Method, device, equipment and storage medium for switching communication mode of walkie-talkie

Abstract

The present application relates to the field of data processing technology, and discloses a method, device, equipment, and storage medium for switching the communication mode of a walkie-talkie. The method includes: dynamically fusing heterogeneous data collected by multimodal sensors to obtain a high-dimensional feature vector; inputting the high-dimensional feature vector into a preset topological communication model for plasticity optimization to obtain an initial communication mode; performing progressive switching processing on the initial communication mode to obtain a target communication state; performing scenario parameter analysis on the acquired user behavior pattern and user communication content to obtain target scenario parameters; extracting adjustment parameters for the target communication state based on the target scenario parameters to obtain an adjustment parameter set; constructing a distributed communication switching strategy for multiple walkie-talkies based on the adjustment parameter set to obtain a target communication switching strategy, and controlling the multiple walkie-talkies based on the target communication switching strategy. The present application improves the efficiency of switching the communication mode of walkie-talkies.

Description

Technical Field

[0001] The present application relates to the field of data processing, and in particular to a method, device, equipment and storage medium for switching the communication mode of a walkie-talkie. Background Art

[0002] With the rapid development of communication technology, walkie-talkies, as an important mobile communication device, have gained widespread use across various industries. Traditional walkie-talkie communication methods primarily rely on fixed frequencies and channels, manually configured and switched to adapt to varying communication environments. However, this static communication approach often fails to meet user needs in complex and changing environments. In recent years, intelligent walkie-talkie systems have begun to employ adaptive communication technology, dynamically adjusting communication parameters by sensing environmental changes. These systems typically utilize a library of preset communication modes, matching and switching based on simple environmental parameters.

[0003] However, existing smart walkie-talkie systems still have several limitations. Their limited environmental perception makes it difficult to fully capture complex and ever-changing communication scenarios. Furthermore, communication mode switching is often based on predefined rules, lacking a deep understanding of user behavior and communication content. Furthermore, in multi-device collaborative communication scenarios, existing systems struggle to achieve efficient distributed decision-making and resource scheduling. Existing systems also need improvement in handling heterogeneous data, optimizing communication topologies, and protecting user privacy. These shortcomings limit the communication effectiveness and user experience of walkie-talkies in complex environments. Summary of the Invention

[0004] The present application provides a method, apparatus, device and storage medium for switching the communication mode of a walkie-talkie, which are used to improve the efficiency of switching the communication mode of the walkie-talkie.

[0005] In a first aspect, the present application provides a method for switching a communication mode of a walkie-talkie, the method comprising: dynamically fusing heterogeneous data collected by a multimodal sensor to obtain a high-dimensional feature vector;

[0006] Inputting the high-dimensional feature vector into a preset topological communication model for plasticity optimization to obtain an initial communication mode;

[0007] Performing a progressive switching process on the initial communication mode to obtain a target communication state;

[0008] Perform scenario parameter analysis on the acquired user behavior patterns and user communication content to obtain target scenario parameters;

[0009] Extracting adjustment parameters from the target communication state according to the target scenario parameters to obtain an adjustment parameter set;

[0010] A distributed communication switching strategy is constructed for a plurality of intercoms according to the adjustment parameter set to obtain a target communication switching strategy, and the plurality of intercoms are controlled according to the target communication switching strategy.

[0011] In combination with the first aspect, in a first implementation of the first aspect of the present application, dynamically fusing heterogeneous data collected by a multimodal sensor to obtain a high-dimensional feature vector includes:

[0012] Preprocessing the heterogeneous data collected by the multimodal sensor to obtain a standardized data set, and performing data mapping on the standardized data set to obtain initial feature mapping data;

[0013] Adaptively weighting the initial feature map data to obtain a weighted feature matrix, and constructing a tensor decomposition rank based on the weighted feature matrix to obtain low-rank representation data;

[0014] Performing a nonlinear transformation on the low-rank representation data to obtain a nonlinear feature space, and performing deep convolution feature extraction based on the nonlinear feature space to obtain a multi-scale feature map;

[0015] Performing an attention mechanism analysis on the multi-scale feature map to obtain a target feature region, and performing temporal correlation modeling based on the target feature region to obtain a temporal correlation representation;

[0016] Topological feature processing is performed on the temporal correlation representation to obtain topological structure features, and high-dimensional space mapping is performed on the topological structure features to obtain the high-dimensional feature vector.

[0017] In combination with the first aspect, in a second implementation of the first aspect of the present application, inputting the high-dimensional feature vector into a preset topology communication model for plasticity optimization to obtain an initial communication mode includes:

[0018] Performing dimensionality reduction processing on the high-dimensional feature vector to obtain a compressed feature set, and constructing a graph neural network based on the compressed feature set to obtain an initial topological structure;

[0019] Calculating edge weights on the initial topological structure to obtain a weighted adjacency matrix, and performing spectral clustering analysis based on the weighted adjacency matrix to obtain subnetwork partitioning data;

[0020] Performing dynamic routing construction on the subnetwork partitioning data to obtain an initial routing table, and performing state space analysis based on the initial routing table to obtain a state-action space;

[0021] Performing policy gradient optimization on the state-action space to obtain candidate communication strategies, and reconstructing the topology structure according to the candidate communication strategies to obtain optimized topology structure data;

[0022] Performing plasticity analysis on the optimized topology structure data to obtain a plasticity index set, and adaptively adjusting communication parameters based on the plasticity index set to obtain the initial communication mode.

[0023] In combination with the first aspect, in a third implementation of the first aspect of the present application, the step of performing a progressive switching process on the initial communication mode to obtain a target communication state includes:

[0024] Performing parameter analysis on the initial communication mode to obtain an initial parameter set, and performing Markov decision analysis on the initial parameter set to obtain a state transition matrix;

[0025] Performing value iteration calculation on the state transfer matrix to obtain an optimal value function, and constructing a Bellman equation based on the optimal value function to obtain a dynamic programming strategy;

[0026] Performing gradient descent optimization on the dynamic programming strategy to obtain a local optimal solution, and constructing a genetic algorithm population based on the local optimal solution to obtain a candidate solution set;

[0027] Performing a crossover mutation operation on the candidate solution set to obtain a mutation solution space, and performing a global analysis on the mutation solution space using a preset simulated annealing algorithm to obtain a global search strategy;

[0028] Particle swarm optimization is performed on the global search strategy to obtain a final switching sequence, and a progressive switching analysis is performed on the communication parameters according to the final switching sequence to obtain the target communication state.

[0029] In combination with the first aspect, in a fourth implementation of the first aspect of the present application, performing scenario parameter analysis on the acquired user behavior pattern and user communication content to obtain target scenario parameters includes:

[0030] Performing time series decomposition on the acquired user behavior pattern to obtain a behavior sequence set, and extracting behavior features from the behavior sequence set to obtain a user behavior feature vector;

[0031] Performing semantic analysis on user communication content to obtain a semantic feature matrix, and performing topic distribution analysis based on the semantic feature matrix to obtain a latent semantic distribution;

[0032] Performing tensor fusion on the user behavior feature vector and the latent semantic distribution to obtain a multimodal representation, and performing high-order feature extraction on the multimodal representation to obtain a high-order feature set;

[0033] Performing non-negative matrix decomposition on the high-order feature set to obtain basic semantic units, and performing hierarchical clustering analysis based on the basic semantic units to obtain a scene category tree;

[0034] Association rule mining is performed on the scene category tree to obtain a scene association network, and parameterized modeling is performed based on the scene association network to obtain the target scene parameters.

[0035] In combination with the first aspect, in a fifth implementation manner of the first aspect of the present application, extracting adjustment parameters from the target communication state according to the target scenario parameters to obtain an adjustment parameter set includes:

[0036] Performing principal component analysis on the target scene parameters to obtain a reduced-dimensional feature space, and constructing a regression function analysis based on support vectors on the reduced-dimensional feature space to obtain an initial regression function;

[0037] Performing kernel function mapping on the initial regression function to obtain a high-dimensional feature map, and performing cross-validation based on the high-dimensional feature map to obtain an optimal hyperparameter set;

[0038] Performing Bayesian optimization on the optimal hyperparameter set to obtain a posterior distribution of the parameters, and constructing a Gaussian process regression model based on the posterior distribution of the parameters to obtain a probability prediction function;

[0039] Sampling and sensitivity analysis are performed on the probability prediction function to obtain a target parameter subset, and multi-objective optimization and parameter space mapping are performed on the target parameter subset to obtain the adjustment parameter set.

[0040] In combination with the first aspect, in a sixth implementation manner of the first aspect of the present application, constructing a distributed communication switching strategy for multiple walkie-talkies according to the adjustment parameter set, obtaining a target communication switching strategy, and controlling the multiple walkie-talkies according to the target communication switching strategy, including:

[0041] Performing cluster analysis on the adjustment parameter set to obtain parameter subclusters, and constructing a hierarchical graph structure based on the parameter subclusters to obtain a multi-layer communication network topology;

[0042] Performing edge computing task allocation on the multi-layer communication network topology to obtain an initial task allocation scheme, and performing protocol matching according to the initial task allocation scheme to obtain a global consistency protocol;

[0043] Performing game theory analysis on the global consensus protocol to obtain a Nash equilibrium strategy set, and constructing a multi-agent reinforcement learning model based on the Nash equilibrium strategy set to obtain a collaborative decision distribution data set;

[0044] Performing federated learning training on the collaborative decision-making distributed data set to obtain decentralized model parameters, and performing differential privacy processing based on the decentralized model parameters to obtain a privacy-preserving communication strategy;

[0045] The privacy-preserving communication strategy is robustly optimized to obtain an interference-resistant switching scheme, and distributed deployment and execution are performed according to the interference-resistant switching scheme to obtain the target communication switching strategy.

