Artificial intelligence-based environment semantic reconstruction method and device, equipment and medium

Abstract

The application relates to an environment semantic reconstruction method and device based on artificial intelligence, equipment and a medium. In the application, a downlink pilot signal transmitted by a terahertz base station to a communication coverage area is acquired, and multi-dimensional electromagnetic original data corresponding to the downlink pilot signal is collected; the multi-dimensional electromagnetic original data is arranged in a tensor regular manner to obtain a terahertz electromagnetic holographic tensor; the terahertz electromagnetic holographic tensor is input into a terahertz neural radiation field network to perform multi-scale position coding, multi-layer nonlinear feature transformation and environment semantic comprehensive reasoning, and multi-attribute environment representation data is obtained; the multi-attribute environment representation data is subjected to semantic category mapping, material attribute reasoning and spatial topological relationship analysis, and an environment semantic graph is constructed; multi-target semantic utility optimization solving is performed according to the environment semantic graph, a beam scheduling configuration scheme is obtained, and the beam forming parameters of a holographic intelligent super surface are configured according to the beam scheduling configuration scheme.

Description

Technical Field

[0001] This invention relates to the field of artificial intelligence technology, and specifically to an environment semantic reconstruction method, apparatus, device, and medium based on artificial intelligence. Background Technology

[0002] As sixth-generation mobile communication technology (6G) evolves deeply into the terahertz frequency band (0.1-10THz), terahertz waves, with their natural advantages of extremely short wavelength, ultra-wide bandwidth, and ability to support ultra-high-speed transmission at the Tbps (terabits per second) level, have become the core carrier frequency band of the 6G integrated architecture of sensing, communication, and computing. Environmental semantic reconstruction based on terahertz holographic intelligent metasurfaces is an emerging core technology derived for 6G terahertz communication. It utilizes a large-scale passive reflective unit array of reconfigurable intelligent metasurfaces (RIS) and integrates terahertz holographic electromagnetic sensing and acquisition capabilities to capture full-dimensional electromagnetic information about electromagnetic wave propagation, reflection, scattering, and occlusion interactions in complex spaces. This breaks through the limitations of traditional communication, which only focuses on physical layer channel parameters. It is used to mine deep-level environmental semantic information from electromagnetic observation data, such as scene types, spatial geometry, occlusion attributes, target material characteristics, and spatial relationships among multiple objects. It establishes a mapping relationship between the physical environment and the terahertz communication link, providing environmental context support for beam tracking, channel modeling, and dynamic scheduling of multi-user resources. This is a key technological path to overcome the inherent propagation defects of terahertz communication and achieve intelligent network adaptation.

[0003] Currently, environmental perception and communication optimization technology based on terahertz holographic intelligent metasurfaces has become a research hotspot in the 6G field. Existing technologies often integrate semantic communication, reconfigurable smart metasurfaces, and edge intelligence at the architectural level. While using the holographic detection capabilities of smart metasurfaces to acquire electromagnetic interference holograms, only basic physical information such as target position and incident angle can be extracted. At the channel modeling level, existing technologies only utilize a limited number of physical parameters such as signal amplitude, phase, and delay to construct channel state models, failing to explore the unique polarization, time-domain scattering, and spectral characteristics of terahertz waves. At the resource scheduling level, conventional reconfigurable smart metasurface beam scheduling algorithms generally prioritize maximizing channel capacity and minimizing system interference. While some improved schemes introduce semantic prioritization of communication services and differentiate between multiple users in the power and code domains using non-orthogonal multiple access (NOMA) technology, they only focus on the semantics of the communication data itself, completely ignoring the impact of physical environment semantics on the communication link. Furthermore, existing environment modeling often employs offline static training models or highly complex real-time inference algorithms, lacking online learning capabilities for dynamic scenarios. The models have weak cross-generalization adaptability to different scenarios such as indoor, outdoor, and industrial environments, and the update delay is difficult to match the extremely short coherence time constraints of terahertz channels.

[0004] However, the existing environmental perception and scheduling technologies based on terahertz holographic intelligent metasurfaces can only achieve shallow acquisition and modeling of terahertz electromagnetic physical characteristics. They lack the ability to semantically analyze the physical environment, cannot distinguish between temporary moving obstructions and permanent building structures, and cannot identify environmental scene categories and target material properties. Furthermore, the environmental modeling data has a single dimension and poor real-time adaptability to dynamic environments. At the same time, resource scheduling does not achieve deep coupling between environmental semantics and communication task semantics, resulting in beam scheduling lacking environmental context support, insufficient channel prediction accuracy, and serious semantic task conflicts and resource waste in multi-user scenarios. It is difficult to meet the stringent requirements of semantically sensitive scenarios such as industrial control, remote surgery, and autonomous driving for low latency, high reliability, and differentiated services.

[0005] Therefore, how to achieve semantic-level understanding of the terahertz electromagnetic environment and provide comprehensive environmental context support for communication resource scheduling is an urgent problem to be solved. Summary of the Invention

[0006] The technical problem to be solved by this invention is how to achieve semantic-level understanding of the terahertz electromagnetic environment and provide comprehensive environmental context support for communication resource scheduling.

[0007] The technical solution of the present invention to solve the above-mentioned technical problems is as follows: This application provides an artificial intelligence-based method for reconstructing environmental semantics, including: The downlink pilot signal transmitted by the terahertz base station to the communication coverage area is acquired, and the amplitude data, phase data, and polarization data corresponding to the downlink pilot signal are collected by the holographic sensing module mounted on each reflection unit in the holographic intelligent metasurface array to obtain multi-dimensional electromagnetic raw data. The multi-dimensional electromagnetic raw data is arranged by tensor regularization according to the metasurface unit spatial coordinates and signal sampling time to obtain the terahertz electromagnetic holographic tensor. The terahertz electromagnetic holographic tensor is input into a pre-trained terahertz neural radiation field network to perform multi-scale position encoding, multi-layer nonlinear feature transformation, and environmental semantic comprehensive reasoning to obtain multi-attribute environmental representation data. Semantic category mapping, material attribute reasoning, and spatial topological relationship parsing are performed on the multi-attribute environmental representation data to obtain environmental object categories, material attributes, and spatial relationships. Based on the environmental object categories, material attributes, and spatial relationships, an environmental semantic graph is constructed. Based on the environmental semantic map, the preset semantic attributes of communication tasks, and the channel state information extracted from the terahertz electromagnetic holographic tensor, a multi-objective semantic utility optimization solution is performed to obtain a beam scheduling configuration scheme. The beamforming parameters of the holographic intelligent metasurface are then configured according to the beam scheduling configuration scheme to complete the environmental semantic reconstruction and communication control.

[0008] The beneficial effects of this invention are as follows: By using the holographic sensing module mounted on each reflective unit of the holographic intelligent metasurface array, amplitude, phase, and polarization data corresponding to the downlink pilot signal transmitted by the terahertz base station are collected, obtaining multi-dimensional raw electromagnetic data containing rich electromagnetic information. Arranging this multi-dimensional raw electromagnetic data into a terahertz electromagnetic holographic tensor according to the spatial coordinates of the metasurface unit and the signal sampling time provides an ordered data structure. Inputting the terahertz electromagnetic holographic tensor into a pre-trained terahertz neural radiation field network, through multi-scale position encoding, multi-layer nonlinear feature transformation, and comprehensive environmental semantic reasoning, multi-attribute environmental representation data is obtained. In-depth analysis and processing of the data uncovers the environmental semantics it contains. Semantic category mapping, material attribute reasoning, and spatial topological relationship analysis are performed on the multi-attribute environmental representation data to construct an environmental semantic map, clarifying the categories, material attributes, and spatial relationships of environmental objects, thus achieving a semantic-level understanding of the terahertz electromagnetic environment. Based on the environmental semantic map, the preset semantic attributes of communication tasks, and the channel state information extracted from the terahertz electromagnetic holographic tensor, a multi-objective semantic utility optimization solution is performed to obtain a beam scheduling configuration scheme and configure beamforming parameters. This provides comprehensive environmental context support for communication resource scheduling and realizes a close integration of environmental semantic reconstruction and communication control.

[0009] Based on the above technical solution, the present invention can be further improved as follows.

[0010] Furthermore, the acquisition of amplitude data, phase data, and polarization data corresponding to the downlink pilot signal yields multi-dimensional raw electromagnetic data, including: The electric field strength of the downlink pilot signal is detected by induction to obtain the amplitude data corresponding to the downlink pilot signal; The phase of the signal corresponding to the amplitude data is synchronously sampled to obtain the phase data corresponding to the downlink pilot signal; The polarization state of the electromagnetic wave corresponding to the phase data is identified and detected to obtain the polarization data corresponding to the downlink pilot signal; The amplitude data, phase data, and polarization data are time-aligned and fused to obtain the multi-dimensional electromagnetic raw data.

[0011] The beneficial effects of adopting the above-mentioned further scheme are as follows: by detecting, sampling, and identifying the electric field strength, signal phase, and electromagnetic wave polarization state of the downlink pilot signal, and by performing time alignment and fusion of the obtained data, it is possible to comprehensively and accurately acquire multi-dimensional raw electromagnetic data after the interaction between the terahertz wave and the environment. Accurate amplitude data helps to understand signal strength changes, thereby analyzing the reflection and absorption of the signal by objects in the environment; precise phase data can be used to determine environmental interference factors on the signal propagation path; and detailed polarization data can help infer the material properties of objects in the environment.

[0012] Furthermore, the process of performing multi-scale location encoding, multi-layer nonlinear feature transformation, and comprehensive environmental semantic reasoning to obtain multi-attribute environmental representation data includes: Multi-scale position encoding is performed on the spatial coordinate information contained in the terahertz electromagnetic holographic tensor to obtain spatial feature data; The spatial feature data is subjected to multi-layer nonlinear feature transformation to obtain electromagnetic bottom-layer feature data; Environmental semantic comprehensive reasoning is performed on the electromagnetic low-level feature data to obtain sub-item environmental feature parameter data; The spatial feature data, the electromagnetic underlying feature data, and the sub-item environmental feature parameter data are correlated, grouped, and rectified to obtain the multi-attribute environmental characterization data.

[0013] The beneficial effects of adopting the above-mentioned further scheme are as follows: multi-scale position encoding of the spatial coordinate information contained in the terahertz electromagnetic holographic tensor can accurately capture the spatial variation characteristics of terahertz waves and provide high-resolution spatial feature data. Multi-layer nonlinear feature transformation further explores the deep relationship between spatial features and electromagnetic properties, obtaining electromagnetic underlying feature data reflecting fundamental electromagnetic properties. Environmental semantic comprehensive reasoning extracts several key environmental feature parameters from the electromagnetic underlying feature data, realizing the conversion from electromagnetic information to environmental semantics. Finally, through association, grouping, and integration processing, data from different levels are integrated into multi-attribute environmental characterization data, comprehensively and systematically describing the terahertz electromagnetic environment.

[0014] Furthermore, the step of performing semantic category mapping, material attribute reasoning, and spatial topological relationship parsing on the multi-attribute environmental representation data to obtain environmental object categories, material attributes, and spatial relationships, and constructing an environmental semantic graph based on the environmental object categories, material attributes, and spatial relationships, includes: The semantic category labels in the multi-attribute environment representation data are mapped to a preset semantic category library, and the category of the environment object is identified according to the semantic category library; Based on the polarization state and temporal scattering characteristics in the multi-attribute environmental characterization data, material attribute inference is performed to obtain the material attributes of each target object in the environment. Based on the volume density and the category of the environment object in the multi-attribute environment characterization data, the occlusion relationship, adjacency relationship and hierarchical relationship between each target object are analyzed, and the spatial association relationship is obtained based on the occlusion relationship, the adjacency relationship and the hierarchical relationship; The environmental object categories, material attributes, and spatial relationships are structured and associated to form the environmental semantic map.

