WO2025175751A1 - 路径损耗的确定方法和装置、存储介质及电子装置 - Google Patents

路径损耗的确定方法和装置、存储介质及电子装置

Info

Publication number
WO2025175751A1
WO2025175751A1 PCT/CN2024/118596 CN2024118596W WO2025175751A1 WO 2025175751 A1 WO2025175751 A1 WO 2025175751A1 CN 2024118596 W CN2024118596 W CN 2024118596W WO 2025175751 A1 WO2025175751 A1 WO 2025175751A1
Authority
WO
WIPO (PCT)
Prior art keywords
image
field image
loss function
determining
target
Prior art date
Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
Pending
Application number
PCT/CN2024/118596
Other languages
English (en)
French (fr)
Inventor
李勇
姜奋宇
李桐
金德鹏
吕星哉
芮华
Current Assignee (The listed assignees may be inaccurate. Google has not performed a legal analysis and makes no representation or warranty as to the accuracy of the list.)
ZTE Corp
Original Assignee
ZTE Corp
Priority date (The priority date is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the date listed.)
Filing date
Publication date
Application filed by ZTE Corp filed Critical ZTE Corp
Publication of WO2025175751A1 publication Critical patent/WO2025175751A1/zh
Pending legal-status Critical Current
Anticipated expiration legal-status Critical

Links

Classifications

    • H—ELECTRICITY
    • H04—ELECTRIC COMMUNICATION TECHNIQUE
    • H04B—TRANSMISSION
    • H04B17/00—Monitoring; Testing
    • H04B17/30—Monitoring; Testing of propagation channels
    • H04B17/309—Measuring or estimating channel quality parameters
    • G—PHYSICS
    • G06—COMPUTING OR CALCULATING; COUNTING
    • G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
    • G06N3/00—Computing arrangements based on biological models
    • G06N3/02—Neural networks
    • G06N3/04—Architecture, e.g. interconnection topology
    • G06N3/0464—Convolutional networks [CNN, ConvNet]
    • G—PHYSICS
    • G06—COMPUTING OR CALCULATING; COUNTING
    • G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
    • G06N3/00—Computing arrangements based on biological models
    • G06N3/02—Neural networks
    • G06N3/08—Learning methods
    • G—PHYSICS
    • G06—COMPUTING OR CALCULATING; COUNTING
    • G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
    • G06V10/00—Arrangements for image or video recognition or understanding
    • G06V10/40—Extraction of image or video features
    • G06V10/44—Local feature extraction by analysis of parts of the pattern, e.g. by detecting edges, contours, loops, corners, strokes or intersections; Connectivity analysis, e.g. of connected components
    • G—PHYSICS
    • G06—COMPUTING OR CALCULATING; COUNTING
    • G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
    • G06V10/00—Arrangements for image or video recognition or understanding
    • G06V10/40—Extraction of image or video features
    • G06V10/54—Extraction of image or video features relating to texture
    • G—PHYSICS
    • G06—COMPUTING OR CALCULATING; COUNTING
    • G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
    • G06V10/00—Arrangements for image or video recognition or understanding
    • G06V10/40—Extraction of image or video features
    • G06V10/56—Extraction of image or video features relating to colour
    • G—PHYSICS
    • G06—COMPUTING OR CALCULATING; COUNTING
    • G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
    • G06V10/00—Arrangements for image or video recognition or understanding
    • G06V10/70—Arrangements for image or video recognition or understanding using pattern recognition or machine learning
    • G06V10/74—Image or video pattern matching; Proximity measures in feature spaces
    • G06V10/75—Organisation of the matching processes, e.g. simultaneous or sequential comparisons of image or video features; Coarse-fine approaches, e.g. multi-scale approaches; using context analysis; Selection of dictionaries
    • G—PHYSICS
    • G06—COMPUTING OR CALCULATING; COUNTING
    • G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
    • G06V10/00—Arrangements for image or video recognition or understanding
    • G06V10/70—Arrangements for image or video recognition or understanding using pattern recognition or machine learning
    • G06V10/74—Image or video pattern matching; Proximity measures in feature spaces
    • G06V10/75—Organisation of the matching processes, e.g. simultaneous or sequential comparisons of image or video features; Coarse-fine approaches, e.g. multi-scale approaches; using context analysis; Selection of dictionaries
    • G06V10/757—Matching configurations of points or features
    • G—PHYSICS
    • G06—COMPUTING OR CALCULATING; COUNTING
    • G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
    • G06V10/00—Arrangements for image or video recognition or understanding
    • G06V10/70—Arrangements for image or video recognition or understanding using pattern recognition or machine learning
    • G06V10/77—Processing image or video features in feature spaces; using data integration or data reduction, e.g. principal component analysis [PCA] or independent component analysis [ICA] or self-organising maps [SOM]; Blind source separation
    • G06V10/80—Fusion, i.e. combining data from various sources at the sensor level, preprocessing level, feature extraction level or classification level
    • G—PHYSICS
    • G06—COMPUTING OR CALCULATING; COUNTING
    • G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
    • G06V10/00—Arrangements for image or video recognition or understanding
    • G06V10/70—Arrangements for image or video recognition or understanding using pattern recognition or machine learning
    • G06V10/77—Processing image or video features in feature spaces; using data integration or data reduction, e.g. principal component analysis [PCA] or independent component analysis [ICA] or self-organising maps [SOM]; Blind source separation
