Spectrum map communication organization method, device and equipment based on semantic information
Through the spectrum map communication organization method based on semantic information, the bandwidth consumption problem during spectrum map construction and distribution is solved by using the variational autoencoder and denoising diffusion implicit model and UNet network prediction, and the bandwidth consumption problem during spectrum map construction and distribution is achieved efficient spectrum Resource management and signal encoding to adapt to complex electromagnetic environments.
Patent Information
- Application Number
- CN202410975587.1
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2024-07-19
- Publication Date
- 2024-11-22
- Estimated Expiration
- 2044-07-19
AI Technical Summary
In a complex electromagnetic spectrum environment, building and distributing spectrum maps requires a lot of bandwidth and time, resulting in a high burden on communication systems, and it is difficult for the existing technology to effectively optimize spectrum resource management and signal coding strategies.
Using a spectrum map communication organization method based on semantic information, the spectrum map is encoded and added noise processing through a variational autoencoder and denoising diffusion implicit model, combined with the UNet network for noise prediction, optimize the image potential spatial vector to generate the real spectrum Map, reduce data transmission volume and improve communication efficiency.
It significantly reduces the data transmission overhead during the construction and distribution of spectrum maps, improves the reorganization efficiency of spectrum maps and the energy efficiency of communication systems, and can reduce redundant information transmission while ensuring accuracy, and adapt to the rapidly changing electromagnetic environment.
Smart Images

Figure CN119006659A_ABST
Abstract
Description
Technical Field
[0001] The present invention relates to the field of wireless communication network technology, and in particular to a spectrum map communication organization method, device and equipment based on semantic information. Background Art
[0002] With the rapid development of 6G, the Internet of Things and smart cities, a large number of wireless communication devices have been widely deployed, resulting in the rapid complication of the electromagnetic spectrum environment and increasingly acute conflicts in the use of frequencies. As a key carrier and valuable resource for network information transmission, the electromagnetic spectrum is of great significance for monitoring and allocating spectrum resources, avoiding channel conflicts and interference, and optimizing transmission power, frequency selection and signal coding strategies to improve communication efficiency and reduce energy consumption.
[0003] Spectrum maps are a key tool for understanding complex electromagnetic environments. By combining geographic information systems, they describe the electromagnetic environment from multiple dimensions such as time, frequency, space, and power, support the effective acquisition of the distribution of electromagnetic spectrum resources, and explore the multi-dimensional reusable value of the spectrum. Spectrum maps assist wireless resource management in an intuitive and flexible way, and have been widely used in areas such as cognitive radio, interference management, and active resource allocation in prospective networks. However, building spectrum maps is a complex computing task, and the distribution of spectrum maps from cloud servers to users consumes a lot of bandwidth and time. In a rapidly changing electromagnetic environment, this large bandwidth consumption may impose a significant burden on the communication system. Summary of the invention
[0004] Based on this, it is necessary to provide a spectrum map communication organization method, device and equipment based on semantic information that can improve the energy efficiency of spectrum map communication organization in response to the above technical problems.
[0005] A spectrum map communication organization method based on semantic information, the method is implemented in the mutual communication between a cloud / edge server and a frequency user, comprising:
[0006] Get spectrum map and text information.
[0007] A variational autoencoder is used to encode the spectrum map to obtain image feature information, which is then mapped to a latent feature space to obtain an image latent space vector.
[0008] The text information is encoded to obtain text features, and the text features are stacked in the feature dimensions of the same batch according to the attributes of the text prompt words in the text features to obtain the text prompt word feature vector.
[0009] Based on the denoising diffusion implicit model, the image latent space vector is denoised to obtain the noisy image features.
[0010] After fusing the noisy image features with the text prompt word feature vector, they are input into the UNet network together with the time step for noise prediction. The image latent space vector is optimized according to the predicted noise data to obtain the real spectrum map.
[0011] In one of the embodiments, it also includes: encoding the text information using CLIP text vectorization of a CLIP text encoder to obtain encoded text information, and inputting the encoded text information into a CLIP text model for feature extraction to obtain text features.