[0015] The beneficial effects of adopting the above-mentioned further scheme are as follows: By performing semantic category mapping, material attribute reasoning, and spatial topological relationship analysis on multi-attribute environmental characterization data, and constructing an environmental semantic map, a deep semantic understanding of the terahertz electromagnetic environment is achieved. Semantic category mapping can accurately identify the categories of environmental objects, clearly identifying the types of objects in the environment; material attribute reasoning determines the material of objects based on polarization state and time-domain scattering characteristics, further enriching the understanding of environmental objects; spatial correlation analysis based on volume density and environmental object categories clarifies the relative positions of each object in space. The above information is structured and correlated to form an environmental semantic map, providing comprehensive, detailed, and accurate environmental context support for communication resource scheduling. This enables the communication system to allocate resources more rationally and optimize beam scheduling based on information such as the category, material, and spatial layout of objects in the environment, thereby improving the performance of the communication system in complex indoor environments, ensuring the stability and efficiency of communication, and effectively solving the problem of unreasonable resource scheduling caused by insufficient environmental understanding in terahertz communication.

[0016] Furthermore, before obtaining the beam scheduling configuration scheme by performing multi-objective semantic utility optimization based on the environmental semantic map, preset communication task semantic attributes, and channel state information extracted from the terahertz electromagnetic holographic tensor, the method further includes: The amplitude features, phase features, polarization features, time domain features, and frequency domain features of the terahertz electromagnetic holographic tensor are extracted to obtain discrete electromagnetic feature data; The discrete electromagnetic feature data are fused using multimodal methods to obtain fused channel features; Multipath channel parameters are fitted to the fused channel features to obtain the basic channel parameters. The channel state information is obtained by performing state regression correction on the channel basic parameters.

[0017] The beneficial effects of adopting the above-mentioned further scheme are as follows: Multiple discrete electromagnetic feature data are extracted from the terahertz electromagnetic holographic tensor, and multi-mode fusion, multipath channel parameter fitting, and state regression correction are performed to obtain accurate channel state information. Extracting multiple discrete electromagnetic feature data can comprehensively capture the electromagnetic characteristics of terahertz waves in different dimensions, providing rich material for in-depth channel analysis. Multi-mode fusion organically combines these discrete features, fully leveraging the advantages of each feature to obtain a more comprehensive and accurate fused channel feature reflecting channel characteristics. Multipath channel parameter fitting estimates the basic parameters of the channel based on the fused channel features, laying the foundation for describing the channel state. State regression correction further optimizes these parameters to better conform to the actual channel state. Accurate channel state information provides a reliable basis for solving multi-objective semantic utility optimization, helps to formulate more reasonable beam scheduling configuration schemes, improves the adaptability of communication systems to channels in complex environments, optimizes communication resource allocation, and thus improves communication quality and system performance.

[0018] Furthermore, the step of performing multi-objective semantic utility optimization based on the environmental semantic map, preset communication task semantic attributes, and channel state information extracted from the terahertz electromagnetic holographic tensor to obtain a beam scheduling configuration scheme includes: Based on the operational constraints of communication services, a multi-dimensional semantic comprehensive utility function is constructed, which includes throughput utility, semantic matching utility, link robustness utility, and resource allocation fairness utility. Feature association matching is performed on the environmental semantic map, communication task semantic attributes, and channel state information to obtain optimized input feature data; With transmit power, time slot resources, and service quality as constraints, and based on the optimized input feature data, the multi-dimensional semantic comprehensive utility function is optimized by multi-objective semantic utility to obtain the global optimal solution set; According to the preset business decision rules, the optimal configuration is selected from the global optimal solution set, and the beam scheduling configuration scheme is determined based on the optimal configuration.

[0019] The beneficial effects of adopting the above-mentioned further scheme are as follows: Constructing a multi-dimensional semantic comprehensive utility function comprehensively considers the performance requirements of the communication system in multiple key aspects such as throughput, semantic matching, link robustness, and resource allocation fairness, making the optimization objectives more comprehensive and reasonable. Feature association matching of environmental, task, and channel information provides rich and targeted input data for optimization, enabling the optimization process to be closely integrated with actual communication scenarios. Multi-objective optimization is performed with transmit power, time slot resources, and service quality as constraints, ensuring that the obtained resource allocation scheme achieves optimal multi-dimensional semantic comprehensive utility while meeting the actual system limitations. Optimal configurations are selected from the global optimal solution set according to preset service decision rules, enabling precise selection of the most suitable resource configuration for different communication service needs. The resulting beam scheduling configuration can fully leverage the performance of the communication system, improve communication quality, and meet the diverse needs of different services in complex industrial environments.

[0020] Furthermore, after configuring the beamforming parameters of the holographic intelligent metasurface according to the beam scheduling configuration scheme to complete environmental semantic reconstruction and communication control, the method further includes: Collect current environmental semantic features and terahertz channel features, and perform statistical distribution comparison on the current environmental semantic features and terahertz channel features to obtain feature distribution difference data; Drift detection is performed on the feature distribution difference data to obtain the drift detection results; When the drift detection result indicates the presence of feature distribution drift, the parameters of the terahertz neural radiation field network are adjusted using on-site measured samples to obtain an updated terahertz neural radiation field network. The updated terahertz neural radiation field network is used for multi-scale position encoding, multi-layer nonlinear feature transformation, and comprehensive environmental semantic reasoning.

[0021] The advantages of adopting the above-mentioned further scheme are: real-time acquisition of environmental semantics and channel features enables timely capture of changes in the environment and channel, providing an accurate data foundation. Statistical distribution comparison and drift detection can accurately determine whether these changes have reached the point where model adjustments are necessary, avoiding unnecessary model adjustments and improving system stability. When feature distribution drift is detected, the terahertz neural radiation field network parameters are adjusted using field-measured samples, enabling the network to quickly adapt to new environmental and channel conditions and continue to accurately perform multi-scale position coding, multi-layer nonlinear feature transformation, and environmental semantic comprehensive reasoning. This ensures the accuracy of environmental semantic reconstruction and the effectiveness of communication control, enabling the communication system to operate continuously and stably in a dynamically changing environment.

[0022] This application embodiment also provides an environment semantic reconstruction device based on artificial intelligence, the device comprising: The acquisition module is used to acquire the downlink pilot signal transmitted by the terahertz base station to the communication coverage area, and to collect the amplitude data, phase data and polarization data corresponding to the downlink pilot signal through the holographic sensing module mounted on each reflection unit in the holographic intelligent metasurface array, so as to obtain multi-dimensional electromagnetic raw data. The arrangement module is used to arrange the multi-dimensional electromagnetic raw data according to the metasurface unit spatial coordinates and signal sampling time to obtain a terahertz electromagnetic holographic tensor. The reasoning module is used to input the terahertz electromagnetic holographic tensor into a pre-trained terahertz neural radiation field network to perform multi-scale position encoding, multi-layer nonlinear feature transformation and environmental semantic comprehensive reasoning to obtain multi-attribute environmental representation data. The graph construction module is used to perform semantic category mapping, material attribute reasoning, and spatial topological relationship parsing on the multi-attribute environmental representation data to obtain environmental object categories, material attributes, and spatial relationships, and to construct an environmental semantic graph based on the environmental object categories, material attributes, and spatial relationships. The parameter configuration module is used to perform multi-objective semantic utility optimization solution based on the environmental semantic map, the preset communication task semantic attributes and the channel state information extracted by the terahertz electromagnetic holographic tensor, to obtain a beam scheduling configuration scheme, and to configure the beamforming parameters of the holographic intelligent metasurface according to the beam scheduling configuration scheme, so as to complete the environmental semantic reconstruction and communication control.

[0023] This application also provides an electronic device, including a processor and a memory, wherein the memory stores multiple instructions; the processor loads instructions from the memory to execute steps in any of the artificial intelligence-based environmental semantic reconstruction methods provided in this application.

[0024] This application also provides a computer-readable storage medium storing a plurality of instructions adapted for loading by a processor to execute steps in any of the artificial intelligence-based environmental semantic reconstruction methods provided in this application. Attached Figure Description

[0025] Figure 1 A flowchart illustrating the artificial intelligence-based environmental semantic reconstruction method provided in this embodiment of the invention; Figure 2 This is a schematic diagram of the structure of the artificial intelligence-based environmental semantic reconstruction device provided in an embodiment of the present invention; Figure 3 This is a schematic diagram of the structure of an electronic device provided in an embodiment of the present invention. Detailed Implementation

[0026] The principles and features of the present invention are described below. The examples given are only for explaining the present invention and are not intended to limit the scope of the present invention.

[0027] This application provides an environment semantic reconstruction method, apparatus, electronic device, and medium based on artificial intelligence.

[0028] Specifically, this AI-based environmental semantic reconstruction device can be integrated into an electronic device, such as a terminal or server. The terminal can be a mobile phone, tablet, smart Bluetooth device, laptop, or personal computer (PC); the server can be a single server or a server cluster consisting of multiple servers.

[0029] In some embodiments, the AI-based environmental semantic reconstruction device can also be integrated into multiple electronic devices. For example, the AI-based environmental semantic reconstruction device can be integrated into multiple servers, and the AI-based environmental semantic reconstruction method of this application can be implemented by multiple servers.

[0030] In some embodiments, the terminal can also be used as a server to implement some or all of the functions of a server.

[0031] In the embodiments of this application, the terms "module" or "unit" may refer to a computer program or part of a computer program with a predetermined function, which works together with other related parts to achieve a predetermined goal, and can be implemented wholly or partially using software, hardware (such as processing circuitry or memory), or a combination thereof. Similarly, a processor (or multiple processors or memory) may be used to implement one or more modules or units. Furthermore, each module or unit may be part of an overall module or unit that includes the functions of that module or unit.

[0032] The following sections provide detailed descriptions of each example. It should be noted that the order of the following embodiments is not intended to limit the preferred order of the embodiments.

[0033] This embodiment provides an environment semantic reconstruction method based on artificial intelligence. See also Figure 1 , Figure 1 This is a flowchart illustrating the artificial intelligence-based environmental semantic reconstruction method provided in an embodiment of the present invention, as shown below. Figure 1 As shown, the specific process of this AI-based environmental semantic reconstruction method can be summarized as follows: 110. Obtain the downlink pilot signal transmitted by the terahertz base station to the communication coverage area, and collect the amplitude data, phase data, and polarization data corresponding to the downlink pilot signal through the holographic sensing module mounted on each reflection unit in the holographic intelligent metasurface array to obtain multi-dimensional electromagnetic raw data; 120. The multi-dimensional electromagnetic raw data is arranged by tensor regularization according to the metasurface unit spatial coordinates and signal sampling time to obtain the terahertz electromagnetic holographic tensor. 130. Input the terahertz electromagnetic holographic tensor into a pre-trained terahertz neural radiation field network to perform multi-scale position encoding, multi-layer nonlinear feature transformation and environmental semantic comprehensive reasoning to obtain multi-attribute environmental representation data. 140. Perform semantic category mapping, material attribute reasoning, and spatial topology analysis on the multi-attribute environmental representation data to obtain environmental object categories, material attributes, and spatial relationships, and construct an environmental semantic graph based on the environmental object categories, material attributes, and spatial relationships; 150. Based on the environmental semantic map, the preset communication task semantic attributes, and the channel state information extracted from the terahertz electromagnetic holographic tensor, perform multi-objective semantic utility optimization to obtain a beam scheduling configuration scheme, and configure the beamforming parameters of the holographic intelligent metasurface according to the beam scheduling configuration scheme to complete environmental semantic reconstruction and communication control.