    • G06V10/80—Fusion, i.e. combining data from various sources at the sensor level, preprocessing level, feature extraction level or classification level
    • G06V10/806—Fusion, i.e. combining data from various sources at the sensor level, preprocessing level, feature extraction level or classification level of extracted features
    • G—PHYSICS
    • G06—COMPUTING OR CALCULATING; COUNTING
    • G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
    • G06V10/00—Arrangements for image or video recognition or understanding
    • G06V10/70—Arrangements for image or video recognition or understanding using pattern recognition or machine learning
    • G06V10/82—Arrangements for image or video recognition or understanding using pattern recognition or machine learning using neural networks

Definitions

  • the embodiments of the present disclosure relate to the field of communications, and in particular, to a method and device for determining path loss, a storage medium, and an electronic device.
  • Path loss estimation is the process of estimating the path loss of wireless signal transmission given the location information of the base station, engineering parameters, building distribution information in the urban scene, the location where the user receives the signal, and other conditions.
  • the path loss of wireless signal transmission is estimated through a knowledge-driven approach.
  • various channel models under the 3GPP standard (3rd Generation Partnership Project) start from the nature of electromagnetic wave transmission, solve Maxwell's equations at different scales, calculate the electromagnetic field distribution, and obtain the path loss by statistically analyzing the energy attenuation of the signal.
  • This method fully considers the building information and electromagnetic propagation laws in urban scenes, thereby improving the accuracy and generalization of path loss estimation.
  • the embodiments of the present disclosure provide a method and device for determining path loss, a storage medium, and an electronic device to at least solve the problem that the path loss estimation method in the prior art cannot achieve both accuracy and computational efficiency.
  • a method for determining path loss comprising: determining an overlay field image of a target grid scene based on a first image, a second image, and an incident field image of the target grid scene, wherein the first image is used to indicate a position of a first object in the target grid scene, and the second image is used to indicate a position of a second object in the target grid scene; determining a first loss function based on the incident field image and the overlay field image; and determining a second loss function based on the overlay field image and road test data of the target grid scene; determining a target loss function based on the first loss function and the second loss function; and training a neural network model based on the target loss function to obtain a target loss function.
  • the path loss of the target grid scene is determined by the trained neural network model.
  • the method before determining the superimposed field image of the target grid scene based on the first image, the second image and the incident field image of the target grid scene, the method further includes: determining the first position of the first object in the target grid scene based on the first image, and determining the second position of the second object in the target grid scene based on the second image; determining the incident field of each grid in the target grid scene based on the first position and the second position; and determining the incident field image of the target grid scene based on the incident field of each grid.
  • determining a superimposed field image of a target grid scene based on a first image, a second image and an incident field image of the target grid scene includes: determining a first real part in complex form corresponding to the incident field image and a first imaginary part in complex form corresponding to the incident field image; determining a first incident field image based on the first real part, and determining a second incident field image based on the first imaginary part; determining the superimposed field image based on the first image, the second image, the first incident field image and the second incident field image.
  • the superimposed field image is determined based on the first image, the second image, the first incident field image and the second incident field image, including: extracting the first image feature of the first image, the second image feature of the second image, the third image feature of the first incident field image and the fourth image feature of the second incident field image, respectively; determining the associated image features of the first image, the second image, the first incident field image and the second incident field image based on the first image feature, the second image feature, the third image feature and the fourth image feature; and determining the superimposed field image based on the associated image features and the volume integral equation.
  • determining a second loss function based on the overlay field image and the road test data of the target grid scene includes: inputting the overlay field image into a road loss map converter to instruct the road loss map converter to convert the overlay field image into a path loss map corresponding to the target grid scene; correcting the path loss map based on the road test data and the overlay field image to determine a corrected path loss map; and determining the second loss function based on the corrected target path loss map and the road test data.
