Semantic communication method and system, electronic device, storage medium

Abstract

Embodiments of the present application provide a semantic communication method and system, electronic equipment and storage medium, relating to the technical field of artificial intelligence and wireless communication. The method comprises: acquiring multi-modal source data; wherein the multi-modal source data comprises text data and non-text data; performing text semantic mapping on the non-text data through a large model to obtain a mapped text sequence; combining elements in the mapped text sequence to obtain at least two candidate sub-sequences; filtering the candidate sub-sequences according to semantic similarity between the candidate sub-sequences and the mapped text sequence to obtain a semantically equivalent sub-sequence; performing semantic encoding on the semantically equivalent sub-sequence and the text data to obtain semantic features, and sending the semantic features to a receiving end, and executing a target task based on the semantic features by the receiving end. The embodiments of the present application can improve the transmission efficiency of semantic communication.

Description

Technical Field

[0001] This application relates to the fields of artificial intelligence and wireless communication technology, and in particular to a semantic communication method and system, electronic device, and storage medium. Background Technology

[0002] Semantic communication is a communication method that extracts and transmits semantic features of data. Semantic communication can understand and convey the meaning of data, helping to improve communication efficiency. For example, in low-altitude wireless networks in air-to-ground collaborative communication scenarios, airborne nodes (such as drones) and ground nodes (such as ground control stations) typically need to interact with each other for tasks such as target identification and decision support. In this process, airborne nodes can use semantic communication to transmit the semantic features of image, audio, and other data to ground nodes, rather than directly transmitting the raw data.

[0003] Currently, data is typically encoded using artificial intelligence (AI) models to generate semantic features. However, current AI models often introduce a large amount of redundant semantics, such as task-irrelevant semantics, when encoding and generating semantic features. These redundant semantics consume communication bandwidth and computing resources during encoding, transmission, and decoding, increasing the transmission load and affecting data transmission efficiency.

[0004] Therefore, how to improve the transmission efficiency of semantic communication has become an urgent technical problem to be solved. Summary of the Invention

[0005] The main objective of this application is to propose a semantic communication method and system, electronic device, and storage medium, which aims to reduce semantic redundancy in semantic encoding and improve the transmission efficiency of semantic communication.

[0006] To achieve the above objectives, a first aspect of this application proposes a semantic communication method, which is applied at a sending end and includes: Acquire multimodal source data; wherein, the multimodal source data includes text data and non-text data; By performing text semantic mapping on the non-text data using a large model, a mapped text sequence is obtained; Based on the elements in the mapped text sequence, at least two candidate subsequences are obtained; Based on the semantic similarity between the candidate subsequence and the mapped text sequence, the candidate subsequence is filtered to obtain semantically equivalent subsequences; Semantic encoding is performed on the semantically equivalent subsequence and the text data to obtain semantic features, and the semantic features are sent to the receiving end, which then performs the target task based on the semantic features.

[0007] In some embodiments, at least two candidate subsequences have different subsequence lengths; The step of filtering the candidate subsequences based on their semantic similarity to the mapped text sequence to obtain semantically equivalent subsequences includes: Feature extraction is performed on the mapped text sequence to obtain the semantic features of the text sequence; The candidate subsequences are traversed in ascending order of length, and feature extraction is performed on each candidate subsequence to obtain semantic features of the subsequences. Based on a preset similarity threshold, the semantic similarity between the semantic features of each subsequence and the semantic features of the text sequence, and the length of the subsequence, similar sequences are selected for each candidate subsequence of the subsequence length to obtain the semantically equivalent subsequence.

[0008] In some embodiments, the step of selecting similar sequences for candidate subsequences of each subsequence length based on a preset similarity threshold, the semantic similarity between the semantic features of each subsequence and the semantic features of the text sequence, and the subsequence length, to obtain the semantically equivalent subsequence, includes: For each subsequence length, the semantic similarity is calculated based on the semantic features of the subsequence and the semantic features of the text sequence to obtain the semantic similarity. If the semantic similarity of any semantic feature of the subsequence at the subsequence length is greater than or equal to a preset similarity threshold, the subsequence length is determined as the traversal termination length, and the traversal is terminated after similarity calculation is performed on each semantic feature of the subsequence at the traversal termination length. The semantic features of the subsequences whose semantic similarity is greater than or equal to a preset similarity threshold are determined as candidate semantically equivalent subsequence features; The candidate subsequence corresponding to the feature of the candidate semantically equivalent subsequence with the highest semantic similarity is determined as the semantically equivalent subsequence.

[0009] In some embodiments, the sequence length of the mapped text sequence is n, where the sequence length represents the number of elements contained in the mapped text sequence; The combination of elements from the mapped text sequence yields at least two candidate subsequences, including: Let n-1 be the initial value of the initial subsequence length k; Select k elements from the mapped text sequence and combine them to obtain a subsequence set of the initial subsequence length; wherein, the subsequence set of the initial subsequence length includes all the candidate subsequences consisting of k elements; The value of the initial subsequence length minus one is determined as the updated subsequence length; If the length of the updated subsequence is greater than or equal to a preset semantic compression length threshold, the value of the initial subsequence length is updated to the updated subsequence length, and the step of selecting k elements from the mapped text sequence for combination to obtain a subsequence set with the initial subsequence length is executed.

[0010] To achieve the above objectives, a second aspect of this application proposes a semantic communication method, which is applied at a receiving end and includes: Obtain semantic features sent by the sending end; wherein the sending end is used to perform the method described in the first aspect above, and the semantic features include semantic sub-features of multiple data types; Obtain the task hint vector for the target task; Based on each of the semantic sub-features, each of the data types, and the task prompt vector, a strategy is generated to obtain a target fusion strategy; wherein, the target fusion strategy refers to the fusion method of fusing the semantic sub-features; Based on the target fusion strategy, the semantic sub-features are fused to obtain fused semantic features; The target task is executed based on the fused semantic features.

[0011] In some embodiments, the target fusion strategy includes a weighted sum strategy; The process of fusing the semantic sub-features based on the target fusion strategy to obtain fused semantic features includes: If the target fusion strategy is the weighted sum strategy, obtain the signal transmission quality index of each semantic sub-feature; Weights are generated based on the signal transmission quality index to obtain the signal quality weight of each semantic sub-feature. The fused semantic features are obtained by weighting and summing each semantic sub-feature according to the signal quality weights.