[0034] The terahertz base station is responsible for generating and transmitting terahertz communication signals to the communication coverage area. It is equipped with a multi-antenna array, a baseband processing unit, and a power distribution controller. The multi-antenna array can realize beamforming and spatial multiplexing to improve spectrum efficiency. The baseband processing unit executes signal modulation, coding, and beamforming algorithms. The power distribution controller can dynamically adjust the transmission power according to the scheduling results to optimize energy consumption.

[0035] The holographic intelligent metasurface array consists of numerous reflective units that serve as data acquisition and beamforming units. Each reflective unit is equipped with a holographic sensing module, which can collect multi-dimensional electromagnetic information. For example, an array composed of 1024 reflective units, with a unit spacing of half a wavelength at a specific frequency band, has an overall size of approximately 30 centimeters square and can perform related processing on terahertz signals.

[0036] The holographic sensing module is installed on each reflective unit of the metasurface array, enabling it to acquire various electromagnetic data, specifically amplitude, phase, and polarization data corresponding to the downlink pilot signal. For example, the amplitude sensor uses a Schottky diode detector with a certain operating frequency range, and its sensitivity and response time meet the corresponding requirements, allowing it to detect electric field strength; the phase sensor extracts phase information through a zero-difference detection circuit; and the polarization sensor uses a polarization-sensitive detector to detect the polarization state.

[0037] Among them, the multi-dimensional electromagnetic raw data is the initial data set obtained by collecting the amplitude, phase and polarization data of the downlink pilot signal through the holographic sensing module. These data reflect the various physical characteristics of the interaction between terahertz waves and the environment.

[0038] Among them, the terahertz electromagnetic holographic tensor is a data structure formed by regularizing and arranging multi-dimensional electromagnetic raw data according to the spatial coordinates of the metasurface unit and the signal sampling time. It integrates electromagnetic information in both spatial and temporal dimensions, providing an ordered and easy-to-process data input form for the terahertz neural radiation field network, which helps the network to better analyze and extract the environmental semantic information.

[0039] Among them, the Terahertz Neural Radiation Field Network is a pre-trained network specifically designed for processing terahertz electromagnetic holographic tensor data. It performs deep processing on the input data through multi-scale position encoding, multi-layer nonlinear feature transformation, and comprehensive environmental semantic reasoning, thereby obtaining multi-attribute environmental representation data and achieving a semantic-level understanding of the electromagnetic environment.

[0040] Among them, the multi-attribute environmental characterization data is data obtained after being processed by the terahertz neural radiation field network. It contains a variety of information reflecting environmental characteristics, such as volume density, semantic labels, polarization state, temporal features, and terahertz radiation intensity. This information provides a key basis for further constructing an environmental semantic map.

[0041] Among them, the environmental semantic graph is constructed by semantic category mapping, material attribute reasoning and spatial topological relationship parsing of multi-attribute environmental representation data. It clarifies the categories, material attributes and spatial relationships of environmental objects, provides comprehensive environmental context support for communication resource scheduling, and is an important foundation for realizing intelligent communication control.

[0042] Among them, the semantic attributes of communication tasks refer to a series of pre-defined characteristics of communication tasks, such as task priority, latency tolerance, reliability requirements, bandwidth requirements, and task duration. These attributes together define the quality of service requirements of communication tasks and serve as an important decision-making basis when solving multi-objective semantic utility optimization problems.

[0043] Among them, the channel state information is extracted from the terahertz electromagnetic holographic tensor, reflecting the relevant characteristics of the channel, such as the amplitude, phase, polarization, time domain and frequency domain characteristics of the signal. After a series of processing, the data used to describe the channel state is obtained, which has important guiding significance for the rational scheduling of communication resources.

[0044] Among them, the beam scheduling configuration scheme is obtained by solving the multi-objective semantic utility optimization. This scheme comprehensively considers the environmental semantic map, the semantic attributes of the communication task, and the channel state information, and determines the beamforming parameter configuration of the holographic intelligent metasurface to achieve the optimization of environmental semantic reconstruction and communication control.

[0045] Among them, beamforming parameters are parameters used to control how the holographic intelligent metasurface adjusts the beam. After being configured according to the beam scheduling configuration scheme, the user communication link can be optimized to achieve more efficient communication.

[0046] For example, the specific methods for obtaining multi-dimensional electromagnetic raw data can be as follows: A terahertz base station transmits a downlink pilot signal, which propagates across the communication coverage area. Each reflective unit in the holographic intelligent metasurface array senses this signal through its onboard holographic sensing module. For example, an amplitude sensor detects the electric field strength of the downlink pilot signal to obtain amplitude data; a phase sensor synchronously samples the signal phase corresponding to the amplitude data to obtain phase data; and a polarization sensor identifies the electromagnetic wave polarization state corresponding to the phase data to obtain polarization data. Finally, these data are time-aligned and fused to form multi-dimensional raw electromagnetic data.

[0047] For example, the specific method for obtaining the terahertz electromagnetic holographic tensor can be as follows: The acquired multi-dimensional electromagnetic raw data is quite disorganized. To facilitate subsequent processing, it needs to be organized and arranged according to the spatial coordinates of the metasurface units and the temporal order of signal sampling. The amplitude, phase, and polarization data acquired by each reflecting unit at different time points are regularized into a tensor with a pre-defined structure, namely a terahertz electromagnetic holographic tensor, based on its spatial location on the metasurface. This arrangement ensures the data has order in both spatial and temporal dimensions, which is beneficial for analysis and processing by the terahertz neural radiation field network.

[0048] For example, the specific methods for obtaining multi-attribute environmental representation data can be as follows: Terahertz electromagnetic holographic tensors are input into a pre-trained terahertz neural radiation field network. First, multi-scale position encoding is performed. Due to the extremely short wavelength and drastic spatial variations of terahertz waves, high-frequency position encoding techniques are employed to map spatial coordinates to a high-dimensional feature space. By combining basis functions of sine and cosine functions, multiple frequency components are used to extract features at different spatial scales. Different frequency parameters are employed for different frequency bands to balance computational complexity and representational power. Next, multi-layer nonlinear feature transformation is performed. The network processes the position-encoded features through a multi-layer fully connected network to predict multiple environmental attributes, such as volume density, semantic labels, polarization state, temporal features, and terahertz radiation intensity, obtaining the underlying electromagnetic feature data. Finally, comprehensive environmental semantic reasoning is performed. These sub-item environmental feature parameters are associated, combined, and integrated with the spatial feature data to obtain multi-attribute environmental representation data, achieving a preliminary semantic understanding of the environment.

[0049] For example, the specific methods for constructing an environmental semantic graph can be as follows: Semantic category labels in multi-attribute environmental characterization data are mapped to a pre-defined semantic category library to identify environmental object categories. For example, in general scenarios, categories such as empty space, people, vehicles, and building walls can be identified. Based on the polarization state and temporal scattering characteristics in the multi-attribute environmental characterization data, and utilizing the differences in reflection, scattering, and absorption characteristics of terahertz waves by different materials, a material-feature mapping table and a deep learning classifier are combined to perform material attribute inference, deriving the material attributes of each target object in the environment, such as metal or non-metal. By analyzing the volume density and environmental object categories in the multi-attribute environmental characterization data, and based on beam propagation paths, spatial distances, and spatial inclusion relationships, the occlusion, adjacency, and hierarchical relationships between target objects are analyzed to obtain spatial correlation relationships. Finally, the environmental object categories, material attributes, and spatial correlation relationships are structurally correlated and modeled to construct an environmental semantic graph, achieving a semantic-level understanding of the terahertz electromagnetic environment.

[0050] For example, the specific methods for completing environmental semantic reconstruction and communication control can be as follows: First, considering the constraints of communication service operation, a multi-dimensional semantic comprehensive utility function is constructed, encompassing throughput utility, semantic matching utility, link robustness utility, and resource allocation fairness utility. Then, feature association matching is performed on the environmental semantic map, communication task semantic attributes, and channel state information to obtain optimized input feature data. Using transmit power, time slot resources, and service quality as constraints, multi-objective semantic utility optimization is performed on the multi-dimensional semantic comprehensive utility function based on the optimized input feature data, yielding a globally optimal solution set. Next, according to preset service decision rules, the optimal configuration is selected from the globally optimal solution set to determine the beam scheduling configuration scheme. Finally, the beamforming parameters of the holographic intelligent metasurface are configured according to this scheme to optimize the communication link, complete environmental semantic reconstruction and communication control, and achieve the goal of intelligent communication resource scheduling.

[0051] For example, in a communication scenario within an industrial plant, a terahertz base station is deployed on the roof, and a holographic intelligent metasurface array is installed on the ceiling of a key production area. The terahertz base station transmits downlink pilot signals to the communication coverage area below, and the holographic sensing modules mounted on each reflector unit in the holographic intelligent metasurface array begin to operate. Amplitude sensors detect the electric field strength of the pilot signal; for example, amplitude data detected at a certain moment reflects changes in signal strength. Phase sensors synchronously sample phase data, reflecting the phase characteristics of the signal. Polarization sensors identify polarization data, such as detecting polarization states caused by the presence of metal equipment in certain areas. These collected amplitude, phase, and polarization data are integrated into multi-dimensional electromagnetic raw data. Then, according to the spatial coordinates of the reflector units and the signal sampling time, this data is arranged into a terahertz electromagnetic holographic tensor. Subsequently, the tensor is input into a pre-trained terahertz neural radiation field network. The network performs multi-scale position encoding to accurately represent the complex spatial structure within the factory. After multi-layer nonlinear feature transformation, it analyzes the objects that may exist within the factory, such as robotic arms and metal equipment, and obtains their related attributes, such as object volume density and semantic labels. After comprehensive reasoning, multi-attribute environmental representation data is obtained. This data is further processed to map semantic category labels to a pre-defined semantic category library for industrial scenarios, identifying environmental object categories such as robotic arms and workers. Material properties are inferred based on polarization states and temporal scattering characteristics, such as metal equipment being made of metal. Spatial relationships are determined by analyzing volume density and environmental object categories, such as the positional relationship between robotic arms and workers, thereby constructing an environmental semantic map. Simultaneously, based on pre-defined semantic attributes of industrial control communication tasks, such as high priority and low latency tolerance, and channel state information extracted from the terahertz electromagnetic holographic tensor, a multi-dimensional semantic comprehensive utility function is constructed for multi-objective semantic utility optimization. For example, considering the high requirements of industrial control for reliability and real-time performance, the solution focuses on optimizing semantic matching utility and link robustness utility to obtain a beam scheduling configuration scheme. The beamforming parameters of the holographic intelligent metasurface are configured according to this scheme so that the communication signal can accurately cover the required area, ensuring the reliable transmission of industrial control signals and completing environmental semantic reconstruction and communication control.

[0052] As can be seen, in this embodiment, the holographic sensing module mounted on each reflective unit of the holographic intelligent metasurface array collects the amplitude, phase, and polarization data corresponding to the downlink pilot signal transmitted by the terahertz base station, obtaining multi-dimensional raw electromagnetic data containing rich electromagnetic information. Arranging the multi-dimensional raw electromagnetic data into a terahertz electromagnetic holographic tensor according to the spatial coordinates of the metasurface unit and the signal sampling time provides an ordered data structure. The terahertz electromagnetic holographic tensor is input into a pre-trained terahertz neural radiation field network, and through multi-scale position encoding, multi-layer nonlinear feature transformation, and comprehensive environmental semantic reasoning, multi-attribute environmental representation data is obtained. In-depth analysis and processing of the data are performed to mine the environmental semantics contained within. Semantic category mapping, material attribute reasoning, and spatial topological relationship analysis are performed on the multi-attribute environmental representation data to construct an environmental semantic map, clarifying the categories, material attributes, and spatial relationships of environmental objects, thus achieving a semantic-level understanding of the terahertz electromagnetic environment. Based on the environmental semantic map, the preset semantic attributes of communication tasks, and the channel state information extracted from the terahertz electromagnetic holographic tensor, a multi-objective semantic utility optimization solution is performed to obtain a beam scheduling configuration scheme and configure beamforming parameters. This provides comprehensive environmental context support for communication resource scheduling and realizes a close integration of environmental semantic reconstruction and communication control.