  • the path loss map is corrected according to the road test data and the overlay field image to determine a corrected target path loss map, including: inputting the overlay field image into a deep learning network to determine a second real part in complex form corresponding to the overlay field image and a second imaginary part in complex form corresponding to the overlay field image; calculating the path loss data of the path loss map according to the second real part and the second imaginary part; and correcting the path loss map according to the deviation between the path loss data and the road test data to determine the target path loss map.
  • training a neural network model according to the target loss function includes: a calculation step of calculating the gradient of the target loss function; an adjustment step of adjusting the gradient according to a preset optimization algorithm to update the neural network model according to the adjusted gradient; a determination step of determining the calculation result of the target loss function according to the updated neural network model; and looping the adjustment step and the determination step until the final calculation result is less than or equal to a preset threshold.
  • determining a target loss function based on the first loss function and the second loss function includes: determining a first weight of the first loss function in the neural network model, or determining a second weight of the second loss function in the neural network model; when the first weight has been determined, determining a first product of the first weight and the first loss function, and determining the sum of the first product and the second loss function as the target loss function; when the second weight has been determined, determining a second product of the second weight and the second loss function; and determining the sum of the second product and the first loss function as the target loss function.
  • a device for determining path loss including: a first determination module, configured to determine an overlay field image of a target grid scene based on a first image, a second image and an incident field image of the target grid scene, wherein the first image is used to indicate the position of a first object in the target grid scene, and the second image is used to indicate the position of a second object in the target grid scene; a second determination module, configured to determine a first loss function based on the incident field image and the overlay field image; and determine a second loss function based on the overlay field image and road test data of the target grid scene; a third determination module, configured to determine a target loss function based on the first loss function and the second loss function; and a training module, configured to train a neural network model based on the target loss function, so as to determine the path loss of the target grid scene through the trained neural network model.
  • a computer-readable storage medium in which a computer program is stored.
  • the computer program is configured to execute the steps of any one of the above method embodiments when running.
  • an electronic device including a memory and a processor, wherein the memory stores a computer program, and the processor is configured to run the computer program to perform the steps in any one of the above method embodiments.
  • a computer program product including computer instructions, which implement the steps of the method described in various embodiments of the present disclosure when executed by a processor.
  • the superimposed field image of the target grid scene is determined based on the first image for indicating the position of the first object in the target grid scene, the second image for indicating the position of the second object in the target grid scene, and the incident field image of the target grid image.
  • the first loss function is determined based on the incident field image and the superimposed field image
  • the second loss function is determined based on the superimposed field image and the road test data.
  • the target loss function for training the neural network model is determined based on the first loss function and the second loss function, and then the path loss of the target grid scene is determined by the trained neural network model.
  • the embodiment of the present disclosure defines a technical solution for determining the first loss function by the superimposed field image and the incident field image, and determining the second loss function by the road test data and the superimposed field image, and then determining the target loss function for training the neural network model based on the first loss function and the second loss function, so as to calculate the path loss by the trained neural network model.
  • the physical formula describing the incident field and the superimposed field is introduced into the target loss function
  • the road test data is introduced into the target loss function.
  • FIG1 is a hardware structure block diagram of a computer terminal for a method for determining path loss according to an embodiment of the present disclosure
  • FIG2 is a flow chart of a method for determining path loss according to an embodiment of the present disclosure
  • FIG3 is a schematic diagram of a path loss estimation model according to an optional embodiment of the present disclosure.
  • FIG4 is a schematic diagram of a two-dimensional wireless communication scenario according to an optional embodiment of the present disclosure.
  • FIG5 is a network structure diagram of a knowledge-driven module UNet according to an optional embodiment of the present disclosure
  • FIG6 is a network structure diagram of a data-driven module UNet according to an optional embodiment of the present disclosure.
  • FIG1 is a hardware structure block diagram of a computer terminal for a method for determining a path loss according to an embodiment of the present disclosure.
  • the computer terminal may include one or more (only one is shown in FIG1 ) processors 102 (the processor 102 may include but is not limited to a microprocessor (Central Processing Unit, MCU) or a programmable logic device (Field Programmable Gate Array, FPGA) and other processing devices) and a memory 104 for storing data, wherein the above-mentioned computer terminal may also include a transmission device 106 and an input and output device 108 for communication functions.
  • the structure shown in FIG1 is only for illustration and does not limit the structure of the above-mentioned computer terminal.