[0046] In a second aspect, the present application provides a communication mode switching device for a walkie-talkie, the communication mode switching device for the walkie-talkie comprising:

[0047] The fusion module is used to dynamically fuse heterogeneous data collected by multimodal sensors to obtain high-dimensional feature vectors;

[0048] An optimization module, configured to input the high-dimensional feature vector into a preset topology communication model for plasticity optimization to obtain an initial communication mode;

[0049] A switching module, configured to perform a gradual switching process on the initial communication mode to obtain a target communication state;

[0050] An analysis module is used to perform scenario parameter analysis on the acquired user behavior patterns and user communication content to obtain target scenario parameters;

[0051] An extraction module, configured to extract adjustment parameters of the target communication state according to the target scenario parameters to obtain an adjustment parameter set;

[0052] The control module is used to construct a distributed communication switching strategy for multiple intercoms according to the adjustment parameter set, obtain a target communication switching strategy, and control the multiple intercoms according to the target communication switching strategy.

[0053] The third aspect of the present application provides a communication mode switching device for a walkie-talkie, comprising: a memory and at least one processor, wherein the memory stores instructions; the at least one processor calls the instructions in the memory so that the communication mode switching device of the walkie-talkie executes the above-mentioned communication mode switching method of the walkie-talkie.

[0054] A fourth aspect of the present application provides a computer-readable storage medium, wherein the computer-readable storage medium stores instructions, which, when executed on a computer, enable the computer to execute the above-mentioned method for switching the communication mode of the intercom.

[0055] The technical solution provided in this application dynamically fuses heterogeneous data collected by multimodal sensors to obtain high-dimensional feature vectors, achieving comprehensive perception and accurate characterization of complex communication environments, greatly improving the system's adaptability to environmental changes. Secondly, the high-dimensional feature vectors are input into a preset topological communication model for plasticity optimization to obtain the initial communication mode. This method, based on graph neural networks and reinforcement learning, can adaptively construct and optimize the communication topology, significantly improving the flexibility and robustness of the network. Furthermore, the initial communication mode is gradually switched to the target communication state. By combining Markov decision processes, dynamic programming, and evolutionary algorithms, a smooth transition and global optimization of communication parameters are achieved, effectively avoiding the communication interruptions and performance fluctuations encountered in traditional methods. Furthermore, this method performs scenario parameter analysis on the acquired user behavior patterns and user communication content to obtain target scenario parameters. This deep semantic understanding and multimodal fusion method can accurately capture user needs and communication scenario characteristics, providing a precise decision-making basis for subsequent communication optimization. Based on the target scenario parameters, the target communication state is adjusted and extracted to obtain an adjustment parameter set. This step, through the combination of advanced machine learning algorithms such as principal component analysis, support vector regression, and Gaussian processes, achieves refined adjustment and multi-objective optimization of communication parameters, greatly improving the system's adaptability and communication quality. A distributed communication switching strategy is constructed for multiple walkie-talkies based on the adjustment parameter set to obtain the target communication switching strategy, and then multiple walkie-talkies are controlled according to the target communication switching strategy. This distributed decision-making framework based on multi-agent reinforcement learning and federated learning not only achieves collaborative optimization of multiple devices, but also protects the user's communication privacy through differential privacy technology. By integrating multiple advanced artificial intelligence and communication technologies, the walkie-talkie communication system is made intelligent, personalized, and collaborative, significantly improving the system's communication efficiency, reliability, and user experience in complex environments. BRIEF DESCRIPTION OF THE DRAWINGS

[0056] In order to more clearly illustrate the technical solutions of the embodiments of the present invention, the following briefly introduces the drawings required for use in the description of the embodiments. Obviously, the drawings described below are some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without creative work.

[0057] Figure 1 A schematic diagram of an embodiment of a method for switching a communication mode of a walkie-talkie in an embodiment of the present application;

[0058] Figure 2 This is a schematic diagram of an embodiment of a communication mode switching device for a walkie-talkie in an embodiment of the present application. DETAILED DESCRIPTION

[0059] The embodiments of the present application provide a method, apparatus, device and storage medium for switching the communication mode of a walkie-talkie. The terms "first", "second", "third", "fourth", etc. (if any) in the specification and claims of this application and the above-mentioned drawings are used to distinguish similar objects and are not necessarily used to describe a specific order or precedence. It should be understood that the data used in this way can be interchangeable where appropriate, so that the embodiments described herein can be implemented in an order other than that illustrated or described herein. In addition, the terms "including" or "having" and any variations thereof are intended to cover non-exclusive inclusions, for example, a process, method, system, product or device comprising a series of steps or units is not necessarily limited to those steps or units clearly listed, but may include other steps or units that are not clearly listed or inherent to these processes, methods, products or devices.

[0060] For ease of understanding, the specific process of the embodiment of the present application is described below. Figure 1 In one embodiment of the present application, a method for switching the communication mode of a walkie-talkie includes:

[0061] Step S101: dynamically fuse heterogeneous data collected by multimodal sensors to obtain a high-dimensional feature vector;

[0062] It is understandable that the execution subject of the present application can be a communication mode switching device of a walkie-talkie, or a terminal or a server, which is not limited here. The embodiment of the present application is described by taking the server as the execution subject as an example.

[0063] Specifically, heterogeneous data collected by multimodal sensors is preprocessed to obtain a standardized dataset. A multilayer perceptron network is constructed based on the standardized dataset and data mapping is performed to obtain initial feature maps. The learning capabilities of the multilayer perceptron network are then used to extract preliminary features from the data. Adaptive weight adjustment is then performed on the initial feature maps to obtain a weighted feature matrix. Adaptive weight adjustment automatically adjusts the weights of each feature based on the data's characteristics, making feature extraction more accurate and efficient. A tensor decomposition rank is constructed based on the weighted feature matrix to obtain a low-rank representation of the data. Tensor decomposition reduces the data's dimensionality, extracting key features and reducing redundant information. A nonlinear transformation is performed on the low-rank representation to obtain a nonlinear feature space. Nonlinear transformations can capture complex underlying relationships in the data, resulting in richer and more diverse feature representations. A deep convolutional neural network is constructed based on the nonlinear feature space to obtain multi-scale feature maps. Through convolution operations, the deep convolutional neural network extracts local features from the data at different scales, forming multi-scale feature maps. An attention mechanism is applied to the multi-scale feature maps to obtain target feature regions. The attention mechanism focuses on the most important parts of the data, improving the accuracy and efficiency of feature extraction. A recursive neural network model is performed based on the target feature region to obtain a temporal correlation representation. Recursive neural networks excel at processing time series data and, through modeling, can capture temporal correlations within the data. The temporal correlation representation is then processed using a graph convolutional network to obtain topological features. By processing the topological structure of the data, a graph convolutional network can extract information about the relationships between nodes in the data, forming topological features. Based on the topological features, a high-dimensional spatial mapping is performed to obtain a high-dimensional feature vector. This high-dimensional spatial mapping elevates low-dimensional features to a higher-dimensional space, making the feature representation more comprehensive and detailed.

[0064] Step S102: inputting the high-dimensional feature vector into a preset topology communication model for plasticity optimization to obtain an initial communication mode;

[0065] Specifically, dimensionality reduction is performed on high-dimensional feature vectors to obtain a compressed feature set. This reduction in feature dimensionality reduces computational complexity while retaining the key information of the data. A graph neural network is constructed based on the compressed feature set to generate an initial topology. Graph neural networks effectively capture the relationships between nodes and edges in the data, resulting in a more representative initial topology. Edge weights are calculated on the initial topology to obtain a weighted adjacency matrix. The weighted adjacency matrix is a crucial component of graph structures. By calculating edge weights between nodes, it more accurately describes the relationships between nodes. After obtaining the weighted adjacency matrix, spectral clustering analysis is performed to obtain subnetwork partitioning data. Spectral clustering analysis, a clustering method based on graph theory, partitions the network into multiple subnetworks, each with stronger connections between nodes. Dynamic routing is then constructed on the subnetwork partitioning data to obtain an initial routing table. Dynamic routing construction determines the data transmission path based on the network topology and the relationships between nodes, thereby generating the initial routing table. State space analysis is performed on the initial routing table to map different states in the network to corresponding actions, forming a state-action space model. Policy gradient optimization is performed on the state-action space to obtain candidate communication strategies. Policy gradient optimization is a reinforcement learning method that finds the optimal communication strategy by optimizing the policy gradient. The topology is reconstructed based on the candidate communication strategy to obtain optimized topology data. The network structure is adjusted and optimized based on the candidate strategy to obtain a more optimized topology. Plasticity analysis is performed on the optimized topology data to obtain a set of plasticity indicators. Plasticity analysis evaluates the adjustability and adaptability of the network structure. This analysis yields a set of plasticity indicators that reflect the adaptability of the network structure under different conditions. Communication parameters are adaptively adjusted based on the set of plasticity indicators to obtain the initial communication mode. Adaptive adjustment adjusts and optimizes various network parameters based on the plasticity indicators to form an optimal initial communication mode, ensuring that the network can operate efficiently and stably in different environments and conditions.