[0053] Preferably, in this embodiment, the acquisition of amplitude data, phase data, and polarization data corresponding to the downlink pilot signal to obtain multi-dimensional electromagnetic raw data includes: The electric field strength of the downlink pilot signal is detected by induction to obtain the amplitude data corresponding to the downlink pilot signal; The phase of the signal corresponding to the amplitude data is synchronously sampled to obtain the phase data corresponding to the downlink pilot signal; The polarization state of the electromagnetic wave corresponding to the phase data is identified and detected to obtain the polarization data corresponding to the downlink pilot signal; The amplitude data, phase data, and polarization data are time-aligned and fused to obtain the multi-dimensional electromagnetic raw data.

[0054] Electric field strength is a physical quantity describing the strength and direction of an electric field. Terahertz downlink pilot signals, as a type of electromagnetic wave, carry crucial information such as signal energy in their electric field strength. By sensing and detecting the electric field strength, information related to the signal amplitude can be obtained, thus yielding amplitude data. For example, the reflection and scattering of terahertz waves by different environmental objects will cause changes in the electric field strength; detecting these changes can reveal certain characteristics of the objects in the environment.

[0055] Signal phase describes the position of a signal during periodic changes. In terahertz communication, signal phase is crucial for accurately understanding the state of the signal during propagation. By synchronously sampling the signal phase corresponding to the amplitude data, phase data can be obtained. This data helps analyze phase changes caused by environmental factors during signal propagation.

[0056] In this context, the polarization state of an electromagnetic wave refers to the orientation of the electric field vector in space. Different objects have different effects on the polarization characteristics of terahertz waves. By identifying and detecting the polarization state of electromagnetic waves, information about the material of objects in the environment can be obtained. For example, metallic and non-metallic objects exhibit different changes in the polarization state of terahertz waves, which can be used for differentiation.

[0057] For example, the specific method for obtaining amplitude data can be: The electric field strength of the terahertz downlink pilot signal is measured using an inductive detection device corresponding to the amplitude data, such as a Schottky diode detector. This detector operates across the terahertz frequency band, has high sensitivity, and a short response time, enabling it to accurately detect changes in electric field strength and convert them into quantifiable amplitude data. This amplitude data reflects the intensity changes of the signal during propagation.

[0058] For example, the specific method for obtaining phase data can be: After acquiring the amplitude data, synchronous sampling techniques, such as a homodyne detection circuit, are used to acquire the signal phase corresponding to the amplitude data. The homodyne detection circuit accurately extracts the instantaneous phase information of the terahertz wave by mixing the received signal with the local oscillation signal, thus obtaining phase data. This phase data records the phase changes of the signal during propagation and is of great significance for analyzing environmental interference encountered along the signal propagation path.

[0059] For example, the specific method for obtaining polarization data can be: The polarization state of electromagnetic waves corresponding to phase data is identified using a polarization-sensitive detector. This detector can distinguish between different polarization states of terahertz waves, such as horizontal and vertical linear polarization, and left-handed and right-handed circular polarization, thereby acquiring polarization data. This polarization data contains information about the influence of environmental objects on the polarization characteristics of terahertz waves, which helps in inferring the material properties of those objects.

[0060] For example, the specific methods for obtaining multi-dimensional electromagnetic raw data can be as follows: Because there may be slight temporal differences in the acquisition of amplitude, phase, and polarization data, time alignment is required to ensure consistency across the time dimension. Subsequently, the time-aligned data is fused and integrated, merging different data types into a single, multi-dimensional electromagnetic raw data set. This multi-dimensional electromagnetic raw data set integrates the amplitude, phase, and polarization information of the signal.

[0061] In this embodiment, the electric field strength, signal phase, and electromagnetic wave polarization state of the downlink pilot signal are detected, sampled, and identified. The obtained data are then time-aligned and fused to comprehensively and accurately acquire multi-dimensional raw electromagnetic data after the interaction between the terahertz wave and the environment. Accurate amplitude data helps to understand signal strength changes, thereby analyzing the reflection and absorption of the signal by objects in the environment; precise phase data can be used to determine environmental interference factors along the signal propagation path; and detailed polarization data can assist in inferring the material properties of objects in the environment.

[0062] Preferably, in the embodiments, the step of performing multi-scale location encoding, multi-layer nonlinear feature transformation, and comprehensive environmental semantic reasoning to obtain multi-attribute environmental representation data includes: Multi-scale position encoding is performed on the spatial coordinate information contained in the terahertz electromagnetic holographic tensor to obtain spatial feature data; The spatial feature data is subjected to multi-layer nonlinear feature transformation to obtain electromagnetic bottom-layer feature data; Environmental semantic comprehensive reasoning is performed on the electromagnetic low-level feature data to obtain sub-item environmental feature parameter data; The spatial feature data, the electromagnetic underlying feature data, and the sub-item environmental feature parameter data are correlated, grouped, and rectified to obtain the multi-attribute environmental characterization data.

[0063] The spatial coordinate information, contained within the terahertz electromagnetic holographic tensor, is used to determine the spatial position of each reflecting unit on the metasurface. Because terahertz wavelengths are extremely short, spatial variations significantly affect the signal; therefore, accurate spatial coordinate information is crucial for analyzing the interaction between terahertz waves and the environment, serving as the foundational data for multi-scale position coding.

[0064] The spatial feature data is obtained by multi-scale position encoding of the spatial coordinate information contained in the terahertz electromagnetic holographic tensor. This data maps the spatial coordinates to a high-dimensional feature space to accommodate the rapid spatial changes of terahertz waves, providing spatially characteristic input data for multi-layer nonlinear feature transformations and helping the model better understand the relationship between spatial structure and electromagnetic signals.

[0065] Among them, the electromagnetic low-level feature data is the data obtained after performing multi-layer nonlinear feature transformation on the spatial feature data. These data reflect some basic electromagnetic properties of terahertz waves after propagation in space and interaction with the environment, such as the electric field, magnetic field and other related features at different locations, which are important bases for further comprehensive environmental semantic reasoning.

[0066] Among them, the sub-item environmental characteristic parameter data is derived through comprehensive environmental semantic reasoning from the electromagnetic underlying characteristic data. It includes multiple parameters reflecting environmental characteristics, such as volume density, semantic labels, polarization state, temporal characteristics, and terahertz radiation intensity. These parameters describe the characteristics of objects in the environment and the state of the electromagnetic environment from different perspectives, and are a key component in constructing multi-attribute environmental characterization data.

[0067] For example, the specific methods for obtaining spatial feature data can be as follows: Terahertz waves have extremely short wavelengths and exhibit dramatic spatial variations. To accurately capture spatial information, multi-scale position coding technology is employed. This technology utilizes a combination of basis functions of sine and cosine functions to encode spatial coordinates through multiple frequency components. Different frequency components correspond to different spatial scales, thereby achieving a high-resolution representation of spatial coordinates. For example, in the case where the 0.3 terahertz band corresponds to a wavelength of 1 millimeter, and the metasurface unit spacing is typically half a wavelength (0.5 millimeters), appropriate frequency parameters are selected according to different frequency bands to map the spatial coordinates to a high-dimensional feature space, obtaining spatial feature data. This allows the model to better perceive the variations of terahertz waves in space.

[0068] For example, the specific method for obtaining electromagnetic low-level feature data can be: A multi-layer fully connected network is used to process spatial feature data. Through multi-layer nonlinear transformation, spatial feature data is combined with electromagnetic properties to extract the underlying electromagnetic information contained in the propagation of terahertz waves in space. During the processing, the network performs in-depth feature extraction and transformation on the spatial feature data by adjusting the weight parameters, obtaining electromagnetic underlying feature data that reflects the basic electromagnetic properties of terahertz waves.

[0069] For example, the specific methods for obtaining the sub-item environmental characteristic parameter data can be as follows: Based on electromagnetic low-level feature data, comprehensive semantic reasoning of the environment is performed. By analyzing the interaction between terahertz waves and the environment, relevant characteristic parameters of the environment are inferred from the electromagnetic low-level feature data. For example, by analyzing the electromagnetic low-level feature data, the volume density at a certain point in space can be predicted to reflect the probability of the presence of an object at that location; semantic tags can be inferred to identify the semantic category of environmental objects; polarization state can be obtained to infer the material properties of scattering objects; temporal features can be obtained to reflect the dynamic changes of the environment; and terahertz radiation intensity can be obtained, analogous to the radiation characteristics of light waves to describe the reflection and scattering intensity of terahertz waves at that point, thereby obtaining individual environmental characteristic parameter data.

[0070] For example, the specific methods for obtaining multi-attribute environmental representation data can be as follows: Spatial feature data, electromagnetic underlying feature data, and sub-item environmental feature parameter data obtained through different processing steps are integrated. Through preset association rules and combination methods, these data reflecting the terahertz electromagnetic environment from different perspectives are standardized to form a unified multi-attribute environmental characterization data. This data comprehensively integrates spatial, electromagnetic, and environmental semantic information, providing comprehensive and targeted data support for constructing an environmental semantic map and scheduling communication resources.

[0071] In this embodiment, multi-scale position encoding of the spatial coordinate information contained in the terahertz electromagnetic holographic tensor accurately captures the spatial variation characteristics of terahertz waves, providing high-resolution spatial feature data. Multi-layer nonlinear feature transformation further explores the deep relationship between spatial features and electromagnetic properties, obtaining electromagnetic underlying feature data reflecting fundamental electromagnetic characteristics. Environmental semantic comprehensive reasoning extracts several key environmental feature parameters from the electromagnetic underlying feature data, realizing the conversion from electromagnetic information to environmental semantics. Finally, through association, grouping, and integration processing, data from different levels are integrated into multi-attribute environmental characterization data, comprehensively and systematically describing the terahertz electromagnetic environment.

[0072] Preferably, in an embodiment, the step of performing semantic category mapping, material attribute reasoning, and spatial topological relationship parsing on the multi-attribute environmental representation data to obtain environmental object categories, material attributes, and spatial relationships, and constructing an environmental semantic graph based on the environmental object categories, material attributes, and spatial relationships, includes: The semantic category labels in the multi-attribute environment representation data are mapped to a preset semantic category library, and the category of the environment object is identified according to the semantic category library; Based on the polarization state and temporal scattering characteristics in the multi-attribute environmental characterization data, material attribute inference is performed to obtain the material attributes of each target object in the environment. Based on the volume density and the category of the environment object in the multi-attribute environment characterization data, the occlusion relationship, adjacency relationship and hierarchical relationship between each target object are analyzed, and the spatial association relationship is obtained based on the occlusion relationship, the adjacency relationship and the hierarchical relationship; The environmental object categories, material attributes, and spatial relationships are structured and associated to form the environmental semantic map.

[0073] Among them, semantic category labels are a type of identification information used to identify the semantic category of environmental objects in multi-attribute environmental representation data. They provide a basis for identifying the category of environmental objects and can clearly identify the category to which different objects in the environment belong through mapping with a preset semantic category library.

[0074] The semantic category library is a pre-defined collection containing various possible environmental object categories. The content of the semantic category library will vary in different application scenarios. For example, in a general scenario, it may include categories such as empty, person, vehicle, and building wall; in an industrial scenario, it may include categories such as worker, robot, and robotic arm. It is used to provide a reference standard for the mapping of semantic category labels and the recognition of environmental object categories.

[0075] The polarization state describes the orientation characteristics of the electric field vector during the propagation of terahertz waves. Different materials will have different effects on the polarization state of terahertz waves. Therefore, the material properties of the object can be inferred by analyzing the polarization state.