  • the computer terminal may also include more or fewer components than those shown in FIG1 , or have a configuration different from that shown in FIG1 .
  • Memory 104 can be used to store computer programs, such as software programs and modules of application software, such as the computer program corresponding to the path loss determination method in the embodiments of the present disclosure.
  • Processor 102 executes the computer programs stored in memory 104 to execute various functional applications and data processing, thereby implementing the aforementioned method.
  • Memory 104 may include high-speed random access memory (RAM) and may also include non-volatile memory, such as one or more magnetic storage devices, flash memory, or other non-volatile solid-state memory.
  • memory 104 may further include memory located remotely from processor 102, which can be connected to the computer terminal via a network. Examples of such networks include, but are not limited to, the Internet, corporate intranets, local area networks, mobile communication networks, and combinations thereof.
  • FIG2 is a flow chart of the method for determining path loss according to an embodiment of the present disclosure. As shown in FIG2 , the flow chart includes the following steps:
  • Step S202 determining a superimposed field image of the target grid scene based on the first image, the second image, and the incident field image of the target grid scene, wherein the first image is used to indicate the position of the first object in the target grid scene, and the second image is used to indicate the position of the second object in the target grid scene;
  • the first object may be a base station, and thus the first image may be an image indicating the location of the base station in the target grid scene.
  • the second object may be a building, and thus the second image may be an image indicating the location of the building in the target grid scene.
  • the target grid scene includes: base stations, buildings, roads, etc.
  • a two-dimensional space containing at least the base station and the building can be gridded using the moment method to form a target grid scene.
  • the base station position in the target grid scene is set to a first value (optionally, the first value is 1), and the other positions in the target grid scene except the base station position are set to a second value (optionally, the second value is 0), thereby generating a first image.
  • the building position in the target grid scene is set to a first value (optionally, the first value is 1), and the other positions in the target grid scene except the building position are set to a second value (optionally, the second value is 0), thereby generating a second image.
  • Step S204 determining a first loss function based on the incident field image and the superimposed field image; and determining a second loss function based on the superimposed field image and the drive test data of the target grid scene;
  • Step S206 determining a target loss function according to the first loss function and the second loss function
  • Step S208 training a neural network model according to the target loss function to determine the path loss of the target grid scene through the trained neural network model.
  • the superposition field image of the target grid scene is determined based on the first image for indicating the position of the first object in the target grid scene, the second image for indicating the position of the second object in the target grid scene, and the incident field image of the target grid image.
  • a first loss function is determined based on the incident field image and the superposition field image
  • a second loss function is determined based on the superposition field image and the road test data.
  • a target loss function for training the neural network model is determined based on the first loss function and the second loss function, and then the path loss of the target grid scene is determined by the trained neural network model.
  • the embodiment of the present disclosure defines a technical solution for determining the first loss function by the superposition field image and the incident field image, and determining the second loss function by the road test data and the superposition field image, and then determining the target loss function for training the neural network model based on the first loss function and the second loss function, so as to calculate the path loss by the trained neural network model.
  • both the physical formula describing the incident field and the superposition field are introduced into the target loss function
  • the road test data are introduced into the target loss function.
  • the problem that the path loss estimation method in the existing technology cannot take into account both accuracy and computational efficiency is solved. After the neural network model is trained through the target loss function, in the process of determining the path loss, both the accuracy of the path loss estimation and the computational efficiency of the path loss estimation are improved.
  • the method before determining the superimposed field image of the target grid scene based on the first image, the second image and the incident field image of the target grid scene in step S202, the method further includes: determining the first position of the first object in the target grid scene based on the first image, and determining the second position of the second object in the target grid scene based on the second image; determining the incident field of each grid in the target grid scene based on the first position and the second position; and determining the incident field image of the target grid scene based on the incident field of each grid.
  • the incident field image can be determined based on the first image and the second image. Specifically: by formula: Determine the incident field result of the second position (each building) relative to the first position (base station) in the target grid scene, and then determine the incident field image based on the incident field result. represents the incident field result, p n represents the second position, p tx represents the first position, and G( ⁇ ) represents the Green's function in two-dimensional space.
  • the above-mentioned step S202 of determining the superimposed field image of the target grid scene based on the first image, the second image and the incident field image of the target grid scene includes: determining the first real part in complex form corresponding to the incident field image and the first imaginary part in complex form corresponding to the incident field image; determining the first incident field image based on the first real part, and determining the second incident field image based on the first imaginary part; determining the superimposed field image based on the first image, the second image, the first incident field image and the second incident field image.
