An infrared image diffusion super-resolution reconstruction method combining frequency domain enhancement and infrared perception

Abstract

The application discloses an infrared image diffusion super-resolution reconstruction method combining frequency domain enhancement and infrared perception, and belongs to the technical field of infrared image processing and deep learning. The method performs frequency domain enhancement on the infrared image through Fourier transform, and improves the recovery ability of high-frequency details and texture information. In combination with the characteristics of the infrared image, a physically guided diffusion scheduling strategy is designed to ensure the rationality of the generated image in thermal physical consistency. In order to further optimize the image quality, a spectrum consistency and phase consistency loss function is used to strengthen the feature consistency of the infrared image from the frequency domain, and the restoration effect of edges and textures is effectively enhanced. In addition, a self-supervised learning technology is introduced to automatically extract feature information from unlabeled infrared image data, thereby improving the super-resolution reconstruction performance. Through multi-level feature fusion and diffusion model optimization, the super-resolution reconstruction of the infrared image is realized, and the deficiencies of the traditional super-resolution technology, such as the loss of infrared image details and the physical consistency, are solved.

Description

Technical Field

[0001] This invention relates to the fields of infrared image processing and deep learning technology, specifically to an infrared image diffusion super-resolution reconstruction method using frequency domain enhancement and infrared sensing. Background Technology

[0002] With the rapid development of infrared image processing technology and artificial intelligence, infrared image super-resolution reconstruction is playing an increasingly important role in fields such as military reconnaissance, security monitoring, and environmental monitoring. Infrared imaging technology, with its unique advantage of acquiring target thermal radiation information in complex environments such as low light, smoke, and haze, can effectively penetrate atmospheric obstructions, providing all-weather surveillance capabilities unrestricted by visible light. However, infrared images often face problems such as low resolution, loss of detail, and noise interference. Especially in image super-resolution reconstruction, effectively restoring details and textures and improving image quality remains a technical challenge.

[0003] Existing infrared image super-resolution reconstruction methods typically rely on traditional spatial domain methods and frequency domain enhancement methods. Spatial domain methods, such as interpolation and super-resolution reconstruction networks, can improve image quality to some extent, but they often fail to effectively recover high-frequency details in the image, especially in complex dynamic environments, where the recovery of image details and textures is limited. Frequency domain enhancement methods, by processing the image through Fourier transform, can enhance the high-frequency information of the image to some extent, but these methods still have limitations in combining the physical characteristics and feature perception of infrared images, and cannot fully recover key information such as edges and textures in infrared images.

[0004] Furthermore, traditional methods often neglect the physical consistency requirements of infrared images, especially in terms of thermal and radiation properties, resulting in reconstructed images lacking physical plausibility. With the gradual development of self-supervised learning and physically guided models, combining these techniques to automatically learn target features of images from unlabeled data while ensuring the physical consistency of the generated images has become a new direction for solving the problem of infrared image super-resolution reconstruction.

[0005] In recent years, advancements in deep learning-based image super-resolution methods, particularly convolutional neural networks (CNNs) and generative adversarial networks (GANs), have brought new insights into super-resolution reconstruction. These methods can automatically learn multi-level, multi-scale features of images and improve detail recovery capabilities to some extent. However, existing deep learning-based methods in infrared image super-resolution reconstruction still face challenges in effectively fusing frequency domain information with infrared feature perception and incorporating physical models to enhance image detail and consistency. Furthermore, traditional methods often fail to fully integrate multimodal information and self-supervised learning techniques, resulting in limited reconstruction performance.

