Lung cancer image enhancement method based on generative adversarial network

Abstract

The invention discloses a lung cancer image enhancement method based on a generative adversarial network. The method comprises the following steps: S1, obtaining a standardized lung cancer plaque sheet image data set; s2, lung cancer plaque sheet image training data with labels are formed; s3, generating a sheet-shaped multi-scale dynamic feature map of the lung cancer spots; s4, outputting the primarily reconstructed lung cancer plaque sheet image data; s5, constructing a local sensing discriminator; s6, obtaining a trained dilated convolution fusion generative adversarial network model; and S7, inputting newly obtained low-quality lung cancer plaque sheet image data into the trained cavity convolution fusion generative adversarial network model, outputting reconstructed lung cancer plaque sheet image data, and generating a lung cancer plaque sheet image quality evaluation result. According to the invention, the authenticity and credibility of the generated image in the key focus area are improved.

Description

Technical Field

[0001] The present invention relates to the technical field of lung cancer imaging, and in particular to a lung cancer image enhancement method based on a generative adversarial network. Background Art

[0002] With the continuous development of medical imaging technology, early screening and diagnosis of lung cancer increasingly rely on imaging data acquired through computed tomography and magnetic resonance imaging devices. In particular, patchy lung images acquired under low-dose scanning conditions have become an important data source for lung cancer screening. However, due to the reduced signal-to-noise ratio, increased artifacts, and loss of image details caused by low-dose imaging itself, existing lung cancer images have significant deficiencies in actual clinical applications, affecting doctors' accurate identification and analysis of patchy lung cancer lesions.

[0003] Currently, there are two main methods for processing low-quality lung cancer images. One is enhancement methods based on traditional image processing, such as histogram equalization, sharpening filtering, and contrast stretching. Although these methods can improve the overall clarity of the image to a certain extent, they are difficult to effectively restore the structural details of the patchy lesion areas of lung cancer and are prone to introducing noise, resulting in artifact enhancement, which affects subsequent lesion identification and quantitative assessment. The other is image enhancement methods based on deep learning, such as using convolutional neural networks or generative adversarial networks to reconstruct the entire image. However, these methods generally have the following problems: On the one hand, due to the failure to fully consider the local characteristics of the patchy lesion areas of lung cancer, the model is prone to ignore lesion details or produce over-smoothing during the enhancement process, which reduces the recognizability of the lesion area; on the other hand, existing methods often lack specificity in network structure design, ignoring the complex and changing characteristics of the edge texture of patchy lesions, resulting in blurred boundaries or structural distortion, and failing to meet the actual clinical needs for high-precision diagnostic support.

[0004] In addition, in existing deep learning methods, although some studies have attempted to introduce dilated convolution to expand the receptive field, a fixed dilation rate is usually used, and there is a lack of a mechanism to adaptively adjust the dilation parameters according to the local characteristics of the lesion, resulting in poor feature extraction in areas of different scales and texture complexity, further limiting the overall performance of image enhancement. At the same time, the existing generator and discriminator designs are generally based on global discrimination, lacking refined perception and authenticity assessment of the local area of the lesion, resulting in obvious defects in the recovery of fine-grained features in the reconstructed image, especially the lack of authenticity in the transition area of the lesion edge.

[0005] In summary, existing technologies for enhancing patchy lung cancer images generally have the following defects: insufficient perception of local lesion features, weak structural detail reconstruction capabilities, poor boundary preservation, and disconnection between overall enhancement and local authenticity assessment. These defects severely restrict the application value of low-quality images in the clinical early diagnosis of lung cancer. Therefore, a new method that can take into account both local feature perception and overall quality optimization is urgently needed to improve the reconstruction quality and lesion identification accuracy of low-quality patchy lung cancer images, and meet the urgent demand for high-quality lung imaging data in precision medicine. Summary of the Invention

[0006] One object of the present invention is to propose a lung cancer image enhancement method based on a generative adversarial network, which improves the authenticity and credibility of the generated images in key lesion areas.

[0007] A lung cancer image enhancement method based on a generative adversarial network according to an embodiment of the present invention includes the following steps:

[0008] S1. Collecting lung cancer patchy image data and establishing a lung cancer patchy image dataset, preprocessing the lung cancer patchy image dataset, including intensity normalization, artifact removal, and region of interest cropping, to obtain a standardized lung cancer patchy image dataset;

[0009] S2. Generate labeled information for the patchy lung cancer lesion region based on the manual delineation results of the radiologist, and pair the labeled information for the patchy lung cancer lesion region with a standardized patchy lung cancer image dataset to form labeled patchy lung cancer image training data;

[0010] S3. Construct a dilated convolution feature extraction module to perform multi-scale feature extraction on the labeled lung cancer patchy image training data to generate a multi-scale dynamic feature map of lung cancer patchiness;

[0011] S4. Construct a lightweight residual generator network, input the multi-scale dynamic feature map of lung cancer patchiness into the lightweight residual generator network, and output preliminary reconstructed lung cancer patchiness image data;

[0012] S5. Construct a local perception discriminator, which includes a global discrimination branch and a local discrimination branch. The global discrimination branch receives the preliminary reconstructed lung cancer patchy image data, and the local discrimination branch receives the lung cancer patchy lesion area image block cropped according to the lung cancer patchy lesion area annotation information.

[0013] S6. Using a standardized lung cancer patchy image dataset as input, the patchy lung cancer lesion region annotation information as the regional weights, the dilated convolutional feature extraction module and the lightweight residual generator network as the generator, and the local-aware discriminator as the discriminator, we jointly trained a dilated convolutional fusion generative adversarial network model using structure-aware loss, edge-preserving loss, and adversarial loss to obtain the trained dilated convolutional fusion generative adversarial network model.

[0014] S7. Input the newly acquired low-quality lung cancer patchy image data into the trained atrous convolutional fusion generative adversarial network model, output the reconstructed lung cancer patchy image data, input the reconstructed lung cancer patchy image data into the image quality assessment module, and generate the lung cancer patchy image quality assessment results.