  • determining the superimposed field image according to the first image, the second image, the first incident field image and the second incident field image comprises: extracting the first image feature of the first image, the second image feature of the second image, and the superimposed field image respectively; image features, third image features of the first incident field image and fourth image features of the second incident field image; determining associated image features of the first image, the second image, the first incident field image and the second incident field image based on the first image features, the second image features, the third image features and the fourth image features; determining the superimposed field image based on the associated image features and the volume integral equation.
  • the incident field image has a complex form and can be divided into a real image determined by the first real part (i.e., the first incident field image of the present disclosure) and an imaginary image determined by the first imaginary part (i.e., the second incident field image of the present disclosure).
  • the real image can be used to indicate the intensity distribution of light, i.e., the brightness or darkness distribution of light.
  • the imaginary image can be used to indicate phase information, i.e., the phase distribution of the light wave.
  • the first image, the second image, the first incident field image, and the second incident field image are input into a Unified Network Model (i.e., a convolutional neural network with UNet as its core, also referred to as a UNet network model).
  • the encoder extracts the first image features, the second image features, the third image features, and the fourth image features.
  • the image features include, but are not limited to, image color, texture, shape, and edge features.
  • the decoder determines the associated image feature through the first image feature, the second image feature, the third image feature, and the fourth image feature. Specifically:
  • the decoder can complete the determination of associated image features through a series of steps, such as feature matching: determining the association relationship between the first image feature, the second image feature, the third image feature and the fourth image feature; alignment and registration: registering the first image feature, the second image feature, the third image feature and the fourth image feature through an image registration algorithm; and feature fusion: fusing features through weighted averaging, feature cascading, etc.
  • E tot represents the superposition field image
  • E inc represents the incident field image
  • W represents the N ⁇ N coefficient matrix obtained by integrating the Green's function over the unit grid
  • ⁇ represents a diagonal matrix whose diagonal elements represent the contrast of each grid cell.
  • the first loss function can be determined based on the incident field image and the superposition field image.
  • the incident field image is used as a learning label, the superposition field image is continuously trained, and the network parameters are optimized so that the output superposition field is close to the theoretical numerical calculation result.
  • the first loss function E' tot represents the output superposition field.
  • a second loss function can also be determined based on the overlay field image and the road test data of the target grid scene, including: inputting the overlay field image into a road loss map converter to instruct the road loss map converter to convert the overlay field image into a path loss map corresponding to the target grid scene; correcting the path loss map based on the road test data and the overlay field image to determine a corrected path loss map; and determining the second loss function based on the corrected target path loss map and the road test data.
  • the step of correcting the path loss map according to the drive test data and the superimposed field image to determine a corrected target path loss map includes: inputting the superimposed field image into a deep learning network to determine a second real part in a complex form corresponding to the superimposed field image and a second imaginary part in a complex form corresponding to the superimposed field image; Calculating path loss data of the path loss map according to the second real part and the second imaginary part; and correcting the path loss map according to a deviation between the path loss data and the drive test data to determine the target path loss map.
  • the overlay field image is converted into a path loss map of the target grid scene by the path loss map converter. Further, the second real part and the second imaginary part of the overlay field image are determined.
  • Determine the magnitude of the superposition field represents the amplitude of the superposition field, represents the second real part, Represents the second imaginary part.
  • the second loss function is determined by the corrected path loss map and the drive test data.
  • pl' represents the corrected path loss map
  • pl dat represents the drive test data.
  • training a neural network model according to the target loss function includes: a calculation step of calculating the gradient of the target loss function; an adjustment step of adjusting the gradient according to a preset optimization algorithm to update the neural network model according to the adjusted gradient; a determination step of determining the calculation result of the target loss function according to the updated neural network model; and looping the adjustment step and the determination step until the final calculation result is less than or equal to a preset threshold.
  • the role of the objective loss function is to train the neural network model and improve the accuracy, generalization and computational efficiency of the neural network model in estimating path loss.
  • determining the target loss function based on the first loss function and the second loss function includes: determining the first weight of the first loss function in the neural network model, or determining the second weight of the second loss function in the neural network model; when the first weight has been determined, determining the first product of the first weight and the first loss function, and determining the sum of the first product and the second loss function as the target loss function; when the second weight has been determined, determining the second product of the second weight and the second loss function; and determining the sum of the second product and the first loss function as the target loss function.
  • channel modeling is generally based on two aspects: large-scale fading and small-scale fading.
  • Large-scale fading primarily considers path loss, which consists of propagation loss and shadow fading. Propagation loss is related to the distance between the transmitter and receiver, while shadow fading is caused by large obstacles in the scene (such as mountains and buildings).
