool() performs an average pooling operation, then concatenates the pooling results. `Concat()` represents a concatenation operation. The concatenated features are passed through a 3×3 convolutional layer, which helps capture spatial dependencies within local regions. Then, a sigmoid activation function is used to obtain a spatial attention map. To enhance the response in important regions, the output map is element-wise multiplied with the original feature map, and a 3×3 convolutional layer and ReLU activation function are used to further enhance the expressive power of local features. This process can be described as follows: Re() represents the ReLU activation function, Si() represents the sigmoid activation function, and Conv() represents the ReLU activation function. 3×3 () represents a 3×3 convolution. This represents element-wise multiplication; S42. For the lower branch, input features The algorithm is further divided into two branches. The upper branch performs global average pooling, and then uses 1×1 convolution and ReLU activation function to obtain the attention for each channel. GloAvgPool() represents a global average pooling operation. The next branch uses global average pooling and then reverses, swapping the preceding and following channels. This process can be described as follows: Then, a 1×1 convolution and ReLU activation function are used, followed by skip connections to the deflected result, and finally, the result is deflected again. This process can be described as follows: Rev() represents the reverse operation, which finally performs a weighted concatenation of the outputs of the two branches. This process can be described as follows: λ represents the weight, which is equal to 0.7, and + means element-wise addition; S43. After enhancing the high-frequency and low-frequency features, the outputs of the two branches are then compared with the initial input features F. p Element-wise multiplication yields the final output of the spatial channel attention block. This represents element-wise multiplication; S5. Construct a feature processing block. The feature processing block goes through a high-frequency feature enhancement module and a low-frequency feature enhancement module. The outputs of the two modules are connected and then connected through a 1x1 convolution and a spatial channel joint attention block. Then, through a 1x1 convolution and pixel normalization, the output features are finally connected to the input jump connection. S6. Construct a spatial channel joint attention network. Input the image features f0 into the spatial channel joint attention network. The spatial channel joint attention network includes n feature processing blocks and a feature fusion layer. S7. Construct an image reconstruction module, which performs convolution and upsampling on the fused features; S8. Construct an image super-resolution model, including input, feature extraction, spatial channel joint attention network, image reconstruction, and image output; S9. Input the images of automotive parts into the super-resolution image model, and obtain the super-resolution image after processing.

2. The method for super-resolution of automotive component images based on spatial channel joint attention according to claim 1, wherein in step S2, a high-frequency feature enhancement module is constructed, characterized in that: Input the initial features F of the automotive component image in The initial features F of the automotive parts image in The sample is processed by a 7x7 Gaussian kernel and the initial feature F is used. in Subtracting the features processed by the Gaussian kernel, using Gaussian kernel blurring to approximate the low-frequency components, and subtracting the low-frequency components from the initial features yields the high-frequency components, F h =F in -Guassian 7×7 (F in ), Guassian 7×7 () represents a 7x7 Gaussian kernel. High-frequency features are then processed through a 3x3 convolution and a ReLU activation function. The 3x3 convolution extracts the high-frequency features, and the ReLU activation function generates an attention map. Skip connections are then used to multiply the high-frequency features and the attention map pixel-by-pixel to further enhance the high-frequency region. To avoid over-enhancement, residual connections are used to combine the enhanced high-frequency features with the original high-frequency features. Finally, the combined features are processed through a 3x3 convolution and a ReLU activation function. This process can be represented as: Conv 3×3 () represents a 3×3 convolution, and Re() represents the ReLU activation function. + represents element-wise multiplication, and + represents element-wise addition. This is the output of the high-frequency feature enhancement module.

3. The image super-resolution method for automotive parts based on spatial channel joint attention according to claim 2, wherein in step S3, a low-frequency feature enhancement module is constructed, characterized in that: Input the initial features F of the automotive component image in The initial features F of the automotive parts image in Low-frequency features are extracted using a 3x3 convolutional layer. Then, by calculating the range of variation in the low-frequency region, redundancy is removed based on a pre-set threshold. The resulting feature F is the result of this redundancy removal. l =Threshold(Conv 3×3 (F in Threshold() represents thresholding to remove redundancy, where τ represents the preset threshold, and the mathematical model of τ is τ = μ. f_in (F in )+2×σ f_in (F in ), μ f_in (F in ) is the input feature F in The mean, σ f_in (F in ) is the input feature F in The standard deviation of the low-frequency features is used to remove redundancy. For low-frequency features, a difference operation is used to subtract the local mean of the low-frequency region to avoid loss of detail or over-enhancement. Then, an enhancement factor is multiplied to enhance the low-frequency features. Residual connections are then used to combine the enhanced low-frequency features with the original low-frequency features. Finally, a 3×3 convolution and ReLU activation function are applied. This process can be represented as: This is the output of the low-frequency feature enhancement module. Re() represents the ReLU activation function, μ is the local mean of the low-frequency region, and α is the enhancement factor. The mathematical model for α is: σ f_l (F l ) is the standard deviation of the low-frequency features after redundancy removal, σ f_in (F in ) is the standard deviation of the input features.

4. The method for super-resolution of automotive component images based on spatial channel joint attention according to claim 3, wherein in step S5, a feature processing block is constructed, characterized in that: The feature processing block handles the input automotive component image features F. in First, it passes through a high-frequency feature enhancement module and a low-frequency feature enhancement module. After enhancement, the high-frequency features... Hrp() represents the high-frequency feature enhancement module, which enhances low-frequency features. Lrp() represents the low-frequency feature enhancement module. It then concatenates the enhanced high-frequency and low-frequency features and fuses them using a 1×1 convolution. The fused features... Concat() represents a join operation, Conv 1×1 () represents a 1×1 fusion convolution; subsequently, the features are finely adjusted using a spatial channel attention joint module, employing 1×1 convolution for feature transformation, while pixel normalization is added to improve model performance. Finally, long skip connections are used to enhance the model's residual learning ability, yielding the final output F of the feature processing block. out =Norm(Conv 1×1 (Scja(F fusion )))+F in Norm() represents pixel normalization, and Scja() represents spatial channel joint attention block.

5. The method for super-resolution of automotive component images based on spatial channel joint attention according to claim 4, wherein in step S6, a spatial channel joint attention network is constructed, characterized in that: For the input feature F0, we concatenate the outputs of each feature processing block and then perform a concatenation operation on the outputs of each feature processing block. The concatenation results are then fused using a 1x1 convolution, followed by a 3x3 convolution to further refine the features. Finally, the original feature F0 is added to the network output through skip connections, resulting in the final output F of the spatial channel attention network. out =Conv 3×3 (Conv 1×1 (Concat(T1, T2, ..., T) n )))+F0,T n This represents the output of the nth spatial channel attention block.

Images