[0006] To address the aforementioned challenges, this invention proposes an infrared image diffusion super-resolution reconstruction method combining frequency domain enhancement and infrared sensing. By fusing techniques such as frequency domain enhancement, a physically guided diffusion model, and self-supervised learning, this method significantly improves the super-resolution reconstruction effect of infrared images, resolving issues of insufficient detail recovery and physical consistency. This method exhibits stronger robustness and accuracy in complex backgrounds and can be widely applied in practical scenarios such as military reconnaissance, security monitoring, and environmental monitoring. Summary of the Invention

[0007] The purpose of this invention is to provide a super-resolution reconstruction method for infrared images that combines frequency domain enhancement and infrared sensing. This method solves the problems of detail loss and insufficient physical consistency in traditional infrared image super-resolution reconstruction by fusing frequency domain enhancement, a physically guided diffusion model, and self-supervised learning techniques, thereby achieving higher-quality infrared image reconstruction. This method can significantly improve the resolution, sharpness, and detail recovery capability of infrared images, especially in complex backgrounds and low-contrast environments, exhibiting stronger image restoration capabilities and robustness. This method is widely used in infrared image processing fields such as military reconnaissance, security monitoring, and environmental monitoring.

[0008] To achieve the above objectives, the present invention provides a method for infrared image diffusion super-resolution reconstruction using frequency domain enhancement and infrared sensing, comprising the following steps:

[0009] S1. Acquire the input infrared image, perform Fourier transform on the infrared image, and enhance the high-frequency components in the frequency domain.

[0010] S2. Construct physical constraints by combining the thermophysical properties of infrared images, and establish a physically guided diffusion scheduling strategy.

[0011] S3. Construct the spectral consistency loss function and the phase consistency loss function between the generated image and the reference image;

[0012] S4. Input the unlabeled infrared image into the self-supervised learning network and extract image features through reconstruction loss and contrast loss;

[0013] S5. Through multi-level feature fusion and diffusion model optimization, high-quality super-resolution reconstruction of infrared images is achieved.

[0014] Furthermore, in step S1, the input infrared image is enhanced in the frequency domain through Fourier transform, specifically including the following steps:

[0015] S1.1 Acquire infrared image data to obtain the input infrared image to be processed. The input infrared image is represented as follows: ;

[0016] S1.2, Infrared image Perform a Fourier transform to convert the image to the frequency domain, thus obtaining the spectral representation of the image. The formula for calculating the frequency domain enhancement is as follows:

[0017] ;

[0018] in, Represents the original image. For frequency domain representation, These are frequency domain coordinates, Spatial domain coordinates;

[0019] S1.3 Enhance the frequency domain image by utilizing the high-frequency components in the frequency domain. To improve image detail recovery, including enhancing texture and edge information, frequency domain enhancement strategies include weighted or filtered processing of high-frequency regions. The enhancement formula is as follows:

[0020] ;

[0021] in, These are frequency domain enhancement coefficients, used to enhance the response in the high-frequency region;

[0022] S1.4, Enhanced frequency domain image The enhanced infrared image is obtained by converting it back to the spatial domain using inverse Fourier transform. This image has stronger high-frequency details and texture information. The specific inverse Fourier transform formula is as follows:

[0023] .

[0024] Furthermore, the physically guided diffusion scheduling strategy designed in S2 specifically includes the following:

[0025] S2.1. Based on the thermophysical characteristics of infrared images, construct a physically guided diffusion model, where thermophysical consistency is represented as... This function describes the physical consistency of the image during the heat conduction process, ensuring that the temperature distribution in the generated image conforms to physical laws;

[0026] S2.2 Introduce thermophysical constraints during the diffusion process and design a diffusion scheduling strategy to achieve the desired effect. The scheduling strategy involves gradually adjusting the flow of image information during the diffusion process to maintain the thermophysical consistency of the generated image. The specific calculation formula for the scheduling strategy is as follows:

[0027] ;

[0028] in, This is a scheduling coefficient used to control the degree of influence of physical consistency on the diffusion process;

[0029] S2.3 During the diffusion process, a thermophysical consistency constraint is applied to each layer of the diffusion process to ensure that the thermophysical consistency of the generated image at each layer meets expectations, thus obtaining the updated image features. :

[0030] ;

[0031] in, For the current image features during the diffusion process, To ensure the reasonableness of the generated image in terms of thermophysical consistency in order to account for changes in image features during the diffusion process;

[0032] S2.4. Through its diffusion scheduling strategy, a super-resolution infrared image that conforms to thermophysical consistency is finally generated, thereby improving image quality and ensuring its thermophysical rationality.