[0015] Optionally, the S1 specifically includes the following steps:

[0016] S11. Collect lung cancer patchy image data acquired by computed tomography or magnetic resonance imaging equipment under clinical low-dose scanning conditions, and pair each acquired original lung cancer patchy image with its corresponding basic image metadata to establish a lung cancer patchy image dataset. Each data sample in the lung cancer patchy image dataset consists of an original lung cancer patchy image I i and the corresponding basic image metadata m i The total number of lung cancer patchy image data samples is N;

[0017] S12. Performing intensity normalization on each original lung cancer patchy image in the lung cancer patchy image dataset, linearly mapping the pixel intensity of each original lung cancer patchy image according to an interval, using the minimum pixel value and the maximum pixel value of the original lung cancer patchy image as normalization references during the normalization process, so that the pixel intensity values of the normalized lung cancer patchy images fall within the interval, thereby obtaining a normalized lung cancer patchy image dataset;

[0018] S13. Performing artifact removal processing on each patchy normalized lung cancer image in the patchy normalized lung cancer image dataset, using a global histogram threshold segmentation method, setting an artifact removal intensity threshold, and marking a pixel as a valid pixel if the pixel intensity of a pixel in the patchy normalized lung cancer image is greater than or equal to the artifact removal intensity threshold; otherwise, marking the pixel as an invalid pixel, removing artifact regions from the patchy normalized lung cancer image to obtain a patchy artifact-removed lung cancer image dataset;

[0019] S14. Based on the lung cancer patchy artifact removal image dataset, the corresponding lung area is extracted as the region of interest according to the lung anatomical structure detection results in the image. The region of interest consists of all pixel coordinates (x, y) belonging to the lung area mask. The extracted area is defined as the lung cancer patchy region of interest R i , forming a standardized lung cancer patchy image dataset D roi .

[0020] Optionally, the step S2 specifically includes the following steps:

[0021] S21. A radiologist is responsible for the image of each lung cancer patchy region of interest R in the standardized lung cancer patchy image dataset. i Perform manual delineation operation, for each lung cancer patchy region of interest image R i Determine the corresponding lung cancer patchy lesion area contour and generate the lung cancer patchy lesion area mask M i :

[0022]

[0023] Among them, M i (x, y) represents the mask of the patchy lung cancer lesion area at the coordinate position (x, y) in the i-th patchy lung cancer region of interest image;

[0024] S22. Each lung cancer patchy region of interest image R in the standardized lung cancer patchy image dataset i The corresponding lung cancer patchy lesion area mask M i Pairing is performed to construct a pair of lung cancer patchy images and annotation information (R i ,M i );

[0025] S23. Based on the patchy lesion area mask M of lung cancer i Calculate the area ratio A of the lesion area in each lung cancer patchy region of interest image i :

[0026]

[0027] Among them, ∑ (x,y) M i (x,y) represents the number of all pixels belonging to the patchy lesion area of lung cancer in the i-th patchy region of interest image of lung cancer, ∑ (x,y) 1 represents the total number of pixels in the entire lung cancer patchy region of interest image;

[0028] S24. According to the area ratio A iThe statistical distribution of the lung cancer patchy images and annotation information pairs are obtained by filtering out abnormal samples whose area ratio is lower than the set threshold, retaining the lung cancer patchy images and annotation information pairs that meet the requirements, and forming the labeled lung cancer patchy image training dataset D with the retained lung cancer patchy images and annotation information pairs. train .

[0029] Optionally, the S3 specifically includes the following steps:

[0030] S31. Read each pair of lung cancer patchy region of interest image and lung cancer patchy lesion area mask (R i ,M i ), for the patchy region of interest image R of lung cancer i Patchy lesion area of lung cancer mask M i Perform pixel-by-pixel product operations to obtain enhanced images of patchy lesions of lung cancer

[0031] S32. Based on enhanced imaging of patchy lesions in lung cancer Calculate the local texture gradient intensity map G i , local texture gradient intensity map G i The gradient intensity value G of each pixel position (x, y) i (x,y) is defined as the enhanced image of the patchy lung cancer lesion along the transverse direction at that location The square of the rate of change and the enhanced image of lung cancer patchy lesions along the longitudinal direction The result of taking the square root of the sum of the squares of the rates of change;

[0032]

[0033] S33. According to the local texture gradient intensity map G i Dynamically generate void ratio map D i , void ratio graph D i The hole rate value D of each pixel position (x, y) i (x,y) is based on the preset minimum void ratio d min and the maximum void ratio d max The difference multiplied by one minus the local texture gradient intensity value G i (x, y) and the maximum value of the local texture gradient intensity map G max After adding the minimum void ratio d min to obtain;

[0034]

[0035] Among them, d min and d max are the preset minimum void rate and maximum void rate, Gmax Represents the local texture gradient intensity map G i The maximum value in ;

[0036] S34. Based on enhanced imaging of patchy lesions in lung cancer and dynamic void ratio graph D i (x, y), adaptively apply the corresponding dilated convolution operation to each position (x, y) to extract the patchy multi-scale dynamic feature map F of lung cancer i (x,y);

[0037]

[0038] in, Indicated by the void ratio D i The local neighborhood of (x,y) is sampled, W(u,v) is the convolution kernel weight, and σ(·) is the activation function.

[0039] Optionally, the S4 specifically includes the following steps:

[0040] S41. The multi-scale dynamic feature map of lung cancer patchy shape extracted by dynamic void ratio mechanism F i Input to the feature input end of the lightweight residual generator network, the lightweight residual generator network includes multiple stacked lightweight residual units;

[0041] S42. Multi-scale dynamic feature map of lung cancer patchiness in the kth lightweight residual unit Perform convolution transformation operation, use convolution kernel weight and bias term in the convolution process, process the convolution transformation operation result through lightweight nonlinear activation function, obtain the convolution feature output of the kth lightweight residual unit, and compare the convolution feature output with the input lung cancer patchy multi-scale dynamic feature map Add element-wise to generate the output feature map of the kth lightweight residual unit

[0042] S43. After all lightweight residual units are processed, the final feature map is obtained The final feature map Input to the feature reconstruction module, which generates preliminary reconstruction of lung cancer patchy image data through a layer of standard convolution and Tanh activation function

[0043] Optionally, in each lightweight residual unit, the convolution transformation operation uses depth-wise separable convolution instead of standard convolution, and the depth-wise separable convolution includes two steps: depth-wise convolution and point-wise convolution.

[0044] Optionally, the S5 specifically includes the following steps:

[0045] S51. Preliminary reconstruction of lung cancer patchy imaging data The corresponding lung cancer patchy lesion area mask M i The local perception discriminator is used to identify the patchy lesion area of lung cancer. i Calculating the distance transformation map U of patchy lung cancer lesions i :

[0046]

[0047] in, Represents the boundary between pixel (x,y) and the patchy lesion area of lung cancer The Euclidean distance of

[0048] S52. Determine the distance transformation map U of the patchy lung cancer lesions i The pixel coordinate with the largest normalized Euclidean distance Centered on pixel coordinates, multi-scale image blocks of lung cancer patchy lesions are cropped according to the scale set.