[0066] Step S103: Perform a gradual switching process on the initial communication mode to obtain a target communication state;

[0067] Specifically, the initial communication method is parameterized to obtain an initial parameter set. Markov decision analysis is performed on this initial parameter set to obtain a state transition matrix. Markov decision analysis is a probability-based analysis method that determines the transition probabilities between states, thereby forming a state transition matrix. Value iteration is performed on the state transition matrix to obtain the optimal value function. Value iteration is a dynamic programming method that, through repeated iterations, finds the optimal value for each state, forming the optimal value function. Based on the optimal value function, the Bellman equation is constructed, and a dynamic programming strategy is constructed based on the Bellman equation. The Bellman equation describes the optimal strategy, and solving it yields a dynamic programming strategy. Gradient descent optimization is performed on the dynamic programming strategy to obtain a local optimal solution. Gradient descent optimization is a commonly used optimization method that continuously adjusts parameters to gradually reduce the value of the objective function, thereby finding a local optimal solution. Based on the local optimal solution, a genetic algorithm population is constructed to obtain a set of candidate solutions. The genetic algorithm is an optimization algorithm based on natural selection and genetic mechanisms. By constructing a population, multiple candidate solution sets are obtained. Crossover and mutation operations are performed on the candidate solution sets. Through crossover and mutation, new solutions are generated, forming a mutant solution space. A global analysis of the mutant solution space is performed using a preset simulated annealing algorithm to obtain a global search strategy. The simulated annealing algorithm is an optimization algorithm based on the physical annealing process. By controlling temperature parameters, a global search is performed to obtain the global optimal solution. Particle swarm optimization is performed on the global search strategy to obtain the final switching sequence. Particle swarm optimization is an optimization algorithm based on swarm intelligence. By simulating the flight of particles in the search space, an optimal switching sequence is obtained. Based on the final switching sequence, a progressive switching analysis is performed on the communication parameters to obtain the target communication state. Progressive switching analysis is the process of gradually adjusting and optimizing the communication parameters. This method allows for a smooth transition to the target communication state, ensuring a smooth and efficient communication mode switching process for the walkie-talkie.

[0068] Step S104: Analyze the acquired user behavior pattern and user communication content using scenario parameters to obtain target scenario parameters.

[0069] Specifically, the acquired user behavior patterns are subjected to temporal decomposition, breaking down user behavior into a series of behavioral events in chronological order to form a behavior sequence set. Behavioral features are extracted from the behavior sequence set to obtain a user behavior feature vector. By analyzing each event in the behavior sequence, a vector representation reflecting the user behavior characteristics is extracted. Semantic analysis is performed on the user's communication content to obtain a semantic feature matrix. Natural language processing techniques are used to deeply analyze the user's communication content, extracting semantic features and forming a matrix representation. Topic distribution analysis is performed based on the semantic feature matrix. Using statistical methods, the topic distribution in the communication content is analyzed to extract latent semantic topics and obtain a latent semantic distribution. Tensor fusion is performed on the user behavior feature vector and the latent semantic distribution to obtain a multimodal representation. Data from multiple modalities is integrated to form a unified representation that better reflects user behavior and communication content. High-order features are extracted from the multimodal representation to obtain a high-order feature set. Non-negative matrix factorization is performed on the high-order feature set. Non-negative matrix factorization is a matrix decomposition method that decomposes the high-order feature set into several non-negative matrices to extract the basic semantic units. Hierarchical clustering analysis is performed based on basic semantic units, clustering them hierarchically to generate a scenario category tree, which reflects the hierarchical relationships between different scenarios. Association rule mining is performed on the scenario category tree, analyzing the relationships between nodes in the tree to extract association rules and generate a scenario association network. Parameterized modeling is performed based on the scenario association network, modeling each node and edge in the network to form a parameterized representation and obtain the target scenario parameters. These parameters reflect the communication characteristics and user behavior in different scenarios.

[0070] Step S105: extract adjustment parameters for the target communication state according to the target scenario parameters to obtain an adjustment parameter set;

[0071] Specifically, principal component analysis is performed on the target scene parameters to obtain a reduced-dimensional feature space. By converting high-dimensional data into a few principal components in a low-dimensional space, the data's key information is retained while reducing noise and redundancy. Support vector regression is then performed on the reduced-dimensional feature space to obtain an initial regression function. Support vector regression is a regression method that finds relationships between data in a high-dimensional space, forming an initial regression function. Kernel function mapping is then applied to the initial regression function, mapping the data from the low-dimensional space to a high-dimensional space to better capture nonlinear relationships in the data, resulting in a high-dimensional feature map. Cross-validation is performed based on the high-dimensional feature map. Cross-validation is a model validation method that divides the data into training and validation sets, repeatedly training and validating the model to find the optimal hyperparameters and obtain the optimal hyperparameter set. Bayesian optimization is then performed on this optimal hyperparameter set. Bayesian optimization is an optimization method based on Bayesian statistics that optimizes the posterior distribution of hyperparameters to find the optimal parameter combination, thereby obtaining the posterior distribution of the parameters. A Gaussian process regression model is constructed based on the posterior distribution of the parameters. Gaussian process regression is a nonparametric Bayesian regression method that constructs a Gaussian process to perform probabilistic predictions on data, generating a probabilistic prediction function. Sampling and sensitivity analysis are performed on the probabilistic prediction function to obtain a target parameter subset. Sampling and sensitivity analysis involve sampling the probabilistic prediction function and analyzing the impact of each parameter on the prediction results, thereby identifying the most important parameters for the model and forming a target parameter subset. Multi-objective optimization and parameter space mapping are then performed on the target parameter subset to obtain an adjusted parameter set. Multi-objective optimization simultaneously optimizes multiple objective functions to find the optimal parameter combination, resulting in the adjusted parameter set. Parameter space mapping maps the optimized parameters to the actual communication system to facilitate adjustment and optimization of communication parameters.

[0072] Step S106: constructing a distributed communication switching strategy for the multiple intercoms according to the adjustment parameter set to obtain a target communication switching strategy, and controlling the multiple intercoms according to the target communication switching strategy.

[0073] Specifically, cluster analysis is performed on the adjustment parameter set. By grouping similar parameters within the adjustment parameter set, several parameter subclusters are formed. A hierarchical graph structure is constructed based on the parameter subclusters. A hierarchical graph structure is a network representation method that, by dividing the network into layers, can better describe the relationships between nodes at different levels, resulting in a multi-layer communication network topology. Edge computing tasks are then allocated within the multi-layer communication network topology. Computing tasks are rationally assigned to edge nodes based on the capabilities and load of each node in the network topology, resulting in an initial task allocation scheme. Protocol matching is performed based on the initial task allocation scheme. By selecting an appropriate communication protocol, each node in the network can work in a coordinated and consistent manner, resulting in a global consensus protocol. Game theory analysis is performed on the global consensus protocol to obtain a Nash equilibrium strategy set. Game theory analysis is a mathematical analysis method that analyzes the strategy choices of each node to find a strategy combination that satisfies all nodes, forming a Nash equilibrium strategy set. Based on the Nash equilibrium strategy set, a multi-agent reinforcement learning model is constructed. The multi-agent reinforcement learning model is a learning method based on multi-agent collaboration. This method simulates the decision-making process of each node and obtains a distributed collaborative decision-making dataset. Federated learning training is then performed on the distributed collaborative decision-making dataset. Federated learning is a distributed machine learning method that trains models on multiple nodes to obtain a decentralized set of model parameters, thus avoiding the data privacy issues associated with centralized training. Differential privacy processing is performed on the decentralized model parameters. Differential privacy processing is a method for protecting data privacy that effectively protects user privacy by adding noise to the data, resulting in a privacy-preserving communication strategy. Robustness optimization is performed on the privacy-preserving communication strategy to ensure efficient and stable performance in the face of various interferences, resulting in an interference-resistant switching solution. The interference-resistant switching solution is distributedly deployed and executed to obtain the target communication switching strategy. The resulting communication switching strategy is distributed to each walkie-talkie and executed in a real-world environment, ensuring that multiple walkie-talkies can communicate efficiently according to the target communication switching strategy.

[0074] In the embodiment of the present application, by dynamically fusing heterogeneous data collected by multimodal sensors, a high-dimensional feature vector is obtained, which achieves comprehensive perception and accurate characterization of complex communication environments, greatly improving the system's adaptability to environmental changes. Secondly, the high-dimensional feature vector is input into a preset topological communication model for plasticity optimization to obtain the initial communication mode. This method based on graph neural networks and reinforcement learning can adaptively construct and optimize the communication topology, significantly improving the flexibility and robustness of the network. Furthermore, the initial communication mode is gradually switched to obtain the target communication state. By combining Markov decision processes, dynamic programming and evolutionary algorithms, a smooth transition and global optimization of communication parameters are achieved, effectively avoiding communication interruptions and performance fluctuations in traditional methods. In addition, this method performs scenario parameter analysis on the acquired user behavior patterns and user communication content to obtain target scenario parameters. This method of deep semantic understanding and multimodal fusion can accurately capture user needs and communication scenario characteristics, providing an accurate decision-making basis for subsequent communication optimization. Based on the target scenario parameters, the target communication state is adjusted and extracted to obtain an adjustment parameter set. This step, through the combination of advanced machine learning algorithms such as principal component analysis, support vector regression, and Gaussian processes, achieves refined adjustment and multi-objective optimization of communication parameters, greatly improving the system's adaptability and communication quality. A distributed communication switching strategy is constructed for multiple walkie-talkies based on the adjustment parameter set to obtain the target communication switching strategy, and then multiple walkie-talkies are controlled according to the target communication switching strategy. This distributed decision-making framework based on multi-agent reinforcement learning and federated learning not only achieves collaborative optimization of multiple devices, but also protects the user's communication privacy through differential privacy technology. By integrating multiple advanced artificial intelligence and communication technologies, the walkie-talkie communication system is made intelligent, personalized, and collaborative, significantly improving the system's communication efficiency, reliability, and user experience in complex environments.