[0033] Furthermore, in S3, spectral consistency and phase consistency loss functions are used to constrain the feature consistency of the image from the frequency domain perspective to enhance edges and textures, specifically including the following:

[0034] S3.1 First, feature extraction is performed on the infrared image in the frequency domain to obtain the spectral representation. and phase information ,in This refers to amplitude information in the frequency domain. Phase information in the frequency domain;

[0035] S3.2, Define the spectrum consistency loss function This loss function is used to constrain the consistency between the spectrum of the generated image and the original image, ensuring that the spectral features of the generated image do not deviate significantly. The specific calculation formula is as follows:

[0036] ;

[0037] in, and The frequency domain amplitude information of the generated image and the reference image are respectively used to ensure that the spectrum of the generated image is consistent with the spectrum of the original image;

[0038] S3.3, Define the phase consistency loss function This loss function is used to constrain the consistency of the generated image and the reference image in terms of frequency domain phase information, thereby enhancing the ability to restore image edges and texture details. The specific calculation formula is as follows:

[0039] ;

[0040] in, and The frequency domain phase information of the generated image and the reference image are respectively used to ensure that the generated image is consistent with the original image in terms of phase.

[0041] S3.4. Weighted summation of the spectral consistency loss function and the phase consistency loss function yields the final total loss function. This loss function is used to optimize the consistency of frequency domain features in the generated image, and the specific formula is as follows:

[0042] ;

[0043] in, and These are weighting coefficients used to balance the contributions of spectral consistency loss and phase consistency loss to the optimization process;

[0044] S3.5, By optimizing the total loss function The spectral and phase consistency of the generated image are enhanced, resulting in a significant improvement in the recovery of edge and texture details, ensuring the accuracy of image detail and texture restoration.

[0045] Furthermore, in step S4, self-supervised learning techniques are used to automatically extract target feature information from unlabeled infrared image data, specifically including the following:

[0046] S4.1 Extracting the raw image from unlabeled infrared image data As training samples, among them This represents the original input image, which serves as input data for self-supervised learning, used to learn the target feature information of the image.

[0047] S4.2, Define the loss function for self-supervised learning. This loss function is used to generate images. With the original image The differences between them are used to optimize the network's learning. The specific calculation formula is as follows:

[0048] ;

[0049] in, For the image generated through super-resolution reconstruction, Given the original image, the learning ability of the target features is improved by minimizing this loss function;

[0050] S4.3 Introducing self-supervised networks This network is used to extract data from unlabeled images. Extracting feature information and optimizing the feature extraction process; self-supervised network. It will generate feature maps of the image. And by adjusting the image feature map compared with the original image feature map The differences between them guide the extraction of target features, ensuring that the learned features can effectively reflect image details and textures;

[0051] S4.4 To further enhance the effectiveness of self-supervised learning, a contrastive loss function is introduced. This loss function is used to optimize the feature contrast of the generated image, making the target features in the image more prominent. The calculation formula is as follows:

[0052] ;

[0053] in, Feature maps generated by a self-supervised network. Given the feature map of the original image, by minimizing this loss function, we ensure that the generated image is consistent with the original image in the feature space;

[0054] S4.5, through joint optimization and Self-supervised learning networks can effectively learn from unlabeled infrared image data. By learning target feature information, the performance of super-resolution reconstruction is further improved; ultimately, the generated image is significantly improved in terms of detail and texture restoration.

[0055] Furthermore, in S5, high-quality super-resolution reconstruction of infrared images is achieved through multi-level feature fusion and diffusion model optimization, specifically including the following:

[0056] S5.1 Combining the frequency domain enhancement, physically guided diffusion, and self-supervised learning features obtained in the preceding steps, a multi-level feature fusion strategy is applied to integrate the feature maps from different stages. and By fusing the data, multi-scale image feature representations can be obtained. The calculation formula is as follows:

[0057] ;

[0058] in, For the first Image features of the layer, These are the weighting coefficients of the features at this layer. The number of layers for feature fusion;

[0059] S5.2 After feature fusion is completed, the optimized diffusion model is used to further enhance the image detail recovery capability. The optimized diffusion process is based on the fused feature map. By adjusting the detail recovery capability of the image through a diffusion model, high-frequency information in the image is fully recovered.