[0012] In some embodiments, the target fusion strategy includes an attention-based fusion strategy; The process of fusing the semantic sub-features based on the target fusion strategy to obtain fused semantic features includes: If the target fusion strategy is the attention-based fusion strategy, the attention weight of each semantic sub-feature is obtained by performing attention calculation on each semantic sub-feature; Attention fusion is performed on each of the semantic sub-features according to the attention weights to obtain the fused semantic features.

[0013] To achieve the above objectives, a second aspect of this application provides a semantic communication system, the semantic communication system comprising: Sending end and receiving end; the sending end is communicatively connected to the receiving end; The sending end is used to execute the semantic communication method described in the first aspect above; The receiving end is used to execute the semantic communication method described in the second aspect above.

[0014] To achieve the above objectives, a third aspect of the present application provides an electronic device, the electronic device including a memory and a processor, the memory storing a computer program, and the processor executing the computer program to implement the method described in the first aspect or the second aspect.

[0015] To achieve the above objectives, a fourth aspect of the present application provides a computer-readable storage medium storing a computer program that, when executed by a processor, implements the method described in the first aspect or the second aspect.

[0016] The semantic communication method, system, electronic device, and storage medium proposed in this application acquire multimodal source data, including text and non-text data, and perform text semantic mapping on the non-text data (such as images and audio) using a large model to obtain a mapped text sequence. This first transforms the non-text data into a text form with a clear semantic structure, facilitating subsequent semantic encoding and transmission. Then, considering the contextual generalization characteristics of the large model during text generation, its output text semantic description often contains redundant expressions with weak relevance to the target task. Therefore, multiple candidate subsequences are obtained by combining elements from the mapped text sequence. Based on the semantic similarity between the candidate subsequences and the mapped text sequence, semantically equivalent subsequences are selected from the candidate subsequences. This allows for the removal of redundant expressions with weak relevance to the original text in the large model-generated text by compressing semantics, improving the compactness of the semantic representation, and thus reducing the burden of subsequent encoding and transmission, rather than directly encoding the original description containing a large amount of redundant semantics. Semantic features are obtained by semantically encoding the semantically equivalent subsequences and text data, and then sent to the receiving end to execute the target task. This allows the generation of high-information-density semantic features using compressed semantically equivalent subsequences and text data, reducing semantic redundancy in semantic coding and improving the transmission efficiency of semantic communication. Attached Figure Description

[0017] Figure 1 This is a flowchart of the semantic communication method provided in the embodiments of this application; Figure 2 yes Figure 1 The flowchart for step 103 in the text; Figure 3 yes Figure 1 The flowchart for step 104 in the document; Figure 4 yes Figure 3 The flowchart for step 303 in the document; Figure 5 This is a flowchart of a semantic communication method provided in another embodiment of this application; Figure 6 This is a schematic diagram of the semantic decoding process of the receiving end provided in the embodiments of this application; Figure 7 This is an embodiment provided by this application. Figure 5 The flowchart for step 504 in the document; Figure 8 This is provided by another embodiment of the present application. Figure 5 The flowchart for step 504 in the document; Figure 9 This is a flowchart illustrating an application example provided in an embodiment of this application; Figure 10 This is a schematic diagram of the hardware structure of the electronic device provided in the embodiments of this application. Detailed Implementation

[0018] To make the objectives, technical solutions, and advantages of this application clearer, the following detailed description is provided in conjunction with the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are merely illustrative and not intended to limit the scope of this application.

[0019] It should be noted that although functional modules are divided in the device schematic diagram and a logical order is shown in the flowchart, in some cases, the steps shown or described may be performed in a different order than the module division in the device or the order in the flowchart. The terms "first," "second," etc., in the specification, claims, and the aforementioned drawings are used to distinguish similar objects and are not necessarily used to describe a specific order or sequence.

[0020] Unless otherwise defined, all technical and scientific terms used herein have the same meaning as commonly understood by one of ordinary skill in the art to which this application belongs. The terminology used herein is for the purpose of describing embodiments of this application only and is not intended to limit this application.

[0021] First, let's analyze some of the terms used in this application: Artificial Intelligence (AI) is a new technical science that studies and develops theories, methods, technologies, and application systems for simulating, extending, and expanding human intelligence. AI is a branch of computer science that attempts to understand the essence of intelligence and produce intelligent machines that can react in a way similar to human intelligence. Research in this field includes robotics, speech recognition, image recognition, natural language processing, and expert systems. AI can be a simulation of the information processes of human consciousness and thought. AI can also be the theory, methods, technologies, and application systems that use digital computers or machines controlled by digital computers to simulate, extend, and expand human intelligence, perceive the environment, acquire knowledge, and use that knowledge to obtain optimal results. Basic AI technologies generally include sensors, dedicated AI chips, cloud computing, distributed storage, big data processing technology, operating / interactive systems, and mechatronics. AI software technologies mainly include computer vision, robotics, biometrics, speech processing, natural language processing, and machine learning / deep learning. This application can acquire and process relevant data based on AI technology.

[0022] In air-to-ground collaborative communication scenarios, exemplified by low-altitude wireless networks, airborne and ground nodes typically need to engage in multimodal information exchange around specific tasks to support intelligent applications such as target recognition and decision support. Current communication systems mostly transmit raw data directly, which struggles to meet the demands for low latency, high reliability, and intelligent services. Therefore, semantic communication, by understanding the connotations of data and transmitting task-related semantics, helps improve communication efficiency and has become an important research direction in the integration of wireless communication and artificial intelligence. Current semantic communication systems are usually built on deep learning networks, and their semantic representation capabilities are limited by the distribution of training data. When communication scenarios or task requirements change, semantic communication systems struggle to maintain stable generalization performance in cross-domain migration or open environments, often requiring retraining or fine-tuning, making them ill-suited to the dynamically changing low-altitude wireless network environment. To address these issues, large-scale AI models pre-trained on massive corpora can be used for semantic communication. These AI models can learn universally applicable semantic patterns from complex distributions, providing support for cross-task and cross-scenario semantics. However, multimodal semantic communication still faces two limitations. On the one hand, semantic encoding methods based on large AI models introduce a large amount of redundant information with limited contribution to the task when generating semantic features. This redundant semantics consumes communication bandwidth and computing resources during encoding, transmission, and decoding, increasing the transmission and computing load. On the other hand, current semantic decoding processes are based on predefined, manually designed multimodal fusion mechanisms, which are difficult to dynamically adjust according to changes in task requirements, channel conditions, and semantic quality, resulting in insufficient adaptability.