[0076] Among them, the time-domain scattering characteristics reflect the scattering response characteristics of terahertz waves in the time domain. Since objects of different materials scatter terahertz waves in different ways, the time-domain scattering characteristics also contain information about the material of the object and are often used together with the polarization state for inference of material properties.

[0077] In multi-attribute environmental characterization data, volume density represents the probability of an object existing at a certain point in space. By analyzing volume density and environmental object categories, the spatial relationships between various target objects can be further inferred.

[0078] Among them, occlusion relationship refers to a spatial positional relationship between target objects in the environment, that is, whether one object blocks the path of terahertz waves to another object. This relationship can be determined by analyzing volume density and environmental object categories.

[0079] Among them, the adjacency relationship represents the spatial proximity and adjacent relationship between target objects in the environment. It is also based on volume density and environmental object category, and is determined by judging factors such as spatial distance.

[0080] Among them, the hierarchical relationship reflects the spatial inclusion, being included, or hierarchical relationship between target objects in the environment. For example, a large object contains a small object, or one object is above another. It is also derived by analyzing volume density and environmental object categories.

[0081] In structured association modeling, environmental object categories, material attributes, and spatial relationships are organized and associated according to certain structures and rules to establish a model that forms an environmental semantic map that can clearly reflect environmental semantic information.

[0082] For example, the specific way to obtain the environment object category can be: Semantic category labels are extracted from multi-attribute environmental representation data, and then these labels are compared and matched with a pre-defined semantic category library. The semantic category library has defined various possible environmental object categories. Through this mapping and matching, the specific environmental object category corresponding to each semantic category label can be clearly identified, such as determining whether a certain area corresponds to the category of "person", "vehicle", or "building wall".

[0083] For example, the specific way to obtain material properties can be: Different materials have varying effects on the polarization and time-domain scattering characteristics of terahertz waves, resulting in different polarization states and time-domain scattering features. By utilizing the correspondence between these characteristics and materials, the polarization states and time-domain scattering features in multi-attribute environmental characterization data can be analyzed and inferred. For example, metallic objects typically cause terahertz waves to exhibit strong polarization retention characteristics, with rapid decay in the time domain; non-metallic media, on the other hand, show polarization changes, exhibiting complex scattering patterns in the time domain. By establishing such a material-feature mapping table and combining it with tools such as deep learning classifiers, the material properties of various target objects in the environment can be inferred.

[0084] For example, the specific methods for obtaining spatial relationships can be: Volume density reflects the probability of an object existing at a point in space. Combined with the identified environmental object categories, factors such as beam propagation paths, spatial distances, and spatial containment relationships are analyzed to determine the occlusion, adjacency, and hierarchical relationships between target objects. For example, if the volume density at location A indicates the presence of an object of type "building wall," and location B is of type "person," and B is located in front of A on the beam propagation path, then it can be inferred that A occludes B. If the objects at locations C and D are of the same category and the spatial distance is less than a preset threshold, then it can be inferred that they belong to the same object or are adjacent objects of the same category. If a building wall occludes a window, and the window in turn occludes furniture, a hierarchical relationship is formed. By combining these relationships, spatial associations are obtained.

[0085] For example, the specific methods for obtaining an environmental semantic graph can be as follows: The identified environmental object categories, inferred material properties, and analyzed spatial relationships are organized and associated according to a certain logical structure and rules. For example, an environmental object category can be used as a node, and its corresponding material properties and spatial relationships can be used as the attributes and connections of the node to construct a graphical model. This model is the environmental semantic graph, which comprehensively and intuitively displays the semantic information of each object in the environment and the relationships between them.

[0086] In this embodiment, a deep semantic understanding of the terahertz electromagnetic environment is achieved by performing semantic category mapping, material attribute reasoning, and spatial topological relationship analysis on multi-attribute environmental characterization data, and constructing an environmental semantic graph. Semantic category mapping accurately identifies the categories of environmental objects, clearly identifying the types of objects in the environment; material attribute reasoning determines the material of objects based on polarization state and temporal scattering characteristics, further enriching the understanding of environmental objects; spatial correlation analysis using volume density and environmental object categories clarifies the relative positions of objects in space. The above information is then structured and modeled to form an environmental semantic graph, providing comprehensive, detailed, and accurate environmental context support for communication resource scheduling. This enables the communication system to allocate resources more rationally and optimize beam scheduling based on information such as the category, material, and spatial layout of objects in the environment, thereby improving the performance of the communication system in complex indoor environments, ensuring communication stability and efficiency, and effectively solving the problem of unreasonable resource scheduling caused by insufficient environmental understanding in terahertz communication.

[0087] Preferably, in the embodiment, before performing multi-objective semantic utility optimization based on the environmental semantic map, preset communication task semantic attributes, and channel state information extracted from the terahertz electromagnetic holographic tensor to obtain the beam scheduling configuration scheme, the method further includes: The amplitude features, phase features, polarization features, time domain features, and frequency domain features of the terahertz electromagnetic holographic tensor are extracted to obtain discrete electromagnetic feature data; The discrete electromagnetic feature data are fused using multimodal methods to obtain fused channel features; Multipath channel parameters are fitted to the fused channel features to obtain the basic channel parameters. The channel state information is obtained by performing state regression correction on the channel basic parameters.

[0088] The amplitude characteristic, which reflects the signal intensity distribution in the terahertz electromagnetic holographic tensor, is typically represented by a logarithmic domain amplitude, with the unit being decibels (dB). It visually demonstrates the changes in signal intensity during terahertz wave propagation. Different environmental scenarios and objects can cause changes in the amplitude characteristic due to reflection and absorption of terahertz waves, making it crucial for analyzing signal propagation losses and the influence of objects in the environment on the signal.

[0089] Phase characteristics are features that characterize the phase changes of terahertz wave signals, including instantaneous phase and phase difference. Instantaneous phase reflects the phase state of the signal at a specific moment, while phase difference reflects the rate of phase change over time. Phase characteristics play a crucial role in determining signal propagation paths and analyzing interference and scattering phenomena in the environment, helping to accurately understand the state changes of terahertz waves during propagation.

[0090] Polarization features describe the polarization state of terahertz waves and are represented using Stokes parameters. They include information such as horizontal and vertical polarization differences, linear polarization, and circular polarization components. Polarization features can be used to infer the material properties of objects in the environment. Different materials have different effects on the polarization characteristics of terahertz waves, and analyzing polarization features can yield important information about objects in the environment.

[0091] Among these, time-domain characteristics reflect the statistical properties of terahertz waves over time, typically represented by power spectral density and autocorrelation function. Power spectral density reflects the distribution of signal power with frequency, while the autocorrelation function describes the similarity of the signal at different times. Time-domain characteristics help to understand the stability, periodicity, and dynamic changes of terahertz wave signals after interaction with the environment.

[0092] Frequency domain characteristics represent the properties of terahertz waves in the frequency domain, including parameters such as Doppler shift and spectral flatness. Doppler shift reflects the change in signal frequency caused by the relative motion of an object and can be used to detect the motion state of an object; spectral flatness describes the uniformity of the signal spectrum and plays an important role in evaluating signal quality and channel frequency selectivity.

[0093] Among them, discrete electromagnetic feature data is a set of data composed of amplitude features, phase features, polarization features, time domain features and frequency domain features extracted from the terahertz electromagnetic holographic tensor. These features describe the electromagnetic properties of terahertz waves from different aspects, but they are in an independent state after extraction and need to be further fused.

[0094] Multimodal fusion is the process of integrating discrete electromagnetic feature data of different types (modes) to fully utilize the advantages of each feature and obtain more comprehensive and representative channel characteristics. Through preset algorithms and models, features such as amplitude, phase, polarization, time domain, and frequency domain are organically combined, so that the fused data can more accurately reflect the complex characteristics of terahertz channels.

[0095] Among them, the fused channel features are data features obtained after multi-mode fusion. They integrate multiple information from discrete electromagnetic feature data. Compared with single features, they can more comprehensively describe the state of the terahertz channel and provide a richer and more accurate data foundation for channel parameter fitting and state estimation.

[0096] Multipath channel parameters describe the multipath propagation characteristics of terahertz channels, including amplitude attenuation coefficient, time delay, and angle of arrival. Terahertz waves encounter reflections and scattering from various objects during propagation, forming multiple propagation paths. These parameters characterize the signal properties along each path and are crucial for accurately modeling terahertz channels.

[0097] Among them, the channel fundamental parameters are data obtained by fitting multipath channel parameters to the fused channel characteristics. It initially determines the basic characteristic parameters of the terahertz channel and provides a basis for further accurate description of the channel state, but there may still be some errors, which need to be corrected.

[0098] State regression correction is a process of adjusting and optimizing the basic channel parameters. By using a preset algorithm and reference standard, the basic channel parameters are compared and corrected with the actual channel state, so that the obtained channel state information more accurately reflects the real situation of the terahertz channel.

[0099] For example, the specific method for obtaining discrete electromagnetic characteristic data can be as follows: Using pre-defined signal processing algorithms and techniques, features in different aspects, such as amplitude, phase, polarization, time domain, and frequency domain, are extracted from the terahertz electromagnetic holographic tensor. For example, a pre-defined transformation algorithm is used to obtain the logarithmic domain amplitude representation of the amplitude feature; phase detection technology is used to obtain the instantaneous phase and phase difference of the phase feature; polarization analysis methods are used to derive the Stokes parameters of the polarization feature; power spectral density and autocorrelation function are calculated through time-domain statistical analysis to obtain time-domain features; and frequency-domain analysis methods are used to obtain frequency-domain features such as Doppler shift and spectral flatness. These features, which describe the electromagnetic properties of terahertz waves from different dimensions, are combined to form discrete electromagnetic feature data.

[0100] For example, the specific method for obtaining fused channel features can be: Multimodal deep neural networks and other technologies are employed to input discrete electromagnetic feature data into the network for processing. Different modules within the network analyze and process different features. For example, the Transformer module (a deep learning model architecture based on a self-attention mechanism) is used to capture long-range dependencies, grasping the overall relationship between features; convolutional neural networks are used to extract local features, mining spatial information from amplitude and phase images; and LSTM (Long Short-Term Memory) neural networks are used to perform time-series modeling of time-domain and frequency-domain features, capturing temporal dynamics. Then, through a pre-defined fusion strategy, these processed features are integrated to form a fused channel feature, which comprehensively reflects the multifaceted characteristics of the terahertz channel.

[0101] For example, the specific method for obtaining basic channel parameters can be as follows: Based on the rich information contained in the fused channel features, a fitting algorithm is used to estimate the multipath channel parameters. By analyzing the intrinsic relationship between the fused channel features and the multipath channel parameters, such as using information on signal strength, phase change, and polarization characteristics, parameters like amplitude attenuation coefficient, time delay, and angle of arrival are inferred. By continuously adjusting the fitting parameters, the estimated multipath channel parameters are made to reflect the multipath propagation characteristics of the terahertz channel as accurately as possible, thus obtaining the fundamental channel parameters.

[0102] For example, the specific methods for obtaining channel state information can be as follows: Using actual measurement data or known channel models as reference standards, state regression correction is performed on the fundamental channel parameters. By comparing the differences between the fundamental channel parameters and the reference standard, a regression algorithm is used to adjust the fundamental channel parameters so that the adjusted parameters better reflect the actual state of the terahertz channel, ultimately obtaining accurate channel state information. For example, if a deviation is found between the estimated time delay value and the actual measured value in the fundamental channel parameters, the time delay parameter is corrected using a regression algorithm to improve the accuracy of the channel state information.