[0060] S5.3 Define the optimized diffusion model loss function This loss function is used to constrain the differences between the features of the generated image and the fused image, thereby further improving image quality. The formula for calculating the loss function is as follows:

[0061] ;

[0062] in, To generate an image, ensure that the generated super-resolution image is consistent with the fused feature map;

[0063] S5.4 Combining multi-level feature fusion and diffusion model optimization, a high-quality super-resolution image is finally generated. The image shows significant improvements in detail restoration, edge sharpness, and resolution, and the final image features generated during the process are... By minimizing To achieve the best results;

[0064] S5.5 Through iterative optimization and feature fusion, super-resolution reconstruction of infrared images was finally achieved, significantly improving image resolution, sharpness, and texture details, ensuring excellent performance in visual quality and physical consistency of the generated images. The super-resolution images during the iterative process... By minimizing the loss function To achieve the best results, the specific iterative optimization formula is as follows:

[0065] ;

[0066] in, For the first The super-resolution image generated after the next iteration. This is the result of the previous iteration. For learning rate, loss function The gradient of the image represents the optimization direction of the current iteration step; through iterative optimization, the final generated image is obtained. It achieves optimal results in the restoration of details and textures, and significantly improves the visual quality and physical consistency of images.

[0067] Beneficial Effects: This invention effectively improves the quality of super-resolution reconstruction of infrared images by combining frequency domain enhancement, a physically guided diffusion model, and self-supervised learning techniques. Frequency domain enhancement of infrared images via Fourier transform significantly improves the ability to recover high-frequency details and texture information. A physically guided diffusion scheduling strategy is designed based on the physical characteristics of infrared images to ensure the reasonableness of the generated image in terms of thermophysical consistency, enhancing the realism and physical plausibility of the image. Spectral consistency and phase consistency loss functions are used to further constrain image feature consistency from a frequency domain perspective, strengthening the restoration effect of image edges and textures. Self-supervised learning techniques are introduced to automatically extract target features from unlabeled infrared image data, further optimizing super-resolution reconstruction performance. Through multi-level feature fusion and diffusion model optimization, the resolution, sharpness, and detail recovery capabilities of the image are significantly improved, ensuring excellent performance in visual quality and physical consistency of the generated image. The method of this invention effectively solves the shortcomings of traditional super-resolution techniques in detail recovery and physical consistency of infrared images, significantly improving image quality, especially under low contrast and complex background conditions, where it has outstanding advantages. This method is widely used in infrared image processing fields such as military reconnaissance, security monitoring, and environmental monitoring, providing an effective solution for high-quality infrared image reconstruction. Attached Figure Description

[0068] Figure 1 This is a schematic diagram of the overall process of the present invention;

[0069] Figure 2 This is a schematic diagram of frequency domain enhancement processing;

[0070] Figure 3 This is a schematic diagram of a physical-guided diffusion scheduling strategy;

[0071] Figure 4 This is a schematic diagram of the loss functions for spectral consistency and phase consistency;

[0072] Figure 5 This is a schematic diagram of the self-supervised learning process;

[0073] Figure 6 This is a schematic diagram of the optimization of a multi-level feature fusion and diffusion model; Detailed Implementation

[0074] The invention will now be further described with reference to the accompanying drawings.

[0075] Example

[0076] Furthermore, such as Figure 1 As shown, an infrared image diffusion super-resolution reconstruction method combining frequency domain enhancement and infrared sensing includes the following steps:

[0077] S1. Acquire the input infrared image, perform Fourier transform on the infrared image, and enhance the high-frequency components in the frequency domain.

[0078] S2. Construct physical constraints by combining the thermophysical properties of infrared images, and establish a physically guided diffusion scheduling strategy.

[0079] S3. Construct the spectral consistency loss function and the phase consistency loss function between the generated image and the reference image;

[0080] S4. Input the unlabeled infrared image into the self-supervised learning network and extract image features through reconstruction loss and contrast loss;

[0081] S5. Through multi-level feature fusion and diffusion model optimization, high-quality super-resolution reconstruction of infrared images is achieved.