[0075] In a specific embodiment, the process of executing step S101 may specifically include the following steps:

[0076] Preprocessing the heterogeneous data collected by the multimodal sensor to obtain a standardized data set, and performing data mapping on the standardized data set to obtain initial feature mapping data;

[0077] Adaptively weighting the initial feature map data to obtain a weighted feature matrix, and constructing a tensor decomposition rank based on the weighted feature matrix to obtain low-rank representation data;

[0078] Performing a nonlinear transformation on the low-rank representation data to obtain a nonlinear feature space, and performing deep convolution feature extraction based on the nonlinear feature space to obtain a multi-scale feature map;

[0079] Performing an attention mechanism analysis on the multi-scale feature map to obtain a target feature region, and performing temporal correlation modeling based on the target feature region to obtain a temporal correlation representation;

[0080] Topological feature processing is performed on the temporal correlation representation to obtain topological structure features, and high-dimensional space mapping is performed on the topological structure features to obtain the high-dimensional feature vector.

[0081] Specifically, the heterogeneous data collected by multimodal sensors is normalized, converting the data collected by different sensors to the same scale range to eliminate data differences caused by different dimensions. A multilayer perceptron network is constructed based on the standardized data set, and data mapping is performed. The multilayer perceptron is a feedforward neural network consisting of an input layer, a hidden layer, and an output layer. By training the multilayer perceptron network, the input data is mapped to a new feature space, obtaining the initial feature map data. Assuming the input data is X, the mapping process through the MLP network can be expressed as:

[0082] H=σ(WX+b);

[0083] Where H is the initial feature map data, W is the weight matrix, b is the bias vector, and σ is the activation function (such as ReLU or Sigmoid). Adaptive weight adjustment is performed on the initial feature map data to obtain a weighted feature matrix. Adaptive weight adjustment learns the importance of each feature and assigns different weights to it, thereby enhancing the target features and suppressing irrelevant features. Assuming that the initial feature map data is H, the process of adaptive weight adjustment can be expressed as:

[0084] H′=α⊙H;

[0085] Among them, H′ is the weighted feature matrix, α is the adaptive weight vector, Represents the bit-by-bit multiplication operation of elements. The tensor decomposition rank is constructed based on the weighted feature matrix to obtain a low-rank representation of the data. Tensor decomposition is a method of representing high-dimensional data in a low-dimensional form, which can effectively reduce data redundancy and noise. Assuming that the weighted feature matrix is H′, its tensor decomposition can be expressed as:

[0086] H′≈U·S·V T ;

[0087] Where U and V are orthogonal matrices, S is a diagonal matrix, and the low-rank representation of the data is U·S. A nonlinear transformation is performed on the low-rank representation to obtain a nonlinear feature space. Nonlinear transformations use nonlinear functions to map low-rank data into a higher-dimensional feature space, capturing more complex relationships within the data. Common nonlinear transformations include kernel methods, such as the radial basis function (RBF) kernel. Based on the nonlinear feature space, a deep convolutional neural network is constructed to obtain multi-scale feature maps. Through multiple layers of convolution and pooling operations, deep convolutional neural networks can extract local features from the data and process them at different scales. An attention mechanism is applied to the multi-scale feature maps to obtain target feature regions. The attention mechanism assigns different weights to different regions, highlighting important feature regions and suppressing less important ones. A recurrent neural network is then modeled based on the target feature regions to obtain a temporal correlation representation. Recurrent neural networks, through their recurrent structure, can handle temporal correlation in sequential data. The temporal correlation representation is then processed using a graph convolutional network to obtain topological structural features. Through convolution operations, graph convolutional networks effectively capture the relationships between nodes and edges in graph-structured data. Based on the topological structure features, high-dimensional space mapping is performed to obtain high-dimensional feature vectors. By lifting the topological structure features to high-dimensional space, the representation ability of the features is enhanced.

[0088] In a specific embodiment, the process of executing step S102 may specifically include the following steps:

[0089] Performing dimensionality reduction processing on the high-dimensional feature vector to obtain a compressed feature set, and constructing a graph neural network based on the compressed feature set to obtain an initial topological structure;

[0090] Calculating edge weights on the initial topological structure to obtain a weighted adjacency matrix, and performing spectral clustering analysis based on the weighted adjacency matrix to obtain subnetwork partitioning data;

[0091] Performing dynamic routing construction on the subnetwork partitioning data to obtain an initial routing table, and performing state space analysis based on the initial routing table to obtain a state-action space;

[0092] Performing policy gradient optimization on the state-action space to obtain candidate communication strategies, and reconstructing the topology structure according to the candidate communication strategies to obtain optimized topology structure data;

[0093] Performing plasticity analysis on the optimized topology structure data to obtain a plasticity index set, and adaptively adjusting communication parameters based on the plasticity index set to obtain the initial communication mode.

[0094] Specifically, dimensionality reduction techniques, such as principal component analysis or t-SNE, are used to reduce the dimensionality of the feature vector. The purpose of dimensionality reduction is to reduce the dimensionality of the data while retaining the most important information in the data, thereby simplifying subsequent processing steps. A graph neural network is constructed based on the compressed feature set to obtain an initial topological structure. A graph neural network is a type of neural network that can process graph-structured data. By capturing the relationship between nodes and edges, it forms a network that represents the topological structure of the data. Assuming that the compressed feature set is X′, the initial topological structure can be obtained through the graph convolution layer of the graph neural network:

[0095] H=σ(D -1 / 2 AD -1 / 2 X′W);

[0096] Where H is the initial topology representation, A is the adjacency matrix, D is the degree matrix, W is the weight matrix, and σ is the activation function. The edge weight calculation is performed on the initial topology to obtain the weighted adjacency matrix. The purpose of edge weight calculation is to determine the weight of each edge in the graph, so as to more accurately represent the relationship between nodes. Assuming that the initial topology is H, the result of edge weight calculation can be expressed as a weighted adjacency matrix A′, whose element A i ' j Denotes the edge weight between node i and node j:

[0097] A ij ′=f(H i ,H j );

[0098] Among them, H i and H j are the feature representations of node i and node j respectively, and f is the function for calculating edge weights, which can usually be measured using dot product or other similarity metrics. Spectral clustering analysis is performed based on the weighted adjacency matrix to obtain subnetwork partitioning data. Spectral clustering is a clustering method based on graph structure. By analyzing the spectrum (eigenvalue) of the weighted adjacency matrix, the nodes are divided into several subnetworks. Dynamic routing is constructed on the subnetwork partitioning data to obtain an initial routing table. Dynamic routing construction determines the path of data transmission based on the relationship between each node in the subnetwork, thereby forming an initial routing table. The initial routing table can be determined by the shortest path algorithm or other routing algorithms. State space analysis is performed based on the initial routing table, and different states in the network are mapped to corresponding actions to obtain a state-action space model. Assuming that the initial routing table is R, the state-action space can be expressed as:

[0099] SA={(s i ,a i )};

[0100] Among them, s i is the state, ai Is an action, indicating that in state s i Next, perform action a i The mapping relationship is established. Policy gradient optimization is performed on the state-action space to obtain candidate communication strategies. Policy gradient optimization is a reinforcement learning method that finds the optimal communication strategy by optimizing the policy gradient. Assuming that the state-action space is SA, the goal of policy gradient optimization is to find an optimal strategy π:

[0101]

[0102] Where J(π) is the expected reward of the strategy, and Q(s,a) is the state-action value function, representing the reward obtained by performing action a in state s. After obtaining a candidate communication strategy, the topology is reconstructed based on the strategy to obtain optimized topology data. Topology reconstruction involves adjusting and optimizing the network structure based on the candidate communication strategy to obtain a more optimized topology. Optimizing topology data can be achieved by adjusting edge weights and node connectivity. A plasticity analysis is performed on the optimized topology data to evaluate the adjustability and adaptability of the network structure. This analysis yields a set of plasticity indices that reflect the network's adaptability under different conditions. The plasticity indices can be determined by calculating the elasticity coefficients of nodes and edges. Based on the plasticity indices, communication parameters are adaptively adjusted to obtain the initial communication mode. Adaptive adjustment adjusts and optimizes various network parameters based on the plasticity indices to obtain the optimal initial communication mode.

[0103] In a specific embodiment, the process of executing step S103 may specifically include the following steps:

[0104] Performing parameter analysis on the initial communication mode to obtain an initial parameter set, and performing Markov decision analysis on the initial parameter set to obtain a state transition matrix;

[0105] Performing value iteration calculation on the state transfer matrix to obtain an optimal value function, and constructing a Bellman equation based on the optimal value function to obtain a dynamic programming strategy;

[0106] Performing gradient descent optimization on the dynamic programming strategy to obtain a local optimal solution, and constructing a genetic algorithm population based on the local optimal solution to obtain a candidate solution set;

[0107] Performing a crossover mutation operation on the candidate solution set to obtain a mutation solution space, and performing a global analysis on the mutation solution space using a preset simulated annealing algorithm to obtain a global search strategy;

[0108] Particle swarm optimization is performed on the global search strategy to obtain a final switching sequence, and a progressive switching analysis is performed on the communication parameters according to the final switching sequence to obtain the target communication state.