[0012] In one of the embodiments, the method further includes: classifying the prompt words in the text information according to preset rules to obtain positive prompt words and negative prompt words, and stacking the positive prompt words and the negative prompt words in the same batch of feature dimensions to obtain a text prompt word feature vector.
[0013] In one embodiment, the method further includes: setting a noise value and a total number of noise adding iteration steps, adding noise to the image latent space vector according to a product result of the noise value and the total number of noise adding iteration steps, calculating a difference between the total number of noise adding iteration steps and the product result, obtaining a remaining number of iteration steps, and not adding noise to the image latent space vector within the remaining number of iteration steps.
[0014] In one of the embodiments, it also includes: selecting a value that is the same as the reconstruction ratio of the spectrum map as the noise value, adding noise to the image latent space vector based on a denoising diffusion implicit model to reconstruct the noise value percentage, and obtaining a noisy image feature corresponding to the noise percentage feature.
[0015] In one of the embodiments, it also includes: fusing the noisy image features with the text prompt word feature vector through a neural network model, and inputting them together with the time steps obtained through the fully connected layer of the neural network into the UNet network for noise prediction.
[0016] In one of the embodiments, it also includes: after each step of noise prediction by the UNet network, new Gaussian noise is obtained, the Gaussian noise is input into the Unet network according to a preset ratio for denoising iteration, the image latent space vector is optimized by sampling in the latent space multiple times, and a real spectrum map is reconstructed according to the optimized image latent space vector.
[0017] A spectrum map communication organization device based on semantic information, loaded in a mutual communication system implemented between a cloud / edge server and a frequency user, the device comprising:
[0018] The information acquisition module is used to obtain spectrum maps and text information.
[0019] The image latent space vector acquisition module is used to encode the spectrum map using a variational autoencoder to obtain image feature information, map the image feature information to a latent feature space, and obtain an image latent space vector.
[0020] The text prompt word feature vector acquisition module is used to encode text information to obtain text features, and stack the text prompt words in the same batch of feature dimensions according to the attributes of the text prompt words in the text features to obtain the text prompt word feature vector.
[0021] The noisy image feature acquisition module is used to add noise to the image latent space vector based on the denoising diffusion implicit model to obtain the noisy image feature.
[0022] The real spectrum map construction module is used to fuse the noisy image features with the text prompt word feature vector, and input them into the UNet network together with the time step for noise prediction. The image latent space vector is optimized according to the predicted noise data to obtain the real spectrum map.
[0023] A computer device comprises a memory and a processor, wherein the memory stores a computer program, and when the processor executes the computer program, the following steps are implemented:
[0024] Get spectrum map and text information.
[0025] A variational autoencoder is used to encode the spectrum map to obtain image feature information, which is then mapped to a latent feature space to obtain an image latent space vector.
[0026] The text information is encoded to obtain text features, and the text features are stacked in the feature dimensions of the same batch according to the attributes of the text prompt words in the text features to obtain the text prompt word feature vector.
[0027] Based on the denoising diffusion implicit model, the image latent space vector is denoised to obtain the noisy image features.
[0028] After fusing the noisy image features with the text prompt word feature vector, they are input into the UNet network together with the time step for noise prediction. The image latent space vector is optimized according to the predicted noise data to obtain the real spectrum map.