[0082] Furthermore, such as Figure 2 As shown, in step S1, the input infrared image is enhanced in the frequency domain through Fourier transform. The specific steps are as follows:

[0083] S1.1 Acquire infrared image data to obtain the input infrared image to be processed. The input infrared image is represented as follows: ;

[0084] S1.2, transfer the infrared image Transforming to the frequency domain and using Fourier transform to process the image yields its spectral representation. The formula for calculating frequency domain enhancement is as follows:

[0085] ;

[0086] in, For the original image, For frequency domain representation, For frequency domain coordinates, Spatial domain coordinates;

[0087] S1.3 Enhance the frequency domain image by utilizing high-frequency components within the frequency domain. To improve image detail recovery, including texture and edge information, frequency domain enhancement strategies include weighted or filtered operations on high-frequency regions. The enhancement formula is as follows:

[0088] ;

[0089] in, This is a frequency domain enhancement factor used to improve the response in the high-frequency region;

[0090] S1.4, Enhanced frequency domain image The enhanced infrared image is obtained by converting it back to the spatial domain using inverse Fourier transform. This image has stronger high-frequency details and texture information. The specific inverse Fourier transform formula is as follows:

[0091] ;

[0092] Furthermore, such as Figure 3 As shown, the physical guidance diffusion scheduling strategy designed in S2 specifically includes the following:

[0093] S2.1. Based on the thermophysical characteristics of infrared images, construct a physically guided diffusion model, where thermophysical consistency is represented as... This function describes the physical consistency of the image during the heat conduction process, ensuring that the temperature distribution in the generated image conforms to physical laws;

[0094] S2.2 Introduce thermophysical constraints during the diffusion process and design a diffusion scheduling strategy to achieve the desired effect. The scheduling strategy involves gradually adjusting the flow of image information during the diffusion process to maintain the thermophysical consistency of the generated image. The specific calculation formula for the scheduling strategy is as follows:

[0095] ;

[0096] in, This is a scheduling coefficient used to control the degree of influence of physical consistency on the diffusion process;

[0097] S2.3 During the diffusion process, a thermophysical consistency constraint is applied to each layer of the diffusion process to ensure that the thermophysical consistency of the generated image at each layer meets expectations, thus obtaining the updated image features. :

[0098] ;

[0099] in, For the current image features during the diffusion process, To ensure the reasonableness of the generated image in terms of thermophysical consistency in order to account for changes in image features during the diffusion process;

[0100] S2.4. Through its diffusion scheduling strategy, a super-resolution infrared image that conforms to thermophysical consistency is finally generated, thereby improving image quality and ensuring its thermophysical rationality.

[0101] Furthermore, such as Figure 4 As shown, step S3 employs spectral consistency and phase consistency loss functions to constrain the feature consistency of the image from a frequency domain perspective to enhance edges and textures. Specifically, it includes the following:

[0102] S3.1. Extract features from the infrared image in the frequency domain to obtain the spectral representation. and phase information ,in This refers to amplitude information in the frequency domain. This refers to phase information in the frequency domain.

[0103] S3.2, Define the spectrum consistency loss function This loss function is used to constrain the consistency between the spectrum of the generated image and the original image, ensuring that the spectral features of the generated image tend to be consistent with the spectrum of the original image. The calculation formula is as follows:

[0104] ;

[0105] in, and These are the frequency domain amplitude information of the generated image and the reference image, respectively;

[0106] S3.3, Define the phase consistency loss function This loss function is used to constrain the consistency of the generated image and the reference image in terms of frequency domain phase information, thereby enhancing the image's ability to restore edge and texture details. The calculation formula is as follows:

[0107] ;

[0108] in, and These are the frequency domain phase information of the generated image and the reference image, respectively;

[0109] S3.4. Weighted summation of the spectral consistency loss function and the phase consistency loss function yields the final total loss function. This loss function is used to optimize the consistency of frequency domain features in the generated image, and its calculation formula is as follows:

[0110] ;

[0111] in, and These are weighting coefficients used to balance the contributions of spectral consistency loss and phase consistency loss to the optimization process;

[0112] S3.5, By optimizing the total loss function The spectral and phase consistency of the generated image are enhanced, resulting in a significant improvement in the recovery of edge and texture details, ensuring the accuracy of image detail and texture restoration.