  • Small-scale fading is primarily caused by multipath effects and is related to the dynamic changes in the wireless transmission environment.
  • path loss estimation methods based on traditional empirical formulas and protocol standards has deficiencies in accuracy and generalization; the path loss estimation method based on electromagnetic calculations needs to numerically solve Maxwell's equations, which occupies a large area of computing space.
  • the use of data-driven neural network methods lacks generalizability because the model's effectiveness depends on the training data. In other words, existing path loss estimation methods cannot achieve a balanced combination of accuracy, generalizability, and computational efficiency for path loss calculations.
  • An optional embodiment of the present disclosure proposes a knowledge-data collaborative path loss estimation model.
  • This optional embodiment of the present disclosure uses volume integral equations as the physical knowledge of electromagnetic calculations and replaces traditional numerical solution methods by designing and constructing neural networks. Specifically:
  • FIG3 is a schematic diagram of a path loss estimation model according to an optional embodiment of the present disclosure.
  • the path loss estimation model includes: an input part, a knowledge-driven module, and a data-driven module.
  • FIG. 4 is a schematic diagram of a two-dimensional wireless communication scenario according to an optional embodiment of the present disclosure. As shown in Figure 4, the wireless communication scenario in an urban area can be considered an electromagnetic propagation scenario, with base stations as the excitation source and buildings, vegetation, etc. in the city as scatterers.
  • the grid where the base station location (i.e., the first object of the present disclosure) is located is set to 1, and the remaining grids are set to 0.
  • the grid covered by the building i.e., the second object of the present disclosure
  • the uncovered grid is set to 0.
  • the input images of the input part are: base station location map, building distribution map, incident field real part image and incident field imaginary part image.
  • E tot represents the target electric field superposition field.
  • Figure 5 is a network structure diagram of the knowledge-driven module UNet according to an optional embodiment of the present disclosure. As shown in Figure 5, in the knowledge-driven module, the incident field E inc is used as a learning label, and the network parameters are optimized through continuous training so that the output superposition field E' tot is close to the theoretical numerical calculation result.
  • the first loss function is determined based on the incident field and the superposition field after training
  • Data driven module The electric field strength calculated by electromagnetics is converted into a path loss map through the path loss map converter.
  • the UNet network outputs the results of two channels, namely the real part of the estimated superposition field and the real part of the superposition field. and the imaginary part Calculate the amplitude of the superposition field by formula: The energy of the electric field is then calculated and compared with the transmission power of the transmitting source to determine the preliminary distribution of the path loss.
  • FIG. 6 is a diagram of the network structure of the data-driven module UNet according to an optional embodiment of the present disclosure.
  • the present disclosure adopts the UNet network structure shown in Figure 6 to learn the deviation between the preliminary distribution results of the path loss and the actual road test data, and corrects the intermediate output to obtain the corrected path loss map pl' (i.e., the corrected path loss map of the present disclosure).
  • the path loss map data pl dat i.e., the road test data of the present disclosure
  • the gap between them is narrowed, making the estimated results closer to the distribution of the real data.
  • the two loss functions can be combined to form the loss function of the entire network. Specifically, the relative weight corresponding to the first loss function or the relative weight of the second loss function can be determined to determine the target loss function.
  • the neural network model After determining the target loss function, the neural network model needs to be trained. By calculating the gradient of the target loss function and performing gradient backpropagation, while continuously reducing the result of the target loss function, the purpose of optimizing the network parameters is achieved, so that the output result of the neural network model gradually approaches the true value of the road test data.
  • the trained neural network model estimates path loss.
  • the target scene i.e., the target grid scene disclosed herein
  • the target grid scene is preprocessed according to the input dimensions to produce a base station location map, a building distribution map, and an electric field incident distribution map.
  • These features are then fed into the trained neural network model, and the resulting corrected path loss map serves as the model's path loss estimation result.
  • the RadioMapSeer dataset is used as a dataset, which includes 700 areas in six different cities around the world, each with 80 different base station locations.
  • the dataset includes a base station location map (the first image in this disclosure), a building distribution map (the second image in this disclosure), and a road damage map.
  • the relative dielectric constant of the building and the relative dielectric constant of the air are set, and the real image (i.e., the first incident field image disclosed in the present invention) and the imaginary image (i.e., the second incident field image disclosed in the present invention) of the incident field are calculated based on the location information of the base station.
  • 500 of the 700 different areas are randomly selected as training sets, 100 as validation sets, and the remaining 100 as test sets. This division allows the model to face different building distribution characteristics during training and testing, and while accurately evaluating the model performance, it can also ensure the verification of the model's generalization ability.