[0049] S53. In the global discriminant branch, the preliminarily reconstructed lung cancer patchy image data is processed by convolution-global average pooling, and then the global authenticity score is obtained by the fully connected layer and Sigmoid activation function.

[0050] S54. In the multi-scale local discriminant branch, each multi-scale image block of the lung cancer patch is Input convolutional network to obtain local feature map Output scale-local authenticity score through fully connected layer and Sigmoid activation function

[0051] S55. Based on the distance transformation map U of the patchy lesions of lung cancer i Calculate the multi-scale image blocks of each lung cancer patch The scale attention weight The scale attention weight is obtained by averaging the normalized Euclidean distance of all pixels in the patchy lesion area of lung cancer and then normalizing it. The weighted sum of each scale-local authenticity score is calculated according to the scale attention weight to obtain the weighted local authenticity score.

[0052] S56. Calculate the average gradient intensity of the edge gradient map of the patchy lung cancer lesions Get the edge integrity adjustment factor γ i , using the edge integrity adjustment factor to score the global authenticity and weighted local authenticity score Perform linear fusion to obtain a joint authenticity score

[0053]

[0054] Optionally, the S6 specifically includes the following steps:

[0055] S61. Standardize the lung cancer patchy image dataset D roi Each lung cancer patchy region of interest image R i As the generator input, the lung cancer patchy lesion area mask M i As a reference for regional weights, it is input into the generator composed of the dilated convolution feature extraction module and the lightweight residual generator network to generate preliminary reconstructed lung cancer patchy image data.

[0056] S62. Preliminary reconstruction of lung cancer patchy imaging data and the corresponding lung cancer patchy region of interest image R i Input the local perception discriminator to obtain the joint authenticity score

[0057] S63. Calculate the structure-aware loss L struct , the structure-aware loss aims to maintain the continuity of the patchy image structure of lung cancer, which is defined as the patchy region of interest image R i and preliminary reconstruction of lung cancer patchy imaging data The weighted squared error of the local gradient difference:

[0058]

[0059] in, and Represents the gradient values of the original image and the reconstructed image at position (x, y), U i (x, y) is the distance transformation map weight of lung cancer patchy lesions;

[0060] S64. Calculate edge preservation loss L edge , the edge preservation loss aims to enhance the boundary clarity of the lung cancer patchy lesion area, which is defined as the lung cancer patchy lesion area mask M i Edge detection error under the action of:

[0061]

[0062] Among them, E i (x,y) and Represent the edge gradient values of the original image and the preliminary reconstructed image at position (x, y);

[0063] S65. Calculate the adversarial loss Ladv , the adversarial loss aims to improve the realism of lung cancer patchy images and is defined as the joint realism score Negative log loss of

[0064] S66. Integrate structure-aware loss, edge-preserving loss, and adversarial loss to construct the total loss function L total :

[0065] L total =λ struct L struct +λ edge L edge +λ adv L adv ,

[0066] Among them, λ struct ,λ edge ,λ adv are the weighted coefficients of each loss item respectively;

[0067] S67. Take the total loss function L total To achieve the optimization goal, the parameters of the dilated convolution feature extraction module, the lightweight residual generator network and the local perception discriminator are jointly optimized until the training converges, and the trained dilated convolution fusion generative adversarial network model is obtained.

[0068] The beneficial effects of the present invention are:

[0069] (1) The present invention adopts a dilated convolution feature extraction module with dynamic dilation rate control. It dynamically generates a dilation rate map based on the local texture gradient intensity map of the patchy lesion area of lung cancer, and adaptively applies a dilated convolution operation based on the dilation rate of each pixel position, thereby realizing multi-scale dynamic feature extraction. It can adaptively adjust the receptive field size according to the texture complexity of different positions in the lesion area, apply a small dilation rate to finely extract features in areas rich in details, and apply a large dilation rate to enhance contextual information fusion in areas with simple structures, effectively avoiding the problem of detail loss or insufficient contextual information.

[0070] (2) The present invention constructs a lightweight residual generator network, adopts depthwise separable convolution to replace traditional standard convolution in the feature reconstruction process, and introduces stacked lightweight residual units, which effectively reduces the number of model parameters and computational complexity while maintaining efficient information flow and feature fusion capabilities. When processing the patchy multi-scale dynamic feature map of lung cancer, the lightweight residual generator can more fully retain local detail information and overall structural continuity, avoiding the phenomenon of excessive smoothing or artifact diffusion in the reconstructed image.

[0071] (3) The present invention introduces a local perception discriminator and combines it with a scale attention weighting mechanism. By calculating the distance transformation map of the patchy lesion area of lung cancer, multi-scale local image blocks are extracted, and weights are dynamically assigned according to the importance of the local areas at each scale. The local authenticity of the generated image is finely judged, which can effectively evaluate the authenticity of the lesion edge and internal structure at a fine-grained level, further improving the authenticity and credibility of the generated image in the key lesion area. BRIEF DESCRIPTION OF THE DRAWINGS

[0072] The accompanying drawings are used to provide a further understanding of the present invention and constitute a part of the specification. Together with the embodiments of the present invention, they are used to explain the present invention and do not constitute a limitation of the present invention. In the accompanying drawings:

[0073] Figure 1 This is a flowchart of a lung cancer image enhancement method based on generative adversarial networks proposed in the present invention. DETAILED DESCRIPTION

[0074] The present invention will now be described in further detail with reference to the accompanying drawings, which are simplified schematic diagrams that illustrate the basic structure of the present invention in a schematic manner.