[0113] Furthermore, such as Figure 5 As shown, step S4 uses self-supervised learning technology to automatically extract target feature information from unlabeled infrared image data, specifically including the following:

[0114] S4.1 Extracting the raw image from unlabeled infrared image data As training samples, among them The original input image serves as input data for self-supervised learning, used to learn the target feature information of the image;

[0115] S4.2, Define the loss function for self-supervised learning. This loss function is used to generate images. With the original image The differences between them are used to optimize the network's learning. The specific calculation formula is as follows:

[0116] ;

[0117] in, For the image generated through super-resolution reconstruction, Given the original image, the learning ability of the target features is improved by minimizing this loss function;

[0118] S4.3 Introducing self-supervised networks This network is used to extract data from unlabeled images. Extracting feature information and optimizing the feature extraction process; self-supervised network. It will generate feature maps of the image. And by adjusting the image feature map compared with the original image feature map The differences between them are used to guide the extraction of target features, ensuring that the learned features can effectively reflect image details and textures:

[0119] S4.4 To further enhance the effectiveness of self-supervised learning, a contrastive loss function is introduced. This loss function is used to optimize the feature contrast of the generated image, making the target features in the image more prominent. The calculation formula is as follows:

[0120] ;

[0121] in, Feature maps generated by a self-supervised network. Given the feature map of the original image, by minimizing this loss function, we ensure that the generated image is consistent with the original image in the feature space;

[0122] S4.5, through joint optimization and Self-supervised learning networks can effectively learn from unlabeled infrared image data. By learning target feature information, the performance of super-resolution reconstruction is further improved; ultimately, the generated image is significantly improved in terms of detail and texture restoration.

[0123] Furthermore, such as Figure 6 As shown, in step S5, high-quality super-resolution reconstruction of infrared images is achieved through multi-level feature fusion and diffusion model optimization, specifically including the following:

[0124] S5.1 Combining the features obtained from the previous steps—frequency domain enhancement, physical guided diffusion, and self-supervised learning—a multi-level feature fusion strategy is used to integrate the feature maps from different stages. and By fusing the data, multi-scale image feature representations can be obtained. The calculation formula is as follows:

[0125] ;

[0126] in, For the first Image features of the layer, These are the weighting coefficients for the features of this layer. The number of layers for feature fusion;

[0127] S5.2 After feature fusion, the ability to restore image details is further improved by optimizing the diffusion model. The optimized diffusion process is based on the fused feature map. By adjusting the image's detail recovery capability through a diffusion model, high-frequency information is ensured to be fully recovered.

[0128] S5.3 Define the optimized diffusion model loss function This loss function is used to constrain the differences between the features of the generated image and the fused image, thereby further improving image quality. The formula for calculating the loss function is as follows:

[0129] ;

[0130] in, To generate an image, ensure that the generated super-resolution image is consistent with the fused feature map;

[0131] S5.4 Combining multi-level feature fusion and diffusion model optimization, a high-quality super-resolution image is finally generated. The image shows significant improvements in detail restoration, edge sharpness, and resolution, and the final image features generated during the process are... By minimizing To achieve the best results;

[0132] S5.5 Through iterative optimization and feature fusion, super-resolution reconstruction of infrared images was finally achieved, significantly improving image resolution, sharpness, and texture details, ensuring excellent performance in visual quality and physical consistency of the generated images. The super-resolution images during the iterative process... By minimizing the loss function To achieve the best results, the specific iterative optimization formula is as follows:

[0133] ;

[0134] in, For the first The super-resolution image generated after the next iteration. This is the result of the previous iteration. For learning rate, loss function The gradient of the image represents the optimization direction of the current iteration step. Through iterative optimization, the final generated image is obtained. It achieves optimal results in the restoration of details and textures, and significantly improves the visual quality and physical consistency of images.