[0029] The above-mentioned spectrum map communication organization method, device and equipment based on semantic information first encode the spectrum map through VAE, extract image features and map them into the latent feature space, and generate the image latent space vector. This process not only effectively compresses the image information, but also extracts the most representative features, making the subsequent processing more efficient. At the same time, the text information is encoded to extract its features, and stacked on the feature dimension of the same batch according to the attributes of the text prompt words to form a text prompt word feature vector. This step ensures the accurate expression of the text information in the feature space, so that it can be seamlessly integrated with the image features. Next, the image latent space vector is denoised based on the denoising diffusion implicit model. The denoising process simulates the noise interference in the actual environment, so that the model can better adapt to and handle the noise conditions in the real world. In this way, the model can learn how to extract useful signals from the noise, improving the robustness and generalization ability of the model. After the denoised image features are fused with the text prompt word feature vector, they are input into the UNet network together with the time step for noise prediction. The UNet network is a commonly used deep learning architecture that is good at processing image segmentation and reconstruction tasks. In this technical solution, the UNet network is used to predict noise data and optimize the image latent space vector based on the prediction results. Through multiple iterations and optimizations, a realistic and high-quality spectrum map can be generated. BRIEF DESCRIPTION OF THE DRAWINGS
[0030] Figure 1 A flowchart of a spectrum map communication organization method based on semantic information in one embodiment;
[0031] Figure 2 A schematic diagram of a spectrum map communication organization processing framework based on semantic information in one embodiment;
[0032] Figure 3 A schematic diagram of a flowchart of image feature processing in a spectrum map based on semantic information in one embodiment;
[0033] Figure 4 A schematic diagram of a flow chart of text feature processing in a spectrum map based on semantic information in one embodiment;
[0034] Figure 5 It is a structural block diagram of a spectrum map communication organization device based on semantic information in one embodiment;
[0035] Figure 6 FIG. 4 is a diagram showing the internal structure of a computer device in one embodiment. DETAILED DESCRIPTION
[0036] In order to make the purpose, technical solution and advantages of the present invention more clearly understood, the present invention is further described in detail below in conjunction with the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are only used to explain the present invention and are not intended to limit the present invention.
[0037] In one embodiment, Figure 1 As shown, a spectrum map communication organization method based on semantic information is provided, and the method is implemented in the mutual communication between the cloud / edge server and the frequency user, including the following steps:
[0038] Step 102: Obtain spectrum map and text information.
[0039] Step 104, using a variational autoencoder to encode the spectrum map to obtain image feature information, and mapping the image feature information to a latent feature space to obtain an image latent space vector.
[0040] Step 106, encode the text information to obtain text features, and stack the text features in the same batch according to the attributes of the text prompt words in the text features to obtain a text prompt word feature vector.
[0041] Step 108: adding noise to the image latent space vector based on the denoising diffusion implicit model to obtain a noisy image feature.
[0042] Step 110, after fusing the noisy image feature vector with the text prompt word feature vector, the feature vector and the time step are input into the UNet network for noise prediction, and the image latent space vector is optimized according to the predicted noise data to obtain the real spectrum map.
[0043] In the above-mentioned spectrum map communication organization method based on semantic information, first, the spectrum map is encoded by VAE, the image features are extracted and mapped into the latent feature space, and the image latent space vector is generated. This process not only effectively compresses the image information, but also extracts the most representative features, making the subsequent processing more efficient. At the same time, the text information is encoded to extract its features, and stacked on the feature dimension of the same batch according to the attributes of the text prompt words to form the text prompt word feature vector. This step ensures the accurate expression of the text information in the feature space, so that it can be seamlessly integrated with the image features. Next, the image latent space vector is denoised based on the denoising diffusion implicit model. The denoising process simulates the noise interference in the actual environment, so that the model can better adapt to and handle the noise conditions in the real world. In this way, the model can learn how to extract useful signals from the noise, improving the robustness and generalization ability of the model. After the denoised image features are fused with the text prompt word feature vector, they are input into the UNet network together with the time step for noise prediction. The UNet network is a commonly used deep learning architecture that is good at processing image segmentation and reconstruction tasks. In this technical solution, the UNet network is used to predict noise data and optimize the image latent space vector based on the prediction results. Through multiple iterations and optimizations, a realistic and high-quality spectrum map can be generated.