[0109] Specifically, the parameters in the initial communication mode are parsed. These parameters may include communication frequency, bandwidth, signal strength, interference level, etc. Through parsing, the initial parameter set is obtained. Assume that these parameters are recorded as P = {p1, p2, ..., p n}, where each p i Represents a specific communication parameter. Perform Markov decision analysis on the initial parameter set to obtain the state transition matrix. Markov decision analysis is a probability-based analysis method that forms a state transition matrix by analyzing the transition probability of the system in different states. Assume that the state set of the system is S = {s1, s2, ..., s m}, element T of the state transition matrix T ij Indicates that the system changes from state s i Transfer to state s j Probability of:

[0110] T ij =P(s t+1 =s j |s t =s i );

[0111] Perform value iteration on the state transition matrix to obtain the optimal value function. Value iteration is a dynamic programming method that calculates the optimal value of each state through repeated iterations. Assuming V(s) represents the value function of state s, the update formula of value iteration is:

[0112]

[0113] Where a is the action, s′ is the next state, R(s,a,s′) is the reward for taking action a in state s and transitioning to state s′, and γ is the discount factor. The Bellman equation is constructed based on the optimal value function, and a dynamic programming strategy is constructed based on the Bellman equation. The Bellman equation describes the recursive nature of the optimal strategy and can be used to calculate the optimal strategy. The Bellman equation is:

[0114] Q(s,a)=R(s,a)+γ∑ s′ T(s,a,s′)V(s′);

[0115] By solving the Bellman equation, a dynamic programming strategy is obtained. Gradient descent optimization is performed on the dynamic programming strategy to obtain a local optimal solution. Gradient descent is an optimization method that repeatedly adjusts parameters to gradually reduce the value of the objective function, thereby finding a local optimal solution. Assuming that the policy parameter is θ and the objective function is J(θ), the update formula for gradient descent is:

[0116]

[0117] Where α is the learning rate, is the gradient of the objective function. Based on the local optimal solution, a genetic algorithm population is constructed to obtain a candidate solution set. Genetic algorithm is an optimization algorithm based on natural selection and genetic mechanism. By constructing a population, multiple candidate solutions are obtained. Crossover and mutation operations are performed on the candidate solution set, and new solutions are generated through crossover and mutation to obtain a mutant solution space. A global analysis of the mutant solution space is performed using a preset simulated annealing algorithm to obtain a global search strategy. Simulated annealing is an optimization algorithm based on the physical annealing process. By controlling the temperature parameters, a global search is performed to find a global optimal solution. Assume that the current solution is p and the new solution is p new , the probability of accepting the new solution is:

[0118]

[0119] Where E is the objective function value and T is the temperature. Particle swarm optimization is performed on the global search strategy to obtain the final switching sequence. Particle swarm optimization is an optimization algorithm based on swarm intelligence. It finds an optimal switching sequence by simulating the flight of particles in the search space. Assume that the position of the particle is x i , the speed is v i , the update formula is:

[0120]

[0121] Where ω is the inertia weight, c1 and c2 are acceleration constants, r1 and r2 are random numbers, and p i is the optimal position of the particle, and g is the global optimal position. Based on the final switching sequence, a progressive switching analysis of the communication parameters is performed to obtain the target communication state. Progressive switching analysis is the process of gradually adjusting and optimizing the communication parameters. This method allows for a smooth transition to the target communication state, ensuring a smooth and efficient communication mode switching process for the walkie-talkie.

[0122] In a specific embodiment, the process of executing step S104 may specifically include the following steps:

[0123] Performing time series decomposition on the acquired user behavior pattern to obtain a behavior sequence set, and extracting behavior features from the behavior sequence set to obtain a user behavior feature vector;

[0124] Performing semantic analysis on user communication content to obtain a semantic feature matrix, and performing topic distribution analysis based on the semantic feature matrix to obtain a latent semantic distribution;

[0125] Performing tensor fusion on the user behavior feature vector and the latent semantic distribution to obtain a multimodal representation, and performing high-order feature extraction on the multimodal representation to obtain a high-order feature set;

[0126] Performing non-negative matrix decomposition on the high-order feature set to obtain basic semantic units, and performing hierarchical clustering analysis based on the basic semantic units to obtain a scene category tree;

[0127] Association rule mining is performed on the scene category tree to obtain a scene association network, and parameterized modeling is performed based on the scene association network to obtain the target scene parameters.

[0128] Specifically, user behavior data is decomposed chronologically to form continuous behavior sequences. This behavior data may include user operation logs, access records, and device usage. Time series decomposition transforms complex behavior data into multiple chronologically ordered behavioral events, better capturing the dynamics of user behavior. Behavioral features are extracted from the behavior sequence set to generate a user behavior feature vector. By analyzing the target behaviors and patterns within each behavior sequence, a vector representation that reflects the user's behavior characteristics is extracted. For example, features such as the user's operation frequency, operation type distribution, and usage duration within a specific time period are extracted to form a high-dimensional user behavior feature vector. Semantic analysis is performed on user communication content to generate a semantic feature matrix. Natural language processing techniques are used to deeply analyze user communication content and extract semantic information. Using methods such as word embedding models and topic models, user communication content is converted into a semantic feature matrix, where each row represents a semantic feature of the communication content. Topic distribution analysis is performed on the semantic feature matrix to generate a latent semantic distribution. Statistical methods are used to analyze the semantic feature matrix and extract the topic distribution within it. For example, using the Latent Dirichlet Allocation model, each communication content is represented as a distribution over several topics, resulting in a latent semantic distribution. Tensor fusion is performed on the user behavior feature vector and the latent semantic distribution to obtain a multimodal representation. High-order features are extracted from the multimodal representation to obtain a high-order feature set. For example, a deep neural network is used to process the multimodal representation to extract a high-order feature set that can capture complex relationships and patterns in the data. Non-negative matrix factorization is performed on the high-order feature set to obtain basic semantic units. Non-negative matrix factorization is a matrix factorization method that decomposes the high-order feature set into several non-negative matrices to extract basic semantic units. These basic semantic units reflect the essential semantic information in the data. Hierarchical clustering analysis is performed on the basic semantic units to obtain a scene category tree. Hierarchical clustering analysis clusters basic semantic units according to a hierarchical structure to form a scene category tree. For example, behaviors with similar semantic units are clustered together to form a hierarchical scene category tree that reflects the hierarchical relationships between different scenes. Association rule mining is performed on the scene category tree to obtain a scene association network. Association rule mining analyzes the relationships between nodes in a scenario category tree, extracts association rules, and forms a scenario association network. For example, it can be used to identify the high frequency of certain behaviors in certain scenarios, thereby extracting valuable association rules. Parameterized modeling is performed based on the scenario association network. By modeling each node and edge in the scenario association network, a parameterized representation is formed, and the target scenario parameters are obtained. For example, a model can be built to parameterize the associations between different scenarios to predict the likelihood and patterns of user behavior in a specific scenario.

[0129] Furthermore, the acquired user behavior patterns are decomposed into time series. This step aims to convert complex user behavior data into a series of behavioral events in a time sequence. The specific implementation method is as follows: Data collection: Collect user behavior data through various sensors built into the walkie-talkie (such as GPS, accelerometer, microphone, etc.). This data may include changes in the user's location, device posture, voice commands, etc. Time window division: Split the continuous behavior data into fixed time windows (such as 5 seconds, 10 seconds, etc.), and the data within each time window is regarded as a basic behavior unit. Behavior encoding: Encode the behavior within each time window. Methods such as one-hot encoding or word embedding can be used to convert the behavior into a vector representation. Sequence construction: Arrange the encoded behavior units in time sequence to form a behavior sequence set. Behavior feature extraction is performed on the behavior sequence set to obtain a user behavior feature vector. The purpose of this step is to extract features from the behavior sequence that can represent the user behavior pattern. The specific implementation method is as follows: Statistical feature extraction: Calculate the basic statistical features of the behavior sequence, such as the frequency, duration, and transition probability of each behavior. Temporal Pattern Mining: Uses sequential pattern mining algorithms (such as PrefixSpan and SPADE) to discover frequently occurring behavioral patterns. Recurrent Neural Network Encoding: Uses recurrent neural network models such as LSTM or GRU to encode behavioral sequences and capture long-term dependencies between behaviors. Attention Mechanism: Introduces an attention mechanism to identify key behaviors and important moments in a sequence. Feature Fusion: Fuses the various features extracted above to generate a final user behavior feature vector. Performs semantic analysis on user communication content to generate a semantic feature matrix. This step aims to understand and represent the semantic information of user communication content. The specific implementation methods are as follows: Text Preprocessing: Performs preprocessing operations such as word segmentation, stop word removal, and lemmatization on communication content. Word Vector Representation: Uses pretrained word vector models (such as Word2Vec and GloVe) to convert words into dense vector representations. Sentence Encoding: Uses models such as Convolutional Neural Networks (CNN) or Transformers to encode word vector sequences into sentence vectors. Topic Modeling: Uses topic models such as Latent Dirichlet Allocation (LDA) to extract the topic distribution of communication content. Sentiment Analysis: Use a sentiment analysis model to extract the sentiment characteristics of the communication content. Matrix Construction: Combine the various semantic features extracted above into a semantic feature matrix. Perform topic distribution analysis based on the semantic feature matrix to obtain the latent semantic distribution. The purpose of this step is to deeply explore the latent topic structure of the communication content. Specific implementation methods are as follows: Non-negative Matrix Factorization (NMF): Perform NMF on the semantic feature matrix to obtain a document-topic matrix and a topic-word matrix. Hierarchical Topic Modeling: Use hierarchical topic models such as hLDA (Hierarchical LDA) to discover the hierarchical structure between topics.Dynamic Topic Model: Use dynamic topic models (such as the Dynamic Topic Model) to capture the evolution of topics over time. Topic Correlation Analysis: Calculate the correlation between topics and construct a topic relationship network. Topic Visualization: Use dimensionality reduction techniques such as t-SNE to visualize topic distributions for intuitive understanding. Perform tensor fusion of user behavior feature vectors and latent semantic distributions to obtain a multimodal representation. This step aims to comprehensively represent information about user behavior and communication content. The specific implementation methods are as follows: Tensor Construction: Construct the user behavior feature vectors and latent semantic distributions into a third-order tensor, where the three dimensions represent user, behavioral, and semantic features, respectively. Tensor Decomposition: Decompose the tensor using methods such as CP decomposition or Tucker decomposition to obtain a low-rank representation. Multimodal Attention: Introduce a multimodal attention mechanism to learn the interaction between behavioral and semantic features. Cross-modal Learning: Use cross-modal learning methods (such as CCA and DCCA) to learn a common representation space for behavioral and semantic features. Fusion Strategy: Design a fusion strategy (such as weighted summation and gating mechanisms) to effectively fuse features from different modalities. Perform parameterized modeling based on the scenario association network to obtain the target scenario parameters. The purpose of this step is to convert scenario association information into specific parameters that can be used to switch communication modes. The specific implementation methods are as follows: Parameter definition: Based on the communication characteristics of the walkie-talkie, define a set of scenario-related parameters, such as communication frequency, bandwidth, transmission power, encoding method, etc. Association rule conversion: Convert the rules in the scenario association network into constraints on parameter values. Fuzzy logic modeling: Use fuzzy logic methods to establish a mapping relationship between scenario characteristics and parameter values. Bayesian network: Construct a Bayesian network model to represent the probabilistic dependency between scenario characteristics, communication requirements and parameter values. Deep neural network: Train a deep neural network model to predict the optimal parameter configuration directly from the scenario characteristics. Parameter optimization: Use evolutionary algorithms or reinforcement learning methods to optimize the parameter configuration in a given scenario. Verification and adjustment: Verify the effectiveness of the parameter model through simulation or actual testing, and make necessary adjustments.