[0135] The above embodiments are only used to illustrate the technical solutions of the present invention, and are not intended to limit it. Although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some of the technical features; and these modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the embodiments of the present invention. The scope of protection of the present invention should be determined by the scope of protection of the appended claims.

Claims

1. A method for super-resolution reconstruction of infrared images using frequency domain enhancement and infrared sensing, characterized in that, Includes the following steps: S1. Acquire the input infrared image, perform Fourier transform on the infrared image, and enhance the high-frequency components in the frequency domain. S2. Construct physical constraints by combining the thermophysical properties of infrared images, and establish a physically guided diffusion scheduling strategy. S3. Construct the spectral consistency loss function and the phase consistency loss function between the generated image and the reference image; S4. Input the unlabeled infrared image into the self-supervised learning network and extract image features through reconstruction loss and contrast loss; S5. Through multi-level feature fusion and diffusion model optimization, high-quality super-resolution reconstruction of infrared images is achieved.

2. The infrared image diffusion super-resolution reconstruction method based on frequency domain enhancement and infrared sensing according to claim 1, characterized in that, S1 includes the following steps: S1.1 Acquire infrared image data to obtain the input infrared image to be processed. The input infrared image is represented as follows: ; S1.2, transfer the infrared image Transforming to the frequency domain and using Fourier transform to process the image yields its spectral representation. The specific formula for frequency domain enhancement is as follows; ； in, Represents the original image. Frequency domain representation, These are frequency domain coordinates, Spatial domain coordinates; S1.3 Enhancement processing of the frequency domain image, utilizing high-frequency components in the frequency domain. To enhance image detail recovery, including the restoration of texture and edge information, frequency domain enhancement strategies include weighted or filtered operations on high-frequency regions. The enhancement formula is as follows: ； in, This is a frequency domain enhancement factor used to improve the response in the high-frequency region; S1.4, Enhanced frequency domain image The enhanced infrared image is obtained by converting it back to the spatial domain using inverse Fourier transform. This image has stronger high-frequency details and texture information. The specific inverse Fourier transform formula is as follows: 。 3. The infrared image diffusion super-resolution reconstruction method based on frequency domain enhancement and infrared sensing according to claim 1, characterized in that, S2 includes the following steps: S2.

1. Based on the thermophysical characteristics of infrared images, construct a physically guided diffusion model, where thermophysical consistency is represented as... This function describes the physical consistency of the image during the heat conduction process, ensuring that the temperature distribution in the generated image conforms to physical laws; S2.2 Introduce thermophysical constraints during the diffusion process and design a diffusion scheduling strategy to achieve the desired effect. The scheduling strategy involves gradually adjusting the flow of image information during the diffusion process to maintain the thermophysical consistency of the generated image. The specific calculation formula for the scheduling strategy is as follows: ； in, This is a scheduling coefficient used to control the degree of influence of physical consistency on the diffusion process; S2.3 During the diffusion process, a thermophysical consistency constraint is applied to each layer of the diffusion process to ensure that the thermophysical consistency of the generated image at each layer meets expectations, thus obtaining the updated image features. : ； in, For the current image features during the diffusion process, To ensure the reasonableness of the generated image in terms of thermophysical consistency in order to account for changes in image features during the diffusion process; S2.

4. Through its diffusion scheduling strategy, a super-resolution infrared image that conforms to thermophysical consistency is finally generated, thereby improving image quality and ensuring its thermophysical rationality.

4. The infrared image diffusion super-resolution reconstruction method based on frequency domain enhancement and infrared sensing according to claim 1, characterized in that, S3 includes the following steps: S3.

1. Extract features from the infrared image in the frequency domain to obtain the spectral representation. and phase information ,in This refers to amplitude information in the frequency domain. This refers to phase information in the frequency domain. S3.2, Define the spectrum consistency loss function This loss function is used to constrain the consistency between the spectrum of the generated image and the original image, ensuring that the spectral features of the generated image do not deviate significantly. The specific calculation formula is as follows: ； in, and The frequency domain amplitude information of the generated image and the reference image are respectively used to ensure that the spectrum of the generated image is consistent with the spectrum of the original image; S3.3, Define the phase consistency loss function This loss function is used to constrain the consistency of the generated image and the reference image in terms of frequency domain phase information, thereby enhancing the ability to restore image edges and texture details. The specific calculation formula is as follows: ； in, and The frequency domain phase information of the generated image and the reference image are respectively used to ensure that the generated image is consistent with the original image in terms of phase. S3.