[0044] In one embodiment, if Figure 2 As shown, a spectrum map communication organization processing block diagram based on semantic information is provided, and the method is implemented in the mutual communication between the cloud / edge server and the frequency user. The processing block diagram includes the following steps:
[0045] The text information is encoded by the CLIP text encoder, and the original spectrum image is encoded by the variational autoencoder and then added to the scheduler of the diffusion model. The obtained text features and the semantic image after the generated image noise features are transmitted to the frequency user through the channel. The user combines the time step and text embedding to calculate the current noise, predict the noise, remove the noise, and restore the original high-definition spectrum map. Noise is added based on the DDIM diffusion model. The noise value can be understood as the proportion that needs to be noised, that is, the proportion of the image that needs to be reconstructed. It can be considered as the percentage of the reconstructed noise value, and the (1-noise value) percentage feature is retained. Set the noise value and the total number of steps. For the input image, the noise value × total number of steps will be added, and the remaining steps will not be added. If the spectrum map is constructed based on the text information only or based on the text information and the image semantic information using the unconditional restricted diffusion model, it is meaningless for the task of constructing an accurate spectrum map; only the conditional restricted diffusion model can be used to construct an accurate spectrum map. Compared with the transmission of the original spectrum map, the amount of data required to transmit text information and image semantic information is effectively reduced, which can improve the energy efficiency of spectrum map reconstruction.
[0046] In one embodiment, if Figure 3 As shown, a text feature processing process in a spectrum map based on semantic information is provided, which is implemented in the communication between a cloud / edge server and a frequency user, including the following steps:
[0047] Step S300, obtaining a high-definition spectrum map and text information that can represent the characteristics of the spectrum map, such as (transmitter location, power, etc.).
[0048] Step S310, encoding is performed using a variational autoencoder to obtain feature information of the image, and the image feature information is mapped to a potential feature space.
[0049] Step S320, using the contrastive language-image pre-training model CLIP (Contrastive Language-Image Pre-Training, CLIP) to encode the input text and obtain text features.
[0050] Step S330, after obtaining the features of the high-definition spectrum map image, noise is added based on the diffusion model DDIM (Denoising Diffusion Implicit Models, DDIM) to obtain the features of the input image after noise is added.
[0051] Step S340, the obtained text features are fused with the generated image noise features and input into the UNet network with the time step, and the current moment noise is calculated by combining the time step number and text embedding to predict the noise.
[0052] Step S150, sampling is performed based on the obtained latent space noise-added image, and the latent space features are decoded using a variational autoencoder to generate an accurate high-definition spectrum image.
[0053] In this embodiment, the communication between the cloud / edge server and the frequency user transmits the semantic information of the spectrum map, and the user uses the diffusion model to construct the spectrum map based on the text information and the semantic noise image information. The conditional restricted diffusion model is used to use the text information and the captured image semantic information to perform accurate spectrum map reconstruction tasks. This method can reduce other redundant information when the spectrum map is constructed and distributed. When only the semantic information is transmitted, the spectrum map can be completely and accurately constructed, thereby significantly reducing the communication overhead of data transmission while ensuring accuracy.
[0054] In one embodiment, if Figure 4 As shown, a process for processing image features in a spectrum map based on semantic information is provided. The method is implemented in the communication between a cloud / edge server and a frequency user, and specifically includes:
[0055] Step S400: for the input text prompt word information including positive prompt words and negative prompt words, the input positive prompt word feature vectors and negative prompt word feature vectors are stacked in the feature dimension of batch.
[0056] Step S410, after the stacking is completed, the obtained image latent space vector, step number and text prompt word feature vector are input into the UNet network together, and the result is segmented in the batch feature dimension.
[0057] Step S420: For negative prompt words, we hope that the constructed image can be away from its influence and be closer to the influence of positive prompt words. To this end, we hope to expand the distance between the two through calculation.
[0058] In step S430, after each step of noise prediction by the UNet network, new Gaussian noise needs to be obtained and added to the network in a certain proportion. By introducing new noise and continuously iterating the prediction noise denoising, the robustness of the network can be increased.
[0059] Step S440 , sampling is performed multiple times in the latent space, and the obtained latent vector is decoded by using a variational autoencoder to obtain the required spectrum map.
[0060] In this embodiment, the frequency user obtains the required frequency spectrum map by decoding according to the characteristics of the text information and the characteristics of the image information after the noise is added.
[0061] In one of the embodiments, the text information is encoded using CLIP text vectorization of a CLIP text encoder to obtain encoded text information, and the encoded text information is input into a CLIP text model for feature extraction to obtain text features.