[0103] In this embodiment, multiple discrete electromagnetic feature data are extracted from the terahertz electromagnetic holographic tensor, and multimodal fusion, multipath channel parameter fitting, and state regression correction are performed to obtain accurate channel state information. Extracting multiple discrete electromagnetic feature data comprehensively captures the electromagnetic characteristics of terahertz waves in different dimensions, providing rich material for in-depth channel analysis. Multimodal fusion organically combines these discrete features, fully leveraging the advantages of each feature to obtain a more comprehensive and accurate fused channel feature reflecting channel characteristics. Multipath channel parameter fitting estimates the basic parameters of the channel based on the fused channel feature, laying the foundation for describing the channel state. State regression correction further optimizes these parameters to better match the actual channel state. Accurate channel state information provides a reliable basis for solving multi-objective semantic utility optimization, helps to formulate more reasonable beam scheduling configuration schemes, improves the adaptability of communication systems to channels in complex environments, optimizes communication resource allocation, and thus improves communication quality and system performance.

[0104] Preferably, in the embodiment, the step of performing multi-objective semantic utility optimization based on the environmental semantic map, preset communication task semantic attributes, and channel state information extracted from the terahertz electromagnetic holographic tensor to obtain a beam scheduling configuration scheme includes: Based on the operational constraints of communication services, a multi-dimensional semantic comprehensive utility function is constructed, which includes throughput utility, semantic matching utility, link robustness utility, and resource allocation fairness utility. Feature association matching is performed on the environmental semantic map, communication task semantic attributes, and channel state information to obtain optimized input feature data; With transmit power, time slot resources, and service quality as constraints, and based on the optimized input feature data, the multi-dimensional semantic comprehensive utility function is optimized by multi-objective semantic utility to obtain the global optimal solution set; According to the preset business decision rules, the optimal configuration is selected from the global optimal solution set, and the beam scheduling configuration scheme is determined based on the optimal configuration.

[0105] Among them, communication service operation constraints refer to the various restrictions that must be followed to ensure the normal and efficient operation of the communication system during the actual operation of communication services. These conditions cover multiple aspects, such as the performance limitations of system hardware equipment, the standard requirements stipulated by communication protocols, and the pre-set requirements of the service itself for communication quality, resource usage, etc., and are an important basis for constructing a multi-dimensional semantic comprehensive utility function.

[0106] The multi-dimensional semantic comprehensive utility function is a function that comprehensively considers various communication-related factors. By integrating throughput utility, semantic matching utility, link robustness utility, and resource allocation fairness utility, it comprehensively measures the performance of a communication system in different aspects. This function provides a quantifiable objective for solving multi-objective semantic utility optimization, enabling a comprehensive balance of various needs in complex communication environments to achieve better communication results.

[0107] Throughput utility is a measure of the overall data transmission capacity of a communication system. In communication, higher throughput means more data can be transmitted per unit of time, making it a crucial indicator of communication system performance. Throughput utility is calculated by considering factors such as communication channel capacity and data transmission rate to evaluate the effectiveness of the communication system in data transmission.

[0108] Semantic matching utility is a utility metric that measures the degree of matching between the semantic attributes of a communication task and the semantic features of its environment. Different communication tasks have different semantic requirements; for example, emergency communication may have high requirements for reliability and real-time performance, while video streaming communication has high requirements for bandwidth and stability. Communication performance is better when the semantic attributes of the communication task match the semantic features of its environment. Semantic matching utility quantifies the degree of correlation between the two, providing a basis for optimizing communication resource allocation and ensuring that resource allocation aligns with task requirements.

[0109] Link robustness utility is a metric used to evaluate the ability of a communication link to maintain stable communication in the face of adverse factors such as environmental changes and interference. Terahertz communication environments are complex and variable, and signals may be affected by obstruction, multipath effects, and other factors. Link robustness utility considers environmental dynamics and the rate of change in obstruction, and by quantitatively analyzing the stability of the communication link under various conditions, it helps optimize communication schemes and improve the anti-interference capability and reliability of communication systems in complex environments.

[0110] Resource allocation fairness utility is a utility metric that ensures fair treatment for different users or communication tasks when acquiring communication resources. In multi-user access scenarios, each user or task has its own communication needs. If resource allocation is unreasonable, some users or tasks may not receive enough resources, affecting the overall communication effect. Resource allocation fairness utility, through certain algorithms and rules, ensures that resources are allocated reasonably while meeting the basic needs of all parties, so that different users or tasks have fairness in acquiring communication resources.

[0111] The optimized input feature data is obtained by feature association and matching of the environmental semantic graph, the semantic attributes of the communication task, and the channel state information. This data integrates key information from the three aspects of environment, task, and channel, providing comprehensive and targeted input for optimizing the multi-dimensional semantic comprehensive utility function. This allows the optimization process to fully consider various factors in actual communication scenarios, thereby obtaining results that better meet practical needs.

[0112] The global optimal solution set is a set of solutions obtained by performing multi-objective semantic utility optimization on the multi-dimensional semantic comprehensive utility function under constraints such as transmit power, time slot resources, and service quality. This set of solutions represents different resource allocation schemes that can optimize the multi-dimensional semantic comprehensive utility function while satisfying various constraints. Since multi-objective optimization problems usually involve multiple conflicting objectives, the global optimal solution set includes various possible schemes that weigh different objectives.

[0113] The preset business decision rules are a series of rules pre-defined based on the characteristics and requirements of different communication services. These rules are used to select the configuration scheme that best meets the current business needs from the global optimal solution set. These rules may be based on semantic attributes such as service priority, latency tolerance, and reliability requirements. For example, for high-priority and latency-sensitive services, configuration schemes that guarantee low latency are prioritized; for services with extremely high reliability requirements, configuration schemes with strong link robustness are prioritized. Through these rules, the optimal solution that meets the preset business requirements can be quickly and accurately selected from numerous possible resource configuration schemes.

[0114] For example, the specific way to construct a multi-dimensional semantic comprehensive utility function can be as follows: This study delves into the various constraints and requirements of communication services during actual operation, such as power limitations of hardware equipment, time slot allocation rules, and specific service quality requirements of different services. Based on these constraints, the calculation methods and weight allocations for throughput utility, semantic matching utility, link robustness utility, and resource allocation fairness utility are determined. For example, for services with high real-time requirements, the weights of semantic matching utility and link robustness utility may be appropriately increased; for big data transmission services, throughput utility may be given more emphasis. These four utility indicators are then combined according to certain mathematical relationships to construct a multi-dimensional semantic comprehensive utility function, which can comprehensively reflect the overall performance requirements of communication services in complex environments.

[0115] For example, the specific method for obtaining optimized input feature data can be: This study analyzes the environmental semantic graph, including object categories, material attributes, and spatial relationships; the semantic attributes of communication tasks, such as task priority, latency tolerance, and reliability requirements; and the channel state information, including amplitude characteristics, phase characteristics, and multipath parameters. It seeks to identify the intrinsic connections and mutual influences among these information; for example, environmental occlusion may affect channel multipath parameters, thereby impacting the reliability requirements of the communication task. Using pre-defined algorithms and models, this information is correlated and matched to extract key features that significantly influence the optimization of communication resource allocation, forming the optimization input feature data.

[0116] For example, the specific method for obtaining the globally optimal solution set can be: The maximum transmit power, the total amount of time slot resources, and the minimum service quality standard are used as constraints. The optimized input feature data is substituted into a multi-dimensional semantic comprehensive utility function, and multi-objective optimization algorithms, such as Pareto optimality and weighted summation methods, are used to solve the function. During the solution process, the algorithm tries different resource allocation schemes while satisfying the constraints of transmit power, time slot resources, and service quality, in order to find the solution that optimizes the multi-dimensional semantic comprehensive utility function. Due to the complexity of the multi-objective optimization problem, a set of solutions, namely the global optimal solution set, is ultimately obtained. These solutions represent various resource allocation schemes that weigh different objectives, providing multiple choices to meet different business needs.

[0117] For example, the specific method for obtaining the beam scheduling configuration scheme can be as follows: Based on pre-defined business decision-making rules, for example, in industrial control services, if the task priority is greater than or equal to a certain value and there is permanent obstruction, the solution with a preset emergency beam configuration is selected first; for video streaming services, the solution that can improve throughput and ensure a certain degree of fairness is selected first. The global optimal solution set is filtered according to these rules, selecting the configuration that best meets the current business needs from numerous resource configuration schemes. Then, based on this optimal configuration, specific beam scheduling configuration schemes such as beamforming parameters, power allocation, and time slot allocation of the holographic intelligent metasurface are determined to achieve the best communication effect for the preset services.

[0118] In this embodiment, the construction of a multi-dimensional semantic comprehensive utility function comprehensively considers the performance requirements of the communication system in multiple key aspects such as throughput, semantic matching, link robustness, and resource allocation fairness, making the optimization objectives more comprehensive and reasonable. Feature association matching of environment, task, and channel information provides rich and targeted input data for optimization, enabling the optimization process to be closely integrated with actual communication scenarios. Multi-objective optimization is performed with transmit power, time slot resources, and service quality as constraints, ensuring that the obtained resource configuration scheme achieves optimal multi-dimensional semantic comprehensive utility while meeting the actual system limitations. Optimal configuration is selected from the global optimal solution set according to preset service decision rules, enabling precise selection of the most suitable resource configuration for different communication service needs. The determined beam scheduling configuration scheme can fully leverage the performance of the communication system, improve communication quality, and meet the diverse needs of different services in complex industrial environments.

[0119] Preferably, in the embodiment, after configuring the beamforming parameters of the holographic intelligent metasurface according to the beam scheduling configuration scheme to complete the environmental semantic reconstruction and communication control, the method further includes: Collect current environmental semantic features and terahertz channel features, and perform statistical distribution comparison on the current environmental semantic features and terahertz channel features to obtain feature distribution difference data; Drift detection is performed on the feature distribution difference data to obtain the drift detection results; When the drift detection result indicates the presence of feature distribution drift, the parameters of the terahertz neural radiation field network are adjusted using on-site measured samples to obtain an updated terahertz neural radiation field network. The updated terahertz neural radiation field network is used for multi-scale position encoding, multi-layer nonlinear feature transformation, and comprehensive environmental semantic reasoning.

[0120] Among them, the current environmental semantic features refer to the features obtained at the current moment through the perception and analysis of the environment, which can describe the semantic information such as the category, material, and spatial relationship of objects in the environment. These features are derived from the real-time monitoring and processing of the terahertz electromagnetic environment, reflect the current state of the environment, and are of great significance for judging whether the environment has changed.

[0121] Terahertz channel characteristics are a set of parameters characterizing the current properties of a terahertz communication channel, including signal amplitude, phase, polarization, and characteristics in the time and frequency domains. Terahertz channel characteristics change with environmental variations and various factors during communication. Real-time acquisition of these characteristics helps to understand the real-time state of the channel, thus providing a basis for communication control.

[0122] In the statistical distribution comparison, the statistical characteristics of the current environmental semantic features and the terahertz channel features are compared, such as the mean, variance, and probability distribution, to determine the degree of difference between the two sets of data. This comparison can quantify the changes in environmental and channel features.

[0123] Among them, the feature distribution difference data is obtained by statistically comparing the current environmental semantic features and terahertz channel features, and is used to describe the differences between the two. These data reflect the changes in the statistical properties of environmental semantics and channel features, and are an important basis for judging whether the environment and channel have changed significantly.

[0124] Drift detection involves analyzing feature distribution difference data to determine whether the distribution of environmental semantic features and terahertz channel features has undergone significant and persistent changes, i.e., whether drift has occurred. Through pre-defined algorithms and indicators, the difference data is evaluated to determine whether the feature distribution deviates from the normal range, thus yielding the drift detection result.

[0125] The drift detection results are conclusions drawn from drift detection, indicating whether there is a drift in the distribution of environmental semantic features and terahertz channel features. If the detection results show that drift exists, it means that the environment or channel has changed significantly, and the relevant models need to be adjusted; if the results show that there is no drift, it means that the environment and channel are relatively stable, and the current communication control strategies and model parameters may still be applicable.