[0023] Based on this, embodiments of this application propose a semantic communication method and system, electronic device, and storage medium, aiming to solve the problems of semantic redundancy and lack of adaptability of receiver fusion mechanism in multimodal semantic coding based on AI large model in low-altitude wireless networks, thereby reducing semantic redundancy in semantic coding and improving the transmission efficiency of semantic communication.

[0024] The semantic communication method, system, electronic device, and storage medium provided in this application are specifically described through the following embodiments. First, the semantic communication method in the embodiments of this application is described.

[0025] The semantic communication method provided in this application can be applied to a terminal, a server, or software running on either a terminal or a server. In some embodiments, the terminal can be a smartphone, tablet, laptop, desktop computer, etc.; the server can be configured as an independent physical server, a server cluster or distributed system composed of multiple physical servers, or a cloud server providing basic cloud computing services such as cloud services, cloud databases, cloud computing, cloud functions, cloud storage, network services, cloud communication, middleware services, domain name services, security services, CDN (Content Delivery Network), and big data and artificial intelligence platforms; the software can be an application that implements the semantic communication method, but is not limited to the above forms.

[0026] This application can be used in a wide variety of general-purpose or special-purpose computer system environments or configurations. Examples include: personal computers, server computers, handheld or portable devices, tablet devices, multiprocessor systems, microprocessor-based systems, set-top boxes, programmable consumer electronics, network PCs, minicomputers, mainframe computers, and distributed computing environments including any of the above systems or devices. This application can be described in the general context of computer-executable instructions executed by a computer, such as program modules. Generally, program modules include routines, programs, objects, components, data structures, etc., that perform specific tasks or implement specific abstract data types. This application can also be practiced in distributed computing environments where tasks are performed by remote processing devices connected via a communication network. In distributed computing environments, program modules can reside in local and remote computer storage media, including storage devices.

[0027] It should be noted that in all specific embodiments of this application, when processing data related to user identity or characteristics, such as user information, user behavior data, user historical data, and user location information, user permission or consent is obtained first. Furthermore, the collection, use, and processing of this data comply with relevant laws, regulations, and standards. In addition, when embodiments of this application require access to sensitive personal information of users, separate permission or consent from the user is obtained through pop-ups or redirection to confirmation pages. Only after obtaining the user's separate permission or consent is the necessary user-related data required for the proper functioning of these embodiments acquired.

[0028] Figure 1 This is an optional flowchart of the semantic communication method provided in the embodiments of this application. Figure 1 The method described above is applied to the sending end, and the method may include, but is not limited to, steps 101 to 105.

[0029] Step 101: Obtain multimodal source data; wherein, multimodal source data includes text data and non-text data; Step 102: Perform text semantic mapping on non-text data using a large model to obtain a mapped text sequence; Step 103: Combine the elements in the mapped text sequence to obtain at least two candidate subsequences; Step 104: Based on the semantic similarity between the candidate subsequence and the mapped text sequence, the candidate subsequence is filtered to obtain semantically equivalent subsequences; Step 105: Semantic encoding is performed on the semantically equivalent subsequence and text data to obtain semantic features, and the semantic features are sent to the receiving end, which then performs the target task based on the semantic features.

[0030] The beneficial effects of this application's embodiments include, but are not limited to: acquiring multimodal source data including text and non-text data, and performing text semantic mapping on non-text data (such as images, audio, etc.) using a large model to obtain mapped text sequences. This allows non-text data to be uniformly converted into text forms with clear semantic structures, facilitating subsequent semantic encoding and transmission. Then, considering the contextual generalization characteristics of the large model during text generation, its output text semantic descriptions often contain redundant expressions with weak relevance to the target task. Therefore, multiple candidate subsequences are obtained by combining elements in the mapped text sequence. Based on the semantic similarity between the candidate subsequences and the mapped text sequence, semantically equivalent subsequences are selected from the candidate subsequences. This allows for the removal of redundant expressions with weak relevance to the original text in the large model-generated text by compressing semantics, improving the compactness of semantic representation, and thus reducing the burden of subsequent encoding and transmission, rather than directly encoding the original description containing a large amount of redundant semantics. Semantic features are obtained by semantically encoding the semantically equivalent subsequences and text data, and then sent to the receiving end to execute the target task. This allows the generation of high-information-density semantic features using compressed semantically equivalent subsequences and text data, reducing semantic redundancy in semantic coding and improving the transmission efficiency of semantic communication.

[0031] In step 101 of some embodiments, the transmitting end refers to a device used to collect and transmit multimodal source data. Multimodal source data is data with multiple modalities (i.e., data types), which may include text data and non-text data. Specifically, non-text data may include image data and audio data, etc. For example, in an air-to-ground cooperative communication scenario represented by a low-altitude wireless network, the transmitting end may be an air node, such as a drone; the receiving end may be a ground node, such as a ground control station.

[0032] In step 102 of some embodiments, the large model can be an artificial intelligence (AI) large model for performing text semantic mapping, such as a large language model (LLM). Other types of AI models can also be used for text semantic mapping, and are not limited to this. The mapped text sequence is text data obtained by transforming non-text data. For example, assuming the non-text data is an image, the text data can be a textual description of that image.

[0033] In some embodiments, it should be noted that non-textual data, such as image and audio data, typically manifests as high-dimensional continuous signals, lacking explicit structured semantic representation, and containing a large amount of edge contours, texture details, and frequency component information. Directly performing semantic modeling on this type of data would lead to complex feature extraction processes and high computational costs, which is not conducive to efficient transmission in resource-constrained low-altitude wireless networks. Therefore, embodiments of this application introduce a multimodal AI large model to perform cross-modal semantic mapping on non-textual data such as image and audio data, uniformly converting them into text form (i.e., mapping text sequences), thereby achieving the representation of different modal data in a unified semantic space. This can transform the original unstructured high-dimensional signals into text sequences with clear semantic structures, facilitating subsequent semantic encoding and transmission.

[0034] In step 103 of some embodiments, the elements in the mapped text sequence can be characters in the mapped text sequence. Elements can also be phrases in the mapped text sequence, etc., and are not limited thereto. It should be noted that each candidate subsequence is a combination of some elements in the mapped text sequence, and is also a possible compressed text of the mapped text sequence.