[0062] In one of the embodiments, the prompt words in the text information are attribute-classified according to preset rules to obtain positive prompt words and negative prompt words, and the positive prompt words and the negative prompt words are stacked within the feature dimensions of the same batch to obtain a text prompt word feature vector.
[0063] In one embodiment, a noise value and a total number of noise adding iteration steps are set, and noise is added to the image latent space vector according to a product result of the noise value and the total number of noise adding iteration steps. A difference between the total number of noise adding iteration steps and the product result is calculated to obtain a remaining number of iteration steps, and the image latent space vector is not noised within the remaining number of iteration steps.
[0064] In one of the embodiments, a value having the same proportion as the spectrum map reconstruction is selected as the noise value, and the image latent space vector is denoised based on a denoising diffusion implicit model to reconstruct the noise value percentage and obtain a noisy image feature corresponding to the noise percentage feature.
[0065] In one embodiment, the noisy image features and the text prompt word feature vector are fused through a neural network model, and input into a UNet network together with the time steps obtained through the fully connected layer of the neural network for noise prediction.
[0066] In one of the embodiments, after each step of noise prediction by the UNet network, new Gaussian noise is obtained, and the Gaussian noise is input into the Unet network according to a preset ratio for denoising iteration, and the image latent space vector is optimized by sampling in the latent space multiple times, and the real spectrum map is reconstructed according to the optimized image latent space vector.
[0067] It should be understood that although Figure 1 , 3 The steps in the flowchart are shown in sequence as indicated by the arrows, but these steps are not necessarily executed in the order indicated by the arrows. Unless otherwise specified in this document, there is no strict order restriction for the execution of these steps, and these steps can be executed in other orders. Moreover, Figure 1 , 3 At least part of the steps may include multiple sub-steps or multiple stages. These sub-steps or stages are not necessarily executed at the same time, but can be executed at different times. The execution order of these sub-steps or stages is not necessarily sequential, but can be executed in turn or alternately with other steps or at least part of the sub-steps or stages of other steps.
[0068] In one embodiment, Figure 5 As shown, a spectrum map communication organization device based on semantic information is provided, including: an information acquisition module 502, an image latent space vector acquisition module 504, a text prompt word feature vector acquisition module 506, a noisy image feature acquisition module 508 and a real spectrum map construction module 510, wherein:
[0069] Information acquisition module 502, used to acquire spectrum map and text information;
[0070] An image latent space vector acquisition module 504 is used to encode the spectrum map using a variational autoencoder to obtain image feature information, and map the image feature information to a latent feature space to obtain an image latent space vector;
[0071] A text prompt word feature vector acquisition module 506 is used to encode the text information to obtain text features, and stack the text prompt words in the text features according to the attributes of the text prompt words in the same batch of feature dimensions to obtain a text prompt word feature vector;
[0072] A noisy image feature acquisition module 508, configured to add noise to the image latent space vector based on a denoising diffusion implicit model to obtain a noisy image feature;
[0073] The real spectrum map construction module 510 is used to fuse the noisy image feature vector with the text prompt word feature vector, and input it into the UNet network together with the time step to perform noise prediction, and optimize the image latent space vector according to the predicted noise data to obtain the real spectrum map.
[0074] For the specific definition of the spectrum map communication organization device based on semantic information, please refer to the definition of the spectrum map communication organization method based on semantic information above, which will not be repeated here. The various modules in the above-mentioned spectrum map communication organization device based on semantic information can be implemented in whole or in part through software, hardware and a combination thereof. The above-mentioned modules can be embedded in or independent of the processor in the computer device in the form of hardware, or can be stored in the memory of the computer device in the form of software, so that the processor can call and execute the operations corresponding to the above modules.