[0130] In a specific embodiment, the process of executing step S105 may specifically include the following steps:

[0131] Performing principal component analysis on the target scene parameters to obtain a reduced-dimensional feature space, and constructing a regression function analysis based on support vectors on the reduced-dimensional feature space to obtain an initial regression function;

[0132] Performing kernel function mapping on the initial regression function to obtain a high-dimensional feature map, and performing cross-validation based on the high-dimensional feature map to obtain an optimal hyperparameter set;

[0133] Performing Bayesian optimization on the optimal hyperparameter set to obtain a posterior distribution of the parameters, and constructing a Gaussian process regression model based on the posterior distribution of the parameters to obtain a probability prediction function;

[0134] Sampling and sensitivity analysis are performed on the probability prediction function to obtain a target parameter subset, and multi-objective optimization and parameter space mapping are performed on the target parameter subset to obtain the adjustment parameter set.

[0135] Specifically, high-dimensional data is subjected to dimensionality reduction. Principal component analysis is an effective dimensionality reduction technique that converts high-dimensional data into a low-dimensional feature space by finding the principal components of the data. A support vector-based regression function analysis is performed on the reduced feature space to obtain an initial regression function. Support vector regression is a regression analysis method that predicts the target variable by finding the optimal hyperplane. Assuming that each row of the reduced feature space Z represents a data point and the target variable is y, the regression function of SVR can be expressed as:

[0136]

[0137] Among them, α i is the weight of the support vector, K is the kernel function, b is the bias term, and Z i is the support vector in the training data. To improve the nonlinear fitting ability of the regression model, the initial regression function is mapped to a kernel function to obtain a high-dimensional feature map. Kernel function mapping maps data from the original feature space to a high-dimensional feature space through nonlinear transformation, thereby capturing the complex relationships in the data. Commonly used kernel functions include radial basis function (RBF) kernel, linear kernel, and polynomial kernel. Assuming that the kernel function is K, the high-dimensional feature map can be expressed as:

[0138] φ(Z)=K(Z,Z′);

[0139] Where φ(Z) is the feature representation mapped to a high-dimensional feature space, and Z′ is the support vector. After obtaining the high-dimensional feature mapping, cross-validation is performed based on this mapping to determine the optimal hyperparameter set. Cross-validation is a model validation method that finds the optimal hyperparameters by splitting the dataset into training and validation sets, and repeatedly training and validating the model. The cross-validation process can be described as follows: divide the dataset into k non-overlapping subsets. Select each subset in turn as the validation set, and the remaining subsets as the training set. Train the model and calculate the validation error. Average all validation errors and select the hyperparameter set with the smallest error as the optimal hyperparameter set. Bayesian optimization is performed on the optimal hyperparameter set to obtain the posterior distribution of the parameters. Bayesian optimization is an optimization method based on Bayesian statistics that optimizes hyperparameters by constructing the posterior distribution of the objective function. Assuming the objective function is f(θ), the goal of Bayesian optimization is to find the optimal hyperparameters θ that maximize f(θ). An initial Gaussian process model is constructed, assuming the prior distribution of the objective function is Gaussian. The posterior distribution of the objective function is updated based on the current observations. Select the next sampling point on the posterior distribution, calculate the value of the objective function, and update the model. Based on the parameter posterior distribution, construct a Gaussian process regression model to obtain a probability prediction function. Gaussian process regression is a non-parametric Bayesian regression method that makes probabilistic predictions on data by constructing a Gaussian process. Sampling and sensitivity analysis are performed on the probability prediction function to obtain a subset of target parameters. Sampling and sensitivity analysis are performed by sampling the probability prediction function, analyzing the impact of each parameter on the prediction results, and identifying the most important parameters for the model. Multi-objective optimization and parameter space mapping are performed on the target parameter subset to obtain an adjusted parameter set. Multi-objective optimization is to find an optimal parameter combination by optimizing multiple objective functions simultaneously. Parameter space mapping is to map the optimized parameters to the actual application scenario in order to adjust and optimize the model.

[0140] In a specific embodiment, the process of executing step S106 may specifically include the following steps:

[0141] Performing cluster analysis on the adjustment parameter set to obtain parameter subclusters, and constructing a hierarchical graph structure based on the parameter subclusters to obtain a multi-layer communication network topology;

[0142] Performing edge computing task allocation on the multi-layer communication network topology to obtain an initial task allocation scheme, and performing protocol matching according to the initial task allocation scheme to obtain a global consistency protocol;

[0143] Performing game theory analysis on the global consensus protocol to obtain a Nash equilibrium strategy set, and constructing a multi-agent reinforcement learning model based on the Nash equilibrium strategy set to obtain a collaborative decision distribution data set;

[0144] Performing federated learning training on the collaborative decision-making distributed data set to obtain decentralized model parameters, and performing differential privacy processing based on the decentralized model parameters to obtain a privacy-preserving communication strategy;

[0145] The privacy-preserving communication strategy is robustly optimized to obtain an interference-resistant switching scheme, and distributed deployment and execution are performed according to the interference-resistant switching scheme to obtain the target communication switching strategy.

[0146] Specifically, the adjustment parameter set is analyzed and classified. Cluster analysis is used to group similar parameters into several parameter subclusters, which reflect the intrinsic connections and structure between the parameters. A hierarchical graph structure is constructed based on the parameter subclusters, resulting in a multi-layer communication network topology. This hierarchical graph structure effectively represents the relationships between nodes at different levels, forming a clear communication network topology. Edge computing tasks are allocated within the multi-layer communication network topology to obtain an initial task allocation scheme. Edge computing task allocation rationally distributes computing tasks to edge nodes based on the capabilities and load of each node in the network topology, thereby improving computing efficiency and resource utilization. Protocol matching is performed based on the initial task allocation scheme to obtain a global consensus protocol. Protocol matching selects an appropriate communication protocol to enable all nodes in the network to work in a coordinated and consistent manner, forming a global consensus protocol. Game theory analysis is performed on the global consensus protocol to obtain a Nash equilibrium strategy set. Game theory analysis is a mathematical analysis method that analyzes the strategy choices of each node to find a strategy combination that satisfies all nodes, forming a Nash equilibrium strategy set. Multi-agent reinforcement learning is performed based on the Nash equilibrium strategy set to obtain a collaborative decision-making distribution dataset. Multi-agent reinforcement learning is a learning method based on the collaboration of multiple agents. This method simulates the decision-making process of each node to form a distributed dataset for collaborative decision-making. Federated learning is performed on this distributed dataset to obtain decentralized model parameters. Federated learning is a distributed machine learning method that trains the model on multiple nodes to obtain a decentralized set of model parameters, thus avoiding the data privacy issues associated with centralized training. Differential privacy processing is performed on the decentralized model parameters to obtain a privacy-preserving communication strategy. Differential privacy processing is a method for protecting data privacy that effectively protects user privacy by adding noise to the data, thus forming a privacy-preserving communication strategy. Robustness optimization is performed on the privacy-preserving communication strategy to obtain an interference-resistant switching scheme. Robustness optimization optimizes the communication strategy to ensure that it maintains efficient and stable performance in the face of various interferences, thus forming an interference-resistant switching scheme. Distributed deployment and execution of the interference-resistant switching scheme results in the target communication switching strategy. Distributed deployment and execution distributes the resulting communication switching strategy to each node and executes it in the real environment, ensuring that the network can communicate efficiently according to the target communication switching strategy.