[0035] In step 104 of some embodiments, the semantically equivalent subsequence is a text sequence that is selected from multiple candidate subsequences and has a high semantic similarity to the mapped text sequence (e.g., greater than a preset threshold).

[0036] It should be noted that due to the contextual generalization characteristics of large AI models during inference and generation, their output text semantic descriptions (i.e., mapped text sequences) often contain redundant expressions with weak relevance to the target task. This redundant information will consume additional communication bandwidth and computing resources during subsequent semantic encoding and transmission, reducing the overall system efficiency. Therefore, this application employs a semantic compression method that requires no additional training, namely, combining elements from the mapped text sequence to obtain multiple candidate subsequences, and then selecting semantically equivalent subsequences from them. This allows for the compression of redundant text semantics while ensuring semantic equivalence, generating a more compact semantic representation.

[0037] In step 105 of some embodiments, the semantic features are feature vectors obtained by semantically encoding the semantically equivalent subsequences and text data together. In some embodiments, the semantically equivalent subsequences and text data can be semantically encoded using a semantic encoder based on a Transformer architecture. In another embodiment, the semantically equivalent subsequences and text data can also be semantically encoded using a text encoder.

[0038] It should be noted that the target task refers to the task that the sending and receiving ends need to complete together. For example, the target task can be target recognition, such as identifying targets in images captured by a drone (i.e., the sending end). The target task can also be target classification, decision support, etc., and is not limited to these.

[0039] In some embodiments, the transmitting end can transmit signals such as semantic features to the receiving end via a physical channel. For example, in an air-to-ground cooperative communication scenario, a drone in a low-altitude wireless network can communicate wirelessly with a ground control station via a physical channel, such as electromagnetic waves in the air. That is, the drone can transmit semantic features to the ground control station via a physical channel.

[0040] Please see Figure 2 In some embodiments, the sequence length of the mapped text sequence is n, where the sequence length represents the number of elements contained in the mapped text sequence; Step 103 may include, but is not limited to, steps 201 to 204: Step 201: Determine n-1 as the initial value of the initial subsequence length k; Step 202: Select k elements from the mapped text sequence and combine them to obtain a subsequence set with an initial subsequence length; wherein, the subsequence set with an initial subsequence length includes all candidate subsequences consisting of k elements; Step 203: Determine the value of the initial subsequence length minus one as the updated subsequence length; Step 204: If the length of the updated subsequence is greater than or equal to the preset semantic compression length threshold, update the value of the initial subsequence length to the updated subsequence length, and perform the step of selecting k elements from the mapped text sequence for combination to obtain a subsequence set with the initial subsequence length.

[0041] The advantage of this embodiment is that by determining the initial value of the initial subsequence length k by subtracting one from the sequence length n of the mapped text sequence, and selecting k elements from the mapped text sequence for combination to obtain a set containing all candidate subsequences composed of k elements, a set of candidate compression schemes for the mapped text sequence can be constructed, providing a selection basis for subsequent screening. Then, by determining the updated subsequence length by subtracting one from the initial subsequence length, and updating the initial subsequence length and repeating the combination step when the updated subsequence length is greater than or equal to a preset semantic compression length threshold, all possible subsequence combinations can be constructed layer by layer with decreasing length. This allows for the automatic and efficient search for shorter and less redundant semantic equivalent representations, provided that the subsequences satisfy basic semantic integrity (i.e., the subsequence length is greater than or equal to the preset semantic compression length threshold), thereby improving the transmission efficiency of semantic communication.

[0042] In step 201 of some embodiments, the sequence length n of the mapped text sequence refers to the length of the redundant sequence (i.e., the mapped text sequence) generated by the large model, which is the number of elements in the mapped text sequence. For example, assuming the mapped text sequence contains 15 character elements, then the sequence length n of the mapped text sequence is 15, and the initial value of the initial subsequence length k is n-1=14.

[0043] In step 202 of some embodiments, the set of subsequences of initial subsequence length k can be the set of all possible candidate subsequences consisting of k elements. For example, suppose the mapped text sequence is... If the given text sequence is {"cat", "sit", "at", "chair", "on"}, then the sequence length n = 5 and k = n - 1 = 4. In this case, the subsequence set with initial subsequence length k is all possible combinations of elements obtained by selecting 4 elements from the mapped text sequence, which is the subsequence set with initial subsequence length k = 4. The set includes: {"cat", "sit", "at", "chair"}, {"cat", "sit", "at", "on"}, {"cat", "sit", "chair", "on"}, {"cat", "at", "chair", "on"}, and {"sit", "at", "chair", "on"}. The number of candidate subsequences in this set is the number of combinations of choosing 4 elements from 5 elements, C(5,4)=5.

[0044] In step 203 of some embodiments, the value of the initial subsequence length minus one is determined as the updated subsequence length. For example, assuming the initial subsequence length k=4, then the updated subsequence length is k-1=3.

[0045] In step 204 of some embodiments, for example, the preset semantic compression length threshold can be 2, or other values, not limited to these. To avoid catastrophic degradation of semantic expression, a minimum subsequence length constraint can be set using the semantic compression length threshold. If the updated subsequence length is greater than or equal to the preset semantic compression length threshold, the initial subsequence length is updated to the updated subsequence length, that is, the value of k is updated to k-1, and step 203 above is executed. In this way, a set of subsequences with decreasing length can be constructed based on the mapped text sequence. .in, Indicates selection from the mapped text sequence All possible subsequences (i.e., candidate subsequences) consisting of elements.

[0046] Please see Figure 3 In some embodiments, at least two candidate subsequences have different subsequence lengths; Step 104 may include, but is not limited to, steps 301 through 303: Step 301: Extract features from the mapped text sequence to obtain semantic features of the text sequence; Step 302: Traverse each candidate subsequence in ascending order of subsequence length, and extract features based on each candidate subsequence to obtain the semantic features of the subsequence; Step 303: Based on the preset similarity threshold, the semantic similarity between the semantic features of each subsequence and the semantic features of the text sequence, and the length of the subsequence, similar sequences are selected for candidate subsequences of each subsequence length to obtain semantically equivalent subsequences.