[0075] refer to Figure 1 , a lung cancer image enhancement method based on generative adversarial network, comprising the following steps:

[0076] S1. Collecting lung cancer patchy image data and establishing a lung cancer patchy image dataset, preprocessing the lung cancer patchy image dataset, including intensity normalization, artifact removal, and region of interest cropping, to obtain a standardized lung cancer patchy image dataset;

[0077] S2. Generate labeled information for the patchy lung cancer lesion region based on the manual delineation results of the radiologist, and pair the labeled information for the patchy lung cancer lesion region with a standardized patchy lung cancer image dataset to form labeled patchy lung cancer image training data;

[0078] S3. Construct a dilated convolution feature extraction module to perform multi-scale feature extraction on the labeled lung cancer patchy image training data to generate a multi-scale dynamic feature map of lung cancer patchiness;

[0079] S4. Construct a lightweight residual generator network, input the multi-scale dynamic feature map of lung cancer patchiness into the lightweight residual generator network, and output preliminary reconstructed lung cancer patchiness image data;

[0080] S5. Construct a local perception discriminator, which includes a global discrimination branch and a local discrimination branch. The global discrimination branch receives the preliminary reconstructed lung cancer patchy image data, and the local discrimination branch receives the lung cancer patchy lesion area image block cropped according to the lung cancer patchy lesion area annotation information.

[0081] S6. Using a standardized lung cancer patchy image dataset as input, the patchy lung cancer lesion region annotation information as the regional weights, the dilated convolutional feature extraction module and the lightweight residual generator network as the generator, and the local-aware discriminator as the discriminator, we jointly trained a dilated convolutional fusion generative adversarial network model using structure-aware loss, edge-preserving loss, and adversarial loss to obtain the trained dilated convolutional fusion generative adversarial network model.

[0082] S7. Input the newly acquired low-quality lung cancer patchy image data into the trained atrous convolutional fusion generative adversarial network model, output the reconstructed lung cancer patchy image data, input the reconstructed lung cancer patchy image data into the image quality assessment module, and generate the lung cancer patchy image quality assessment results.

[0083] In this embodiment, S1 specifically includes the following steps:

[0084] S11. Collect lung cancer patchy image data acquired by computed tomography or magnetic resonance imaging equipment under clinical low-dose scanning conditions, and pair each acquired original lung cancer patchy image with its corresponding basic image metadata to establish a lung cancer patchy image dataset. Each data sample in the lung cancer patchy image dataset consists of an original lung cancer patchy image I i and the corresponding basic image metadata m i The total number of lung cancer patchy image data samples is N;

[0085] S12. Performing intensity normalization on each original lung cancer patchy image in the lung cancer patchy image dataset, linearly mapping the pixel intensity of each original lung cancer patchy image according to an interval, using the minimum pixel value and the maximum pixel value of the original lung cancer patchy image as normalization references during the normalization process, so that the pixel intensity values of the normalized lung cancer patchy images fall within the interval, thereby obtaining a normalized lung cancer patchy image dataset;

[0086] S13. Performing artifact removal processing on each patchy normalized lung cancer image in the patchy normalized lung cancer image dataset, using a global histogram threshold segmentation method, setting an artifact removal intensity threshold, and marking a pixel as a valid pixel if the pixel intensity of a pixel in the patchy normalized lung cancer image is greater than or equal to the artifact removal intensity threshold; otherwise, marking the pixel as an invalid pixel, removing artifact regions from the patchy normalized lung cancer image to obtain a patchy artifact-removed lung cancer image dataset;

[0087] S14. Based on the lung cancer patchy artifact removal image dataset, the corresponding lung area is extracted as the region of interest according to the lung anatomical structure detection results in the image. The region of interest consists of all pixel coordinates (x, y) belonging to the lung area mask. The extracted area is defined as the lung cancer patchy region of interest R i , forming a standardized lung cancer patchy image dataset D roi .

[0088] In this embodiment, S2 specifically includes the following steps:

[0089] S21. A radiologist is responsible for the image of each lung cancer patchy region of interest R in the standardized lung cancer patchy image dataset. i Perform manual delineation operation, for each lung cancer patchy region of interest image R i Determine the corresponding lung cancer patchy lesion area contour and generate the lung cancer patchy lesion area mask M i :

[0090]

[0091] Among them, M i (x, y) represents the mask of the patchy lung cancer lesion area at the coordinate position (x, y) in the i-th patchy lung cancer region of interest image;

[0092] S22. Each lung cancer patchy region of interest image R in the standardized lung cancer patchy image dataset i The corresponding lung cancer patchy lesion area mask M i Pairing is performed to construct a pair of lung cancer patchy images and annotation information (R i ,M i );

[0093] S23. Based on the patchy lesion area mask M of lung cancer i Calculate the area ratio A of the lesion area in each lung cancer patchy region of interest image i :

[0094]

[0095] Among them, ∑ (x,y) M i (x,y) represents the number of all pixels belonging to the patchy lesion area of lung cancer in the i-th patchy region of interest image of lung cancer, ∑ (x,y) 1 represents the total number of pixels in the entire lung cancer patchy region of interest image;

[0096] S24. According to the area ratio A i The statistical distribution of the lung cancer patchy images and annotation information pairs are obtained by filtering out abnormal samples whose area ratio is lower than the set threshold, retaining the lung cancer patchy images and annotation information pairs that meet the requirements, and forming the labeled lung cancer patchy image training dataset D with the retained lung cancer patchy images and annotation information pairs. train .

[0097] In this embodiment, S3 specifically includes the following steps:

[0098] S31. Read each pair of lung cancer patchy region of interest image and lung cancer patchy lesion area mask (R i ,M i ), for the patchy region of interest image R of lung cancer i Patchy lesion area of lung cancer mask M i Perform pixel-by-pixel product operations to obtain enhanced images of patchy lesions of lung cancer

[0099] S32. Based on enhanced imaging of patchy lesions in lung cancer Calculate the local texture gradient intensity map G i , local texture gradient intensity map G i The gradient intensity value G of each pixel position (x, y) i (x,y) is defined as the enhanced image of the patchy lung cancer lesion along the transverse direction at that location The square of the rate of change and the enhanced image of lung cancer patchy lesions along the longitudinal direction The result of taking the square root of the sum of the squares of the rates of change;

[0100]

[0101] S33. According to the local texture gradient intensity map G i Dynamically generate void ratio map D i , void ratio graph D i The hole rate value D of each pixel position (x, y) i (x,y) is based on the preset minimum void ratio d min and the maximum void ratio d max The difference multiplied by one minus the local texture gradient intensity value G i (x, y) and the maximum value of the local texture gradient intensity map Gmax After adding the minimum void ratio d min to obtain;

[0102]

[0103] Among them, d min and d max are the preset minimum void rate and maximum void rate, G max Represents the local texture gradient intensity map G i The maximum value in ;

[0104] S34. Based on enhanced imaging of patchy lesions in lung cancer and dynamic void ratio graph D i (x, y), adaptively apply the corresponding dilated convolution operation to each position (x, y) to extract the patchy multi-scale dynamic feature map F of lung cancer i (x,y);

[0105]

[0106] in, Indicated by the void ratio D i The local neighborhood of (x,y) is sampled, W(u,v) is the convolution kernel weight, and σ(·) is the activation function.