[0126] Among them, the field-measured samples are sample data on environmental semantics and terahertz channels obtained through real-time field measurements. These sample data truly reflect the actual situation of the current environment and channel, and have important reference value for adjusting the parameters of the terahertz neural radiation field network, enabling the network to better adapt to changes in the environment and channel.

[0127] For example, the specific method for obtaining feature distribution difference data can be: By utilizing sensors and data acquisition equipment deployed on-site, real-time environmental semantic features and terahertz channel features are acquired. For example, terahertz wave-related information is collected through a holographic intelligent metasurface array, and after processing, environmental semantic features are obtained, such as identifying object categories and material properties in the current environment; simultaneously, the amplitude and phase characteristics of the terahertz channel at the current moment are acquired. Then, statistical analysis methods are used to compare the statistical distributions of these two sets of features in detail. For example, their mean and variance are calculated, and probability distributions are analyzed. The comparison results are presented in data form, yielding feature distribution difference data. This data can intuitively reflect the changes in environmental semantics and channel features compared to previous states.

[0128] For example, the specific method for obtaining drift detection results can be: Pre-defined drift detection algorithms are used to perform in-depth analysis of feature distribution difference data. These algorithms are typically based on statistical principles and machine learning methods, setting certain thresholds and judgment criteria. For example, by comparing the statistical characteristics of current difference data with historical data, or by using hypothesis testing methods, it is determined whether the distribution of environmental semantic features and terahertz channel features has undergone significant and persistent changes. If the difference data exceeds the set threshold range, or meets the preset drift judgment conditions, feature distribution drift is determined to exist, and the corresponding drift detection result is obtained; otherwise, it is considered that no drift exists.

[0129] For example, the updated terahertz neural radiation field network can be obtained in the following ways: If drift detection results show characteristic distribution drift, it indicates a significant change in the environment or channel, and the current terahertz neural radiation field network may no longer be suitable for the new situation. In this case, the network parameters are adjusted using field-measured samples. The field-measured samples are input into the terahertz neural radiation field network, and through a pre-set optimization algorithm, such as gradient descent, the network's weights and biases are adjusted based on the difference between the sample data and the network's prediction results. After a series of parameter adjustments, an updated terahertz neural radiation field network is obtained. This updated network can better adapt to the new environmental semantics and channel characteristics, and continues to perform multi-scale position encoding, multi-layer nonlinear feature transformation, and comprehensive environmental semantic reasoning on the collected data to ensure the accuracy and effectiveness of environmental semantic reconstruction and communication control.

[0130] In this embodiment, real-time acquisition of environmental semantics and channel features enables timely capture of changes in the environment and channel, providing an accurate data foundation. Statistical distribution comparison and drift detection accurately determine whether these changes have reached the point where model adjustments are necessary, avoiding unnecessary model adjustments and improving system stability. When feature distribution drift is detected, the terahertz neural radiation field network parameters are adjusted using field-measured samples, enabling the network to quickly adapt to new environmental and channel conditions. This allows for continued accurate multi-scale position coding, multi-layer nonlinear feature transformation, and comprehensive environmental semantic reasoning, ensuring the accuracy of environmental semantic reconstruction and the effectiveness of communication control. This allows the communication system to operate stably and continuously in dynamically changing environments.

[0131] To better implement the above methods, this application also provides an artificial intelligence-based environmental semantic reconstruction device, which can be integrated into electronic devices, such as servers or terminals. The terminal may include tablet computers, laptop computers, and / or personal computers.

[0132] For example, see Figure 2 , Figure 2 This is a schematic diagram of the structure of the artificial intelligence-based environmental semantic reconstruction device provided in an embodiment of the present invention, as shown below. Figure 2 As shown, the AI-based environmental semantic reconstruction device may include an acquisition module 201, an arrangement module 202, an inference module 203, a graph construction module 204, and a parameter configuration module 205, as follows: The acquisition module 201 is used to acquire the downlink pilot signal transmitted by the terahertz base station to the communication coverage area, and to collect the amplitude data, phase data and polarization data corresponding to the downlink pilot signal through the holographic sensing module mounted on each reflection unit in the holographic intelligent metasurface array, so as to obtain multi-dimensional electromagnetic raw data. Arrangement module 202 is used to arrange the multi-dimensional electromagnetic raw data according to the metasurface unit spatial coordinates and signal sampling time to obtain a terahertz electromagnetic holographic tensor. Inference module 203 is used to input the terahertz electromagnetic holographic tensor into a pre-trained terahertz neural radiation field network to perform multi-scale position encoding, multi-layer nonlinear feature transformation and environmental semantic comprehensive reasoning to obtain multi-attribute environmental representation data. The graph construction module 204 is used to perform semantic category mapping, material attribute reasoning, and spatial topological relationship parsing on the multi-attribute environmental representation data to obtain environmental object categories, material attributes, and spatial relationships, and to construct an environmental semantic graph based on the environmental object categories, material attributes, and spatial relationships. The parameter configuration module 205 is used to perform multi-objective semantic utility optimization solution based on the environmental semantic map, the preset communication task semantic attributes and the channel state information extracted by the terahertz electromagnetic holographic tensor, to obtain a beam scheduling configuration scheme, and to configure the beamforming parameters of the holographic intelligent metasurface according to the beam scheduling configuration scheme, so as to complete the environmental semantic reconstruction and communication control.

[0133] In practice, the above modules can be implemented as independent entities or combined in any way to be implemented as the same or several entities. For the specific implementation of the above modules, please refer to the previous method implementation examples, which will not be repeated here.

[0134] As described above, in this embodiment, the holographic sensing module mounted on each reflective unit of the holographic intelligent metasurface array collects amplitude, phase, and polarization data corresponding to the downlink pilot signal transmitted by the terahertz base station, obtaining multi-dimensional raw electromagnetic data containing rich electromagnetic information. Arranging this multi-dimensional raw electromagnetic data into a terahertz electromagnetic holographic tensor according to the metasurface unit spatial coordinates and signal sampling time provides an ordered data structure. Inputting the terahertz electromagnetic holographic tensor into a pre-trained terahertz neural radiation field network, through multi-scale position encoding, multi-layer nonlinear feature transformation, and comprehensive environmental semantic reasoning, yields multi-attribute environmental representation data. This data is then analyzed in depth to uncover the environmental semantics it contains. Semantic category mapping, material attribute reasoning, and spatial topological relationship analysis are performed on the multi-attribute environmental representation data to construct an environmental semantic map, clarifying the environmental object categories, material attributes, and spatial relationships, thus achieving a semantic-level understanding of the terahertz electromagnetic environment. Based on the environmental semantic map, the preset semantic attributes of communication tasks, and the channel state information extracted from the terahertz electromagnetic holographic tensor, a multi-objective semantic utility optimization solution is performed to obtain a beam scheduling configuration scheme and configure beamforming parameters. This provides comprehensive environmental context support for communication resource scheduling and realizes a close integration of environmental semantic reconstruction and communication control.

[0135] In some embodiments, the AI-based environmental semantic reconstruction device of the present invention can be implemented in a combination of hardware and software. As an example, the AI-based environmental semantic reconstruction device of the present invention can be a processor in the form of a hardware decoding processor, which is programmed to execute the AI-based environmental semantic reconstruction method of the present invention. For example, the processor in the form of a hardware decoding processor can be one or more application-specific integrated circuits (ASICs), DSPs, programmable logic devices (PLDs), complex programmable logic devices (CPLDs), field-programmable gate arrays (FPGAs), or other electronic components.

[0136] The modules described in the embodiments of this invention can be implemented in software or hardware. The names of the modules are not, in some cases, limiting the scope of the module itself.

[0137] An electronic device according to an embodiment of the present invention includes a processor and a memory, wherein the memory stores multiple instructions; the processor loads instructions from the memory to execute the steps in the artificial intelligence-based environmental semantic reconstruction method described above. That is, an electronic device according to an embodiment of the present invention may include, but is not limited to: a processor and a memory; the memory is used to store a computer program; the processor is used to execute the artificial intelligence-based environmental semantic reconstruction method shown in any embodiment of the present invention by calling the computer program.

[0138] In one alternative embodiment, an electronic device is provided, such as Figure 3 As shown, Figure 3 The illustrated electronic device 3000 includes a processor 3001 and a memory 3003. The processor 3001 and the memory 3003 are connected, for example, via a bus 3002. Optionally, the electronic device 3000 may further include a transceiver 3004, which can be used for data interaction between the electronic device and other electronic devices, such as sending and / or receiving data. It should be noted that in practical applications, the transceiver 3004 is not limited to one type, and the structure of the electronic device 3000 does not constitute a limitation on the embodiments of the present invention.

[0139] Processor 3001 may be a CPU (Central Processing Unit), a general-purpose processor, a DSP (Digital Signal Processor), an ASIC (Application Specific Integrated Circuit), an FPGA (Field Programmable Gate Array), or other programmable logic devices, transistor logic devices, hardware components, or any combination thereof. It can implement or execute the various exemplary logic blocks, modules, and circuits described in conjunction with the disclosure of this invention. Processor 3001 may also be a combination that implements computational functions, such as including one or more microprocessor combinations, a combination of a DSP and a microprocessor, etc.

[0140] Bus 3002 may include a path for transmitting information between the aforementioned components. Bus 3002 may be a PCI (Peripheral Component Interconnect) bus or an EISA (Extended Industry Standard Architecture) bus, etc. Bus 3002 can be divided into address bus, data bus, control bus, etc. For ease of representation, Figure 3The bus 3002 is represented by only one thick line, but this does not mean that there is only one bus or one type of bus.

[0141] The memory 3003 may be ROM (Read Only Memory) or other types of static storage devices capable of storing static information and instructions, RAM (Random Access Memory) or other types of dynamic storage devices capable of storing information and instructions, or EEPROM (Electrically Erasable Programmable Read Only Memory), CD-ROM (Compact Disc Read Only Memory) or other optical disc storage, optical disc storage (including compressed optical discs, laser discs, optical discs, digital universal optical discs, Blu-ray discs, etc.), magnetic disk storage media or other magnetic storage devices, or any other medium capable of carrying or storing desired program code in the form of instructions or data structures and accessible by a computer, but not limited thereto.

[0142] The memory 3003 stores application code (computer program) for executing the present invention, and its execution is controlled by the processor 3001. The processor 3001 executes the application code stored in the memory 3003 to implement the content shown in the foregoing method embodiments.

[0143] Among them, electronic devices can also be terminal devices, which can be any device that can install applications, including at least one of smartphones, tablets, laptops, desktop computers, smart speakers, smartwatches, smart TVs, and smart in-vehicle devices.

[0144] It should be noted that, Figure 3 The electronic device shown is merely an example and should not be construed as limiting the functionality and scope of use of the embodiments of the present invention.

[0145] An embodiment of the present invention provides a computer-readable storage medium storing a plurality of instructions adapted for loading by a processor to execute the steps in any of the above-described artificial intelligence-based environmental semantic reconstruction methods.

[0146] Alternatively, the computer-readable storage medium may be a read-only memory (ROM), a random access memory (RAM), a compact disc read-only memory (CD-ROM), magnetic tape, a floppy disk, and an optical data storage device, etc.

[0147] In an exemplary embodiment, a computer program product or computer program is also provided, which includes computer instructions stored in a computer-readable storage medium. A processor of an electronic device reads the computer instructions from the computer-readable storage medium and executes the computer instructions, causing the electronic device to perform the steps in the AI-based environmental semantic reconstruction method described above.