  • a trained model i.e., the trained neural network model disclosed in the present invention.
  • the training data is gridded according to its pixels and dimensions, and converted into corresponding base station location maps, building distribution maps, and incident field real and imaginary maps. These four images are input into the model, and the corrected path loss map is used as the path loss estimation result of the model output.
  • An optional embodiment of the present disclosure integrates the loss function of the physical formula with the loss function based on the road test data, so that the output result satisfies both the volume integral equation of physical knowledge and the distribution law of actual road test data, thereby improving the accuracy, efficiency and generalization of the road loss estimation.
  • the computer software product is stored in a storage medium (such as a read-only memory/random access memory (ROM/RAM), a magnetic disk, or an optical disk), and includes a number of instructions for enabling a terminal device (which can be a mobile phone, a computer, a server, or a network device, etc.) to execute the methods described in the various embodiments of the present disclosure.
  • a storage medium such as a read-only memory/random access memory (ROM/RAM), a magnetic disk, or an optical disk
  • a terminal device which can be a mobile phone, a computer, a server, or a network device, etc.
  • This embodiment also provides a path loss determination device, which is used to implement the above-mentioned embodiments and preferred implementations. Details already described are omitted for clarity.
  • the term "module” may refer to a combination of software and/or hardware that implements a predetermined function. Although the devices described in the following embodiments are preferably implemented in software, implementation in hardware, or a combination of software and hardware, is also possible and contemplated.
  • FIG7 is a structural block diagram of a device for determining path loss according to an embodiment of the present disclosure. As shown in FIG7 , the device includes:
  • a first determining module 72 configured to determine a superimposed field image of the target grid scene based on a first image, a second image, and an incident field image of the target grid scene, wherein the first image is used to indicate a position of a first object in the target grid scene, and the second image is used to indicate a position of a second object in the target grid scene;
  • a second determination module 74 is configured to determine a first loss function based on the incident field image and the superimposed field image; and determine a second loss function based on the superimposed field image and the drive test data of the target grid scene;
  • a third determining module 76 is configured to determine a target loss function based on the first loss function and the second loss function;
  • the training module 78 is configured to train a neural network model according to the target loss function, so as to determine the path loss of the target grid scene through the trained neural network model.
  • the first determination module 72 is further configured to determine the first position of the first object in the target grid scene based on the first image, and determine the second position of the second object in the target grid scene based on the second image; determine the incident field of each grid in the target grid scene based on the first position and the second position; and determine the incident field image of the target grid scene based on the incident field of each grid.
  • the first determination module 72 is further configured to determine a first real part in complex form corresponding to the incident field image and a first imaginary part in complex form corresponding to the incident field image; determine a first incident field image based on the first real part, and determine a second incident field image based on the first imaginary part; determine the superimposed field image based on the first image, the second image, the first incident field image and the second incident field image.
  • the first determination module 72 is further configured to respectively extract the first image feature of the first image, the second image feature of the second image, the third image feature of the first incident field image, and the fourth image feature of the second incident field image; determine the associated image features of the first image, the second image, the first incident field image, and the second incident field image based on the first image feature, the second image feature, the third image feature, and the fourth image feature; and determine the superimposed field image based on the associated image features and the volume integral equation.
  • the second determination module 74 is further configured to input the overlay field image into a road loss map converter to instruct the road loss map converter to convert the overlay field image into a path loss map corresponding to the target grid scenario; correct the path loss map according to the road test data and the overlay field image to determine a corrected path loss map; and determine the second loss function according to the corrected target path loss map and the road test data.
  • the second determination module 74 is further configured to input the superimposed field image into a deep learning network to determine a second real part of the complex form corresponding to the superimposed field image and a second imaginary part of the complex form corresponding to the superimposed field image; and calculate the path loss according to the second real part and the second imaginary part. path loss data of the map; and correcting the path loss map according to a deviation between the path loss data and the drive test data to determine the target path loss map.
  • the training module 78 is further configured as a calculation step: calculating the gradient of the target loss function; an adjustment step: adjusting the gradient according to a preset optimization algorithm to update the neural network model according to the adjusted gradient; a determination step: determining the calculation result of the target loss function according to the updated neural network model; and looping the adjustment step and the determination step until the final calculation result is less than or equal to a preset threshold.