[0047] The advantage of this embodiment lies in the fact that by specifying that at least two candidate subsequences have different lengths, a series of candidate schemes with different compression levels can be constructed, providing a basis for balancing semantic fidelity and compression efficiency. Then, each candidate subsequence is traversed in ascending order of length, and semantic features are extracted from each candidate subsequence. This prioritizes evaluating shorter subsequences with higher compression rates; if their semantic similarity meets the requirements, they can be considered semantically equivalent subsequences, thus seeking higher compression efficiency while ensuring semantic equivalence. Based on the semantic similarity between the semantic features of each subsequence and the semantic features of the text sequence, as well as the subsequence length, similarity sequence selection is performed on candidate subsequences of each length to obtain semantically equivalent subsequences. This approach, under the condition of satisfying basic semantic similarity, tends to select shorter subsequences, thereby generating semantic representations with higher information density and lower redundancy before semantic encoding, further improving the transmission efficiency of semantic communication.

[0048] In step 301 of some embodiments, the semantic features of the text sequence are feature vectors obtained by feature extraction from the mapped text sequence. In some embodiments, features of the mapped text sequence can be extracted using a pre-trained language model (such as the general semantic representation model BERT).

[0049] In step 302 of some embodiments, the subsequence semantic features are feature vectors obtained by feature extraction from the candidate subsequences. In some embodiments, feature extraction of the candidate subsequences can be performed using a pre-trained language model (such as the general semantic representation model BERT). Feature extraction can also be performed in other ways, and is not limited to these.

[0050] In step 303 of some embodiments, the semantically equivalent subsequence is a text sequence selected from multiple candidate subsequences. The selection process for the semantically equivalent subsequence can be found in the detailed description of steps 401 to 404 below, and will not be repeated here.

[0051] Please see Figure 4In some embodiments, step 303 may include, but is not limited to, steps 401 to 404: Step 401: For the semantic features of each subsequence length, calculate the similarity between the semantic features of the subsequence and the semantic features of the text sequence to obtain the semantic similarity. Step 402: If the semantic similarity of any semantic feature of a subsequence of length is greater than or equal to a preset similarity threshold, the subsequence length is determined as the traversal termination length, and the traversal is terminated after calculating the similarity of the semantic features of each subsequence of the traversal termination length. Step 403: Define the semantic features of subsequences with semantic similarity greater than or equal to a preset similarity threshold as candidate semantically equivalent subsequence features; Step 404: The candidate subsequence corresponding to the candidate semantically equivalent subsequence feature with the highest semantic similarity is determined as the semantically equivalent subsequence.

[0052] The advantage of this embodiment lies in that it calculates the similarity of semantic features for each subsequence length and determines the traversal termination length and terminates the traversal when the semantic similarity is greater than or equal to a preset similarity threshold. This ensures that, while meeting the basic requirements of semantic similarity, invalid calculations on longer and more redundant subsequences are stopped in a timely manner, thereby reducing the computational overhead of the semantic compression process and improving the screening efficiency. Then, the semantic features of the subsequences that meet the threshold requirements are determined as candidate semantically equivalent subsequence features, and the candidate subsequence with the highest semantic similarity is determined as the final semantically equivalent subsequence. This allows the selection of the representation that is closest to the original semantics and has the lowest distortion among all compression candidate schemes that meet the basic semantic fidelity requirements. Thus, while achieving semantic compression and reducing redundancy, it retains the core semantic information related to the task to the greatest extent, ensuring the accuracy of subsequent encoding and task execution. Therefore, this scheme optimizes the decision-making process of semantic compression by introducing a threshold-based early termination mechanism and an optimal similarity screening strategy. It can generate higher quality and less redundant semantic representations while controlling computational complexity, thereby further improving the transmission efficiency of semantic communication.

[0053] In step 401 of some embodiments, cosine similarity can be calculated for the semantic features of the subsequence and the semantic features of the text sequence. The calculated cosine similarity is used as the semantic similarity, that is, the semantic consistency between the semantic features of the subsequence and the semantic features of the text sequence is measured by the cosine similarity.

[0054] In step 402 of some embodiments, since the embodiments of this application traverse each set of subsequences in ascending order of subsequence length k, and when a semantic similarity of a subsequence is found that meets the standard (i.e., is greater than or equal to a preset similarity threshold), the traversal is terminated after traversing all semantic features of the current subsequence length. This can avoid redundant calculations for longer subsequences.

[0055] In step 403 of some embodiments, among multiple subsequence semantic features of the same subsequence length, there may be one or more subsequence semantic features with a semantic similarity greater than or equal to a preset similarity threshold. Embodiments of this application can determine all these subsequence semantic features that satisfy the semantic similarity condition as candidate semantically equivalent subsequence features.

[0056] In step 404 of some embodiments, among the candidate subsequences corresponding to multiple candidate semantically equivalent subsequence features, the candidate subsequence with the highest semantic similarity is determined as the semantically equivalent subsequence.

[0057] Figure 5 This is an optional flowchart of the semantic communication method provided in the embodiments of this application. Figure 5 The method described above is applied to the receiving end, and the method may include, but is not limited to, steps 501 to 505: Step 501: Obtain the semantic features sent by the sending end; wherein the sending end is used to execute the above semantic communication method, and the semantic features include semantic sub-features of multiple data types; Step 502: Obtain the task hint vector for the target task; Step 503: Based on each semantic sub-feature, each data type, and the task prompt vector, a strategy is generated to obtain the target fusion strategy; whereby the target fusion strategy refers to the fusion method of fusing semantic sub-features. Step 504: Based on the target fusion strategy, fuse the various semantic sub-features to obtain fused semantic features; Step 505: Execute the target task based on the fused semantic features.

[0058] The advantage of this embodiment lies in acquiring semantic features containing semantic sub-features of multiple data types transmitted by the sending end after semantic compression, and obtaining the task prompt vector of the target task. This provides the receiving end with low-redundancy semantic features and clear task requirements for task execution. A target fusion strategy is generated based on each semantic sub-feature, its data type, and the task prompt vector. This enables a dynamic, task-adaptive feature fusion mechanism, rather than using a fixed fusion method for all data types or types of data (such as target recognition and classification). Therefore, it intelligently determines the most effective fusion method based on different task requirements and the characteristics of different modalities of data (such as text and images), fully understanding the task-related semantics in the compressed semantic information. The semantic sub-features are fused based on the target fusion strategy to obtain fused semantic features, and the target task is executed based on these fused semantic features. This allows the receiving end to accurately complete the task using the optimized, high-information-density fused semantic features, improving the adaptability of the receiving end's fusion mechanism.

[0059] In step 501 of some embodiments, such as Figure 6 As shown, the receiving end can obtain semantic sub-features of multiple data types from the semantic features sent by the sending end, such as signal 1, signal 2, signal 3, etc. In some embodiments, the Transformer module can be used to perform attention calculation on each semantic sub-feature to capture the global dependencies between different modal information.