[0075] In one embodiment, a computer device is provided. The computer device may be a terminal, and its internal structure diagram may be as follows: Figure 6 As shown. The computer device includes a processor, a memory, a network interface, a display screen and an input device connected through a system bus. Among them, the processor of the computer device is used to provide computing and control capabilities. The memory of the computer device includes a non-volatile storage medium and an internal memory. The non-volatile storage medium stores an operating system and a computer program. The internal memory provides an environment for the operation of the operating system and the computer program in the non-volatile storage medium. The network interface of the computer device is used to communicate with an external terminal through a network connection. When the computer program is executed by the processor, a spectrum map communication organization method based on semantic information is implemented. The display screen of the computer device can be a liquid crystal display screen or an electronic ink display screen, and the input device of the computer device can be a touch layer covering the display screen, or a button, trackball or touchpad set on the computer device housing, or an external keyboard, touchpad or mouse, etc.
[0076] Those skilled in the art will understand that the structure shown in FIG. Y is merely a block diagram of a partial structure related to the solution of the present invention, and does not constitute a limitation on the computer device to which the solution of the present invention is applied. The specific computer device may include more or fewer components than those shown in the figure, or combine certain components, or have a different arrangement of components.
[0077] In one embodiment, a computer device is provided, including a memory and a processor, wherein the memory stores a computer program, and when the processor executes the computer program, the following steps are implemented:
[0078] Get spectrum map and text information.
[0079] A variational autoencoder is used to encode the spectrum map to obtain image feature information, which is then mapped to a latent feature space to obtain an image latent space vector.
[0080] The text information is encoded to obtain text features, and the text features are stacked in the feature dimensions of the same batch according to the attributes of the text prompt words in the text features to obtain the text prompt word feature vector.
[0081] The image latent space vector is denoised based on a denoising diffusion implicit model to obtain a noisy image feature.
[0082] After fusing the noisy image features with the text prompt word feature vector, they are input into the UNet network together with the time step for noise prediction. The image latent space vector is optimized according to the predicted noise data to obtain the real spectrum map.
[0083] Those skilled in the art can understand that all or part of the processes in the above-mentioned embodiment methods can be completed by instructing the relevant hardware through a computer program, and the computer program can be stored in a non-volatile computer-readable storage medium. When the computer program is executed, it can include the processes of the embodiments of the above-mentioned methods. Among them, any reference to memory, storage, database or other media used in the embodiments provided by the present invention can include non-volatile and / or volatile memory. Non-volatile memory can include read-only memory (ROM), programmable ROM (PROM), electrically programmable ROM (EPROM), electrically erasable programmable ROM (EEPROM) or flash memory. Volatile memory can include random access memory (RAM) or external cache memory. As an illustration and not limitation, RAM is available in many forms, such as static RAM (SRAM), dynamic RAM (DRAM), synchronous DRAM (SDRAM), double data rate SDRAM (DDRSDRAM), enhanced SDRAM (ESDRAM), synchronous link (Synchlink) DRAM (SLDRAM), memory bus (Rambus) direct RAM (RDRAM), direct memory bus dynamic RAM (DRDRAM), and memory bus dynamic RAM (RDRAM).
[0084] The technical features of the above embodiments may be arbitrarily combined. To make the description concise, not all possible combinations of the technical features in the above embodiments are described. However, as long as there is no contradiction in the combination of these technical features, they should be considered to be within the scope of this specification.
[0085] The above-mentioned embodiments only express several implementation methods of the present invention, and the descriptions thereof are relatively specific and detailed, but they cannot be understood as limiting the scope of the invention patent. It should be pointed out that, for ordinary technicians in this field, several variations and improvements can be made without departing from the concept of the present invention, and these all belong to the protection scope of the present invention. Therefore, the protection scope of the patent of the present invention shall be subject to the attached claims.
Claims
1. A spectrum map communication organization method based on semantic information, It is characterized in that The method is implemented in the mutual communication between the cloud / edge server and the frequency user, and includes: Get spectrum maps and text information; Using a variational autoencoder to encode the spectrum map to obtain image feature information, and mapping the image feature information to a latent feature space to obtain an image latent space vector; Encoding the text information to obtain text features, and stacking the text features in the same batch according to the attributes of the text prompt words in the text features to obtain a text prompt word feature vector; Adding noise to the image latent space vector based on a denoising diffusion implicit model to obtain a noisy image feature; After the noisy image feature vector is fused with the text prompt word feature vector, it is input into the UNet network together with the time step for noise prediction, and the image latent space vector is optimized according to the predicted noise data to obtain the real spectrum map.