[0107] In this embodiment, S4 specifically includes the following steps:

[0108] S41. The multi-scale dynamic feature map of lung cancer patchy shape extracted by dynamic void ratio mechanism F i Input to the feature input end of the lightweight residual generator network, the lightweight residual generator network includes multiple stacked lightweight residual units;

[0109] S42. Multi-scale dynamic feature map of lung cancer patchiness in the kth lightweight residual unit Perform convolution transformation operation, use convolution kernel weight and bias term in the convolution process, process the convolution transformation operation result through lightweight nonlinear activation function, obtain the convolution feature output of the kth lightweight residual unit, and compare the convolution feature output with the input lung cancer patchy multi-scale dynamic feature map Add element-wise to generate the output feature map of the kth lightweight residual unit

[0110] S43. After all lightweight residual units are processed, the final feature map is obtained The final feature map Input to the feature reconstruction module, which generates preliminary reconstruction of lung cancer patchy image data through a layer of standard convolution and Tanh activation function

[0111] In this embodiment, in each lightweight residual unit, the convolution transformation operation uses depthwise separable convolution instead of standard convolution, and the depthwise separable convolution includes two steps: depthwise convolution and pointwise convolution.

[0112] In this embodiment, S5 specifically includes the following steps:

[0113] S51. Preliminary reconstruction of lung cancer patchy imaging data The corresponding lung cancer patchy lesion area mask M i The local perception discriminator is used to identify the patchy lesion area of lung cancer. i Calculating the distance transformation map U of patchy lung cancer lesions i :

[0114]

[0115] in, Represents the boundary between pixel (x,y) and the patchy lesion area of lung cancer The Euclidean distance of

[0116] S52. Determine the distance transformation map U of the patchy lung cancer lesions i The pixel coordinate with the largest normalized Euclidean distance Centered on pixel coordinates, multi-scale image blocks of lung cancer patchy lesions are cropped according to the scale set.

[0117] S53. In the global discriminant branch, the preliminarily reconstructed lung cancer patchy image data is processed by convolution-global average pooling, and then the global authenticity score is obtained by the fully connected layer and Sigmoid activation function.

[0118] S54. In the multi-scale local discriminant branch, each multi-scale image block of the lung cancer patch is Input convolutional network to obtain local feature map Output scale-local authenticity score through fully connected layer and Sigmoid activation function

[0119] S55. Based on the distance transformation map U of the patchy lesions of lung cancer i Calculate the multi-scale image blocks of each lung cancer patch The scale attention weight The scale attention weight is obtained by averaging the normalized Euclidean distance of all pixels in the patchy lesion area of lung cancer and then normalizing it. The weighted sum of each scale-local authenticity score is calculated according to the scale attention weight to obtain the weighted local authenticity score.

[0120] S56. Calculate the average gradient intensity of the edge gradient map of the patchy lung cancer lesions Get the edge integrity adjustment factor γ i , using the edge integrity adjustment factor to score the global authenticity and weighted local authenticity score Perform linear fusion to obtain a joint authenticity score

[0121]

[0122] In this embodiment, S6 specifically includes the following steps:

[0123] S61. Standardize the lung cancer patchy image dataset D roi Each lung cancer patchy region of interest image R i As the generator input, the lung cancer patchy lesion area mask M i As a reference for regional weights, it is input into the generator composed of the dilated convolution feature extraction module and the lightweight residual generator network to generate preliminary reconstructed lung cancer patchy image data.

[0124] S62. Preliminary reconstruction of lung cancer patchy imaging data and the corresponding lung cancer patchy region of interest image R i Input the local perception discriminator to obtain the joint authenticity score

[0125] S63. Calculate the structure-aware loss L struct , the structure-aware loss aims to maintain the continuity of the patchy image structure of lung cancer, which is defined as the patchy region of interest image R i and preliminary reconstruction of lung cancer patchy imaging data The weighted squared error of the local gradient difference:

[0126]

[0127] in, and Represents the gradient values of the original image and the reconstructed image at position (x, y), U i (x, y) is the distance transformation map weight of lung cancer patchy lesions;

[0128] S64. Calculate edge preservation loss L edge , the edge preservation loss aims to enhance the boundary clarity of the lung cancer patchy lesion area, which is defined as the lung cancer patchy lesion area mask M i Edge detection error under the action of:

[0129]

[0130] Among them, E i (x,y) and Represent the edge gradient values of the original image and the preliminary reconstructed image at position (x, y);

[0131] S65. Calculate the adversarial loss L adv , the adversarial loss aims to improve the realism of lung cancer patchy images and is defined as the joint realism score Negative log loss of

[0132] S66. Integrate structure-aware loss, edge-preserving loss, and adversarial loss to construct the total loss function L total :

[0133] L total =λ struct L struct +λ edge L edge +λ adv L adv ,

[0134] Among them, λ struct ,λ edge ,λ adv are the weighted coefficients of each loss item respectively;

[0135] S67. Take the total loss function L total To achieve the optimization goal, the parameters of the dilated convolution feature extraction module, the lightweight residual generator network and the local perception discriminator are jointly optimized until the training converges, and the trained dilated convolution fusion generative adversarial network model is obtained.

[0136] In this embodiment, the step S7 specifically includes the following steps:

[0137] S71. Acquire newly acquired low-quality patchy lung cancer image data Low-quality patchy lung cancer image data Input the trained dilated convolutional fusion generative adversarial network model and output the reconstructed lung cancer patchy image data.