4. Weighted summation of the spectral consistency loss function and the phase consistency loss function yields the final total loss function. This loss function is used to optimize the consistency of frequency domain features in the generated image, and the specific formula is as follows: ； in, and These are weighting coefficients used to balance the contributions of spectral consistency loss and phase consistency loss to the optimization process; S3.5, By optimizing the total loss function The spectral and phase consistency of the generated image are enhanced, resulting in a significant improvement in the recovery of edge and texture details, ensuring the accuracy of image detail and texture restoration.

5. The infrared image diffusion super-resolution reconstruction method based on frequency domain enhancement and infrared sensing according to claim 1, characterized in that, S4 includes the following steps: S4.1 Extracting the raw image from unlabeled infrared image data The training samples serve as input data for self-supervised learning, used to learn the target feature information of the image; S4.2, Define the loss function for self-supervised learning. This loss function is used to generate images. With the original image The differences between them are used to optimize the network's learning. The specific calculation formula is as follows: ； in, For the image generated through super-resolution reconstruction, Given the original image, the learning ability of the target features is improved by minimizing this loss function; S4.3 Introducing self-supervised networks This network is used to extract data from unlabeled images. Extracting feature information and optimizing the feature extraction process; self-supervised network. It will generate feature maps of the image. And by adjusting the image feature map compared with the original image feature map The differences between them guide the extraction of target features, ensuring that the learned features can effectively reflect image details and textures; S4.4 To further enhance the effectiveness of self-supervised learning, a contrastive loss function is introduced. This loss function is used to optimize the feature contrast of the generated image, making the target features in the image more prominent. The calculation formula is as follows: ； in, Feature maps generated by a self-supervised network. Given the feature map of the original image, by minimizing this loss function, we ensure that the generated image is consistent with the original image in the feature space; S4.5, through joint optimization and Self-supervised learning networks can effectively learn from unlabeled infrared image data. By learning target feature information, the performance of super-resolution reconstruction is further improved; ultimately, the generated image is significantly improved in terms of detail and texture restoration.

6. The infrared image diffusion super-resolution reconstruction method based on frequency domain enhancement and infrared sensing according to claim 1, characterized in that, S5 includes the following steps: S5.1 Combining the features extracted from frequency domain enhancement, physical guided diffusion, and self-supervised learning in steps S1 to S4, a multi-level feature fusion strategy is used to integrate the feature maps from different stages. and By fusing the data, multi-scale image feature representations can be obtained. : ； in, For the first Image features of the layer, These are the weighting coefficients for the features of this layer. The number of layers for feature fusion; S5.2 After feature fusion, the ability to restore image details is further improved by optimizing the diffusion model; the optimized diffusion process is based on the fused feature map. By adjusting the image's detail recovery capability through a diffusion model, high-frequency information is ensured to be fully recovered. S5.3 Define the optimized diffusion model loss function This loss function is used to constrain the differences between the features of the generated image and the fused image, further improving image quality. The formula for calculating the loss function is as follows: ； in, To generate an image, ensure that the generated super-resolution image is consistent with the fused feature map; S5.4 Combining multi-level feature fusion and diffusion model optimization, a high-quality super-resolution image is finally generated. The image shows significant improvements in detail restoration, edge sharpness, and resolution, resulting in improved final image features during the generation process. By minimizing To achieve the best results; S5.5 Through iterative optimization and feature fusion, super-resolution reconstruction of infrared images was finally achieved, significantly improving image resolution, sharpness, and texture details, ensuring excellent performance in visual quality and physical consistency of the generated images. The super-resolution images during the iterative process... By minimizing the loss function To achieve the best results, the specific iterative optimization formula is as follows: ； in, In the first The super-resolution image generated after the next iteration. This is the result of the previous iteration. For learning rate, loss function The gradient of the image indicates the direction of optimization in the current iteration step.

Images