2. The method according to claim 1, It is characterized in that The text information is encoded to obtain text features, including: The text information is encoded by CLIP text vectorization of a CLIP text encoder to obtain encoded text information, and the encoded text information is input into a CLIP text model for feature extraction to obtain text features.
3. The method according to claim 2, It is characterized in that According to the attributes of the text prompt words in the text features, the text prompt word feature vectors are obtained by stacking the feature dimensions of the same batch, including: According to preset rules, attribute classification is performed on the prompt words in the text information to obtain positive prompt words and negative prompt words, and the positive prompt words and the negative prompt words are stacked within the feature dimensions of the same batch to obtain a text prompt word feature vector.
4. The method according to claim 1, It is characterized in that Setting a noise value and a total number of noise adding iteration steps, and adding noise to the image latent space vector according to a product result of the noise value and the total number of noise adding iteration steps; The difference between the total number of iteration steps of the noise addition and the product result is calculated to obtain the remaining number of iteration steps, and the image latent space vector is not denoised within the remaining number of iteration steps.
5. The method according to claim 4, It is characterized in that The image latent space vector is denoised based on a denoising diffusion implicit model to obtain a noisy image feature, including: A value having the same proportion as the spectrum map reconstruction is selected as the noise value, and the image latent space vector is denoised based on a denoising diffusion implicit model to reconstruct the noise value percentage, thereby obtaining a noisy image feature corresponding to the noise percentage feature.
6. The method according to claim 5, It is characterized in that After the noisy image feature vector is fused with the text prompt word feature vector, it is input into the UNet network together with the time step number for noise prediction, including: The noisy image features and the text prompt word feature vector are fused through a neural network model, and are input into a UNet network together with the time steps obtained through the fully connected layer of the neural network for noise prediction.
7. The method according to claim 6, It is characterized in that Optimizing the image latent space vector according to the predicted noise data to obtain a true spectrum map, including: After each step of noise prediction by the UNet network, new Gaussian noise is obtained, and the Gaussian noise is input into the Unet network according to a preset ratio for denoising iteration. The image latent space vector is optimized by sampling in the latent space multiple times, and the real spectrum map is reconstructed according to the optimized image latent space vector.
8. A spectrum map communication organization device based on semantic information, It is characterized in that The device is loaded in a communication system implemented between a cloud / edge server and a frequency user, and includes: Information acquisition module, used to obtain spectrum map and text information; An image latent space vector acquisition module is used to encode the spectrum map using a variational autoencoder to obtain image feature information, and map the image feature information to a latent feature space to obtain an image latent space vector; A text prompt word feature vector acquisition module is used to encode the text information to obtain text features, and stack the text prompt words in the text features according to the attributes of the text prompt words in the same batch of feature dimensions to obtain text prompt word feature vectors; A noisy image feature acquisition module, used for adding noise to the image latent space vector based on a denoising diffusion implicit model to obtain a noisy image feature; The real spectrum map construction module is used to fuse the noisy image feature with the text prompt word feature vector, and input it into the UNet network together with the time step for noise prediction, and optimize the image latent space vector according to the predicted noise data to obtain the real spectrum map.
9. A computer device comprising a memory and a processor, wherein the memory stores a computer program. It is characterized in that When the processor executes the computer program, the steps of the method according to any one of claims 1 to 7 are implemented.
Citation Information
Patent Citations
Image enhancement method and device based on diffusion model, equipment and storage medium
CN116664450A
Image generation method and device, electronic equipment and computer readable storage medium
CN117575894A
Spectrum semantic communication system for sparse data completion and radiation source positioning
CN117915342A
Face image restoration method based on generation diffusion prior
CN118333866A
Structured document generation from text prompts
GB202402040D0
Cited By
A high-precision map reconstruction method
CN119784960A
A high-precision map reconstruction method
CN119784960B