[0060] In step 502 of some embodiments, the task cue vector of the target task is used to characterize the requirements of the task, such as the requirements of classifying or recognizing targets in image data.

[0061] In step 503 of some embodiments, such as Figure 6 As shown, the fusion strategy for semantic sub-features can be dynamically selected through a gating unit. It should be noted that the gating unit generates a corresponding routing decision (i.e., selecting which semantic sub-features to participate in subsequent fusion) based on the received semantic sub-features, their modal composition, and the current task requirements, and maps the semantic sub-features to the best-matching candidate fusion strategy in the expert pool. The expert pool includes multiple preset candidate fusion strategies, which target different modal combinations and semantic association characteristics, thereby fully exploring the semantic complementarity and correlation between heterogeneous modalities. For example, the various candidate fusion strategies in the expert pool may include additive fusion, weighted fusion, and attention-based fusion methods. In some embodiments, the gating unit can use one-hot encoding to activate only the output of the selected candidate fusion strategy (i.e., the target fusion strategy), thereby significantly reducing computational complexity while ensuring semantic fusion effectiveness.

[0062] In some embodiments, the task executor is composed of a multilayer perceptron network. Specifically, the gating unit may consist of two linear layers with a hidden layer dimension of 128. In some embodiments, to avoid policy selection bias caused by unstable gating decisions in the early stages of training, a random routing strategy can be used to initialize the gating unit in the early stages of training, so that each candidate fusion strategy has a balanced opportunity to participate, thereby improving the overall training stability and system robustness.

[0063] In step 504 of some embodiments, the fused semantic features are semantic features obtained by fusing the various semantic sub-features according to the target fusion strategy. For example, assuming the target fusion strategy is additive fusion, then the fused semantic features are feature vectors obtained by summing the various semantic sub-features.

[0064] In step 505 of some embodiments, the task executor can perform task-related reasoning, decision-making, or prediction operations based on fused semantic features, thereby outputting the final task execution result.

[0065] Please see Figure 7 In some embodiments, the target fusion strategy includes a weighted sum strategy; Step 504 may include, but is not limited to, steps 601 through 603: Step 601: If the target fusion strategy is a weighted sum strategy, obtain the signal transmission quality index of each semantic sub-feature. Step 602: Generate weights based on signal transmission quality indicators to obtain the signal quality weights for each semantic sub-feature; Step 603: The semantic sub-features are weighted and summed according to the signal quality weights to obtain the fused semantic features.

[0066] The advantage of this embodiment lies in that, when the target fusion strategy is a weighted sum strategy, it obtains the signal transmission quality index of each semantic sub-feature and generates signal quality weights based on the signal transmission quality index. This allows for a quantitative assessment of the transmission quality of each semantic sub-feature during its transmission from the transmitter to the receiver, such as the degree of interference. The fused semantic features are obtained by weighted summation of each semantic sub-feature according to these weights. This dynamically adjusts the contribution of semantic sub-features recovered under different data types and channel conditions to the final fused representation, giving higher weights to semantic sub-features with better transmission quality and less susceptibility to noise. This proactively suppresses semantic impairment caused by channel noise or interference, thereby improving the completeness and accuracy of the fused semantic features and enhancing the adaptability of the receiver fusion mechanism.

[0067] In step 601 of some embodiments, the signal transmission quality index is an indicator used to characterize the quality of semantic sub-features during transmission. Specifically, the signal transmission quality index may include bit error rate (BER), latency, reference signal received power (RSRP), signal to interference plus noise ratio (SINR), received signal strength indication (RSSI), etc.

[0068] It should be noted that latency refers to the time from data transmission to reception. Bit error rate (BER) refers to the proportion of bits that err during signal transmission. Reference signal received power (RSRP) reflects the signal power received. Signal-to-interference-plus-noise ratio (SINR) reflects the clarity of the signal. Received signal strength indication (RSSI) reflects the signal strength received.

[0069] In step 602 of some embodiments, for example, a larger signal quality weight for a semantic sub-feature indicates less interference experienced by that semantic sub-feature during transmission. The signal quality weight can be obtained by normalizing the signal transmission quality index. Other methods can also be used to generate the signal quality weight, and this is not limited to these.

[0070] In step 603 of some embodiments, if the signal quality weight of a semantic sub-feature is less than a preset signal quality threshold, the semantic sub-feature may not be selected to participate in the fusion operation. For example, Figure 6 As shown, assuming there are three semantic sub-features (i.e., signal 1, signal 2, and signal 3), semantic sub-features with signal quality weights greater than or equal to the signal quality threshold can be selected for fusion decision. For example, assuming only signal 1 and signal 2 have signal quality weights greater than or equal to the signal quality threshold, then signal 1 and signal 2 can be selected, i.e., @1,2. It should be noted that... Figure 6 In the code, @1,2 indicates that signal 1 and signal 2 are selected for fusion; @1,2,3 indicates that signal 1, signal 2 and signal 3 are selected for fusion, and so on.

[0071] Please see Figure 8 In some embodiments, the target fusion strategy includes an attention-based fusion strategy; Step 504 may include, but is not limited to, steps 701 to 702: Step 701: If the target fusion strategy is an attention-based fusion strategy, the attention weight of each semantic sub-feature is obtained by performing attention calculation on each semantic sub-feature. Step 702: Perform attention fusion on each semantic sub-feature according to the attention weight to obtain the fused semantic feature.

[0072] The advantage of this embodiment lies in that, when the target fusion strategy is an attention-based fusion strategy, attention weights are calculated for each semantic sub-feature. This allows the receiver to leverage the attention mechanism to perform fusion based on the semantic relevance and importance of each semantic sub-feature, dynamically evaluating the intrinsic value of each sub-feature to the current fusion task, rather than relying on fixed external metrics. Then, attention fusion is performed on each semantic sub-feature according to the calculated attention weights to obtain the fused semantic features, thereby improving the adaptability of the receiver's fusion mechanism.

[0073] In step 701 of some embodiments, attention can be calculated for each semantic sub-feature using the Transformer algorithm to obtain the attention weight of each semantic sub-feature.

[0074] In step 702 of some embodiments, the fused semantic features can be a feature vector obtained by performing attention fusion on each semantic sub-feature according to attention weights.