[0147] Furthermore, cluster analysis is performed on the adjusted parameter set to obtain parameter subclusters. The purpose of this step is to group similar parameter configurations to facilitate subsequent strategy construction. The specific implementation methods are as follows: Parameter standardization: Parameters of different scales are standardized to ensure that each parameter has the same weight during the clustering process. Distance metric: An appropriate distance metric, such as Euclidean distance, Manhattan distance, or Mahalanobis distance, is selected to calculate the similarity between parameters. Clustering algorithm selection: Based on the characteristics and distribution of the parameters, an appropriate clustering algorithm, such as K-means, DBSCAN, or hierarchical clustering, is selected. Cluster number determination: The optimal number of clusters is determined using methods such as the silhouette coefficient and the elbow rule. Clustering result evaluation: The quality of the clustering results is evaluated by calculating the internal compactness of the clusters and the separation between clusters. Parameter subcluster extraction: Representative parameters are extracted from each cluster to form parameter subclusters. A hierarchical graph structure is constructed based on the parameter subclusters to obtain a multi-layer communication network topology. This step aims to establish a topological model that reflects the hierarchical structure of the walkie-talkie network. The specific implementation method is as follows: Layer Definition: Define the network's hierarchical structure, such as the core layer, aggregation layer, and access layer, based on factors such as the function, performance, and location of the walkie-talkies. Node Assignment: Assign each walkie-talkie to the corresponding network layer based on the characteristics of its parameter subcluster. Connection Establishment: Establish connections between nodes based on factors such as geographic location, communication range, and signal strength. Weight Assignment: Assign weights to edges in the graph structure based on metrics such as connection quality and bandwidth. Topology Optimization: Optimize the network topology using graph theory algorithms (such as minimum spanning tree and shortest path). Dynamic Update Mechanism: Design a dynamic update mechanism for the topology to adapt to changes in the network environment. Perform edge computing task allocation on the multi-layer communication network topology to obtain an initial task allocation plan. The purpose of this step is to fully utilize the computing resources in the network and improve communication efficiency. The specific implementation method is as follows: Resource Assessment: Assess the computing power, storage capacity, and energy status of each walkie-talkie node. Task Decomposition: Decompose the computing tasks related to communication switching into subtasks that can be executed in parallel. Load Balancing: Design a load balancing algorithm to ensure even distribution of tasks across the network. Task scheduling: Use heuristic algorithms (such as genetic algorithms, ant colony algorithms, etc.) to optimize task scheduling. Data localization: Assign computing tasks to the nodes where the data is located as much as possible to reduce data transmission overhead. Fault-tolerant mechanism: Design a task failure detection and reallocation mechanism to improve system reliability. Perform protocol matching based on the initial task allocation plan to obtain a global consistency protocol. This step aims to ensure that all nodes in the network remain consistent when the communication mode is switched. The specific implementation method is as follows: Protocol library construction: Establish a protocol library containing various communication protocols (such as TDMA, CSMA, FDMA, etc.). Protocol evaluation: Evaluate the applicability of each protocol based on the current network status and task requirements. Protocol selection: Use multi-criteria decision-making methods (such as AHP, TOPSIS, etc.) to select the most suitable protocol.Consensus Algorithm: Design a distributed consensus algorithm (such as Paxos or Raft) to ensure that all nodes reach consensus on the selected protocol. Protocol Parameter Optimization: Optimize key parameters of the selected protocol (such as time slot length and frequency bandwidth). Protocol Switching Mechanism: Design a smooth protocol switching mechanism to minimize communication interruptions during the switching process. Conduct game-theoretic analysis of the global consensus protocol to obtain a set of Nash equilibrium strategies. The goal of this step is to find the optimal communication strategy while considering the interests of each node. The specific implementation methods are as follows: Game Model Construction: Model the communication mode switching problem as a non-cooperative or cooperative game. Utility Function Definition: Define a utility function for each intercom node, taking into account factors such as communication quality, energy consumption, and latency. Policy Space Construction: Construct a policy space for each node based on the available communication parameters and protocols. Equilibrium Solving: Use iterative best response, linear complementarity, or other numerical methods to solve the Nash equilibrium. Equilibrium Analysis: Analyze the uniqueness, stability, and Pareto optimality of the resulting Nash equilibrium. Dynamic Adjustment: Design a dynamic game mechanism to adapt to changes in the network environment. A multi-agent reinforcement learning model is constructed based on the Nash equilibrium strategy set to obtain a collaborative decision distribution data set. This step aims to optimize the communication switching strategy through learning. The specific implementation method is as follows: State space definition: Define the state space including factors such as network topology, communication quality, and task requirements. Action space design: Design an action space including operations such as communication parameter adjustment and protocol switching. Reward function construction: Design a reward function that reflects goals such as communication performance and energy efficiency. Multi-agent architecture: Select an appropriate multi-agent reinforcement learning architecture, such as MADDPG, QMIX, etc. Experience replay: Implement a priority experience replay mechanism to improve learning efficiency. Policy optimization: Use policy gradient or value function methods to optimize the decision-making strategy of the agent. Collaboration mechanism: Design a collaboration mechanism between agents.

[0148] The above describes the communication mode switching method of the walkie-talkie in the embodiment of the present application. The following describes the communication mode switching device of the walkie-talkie in the embodiment of the present application. Figure 2 In one embodiment of the present application, a device for switching a communication mode of a walkie-talkie includes:

[0149] A fusion module 201 is used to dynamically fuse heterogeneous data collected by multimodal sensors to obtain a high-dimensional feature vector;

[0150] An optimization module 202 is configured to input the high-dimensional feature vector into a preset topology communication model for plasticity optimization to obtain an initial communication mode;

[0151] The switching module 203 is configured to perform a gradual switching process on the initial communication mode to obtain a target communication state;

[0152] An analysis module 204 is configured to perform scenario parameter analysis on the acquired user behavior patterns and user communication content to obtain target scenario parameters;

[0153] An extraction module 205 is configured to extract adjustment parameters of the target communication state according to the target scenario parameters to obtain an adjustment parameter set;

[0154] The control module 206 is configured to construct a distributed communication switching strategy for the plurality of intercoms according to the adjustment parameter set, obtain a target communication switching strategy, and control the plurality of intercoms according to the target communication switching strategy.

[0155] Through the collaborative efforts of these components, the system dynamically fuses heterogeneous data collected by multimodal sensors to generate high-dimensional feature vectors, enabling comprehensive perception and accurate characterization of complex communication environments, significantly improving the system's adaptability to environmental changes. Secondly, the high-dimensional feature vectors are input into a pre-set topological communication model for plasticity optimization to obtain the initial communication mode. This approach, based on graph neural networks and reinforcement learning, adaptively constructs and optimizes the communication topology, significantly improving the network's flexibility and robustness. Furthermore, the initial communication mode is gradually switched to the target communication state. By combining Markov decision processes, dynamic programming, and evolutionary algorithms, the system achieves smooth transition and global optimization of communication parameters, effectively avoiding the communication interruptions and performance fluctuations common in traditional methods. Furthermore, the method analyzes user behavior patterns and communication content through scenario parameters to obtain target scenario parameters. This deep semantic understanding and multimodal fusion approach accurately captures user needs and communication scenario characteristics, providing a precise decision-making basis for subsequent communication optimization. Based on the target scenario parameters, the target communication state is adjusted and extracted to obtain an adjustment parameter set. This step, through the combination of advanced machine learning algorithms such as principal component analysis, support vector regression, and Gaussian processes, achieves refined adjustment and multi-objective optimization of communication parameters, greatly improving the system's adaptability and communication quality. A distributed communication switching strategy is constructed for multiple walkie-talkies based on the adjustment parameter set to obtain the target communication switching strategy, and then multiple walkie-talkies are controlled according to the target communication switching strategy. This distributed decision-making framework based on multi-agent reinforcement learning and federated learning not only achieves collaborative optimization of multiple devices, but also protects the user's communication privacy through differential privacy technology. By integrating multiple advanced artificial intelligence and communication technologies, the walkie-talkie communication system is made intelligent, personalized, and collaborative, significantly improving the system's communication efficiency, reliability, and user experience in complex environments.

[0156] The present application also provides a communication mode switching device for a walkie-talkie, which includes a memory and a processor. The memory stores computer-readable instructions. When the computer-readable instructions are executed by the processor, the processor executes the steps of the communication mode switching method of the walkie-talkie in the above-mentioned embodiments.

[0157] The present application also provides a computer-readable storage medium, which may be a non-volatile computer-readable storage medium or a volatile computer-readable storage medium. The computer-readable storage medium stores instructions, which, when executed on a computer, cause the computer to execute the steps of the method for switching the communication mode of the intercom.

[0158] Those skilled in the art will clearly understand that, for the convenience and brevity of description, the specific working processes of the above-described systems, systems and units can refer to the corresponding processes in the aforementioned method embodiments and will not be repeated here.

[0159] If the integrated unit is implemented in the form of 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 the present application is essentially 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, and the computer software product is stored in a storage medium, including a number of instructions for enabling a computer device (which can be a personal computer, a server, or a network device, etc.) to execute all or part of the steps of the method described in each embodiment of the present application. The aforementioned storage medium includes: U disk, mobile hard disk, read-only memory (ROM), random access memory (RAM), disk or optical disk, and other media that can store program code.