[0148] Computer program code for performing the operations of this invention can be written in one or more programming languages ​​or a combination thereof, including object-oriented programming languages ​​such as Java, Smalltalk, and C++, and conventional procedural programming languages ​​such as C or similar languages. The program code can be executed entirely on the user's computer, partially on the user's computer, as a standalone software package, partially on the user's computer and partially on a remote computer, or entirely on a remote computer or server. In cases involving remote computers, the remote computer can be connected to the user's computer via any type of network—including a local area network (LAN) or a wide area network (WAN)—or can be connected to an external computer (e.g., via the Internet using an Internet service provider).

[0149] It should be understood that the flowcharts and block diagrams in the accompanying drawings illustrate the implementable architecture, functions, and operations of methods and computer program products according to various embodiments of the present invention. In this regard, each block in a flowchart or block diagram may represent a module, segment, or portion of code, which includes one or more executable instructions for implementing the specified logical function. It should also be noted that in some alternative implementations, the functions indicated in the blocks may occur in a different order than those indicated in the drawings. For example, two consecutively indicated blocks may actually be executed substantially in parallel, and they may sometimes be executed in reverse order, depending on the functions involved. It should also be noted that each block in the block diagrams and / or flowcharts, and combinations of blocks in the block diagrams and / or flowcharts, can be implemented using a dedicated hardware-based system that performs the specified function or operation, or using a combination of dedicated hardware and computer instructions.

[0150] The computer-readable storage medium provided in this invention can be, but is not limited to, an electrical, magnetic, optical, electromagnetic, infrared, or semiconductor system, apparatus, or device, or any combination thereof. More specific examples of a computer-readable storage medium may include, but are not limited to: an electrical connection having one or more wires, a portable computer disk, a hard disk, random access memory (RAM), read-only memory (ROM), erasable programmable read-only memory (EEPROM or flash memory), optical fiber, portable compact disk read-only memory (CD-ROM), optical storage device, magnetic storage device, or any suitable combination thereof. In this invention, a computer-readable storage medium can be any tangible medium that includes or stores a program that can be used by or in conjunction with an instruction execution system, apparatus, or device.

[0151] The aforementioned computer-readable storage medium carries one or more programs, which, when executed by the electronic device, cause the electronic device to perform the method shown in the above embodiments.

[0152] The above description is merely a preferred embodiment of the present invention and an explanation of the technical principles employed. Those skilled in the art should understand that the scope of disclosure in this invention is not limited to technical solutions formed by specific combinations of the above-described technical features, but should also cover other technical solutions formed by arbitrary combinations of the above-described technical features or their equivalents without departing from the above-disclosed concept. For example, technical solutions formed by substituting the above features with (but not limited to) technical features with similar functions disclosed in this invention.

[0153] It should be noted that the terms "first," "second," etc., used in the specification and claims of this application are used to distinguish similar objects and represent a limitation on a specific order or sequence. Where appropriate, the order of use for similar objects can be interchanged so that the embodiments of this application described herein can be implemented in an order other than that shown or described.

[0154] Those skilled in the art will recognize that this invention can be implemented as a system, method, or computer program product. Therefore, this invention can be specifically implemented in the following forms: it can be entirely hardware, entirely software (including firmware, resident software, microcode, etc.), or a combination of hardware and software, generally referred to herein as a "circuit," "module," or "system." Furthermore, in some embodiments, this invention can also be implemented as a computer program product in one or more computer-readable media, the computer-readable media including computer-readable program code.

[0155] Although embodiments of the present invention have been shown and described above, it is understood that the above embodiments are exemplary and should not be construed as limiting the present invention. Those skilled in the art can make changes, modifications, substitutions and variations to the above embodiments within the scope of the present invention.

Claims

1. An environment semantic reconstruction method based on artificial intelligence, characterized in that, include: The downlink pilot signal transmitted by the terahertz base station to the communication coverage area is acquired, and the amplitude data, phase data, and polarization data corresponding to the downlink pilot signal are collected by the holographic sensing module mounted on each reflection unit in the holographic intelligent metasurface array to obtain multi-dimensional electromagnetic raw data. The multi-dimensional electromagnetic raw data is arranged by tensor regularization according to the metasurface unit spatial coordinates and signal sampling time to obtain the terahertz electromagnetic holographic tensor. The terahertz electromagnetic holographic tensor is input into a pre-trained terahertz neural radiation field network to perform multi-scale position encoding, multi-layer nonlinear feature transformation, and environmental semantic comprehensive reasoning to obtain multi-attribute environmental representation data. Semantic category mapping, material attribute reasoning, and spatial topological relationship parsing are performed on the multi-attribute environmental representation data to obtain environmental object categories, material attributes, and spatial relationships. Based on the environmental object categories, material attributes, and spatial relationships, an environmental semantic graph is constructed. Based on the environmental semantic map, the preset semantic attributes of communication tasks, and the channel state information extracted from the terahertz electromagnetic holographic tensor, a multi-objective semantic utility optimization solution is performed to obtain a beam scheduling configuration scheme. The beamforming parameters of the holographic intelligent metasurface are then configured according to the beam scheduling configuration scheme to complete the environmental semantic reconstruction and communication control.

2. The method for reconstructing environmental semantics based on artificial intelligence according to claim 1, characterized in that, The acquisition of amplitude data, phase data, and polarization data corresponding to the downlink pilot signal yields multi-dimensional raw electromagnetic data, including: The electric field strength of the downlink pilot signal is detected by induction to obtain the amplitude data corresponding to the downlink pilot signal; The phase of the signal corresponding to the amplitude data is synchronously sampled to obtain the phase data corresponding to the downlink pilot signal; The polarization state of the electromagnetic wave corresponding to the phase data is identified and detected to obtain the polarization data corresponding to the downlink pilot signal; The amplitude data, phase data, and polarization data are time-aligned and fused to obtain the multi-dimensional electromagnetic raw data.

3. The method for reconstructing environmental semantics based on artificial intelligence according to claim 1, characterized in that, The process involves multi-scale location encoding, multi-layer nonlinear feature transformation, and comprehensive environmental semantic reasoning to obtain multi-attribute environmental representation data, including: Multi-scale position encoding is performed on the spatial coordinate information contained in the terahertz electromagnetic holographic tensor to obtain spatial feature data; The spatial feature data is subjected to multi-layer nonlinear feature transformation to obtain electromagnetic bottom-layer feature data; Environmental semantic comprehensive reasoning is performed on the electromagnetic low-level feature data to obtain sub-item environmental feature parameter data; The spatial feature data, the electromagnetic underlying feature data, and the sub-item environmental feature parameter data are correlated, grouped, and rectified to obtain the multi-attribute environmental characterization data.

4. The method for reconstructing environmental semantics based on artificial intelligence according to claim 1, characterized in that, The process of performing semantic category mapping, material attribute reasoning, and spatial topological relationship parsing on the multi-attribute environmental representation data to obtain environmental object categories, material attributes, and spatial relationships, and constructing an environmental semantic graph based on the environmental object categories, material attributes, and spatial relationships, includes: The semantic category labels in the multi-attribute environment representation data are mapped to a preset semantic category library, and the category of the environment object is identified according to the semantic category library; Based on the polarization state and temporal scattering characteristics in the multi-attribute environmental characterization data, material attribute inference is performed to obtain the material attributes of each target object in the environment. Based on the volume density and the category of the environment object in the multi-attribute environment characterization data, the occlusion relationship, adjacency relationship and hierarchical relationship between each target object are analyzed, and the spatial association relationship is obtained based on the occlusion relationship, the adjacency relationship and the hierarchical relationship; The environmental object categories, material attributes, and spatial relationships are structured and associated to form the environmental semantic map.

5. The method for reconstructing environmental semantics based on artificial intelligence according to claim 1, characterized in that, Before obtaining the beam scheduling configuration scheme by performing multi-objective semantic utility optimization based on the environmental semantic map, preset communication task semantic attributes, and channel state information extracted from the terahertz electromagnetic holographic tensor, the method further includes: The amplitude features, phase features, polarization features, time domain features, and frequency domain features of the terahertz electromagnetic holographic tensor are extracted to obtain discrete electromagnetic feature data; The discrete electromagnetic feature data are fused using multimodal methods to obtain fused channel features; Multipath channel parameters are fitted to the fused channel features to obtain the basic channel parameters. The channel state information is obtained by performing state regression correction on the channel basic parameters.

6. The method for reconstructing environmental semantics based on artificial intelligence according to claim 1, characterized in that, The step involves performing multi-objective semantic utility optimization based on the environmental semantic map, preset communication task semantic attributes, and channel state information extracted from the terahertz electromagnetic holographic tensor to obtain a beam scheduling configuration scheme, including: Based on the operational constraints of communication services, a multi-dimensional semantic comprehensive utility function is constructed, which includes throughput utility, semantic matching utility, link robustness utility, and resource allocation fairness utility. Feature association matching is performed on the environmental semantic map, communication task semantic attributes, and channel state information to obtain optimized input feature data; With transmit power, time slot resources, and service quality as constraints, and based on the optimized input feature data, the multi-dimensional semantic comprehensive utility function is optimized by multi-objective semantic utility to obtain the global optimal solution set; According to the preset business decision rules, the optimal configuration is selected from the global optimal solution set, and the beam scheduling configuration scheme is determined based on the optimal configuration.

7. The method for reconstructing environmental semantics based on artificial intelligence according to claim 1, characterized in that, After configuring the beamforming parameters of the holographic intelligent metasurface according to the beam scheduling configuration scheme to complete environmental semantic reconstruction and communication control, the method further includes: Collect current environmental semantic features and terahertz channel features, and perform statistical distribution comparison on the current environmental semantic features and terahertz channel features to obtain feature distribution difference data; Drift detection is performed on the feature distribution difference data to obtain drift detection results; When the drift detection result indicates the presence of feature distribution drift, the parameters of the terahertz neural radiation field network are adjusted using on-site measured samples to obtain an updated terahertz neural radiation field network. The updated terahertz neural radiation field network is used for multi-scale position encoding, multi-layer nonlinear feature transformation, and comprehensive environmental semantic reasoning.

8. An environment semantic reconstruction device based on artificial intelligence, characterized in that, include: The acquisition module is used to acquire the downlink pilot signal transmitted by the terahertz base station to the communication coverage area, and to collect the amplitude data, phase data and polarization data corresponding to the downlink pilot signal through the holographic sensing module mounted on each reflection unit in the holographic intelligent metasurface array, so as to obtain multi-dimensional electromagnetic raw data. The arrangement module is used to arrange the multi-dimensional electromagnetic raw data according to the metasurface unit spatial coordinates and signal sampling time to obtain a terahertz electromagnetic holographic tensor. The reasoning module is used to input the terahertz electromagnetic holographic tensor into a pre-trained terahertz neural radiation field network to perform multi-scale position encoding, multi-layer nonlinear feature transformation and environmental semantic comprehensive reasoning to obtain multi-attribute environmental representation data. The graph construction module is used to perform semantic category mapping, material attribute reasoning, and spatial topological relationship parsing on the multi-attribute environmental representation data to obtain environmental object categories, material attributes, and spatial relationships, and to construct an environmental semantic graph based on the environmental object categories, material attributes, and spatial relationships. The parameter configuration module is used to perform multi-objective semantic utility optimization solution based on the environmental semantic map, the preset communication task semantic attributes and the channel state information extracted by the terahertz electromagnetic holographic tensor, to obtain a beam scheduling configuration scheme, and to configure the beamforming parameters of the holographic intelligent metasurface according to the beam scheduling configuration scheme, so as to complete the environmental semantic reconstruction and communication control.

9. An electronic device, characterized in that, It includes a processor and a memory, the memory storing multiple instructions; the processor loads instructions from the memory to perform the steps in the AI-based environmental semantic reconstruction method as described in any one of claims 1 to 7.

10. A computer-readable storage medium, characterized in that, The computer-readable storage medium stores a plurality of instructions adapted for loading by a processor to execute the steps of the artificial intelligence-based environmental semantic reconstruction method according to any one of claims 1 to 7.

Images