[0138] S72. Reconstructing patchy lung cancer imaging data Input the image quality assessment module to extract the following four image quality indicators:

[0139] The structural clarity index is used to measure the integrity of the tissue structure in patchy lung cancer lesions;

[0140] Edge sharpness index, used to measure the clarity of the boundaries of patchy lung cancer lesions;

[0141] Texture fidelity, a metric used to measure the level of texture detail recovery within patchy lung cancer lesions;

[0142] The overall visual consistency index is used to measure the overall perceptual consistency between the reconstructed image and the original image;

[0143] S73. Extract classification criteria from historical lung cancer patchy image reconstruction assessment results and define high-quality, medium-quality, and low-quality image sample classification rules based on structural clarity, edge sharpness, texture fidelity, and overall visual consistency, where:

[0144] High-quality samples: structural clarity index Q struct,j ≥T struct , edge sharpness index Q edge,j ≥T edge , texture fidelity index Q texture,j ≥T texture , overall visual consistency index Q vis,j ≥T vis ;

[0145] Medium quality samples: meet the evaluation conditions of any three indicators;

[0146] Low-quality samples: fewer than three indicators meet the evaluation criteria;

[0147] Among them, T struct 、T edge 、T texture 、T vis They are preset thresholds for structural clarity, edge sharpness, texture fidelity, and overall visual consistency evaluation;

[0148] S74. Output the lung cancer patchy image quality assessment results and store and record them according to the classification of high-quality samples, medium-quality samples, and low-quality samples for subsequent clinical application screening and dynamic optimization of model performance.

[0149] Example 1:

[0150] In August 2024, the lung cancer early screening special project team located in the imaging center of Hospital A launched a new technology verification study on the enhancement of patchy images of lung cancer under low-dose CT scanning conditions. The project aims to solve the problems of many artifacts, blurred lesion boundaries, and missing details in current low-dose lung images, improve the visualization of patchy lesion areas of lung cancer, and assist doctors in more accurately identifying early tiny lesions.

[0151] In the actual test, the project team selected 527 low-dose chest CT scan data collected between May 2023 and March 2024. The patients were aged between 45 and 76 years old and were all high-risk lung cancer screening subjects. The imaging data acquisition equipment model was Siemens SOMATOM Force, the scanning conditions were 120kV, the tube current was set to 25-30mAs, the layer thickness was 1.0mm, and the standard low-dose protocol was used, resulting in large image noise and insufficient contrast. After preliminary screening, a total of 482 cases of imaging data met the requirements for patchy lesion characteristics and were included in this study.

[0152] The project team first standardized the raw image data. Specifically, they performed intensity normalization on all images to unify the grayscale distribution to the [0,1] range. At the same time, they applied a global histogram threshold segmentation method to remove large areas of metal artifacts and motion artifacts. After processing, a standardized lung cancer patchy image dataset was formed. Based on the lung anatomical area in each image, an ROI (Region of Interest) mask was extracted, retaining only the lung parenchyma area for subsequent analysis.

[0153] In order to obtain reliable lesion area annotation information, the project team invited 5 radiologists with more than 10 years of clinical experience to outline each ROI image frame by frame and draw lesion masks. The annotation work was completed between August and September 2024. A total of 482 pairs of annotation data were generated, each of which contained detailed lesion contour information, including patchy irregular density shadows, ground glass nodules, and small solid nodules.

[0154] Next, the dilated convolution fusion generative adversarial network (DR-GAN) proposed in the present invention is applied for image enhancement. First, the local texture gradient intensity map is calculated based on the lesion mask and the original image, and the dilation rate map is dynamically generated to guide the dilated convolution feature extraction module. Compared with the traditional method of using a fixed dilation rate (such as a fixed value of 2, 4 or 8) to extract features, the present invention adjusts the dilation rate in real time according to the texture changes in the lesion area, applies a smaller dilation rate (dmin=2) in areas with fine texture, and applies a larger dilation rate (dmax=6) in areas with single texture, effectively maintaining the richness of details.

[0155] After dynamic feature extraction, the multi-scale dynamic feature map is input into the lightweight residual generator. The lightweight residual network uses depthwise separable convolution, which significantly reduces the number of parameters. When the entire model is trained on a Tesla V100 GPU, the memory usage of a single batch is reduced from 3.8GB with the traditional method to 2.2GB, and the training speed is increased by about 1.7 times. The training process uses the Adam optimizer, with an initial learning rate of 1e-4, a batch size of 16, and a total training cycle of 120 epochs.

[0156] In order to further improve the local authenticity of the generated images, the project team constructed a local perception discriminator. Unlike traditional discriminators, the discriminator of the present invention calculates the distance transformation map based on the lesion mask, automatically extracts local image blocks of different scales, scores the authenticity of local details, and integrates the scores based on the scale attention weight, significantly enhancing the attention to the authenticity of the lesion boundary.

[0157] After complete training, the project team used 50 newly collected low-dose CT images from the October 2024 lung cancer screening batch as an independent test set, performed image enhancement processing using the proposed model, and conducted comparative experiments with the traditional GAN method based on the U-Net structure (denoted as Baseline-GAN). The specific comparison is as follows:

[0158] Table 1 Comparative experiments between the present invention and the traditional GAN method based on U-Net structure

[0159] index Baseline-GAN The present invention (DR-GAN) Average PSNR (dB) 27.6 30.2 Average SSIM 0.842 0.893 Dice coefficient of lesion area 0.778 0.846 Detail texture retention rate (custom texture score) 74.5％ 85.3％ Inference speed (fps) 6.2 11.1

[0160] The data shows that the PSNR of the enhanced lung cancer patchy images improved by 2.6dB, indicating that the overall signal-to-noise ratio of the image was significantly improved; the SSIM increased by 5.1%, indicating that structural information was better preserved; the Dice coefficient of the lesion area increased by 8.7%, indicating that the reconstructed image is more conducive to lesion identification. In addition, by introducing dynamic void convolution and lightweight residual design, the inference speed was increased by nearly 79%, meeting the actual clinical requirements for rapid auxiliary diagnosis.

[0161] In the qualitative evaluation phase, the project team organized three chief physicians and two deputy chief physicians to conduct a blind evaluation experiment. Each physician was required to score the image clarity and lesion recognizability generated by Baseline-GAN and the present invention without informing the doctor. The final statistics showed that the average score of Baseline-GAN was 3.2 / 5, while the average score of the present invention was 4.5 / 5. It was unanimously agreed that the present invention was superior to the traditional method in enhancing tiny lesions and blurred boundaries.

[0162] In addition, in the test set, for patchy lesions smaller than 5 mm in size, the method of the present invention successfully improved the sharpness of the lesion edges by an average of 16.4%, significantly improving the problem of severe detail loss in traditional methods when processing small lesions.

[0163] In summary, this example, through specific application in real medical scenarios and detailed data comparison, fully verifies that the method of the present invention can effectively solve the problems of missing lesion details, blurred boundaries, and poor enhancement quality in low-dose patchy lung cancer images, and has good feasibility and significant practical application value.