[0160] As described above, the above embodiments are only used to illustrate the technical solutions of the present application, rather than to limit them. Although the present application has been described in detail with reference to the above embodiments, those skilled in the art should understand that they can still modify the technical solutions described in the above embodiments, or make equivalent replacements for some of the technical features therein. However, these modifications or replacements do not deviate the essence of the corresponding technical solutions from the spirit and scope of the technical solutions of the embodiments of the present application.

Claims

1. A method for switching the communication mode of an intercom, characterized in that: The communication mode switching method of the intercom includes: Dynamically fuse heterogeneous data collected by multimodal sensors to obtain high-dimensional feature vectors; Inputting the high-dimensional feature vector into a preset topological communication model for plasticity optimization to obtain an initial communication mode; specifically comprising: performing dimensionality reduction processing on the high-dimensional feature vector to obtain a compressed feature set, and constructing a graph neural network based on the compressed feature set to obtain an initial topological structure; performing edge weight calculation on the initial topological structure to obtain a weighted adjacency matrix, and performing spectral clustering analysis based on the weighted adjacency matrix to obtain subnetwork partitioning data; performing dynamic routing construction on the subnetwork partitioning data to obtain an initial routing table, and performing state space analysis based on the initial routing table to obtain a state-action space; performing policy gradient optimization on the state-action space to obtain a candidate communication strategy, and reconstructing the topological structure based on the candidate communication strategy to obtain optimized topological structure data; performing plasticity analysis on the optimized topological structure data to obtain a plasticity index set, and adaptively adjusting communication parameters based on the plasticity index set to obtain the initial communication mode; Performing a progressive switching process on the initial communication mode to obtain a target communication state; Perform scenario parameter analysis on the acquired user behavior patterns and user communication content to obtain target scenario parameters; Extracting adjustment parameters from the target communication state according to the target scenario parameters to obtain an adjustment parameter set; A distributed communication switching strategy is constructed for a plurality of intercoms according to the adjustment parameter set to obtain a target communication switching strategy, and the plurality of intercoms are controlled according to the target communication switching strategy.

2. The method for switching the communication mode of an intercom according to claim 1, wherein: The dynamic fusion of heterogeneous data collected by multimodal sensors to obtain a high-dimensional feature vector includes: Preprocessing the heterogeneous data collected by the multimodal sensor to obtain a standardized data set, and performing data mapping on the standardized data set to obtain initial feature mapping data; Adaptively weighting the initial feature map data to obtain a weighted feature matrix, and constructing a tensor decomposition rank based on the weighted feature matrix to obtain low-rank representation data; Performing a nonlinear transformation on the low-rank representation data to obtain a nonlinear feature space, and performing deep convolution feature extraction based on the nonlinear feature space to obtain a multi-scale feature map; Performing an attention mechanism analysis on the multi-scale feature map to obtain a target feature region, and performing temporal correlation modeling based on the target feature region to obtain a temporal correlation representation; Topological feature processing is performed on the temporal correlation representation to obtain topological structure features, and high-dimensional space mapping is performed on the topological structure features to obtain the high-dimensional feature vector.

3. The communication mode switching method of an intercom according to claim 1, characterized in that: The step of gradually switching the initial communication mode to obtain a target communication state includes: Performing parameter analysis on the initial communication mode to obtain an initial parameter set, and performing Markov decision analysis on the initial parameter set to obtain a state transition matrix; Performing value iteration calculation on the state transfer matrix to obtain an optimal value function, and constructing a Bellman equation based on the optimal value function to obtain a dynamic programming strategy; Performing gradient descent optimization on the dynamic programming strategy to obtain a local optimal solution, and constructing a genetic algorithm population based on the local optimal solution to obtain a candidate solution set; Performing a crossover mutation operation on the candidate solution set to obtain a mutation solution space, and performing a global analysis on the mutation solution space using a preset simulated annealing algorithm to obtain a global search strategy; Particle swarm optimization is performed on the global search strategy to obtain a final switching sequence, and a progressive switching analysis is performed on the communication parameters according to the final switching sequence to obtain the target communication state.

4. The method for switching the communication mode of an intercom according to claim 1, wherein: The scenario parameter analysis of the acquired user behavior pattern and user communication content to obtain target scenario parameters includes: Performing time series decomposition on the acquired user behavior pattern to obtain a behavior sequence set, and extracting behavior features from the behavior sequence set to obtain a user behavior feature vector; Performing semantic analysis on user communication content to obtain a semantic feature matrix, and performing topic distribution analysis based on the semantic feature matrix to obtain a latent semantic distribution; Performing tensor fusion on the user behavior feature vector and the latent semantic distribution to obtain a multimodal representation, and performing high-order feature extraction on the multimodal representation to obtain a high-order feature set; Performing non-negative matrix decomposition on the high-order feature set to obtain basic semantic units, and performing hierarchical clustering analysis based on the basic semantic units to obtain a scene category tree; Association rule mining is performed on the scene category tree to obtain a scene association network, and parameterized modeling is performed based on the scene association network to obtain the target scene parameters.

5. The communication mode switching method of an intercom according to claim 1, characterized in that: The extracting adjustment parameters of the target communication state according to the target scenario parameters to obtain an adjustment parameter set includes: Performing principal component analysis on the target scene parameters to obtain a reduced-dimensional feature space, and constructing a regression function analysis based on support vectors on the reduced-dimensional feature space to obtain an initial regression function; Performing kernel function mapping on the initial regression function to obtain a high-dimensional feature map, and performing cross-validation based on the high-dimensional feature map to obtain an optimal hyperparameter set; Performing Bayesian optimization on the optimal hyperparameter set to obtain a posterior distribution of the parameters, and constructing a Gaussian process regression model based on the posterior distribution of the parameters to obtain a probability prediction function; Sampling and sensitivity analysis are performed on the probability prediction function to obtain a target parameter subset, and multi-objective optimization and parameter space mapping are performed on the target parameter subset to obtain the adjustment parameter set.

6. The method for switching the communication mode of an intercom according to claim 1, characterized in that: The step of constructing a distributed communication switching strategy for a plurality of intercoms according to the adjustment parameter set to obtain a target communication switching strategy, and controlling the plurality of intercoms according to the target communication switching strategy includes: Performing cluster analysis on the adjustment parameter set to obtain parameter subclusters, and constructing a hierarchical graph structure based on the parameter subclusters to obtain a multi-layer communication network topology; Performing edge computing task allocation on the multi-layer communication network topology to obtain an initial task allocation scheme, and performing protocol matching according to the initial task allocation scheme to obtain a global consistency protocol; Performing game theory analysis on the global consensus protocol to obtain a Nash equilibrium strategy set, and constructing a multi-agent reinforcement learning model based on the Nash equilibrium strategy set to obtain a collaborative decision distribution data set; Performing federated learning training on the collaborative decision-making distributed data set to obtain decentralized model parameters, and performing differential privacy processing based on the decentralized model parameters to obtain a privacy-preserving communication strategy; The privacy-preserving communication strategy is robustly optimized to obtain an interference-resistant switching scheme, and distributed deployment and execution are performed according to the interference-resistant switching scheme to obtain the target communication switching strategy.

7. A communication mode switching device for an intercom, characterized in that: The method for switching the communication mode of a two-way radio according to any one of claims 1 to 6 is used, wherein the device for switching the communication mode of the two-way radio comprises: The fusion module is used to dynamically fuse heterogeneous data collected by multimodal sensors to obtain high-dimensional feature vectors; An optimization module is used to input the high-dimensional feature vector into a preset topological communication model for plasticity optimization to obtain an initial communication mode; specifically comprising: performing dimensionality reduction processing on the high-dimensional feature vector to obtain a compressed feature set, and constructing a graph neural network based on the compressed feature set to obtain an initial topological structure; performing edge weight calculation on the initial topological structure to obtain a weighted adjacency matrix, and performing spectral clustering analysis based on the weighted adjacency matrix to obtain subnetwork partitioning data; performing dynamic routing construction on the subnetwork partitioning data to obtain an initial routing table, and performing state space analysis based on the initial routing table to obtain a state-action space; performing policy gradient optimization on the state-action space to obtain a candidate communication strategy, and reconstructing the topological structure based on the candidate communication strategy to obtain optimized topological structure data; performing plasticity analysis on the optimized topological structure data to obtain a plasticity index set, and adaptively adjusting communication parameters based on the plasticity index set to obtain the initial communication mode; A switching module, configured to perform a gradual switching process on the initial communication mode to obtain a target communication state; An analysis module is used to perform scenario parameter analysis on the acquired user behavior patterns and user communication content to obtain target scenario parameters; An extraction module, configured to extract adjustment parameters of the target communication state according to the target scenario parameters to obtain an adjustment parameter set; The control module is used to construct a distributed communication switching strategy for multiple intercoms according to the adjustment parameter set, obtain a target communication switching strategy, and control the multiple intercoms according to the target communication switching strategy.

8. A communication mode switching device for an intercom, characterized in that: The communication mode switching device of the intercom includes: a memory and at least one processor, wherein the memory stores instructions; The at least one processor calls the instruction in the memory to enable the communication mode switching device of the intercom to execute the communication mode switching method of the intercom according to any one of claims 1 to 6.

9. A computer-readable storage medium having instructions stored thereon, characterized in that: When the instructions are executed by the processor, the communication mode switching method of the intercom according to any one of claims 1 to 6 is implemented.

Images