[0075] This application embodiment also provides a semantic communication system, the semantic communication system including: Sender and receiver; the sender's communication connection to the receiver; The sending end is used to execute the semantic communication method described above; The receiving end is used to execute the semantic communication method described above.

[0076] Please see Figure 9In one application example, in a semantic communication system, the sender communicates wirelessly with the receiver via a physical channel. The physical channel carries the wireless transmission of semantic features. For example, in a low-altitude wireless network, an airborne node can act as the sender, and a ground node as the receiver. Specifically, based on a large AI model, non-textual data such as images and audio from multimodal source data (e.g., images, audio, and text) can be cross-modal semantically mapped to obtain a text sequence. Then, the text sequence can be semantically compressed to obtain a minimum semantically equivalent subsequence. The text from the multimodal source data and the minimum semantically equivalent subsequence are then input into a semantic encoder based on a Transformer architecture for semantic encoding to obtain semantic features. After the sender wirelessly transmits the semantic features to the receiver via the physical channel, the receiver can input the received semantic features into a multimodal semantic fusion decoder based on dynamic routing for adaptive decoding to obtain fused semantic features, thus obtaining a task-oriented semantic representation. Then, a task executor can execute tasks based on the fused semantic features.

[0077] The specific implementation of this semantic communication system is basically the same as the specific implementation of the semantic communication method described above, and will not be repeated here.

[0078] This application also provides an electronic device, which includes a memory and a processor. The memory stores a computer program, and the processor executes the computer program to implement the aforementioned semantic communication method. This electronic device can include any smart terminal such as a tablet computer or an in-vehicle computer.

[0079] Please see Figure 10 , Figure 10 The hardware structure of an electronic device according to another embodiment is illustrated. The electronic device includes: The processor 901 can be implemented using a general-purpose CPU (Central Processing Unit), microprocessor, application-specific integrated circuit (ASIC), or one or more integrated circuits, and is used to execute relevant programs to implement the technical solutions provided in the embodiments of this application. The memory 902 can be implemented as a read-only memory (ROM), a static storage device, a dynamic storage device, or a random access memory (RAM). The memory 902 can store the operating system and other applications. When the technical solutions provided in the embodiments of this specification are implemented through software or firmware, the relevant program code is stored in the memory 902 and is called and executed by the processor 901 using the semantic communication method of the embodiments of this application. The 903 input / output interface is used to implement information input and output. The communication interface 904 is used to enable communication and interaction between this device and other devices. Communication can be achieved through wired means (such as USB, Ethernet cable, etc.) or wireless means (such as mobile network, WIFI, Bluetooth, etc.). Bus 905 transmits information between various components of the device (e.g., processor 901, memory 902, input / output interface 903, and communication interface 904); The processor 901, memory 902, input / output interface 903, and communication interface 904 are connected to each other within the device via bus 905.

[0080] This application also provides a computer-readable storage medium storing a computer program that, when executed by a processor, implements the above-described semantic communication method.

[0081] Memory, as a non-transitory computer-readable storage medium, can be used to store non-transitory software programs and non-transitory computer-executable programs. Furthermore, memory may include high-speed random access memory, and may also include non-transitory memory, such as at least one disk storage device, flash memory device, or other non-transitory solid-state storage device. In some embodiments, memory may optionally include memory remotely located relative to the processor, and these remote memories can be connected to the processor via a network. Examples of such networks include, but are not limited to, the Internet, intranets, local area networks, mobile communication networks, and combinations thereof.

[0082] It should be noted that the models, software tools, or components appearing in the embodiments of this application are merely illustrative examples and do not represent actual use. All user personal information involved in the embodiments of this application has been authorized (with knowledge and consent) by the relevant parties or has been fully authorized by all parties, and the executing entity can obtain it through various legal and compliant means. The collection, storage, use, processing, transmission, provision, and disclosure of the information, data, and signals involved all comply with the provisions of relevant laws and regulations and do not violate public order and good morals.

[0083] The embodiments described in this application are intended to more clearly illustrate the technical solutions of the embodiments of this application, and do not constitute a limitation on the technical solutions provided by the embodiments of this application. Those skilled in the art will know that with the evolution of technology and the emergence of new application scenarios, the technical solutions provided by the embodiments of this application are also applicable to similar technical problems.

[0084] Those skilled in the art will understand that the technical solutions shown in the figures do not constitute a limitation on the embodiments of this application, and may include more or fewer steps than shown, or combine certain steps, or different steps.

[0085] The device embodiments described above are merely illustrative. The units described as separate components may or may not be physically separate; that is, they may be located in one place or distributed across multiple network units. Some or all of the modules can be selected to achieve the purpose of this embodiment according to actual needs.

[0086] Those skilled in the art will understand that all or some of the steps in the methods disclosed above, as well as the functional modules / units in the systems and devices, can be implemented as software, firmware, hardware, or suitable combinations thereof.

[0087] The terms “first,” “second,” “third,” “fourth,” etc. (if present) in the specification and accompanying drawings of this application are used to distinguish similar objects and are not necessarily used to describe a specific order or sequence. It should be understood that such data can be interchanged where appropriate so that the embodiments of this application described herein can be implemented in orders other than those illustrated or described herein. Furthermore, the terms “comprising” and “having,” and any variations thereof, are intended to cover non-exclusive inclusion; for example, a process, method, system, product, or apparatus that comprises a series of steps or units is not necessarily limited to those steps or units explicitly listed, but may include other steps or units not explicitly listed or inherent to such processes, methods, products, or apparatus.

[0088] It should be understood that in this application, "at least one (item)" means one or more, and "more than" means two or more. "And / or" is used to describe the relationship between related objects, indicating that three relationships can exist. For example, "A and / or B" can represent three cases: only A exists, only B exists, and both A and B exist simultaneously, where A and B can be singular or plural. The character " / " generally indicates that the preceding and following related objects are in an "or" relationship. "At least one (item) of the following" or similar expressions refer to any combination of these items, including any combination of single or plural items. For example, at least one (item) of a, b, or c can represent: a, b, c, "a and b", "a and c", "b and c", or "a and b and c", where a, b, and c can be single or multiple.

[0089] In the several embodiments provided in this application, it should be understood that the disclosed apparatus and methods can be implemented in other ways. For example, the apparatus embodiments described above are merely illustrative; for instance, the division of the units described above is only a logical functional division, and in actual implementation, there may be other division methods. For example, multiple units or components may be combined or integrated into another system, or some features may be ignored or not executed. The coupling or direct coupling or communication connection between the shown or discussed units may be through some interfaces, or indirect coupling or communication connection between the apparatus or units, and may be electrical, mechanical, or other forms.