[0164] The present invention adopts a dilated convolution feature extraction module with dynamic dilation rate regulation. It dynamically generates a dilation rate map based on the local texture gradient intensity map of the patchy lesion area of lung cancer, and adaptively applies the dilated convolution operation based on the dilation rate of each pixel position, thereby realizing multi-scale dynamic feature extraction. It can adaptively adjust the receptive field size according to the texture complexity of different positions in the lesion area, apply a small dilation rate to finely extract features in areas rich in details, and apply a large dilation rate to enhance contextual information fusion in areas with simple structures, effectively avoiding the problem of detail loss or insufficient contextual information.

[0165] The present invention constructs a lightweight residual generator network, adopts depthwise separable convolution instead of traditional standard convolution in the feature reconstruction process, and introduces stacked lightweight residual units, which effectively reduces the number of model parameters and computational complexity while maintaining efficient information flow and feature fusion capabilities. When processing patchy multi-scale dynamic feature maps of lung cancer, the lightweight residual generator can more fully retain local detail information and overall structural continuity, avoiding excessive smoothing or artifact diffusion in the reconstructed image.

[0166] The present invention introduces a local perception discriminator and combines it with a scale-attention weighted mechanism. By calculating the distance transformation map of the patchy lesion area of lung cancer, multi-scale local image blocks are extracted, and weights are dynamically assigned according to the importance of the local areas at each scale. The generated image is subjected to fine local authenticity judgment, which can effectively evaluate the authenticity of the lesion edge and internal structure at a fine-grained level, further improving the authenticity and credibility of the generated image in the key lesion area.

[0167] The above description is only a preferred specific embodiment of the present invention, but the scope of protection of the present invention is not limited thereto. Any technician familiar with the technical field, within the technical scope disclosed by the present invention, who makes equivalent replacements or changes based on the technical solution and inventive concept of the present invention, should be covered by the scope of protection of the present invention.

Claims

1. A lung cancer image enhancement method based on generative adversarial network, characterized in that: The steps include: S1. Collecting lung cancer patchy image data and establishing a lung cancer patchy image dataset, preprocessing the lung cancer patchy image dataset to obtain a standardized lung cancer patchy image dataset; S2. Generate labeled information for the patchy lung cancer lesion region based on the manual delineation results of the radiologist, and pair the labeled information for the patchy lung cancer lesion region with a standardized patchy lung cancer image dataset to form labeled patchy lung cancer image training data; S3. Construct a dilated convolution feature extraction module to perform multi-scale feature extraction on the labeled lung cancer patchy image training data to generate a multi-scale dynamic feature map of lung cancer patchiness; S4. Construct a lightweight residual generator network, input the multi-scale dynamic feature map of lung cancer patchiness into the lightweight residual generator network, and output preliminary reconstructed lung cancer patchiness image data; S5. Construct a local perception discriminator; S6. Using a standardized lung cancer patchy image dataset as input, the patchy lung cancer lesion region annotation information as the regional weights, the dilated convolutional feature extraction module and the lightweight residual generator network as the generator, and the local perception discriminator as the discriminator, we obtained a trained dilated convolutional fusion generative adversarial network model. S7. Input the newly acquired low-quality lung cancer patchy image data into the trained atrous convolutional fusion generative adversarial network model, output the reconstructed lung cancer patchy image data, and generate the lung cancer patchy image quality assessment results.

2. The lung cancer image enhancement method based on generative adversarial network according to claim 1, characterized in that: The S1 specifically includes the following steps: S11. Collecting lung cancer patchy image data acquired by computed tomography or magnetic resonance imaging equipment under clinical low-dose scanning conditions, pairing each acquired original lung cancer patchy image with its corresponding basic image metadata to establish a lung cancer patchy image dataset; S12. performing intensity normalization processing on each original lung cancer patchy image in the lung cancer patchy image dataset to obtain a normalized lung cancer patchy image dataset; S13. Performing artifact removal processing on each patchy normalized lung cancer image in the patchy normalized lung cancer image dataset, using a global histogram threshold segmentation method, setting an artifact removal intensity threshold, and marking a pixel as a valid pixel if the pixel intensity of a pixel in the patchy normalized lung cancer image is greater than or equal to the artifact removal intensity threshold; otherwise, marking the pixel as an invalid pixel, removing artifact regions from the patchy normalized lung cancer image to obtain a patchy artifact-removed lung cancer image dataset; S14. Based on the lung cancer patchy artifact removal image dataset, the corresponding lung area is extracted as the region of interest according to the lung anatomical structure detection results in the image. The region of interest consists of all pixel coordinates (x, y) belonging to the lung area mask. The extracted area is defined as the lung cancer patchy region of interest R i , forming a standardized lung cancer patchy image dataset D roi .

3. The lung cancer image enhancement method based on generative adversarial network according to claim 2, characterized in that: The S2 specifically includes the following steps: S21. A radiologist is responsible for the image of each lung cancer patchy region of interest R in the standardized lung cancer patchy image dataset. i Perform manual delineation operation, for each lung cancer patchy region of interest image R i Determine the corresponding lung cancer patchy lesion area contour and generate the lung cancer patchy lesion area mask M i : Among them, M i (x, y) represents the mask of the patchy lung cancer lesion area at the coordinate position (x, y) in the i-th patchy lung cancer region of interest image; S22. Each lung cancer patchy region of interest image R in the standardized lung cancer patchy image dataset i The corresponding lung cancer patchy lesion area mask M i Pairing is performed to construct a pair of lung cancer patchy images and annotation information (R i ,M i ); S23. Based on the patchy lesion area mask M of lung cancer i Calculate the area ratio A of the lesion area in each lung cancer patchy region of interest image i : Among them, ∑ (x,y) M i (x,y) represents the number of all pixels belonging to the patchy lesion area of lung cancer in the i-th patchy region of interest image of lung cancer, ∑ (x,y) 1 represents the total number of pixels in the entire lung cancer patchy region of interest image; S24. According to the area ratio A i The statistical distribution of the lung cancer patchy images and annotation information pairs are obtained by filtering out abnormal samples whose area ratio is lower than the set threshold, retaining the lung cancer patchy images and annotation information pairs that meet the requirements, and forming the labeled lung cancer patchy image training dataset D with the retained lung cancer patchy images and annotation information pairs. train .

4. The lung cancer image enhancement method based on generative adversarial network according to claim 3, characterized in that: The S3 specifically includes the following steps: S31. Read each pair of lung cancer patchy region of interest image and lung cancer patchy lesion area mask (R i ,M i ), for the patchy region of interest image R of lung cancer i Patchy lesion area of lung cancer mask M i Perform pixel-by-pixel product operations to obtain enhanced images of patchy lesions of lung cancer S32. Based on enhanced imaging of patchy lesions in lung cancer Calculate the local texture gradient intensity map G i , local texture gradient intensity map G i The gradient intensity value G of each pixel position (x, y) i (x,y) is defined as the enhanced image of the patchy lung cancer lesion along the transverse direction at that location The square of the rate of change and the enhanced image of lung cancer patchy lesions along the longitudinal direction The result of taking the square root of the sum of the squares of the rates of change; S33. According to the local texture gradient intensity map G i Dynamically generate void ratio map D i , void ratio graph D i The hole rate value D of each pixel position (x, y) i (x,y) is based on the preset minimum void ratio d min and the maximum void ratio d max The difference multiplied by one minus the local texture gradient intensity value G i (x, y) and the maximum value of the local texture gradient intensity map G max After adding the minimum void ratio d min to obtain; S34. Based on enhanced imaging of patchy lesions in lung cancer and dynamic void ratio graph D i (x, y), adaptively apply the corresponding dilated convolution operation to each position (x, y) to extract the patchy multi-scale dynamic feature map F of lung cancer i (x,y).

5. The lung cancer image enhancement method based on generative adversarial network according to claim 4, characterized in that: The S4 specifically includes the following steps: S41. Inputting the patchy multi-scale dynamic feature map of lung cancer extracted by the dynamic void ratio mechanism into the feature input end of the lightweight residual generator network, the lightweight residual generator network comprising a plurality of stacked lightweight residual units; S42. Multi-scale dynamic feature map of lung cancer patchiness in the kth lightweight residual unit Perform convolution transformation operation, use convolution kernel weight and bias term in the convolution process, process the convolution transformation operation result through lightweight nonlinear activation function, obtain the convolution feature output of the kth lightweight residual unit, and compare the convolution feature output with the input lung cancer patchy multi-scale dynamic feature map Add element-wise to generate the output feature map of the kth lightweight residual unit S43. After all lightweight residual units are processed, the final feature map is obtained The final feature map Input to the feature reconstruction module, which generates preliminary reconstruction of lung cancer patchy image data through a layer of standard convolution and Tanh activation function 6. The lung cancer image enhancement method based on generative adversarial network according to claim 5, characterized in that: In each lightweight residual unit, the convolution transformation operation uses depth-wise separable convolution instead of standard convolution. Depth-wise separable convolution includes two steps: depth-wise convolution and point-wise convolution.

7. The lung cancer image enhancement method based on generative adversarial network according to claim 5, characterized in that: The S5 specifically includes the following steps: S51. Preliminary reconstruction of lung cancer patchy imaging data The corresponding lung cancer patchy lesion area mask M i The local perception discriminator is used to identify the patchy lesion area of lung cancer. i Calculating the distance transformation map U of patchy lung cancer lesions i : in, Represents the boundary between pixel (x,y) and the patchy lesion area of lung cancer The Euclidean distance of S52. Determine the distance transformation map U of the patchy lung cancer lesions i The pixel coordinate with the largest normalized Euclidean distance Centered on pixel coordinates, multi-scale image blocks of lung cancer patchy lesions are cropped according to the scale set. S53. In the global discriminant branch, the preliminarily reconstructed lung cancer patchy image data is processed by convolution-global average pooling, and then the global authenticity score is obtained by the fully connected layer and Sigmoid activation function. S54. In the multi-scale local discriminant branch, each multi-scale image block of the lung cancer patch is Input convolutional network to obtain local feature map Output scale-local authenticity score through fully connected layer and Sigmoid activation function S55. Based on the distance transformation map U of the patchy lesions of lung cancer i Calculate the multi-scale image blocks of each lung cancer patch The scale attention weight The scale attention weight is obtained by averaging the normalized Euclidean distance of all pixels in the patchy lesion area of lung cancer and then normalizing it. The weighted sum of each scale-local authenticity score is calculated according to the scale attention weight to obtain the weighted local authenticity score. S56. Calculate the average gradient intensity of the edge gradient map of the patchy lung cancer lesions Get the edge integrity adjustment factor γ i , using the edge integrity adjustment factor to score the global authenticity and weighted local authenticity score Perform linear fusion to obtain a joint authenticity score 8. The lung cancer image enhancement method based on generative adversarial network according to claim 7, characterized in that: The S6 specifically includes the following steps: S61. Standardize the lung cancer patchy image dataset D roi Each lung cancer patchy region of interest image R i As the generator input, the lung cancer patchy lesion area mask M i As a reference for regional weights, it is input into the generator composed of the dilated convolution feature extraction module and the lightweight residual generator network to generate preliminary reconstructed lung cancer patchy image data. S62. Preliminary reconstruction of lung cancer patchy imaging data and the corresponding lung cancer patchy region of interest image R i Input the local perception discriminator to obtain the joint authenticity score S63. Calculate the structure-aware loss L struct , the structure-aware loss aims to maintain the continuity of the patchy image structure of lung cancer, which is defined as the patchy region of interest image R i and preliminary reconstruction of lung cancer patchy imaging data The weighted squared error of the local gradient difference: in, and Represents the gradient values of the original image and the reconstructed image at position (x, y), U i (x, y) is the distance transformation map weight of lung cancer patchy lesions; S64. Calculate edge preservation loss L edge , the edge preservation loss aims to enhance the boundary clarity of the lung cancer patchy lesion area, which is defined as the lung cancer patchy lesion area mask M i Edge detection error under the action of: Among them, E i (x,y) and Represent the edge gradient values of the original image and the preliminary reconstructed image at position (x, y); S65. Calculate the adversarial loss L adv , the adversarial loss aims to improve the realism of lung cancer patchy images and is defined as the joint realism score Negative log loss of S66. Integrate structure-aware loss, edge-preserving loss, and adversarial loss to construct the total loss function L total : L total =λ struct L struct +λ edge L edge +λ adv L adv , Among them, λ struct ,λ edge ,λ adv are the weighted coefficients of each loss item respectively; S67. Take the total loss function L total To achieve the optimization goal, the parameters of the dilated convolution feature extraction module, the lightweight residual generator network and the local perception discriminator are jointly optimized until the training converges, and the trained dilated convolution fusion generative adversarial network model is obtained.

Images