[0090] The units described above as separate components may or may not be physically separate. The components shown as units may or may not be physical units; that is, they may be located in one place or distributed across multiple network units. Some or all of the units can be selected to achieve the purpose of this embodiment according to actual needs.

[0091] Furthermore, the functional units in the various embodiments of this application can be integrated into one processing unit, or each unit can exist physically separately, or two or more units can be integrated into one unit. The integrated unit can be implemented in hardware or as a software functional unit.

[0092] If the integrated unit is implemented as a software functional unit and sold or used as an independent product, it can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of this application, in essence, or the part that contributes to the prior art, or all or part of the technical solution, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes multiple instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute all or part of the steps of the methods of the various embodiments of this application. The aforementioned storage medium includes various media capable of storing programs, such as USB flash drives, portable hard drives, read-only memory (ROM), random access memory (RAM), magnetic disks, or optical disks.

[0093] The preferred embodiments of the present application have been described above with reference to the accompanying drawings, but this does not limit the scope of the claims of the present application. Any modifications, equivalent substitutions, and improvements made by those skilled in the art without departing from the scope and substance of the embodiments of the present application shall be within the scope of the claims of the present application.

Claims

1. A semantic communication method, characterized in that, The method is applied at the sending end, and the method includes: Acquire multimodal source data; wherein, the multimodal source data includes text data and non-text data; By performing text semantic mapping on the non-text data using a large model, a mapped text sequence is obtained; Based on the elements in the mapped text sequence, at least two candidate subsequences are obtained; the subsequence lengths of the at least two candidate subsequences are different. Feature extraction is performed on the mapped text sequence to obtain the semantic features of the text sequence; The candidate subsequences are traversed in ascending order of length, and feature extraction is performed on each candidate subsequence to obtain semantic features of the subsequences. Based on a preset similarity threshold, the semantic similarity between the semantic features of each subsequence and the semantic features of the text sequence, and the length of the subsequence, similar sequences are selected for each candidate subsequence of each subsequence length to obtain semantically equivalent subsequences. Semantic encoding is performed on the semantically equivalent subsequence and the text data to obtain semantic features, and the semantic features are sent to the receiving end, which then performs the target task based on the semantic features.

2. The method according to claim 1, characterized in that, The step of selecting similar sequences for candidate subsequences of each subsequence length based on a preset similarity threshold, the semantic similarity between the semantic features of each subsequence and the semantic features of the text sequence, and the subsequence length, to obtain the semantically equivalent subsequence, includes: For each subsequence length, the semantic similarity is calculated based on the semantic features of the subsequence and the semantic features of the text sequence to obtain the semantic similarity. If the semantic similarity of any semantic feature of the subsequence at the subsequence length is greater than or equal to a preset similarity threshold, the subsequence length is determined as the traversal termination length, and the traversal is terminated after similarity calculation is performed on each semantic feature of the subsequence at the traversal termination length. The semantic features of the subsequences whose semantic similarity is greater than or equal to a preset similarity threshold are determined as candidate semantically equivalent subsequence features; The candidate subsequence corresponding to the feature of the candidate semantically equivalent subsequence with the highest semantic similarity is determined as the semantically equivalent subsequence.

3. The method according to any one of claims 1 to 2, characterized in that, The sequence length of the mapped text sequence is n, and the sequence length represents the number of elements contained in the mapped text sequence; The combination of elements from the mapped text sequence yields at least two candidate subsequences, including: Let n-1 be the initial value of the initial subsequence length k; Select k elements from the mapped text sequence and combine them to obtain a subsequence set of the initial subsequence length; wherein, the subsequence set of the initial subsequence length includes all the candidate subsequences consisting of k elements; The value of the initial subsequence length minus one is determined as the updated subsequence length; If the length of the updated subsequence is greater than or equal to a preset semantic compression length threshold, the value of the initial subsequence length is updated to the updated subsequence length, and the step of selecting k elements from the mapped text sequence for combination to obtain a subsequence set with the initial subsequence length is executed.

4. A semantic communication method, characterized in that, The method is applied at the receiving end, and the method includes: Obtain semantic features sent by the sending end; wherein the sending end is used to perform the method as described in any one of claims 1 to 3, and the semantic features include semantic sub-features of multiple data types; Obtain the task hint vector for the target task; Based on each of the semantic sub-features, each of the data types, and the task prompt vector, a strategy is generated to obtain a target fusion strategy; wherein, the target fusion strategy refers to the fusion method of fusing the semantic sub-features; Based on the target fusion strategy, the semantic sub-features are fused to obtain fused semantic features; The target task is executed based on the fused semantic features.

5. The method according to claim 4, characterized in that, The target fusion strategy includes a weighted sum strategy; The process of fusing the semantic sub-features based on the target fusion strategy to obtain fused semantic features includes: If the target fusion strategy is the weighted sum strategy, obtain the signal transmission quality index of each semantic sub-feature; Weights are generated based on the signal transmission quality index to obtain the signal quality weight of each semantic sub-feature. The fused semantic features are obtained by weighting and summing each semantic sub-feature according to the signal quality weights.

6. The method according to claim 4 or 5, characterized in that, The target fusion strategy includes an attention-based fusion strategy; The process of fusing the semantic sub-features based on the target fusion strategy to obtain fused semantic features includes: If the target fusion strategy is the attention-based fusion strategy, the attention weight of each semantic sub-feature is obtained by performing attention calculation on each semantic sub-feature; Attention fusion is performed on each of the semantic sub-features according to the attention weights to obtain the fused semantic features.

7. A semantic communication system, characterized in that, The semantic communication system includes: Sending end and receiving end; the sending end is communicatively connected to the receiving end; The sending end is used to execute the semantic communication method as described in any one of claims 1 to 3; The receiving end is used to perform the semantic communication method as described in any one of claims 4 to 6.

8. An electronic device, characterized in that, The electronic device includes a memory and a processor. The memory stores a computer program, and the processor executes the computer program to implement the semantic communication method according to any one of claims 1 to 3 or any one of claims 4 to 6.

9. A computer-readable storage medium storing a computer program, characterized in that, When the computer program is executed by a processor, it implements the semantic communication method according to any one of claims 1 to 3 or any one of claims 4 to 6.

Images