  • the third determination module 76 is further configured to determine a first weight of the first loss function in the neural network model, or to determine a second weight of the second loss function in the neural network model; when the first weight has been determined, determine a first product of the first weight and the first loss function, and determine the sum of the first product and the second loss function as the target loss function; when the second weight has been determined, determine a second product of the second weight and the second loss function; and determine the sum of the second product and the first loss function as the target loss function.
  • the superimposed field image of the target grid scene is determined based on the first image for indicating the position of the first object in the target grid scene, the second image for indicating the position of the second object in the target grid scene, and the incident field image of the target grid image.
  • the first loss function is determined based on the incident field image and the superimposed field image
  • the second loss function is determined based on the superimposed field image and the road test data.
  • the target loss function for training the neural network model is determined based on the first loss function and the second loss function, and then the path loss of the target grid scene is determined by the trained neural network model.
  • the embodiment of the present disclosure defines a technical solution for determining the first loss function by the superimposed field image and the incident field image, and determining the second loss function by the road test data and the superimposed field image, and then determining the target loss function for training the neural network model based on the first loss function and the second loss function, so as to calculate the path loss by the trained neural network model.
  • the physical formula describing the incident field and the superimposed field is introduced into the target loss function
  • the road test data is introduced into the target loss function.
  • the above modules can be implemented through software or hardware. For the latter, it can be implemented in the following ways, but not limited to: the above modules are all located in the same processor; or the above modules are located in different processors in any combination.
  • An embodiment of the present disclosure further provides a computer-readable storage medium, in which a computer program is stored.
  • the computer program is configured to execute the steps of any one of the above method embodiments when run.
  • the above-mentioned computer-readable storage medium may include but is not limited to: USB flash drive, read-only memory (ROM), random access memory (RAM), mobile hard disk, magnetic disk or optical disk, and other media that can store computer programs.
  • An embodiment of the present disclosure further provides an electronic device, including a memory and a processor, wherein the memory stores a computer program, and the processor is configured to run the computer program to perform the steps in any of the above method embodiments.
  • the electronic device may further include a transmission device and an input/output device, wherein the transmission device is connected to the processor, and the input/output device is connected to the processor.
  • the computer program product may further include computer instructions, which, when executed by a processor, implement the steps of the method described in various embodiments of the present disclosure.
  • modules or steps of the present disclosure described above can be implemented using a general-purpose computing device, they can be concentrated on a single computing device, or distributed across a network composed of multiple computing devices, they can be implemented using program code executable by the computing device, and thus, they can be stored in a storage device and executed by the computing device, and in some cases, the steps shown or described can be performed in a different order than herein, or they can be fabricated into separate integrated circuit modules, or multiple modules or steps can be fabricated into a single integrated circuit module for implementation.
  • the present disclosure is not limited to any particular combination of hardware and software.

Landscapes

  • Engineering & Computer Science (AREA)
  • Theoretical Computer Science (AREA)
  • Physics & Mathematics (AREA)
  • General Physics & Mathematics (AREA)
  • Evolutionary Computation (AREA)
  • Multimedia (AREA)
  • Computer Vision & Pattern Recognition (AREA)
  • Computing Systems (AREA)
  • Software Systems (AREA)
  • Health & Medical Sciences (AREA)
  • General Health & Medical Sciences (AREA)
  • Artificial Intelligence (AREA)
  • Medical Informatics (AREA)
  • Databases & Information Systems (AREA)
  • Biophysics (AREA)
  • Mathematical Physics (AREA)
  • General Engineering & Computer Science (AREA)
  • Molecular Biology (AREA)
  • Life Sciences & Earth Sciences (AREA)
  • Biomedical Technology (AREA)
  • Data Mining & Analysis (AREA)
  • Computational Linguistics (AREA)
  • Quality & Reliability (AREA)
  • Electromagnetism (AREA)
  • Computer Networks & Wireless Communication (AREA)
  • Signal Processing (AREA)
  • Image Analysis (AREA)

Abstract

本公开实施例提供了一种路径损耗的确定方法和装置、存储介质及电子装置，上述方法包括：根据第一图像、第二图像和目标网格场景的入射场图像确定目标网格场景的叠加场图像，其中，第一图像用于指示第一对象在目标网格场景中的位置，第二图像用于指示第二对象在目标网格场景中的位置；根据入射场图像和叠加场图像确定第一损失函数；以及根据叠加场图像和目标网格场景的路测数据确定第二损失函数；根据第一损失函数和第二损失函数确定目标损失函数；根据目标损失函数训练神经网络模型，以通过训练后的神经网络模型确定目标网格场景的路径损耗。通过上述方法，可以解决现有技术中进行路径损耗估计的方法无法兼顾准确性与计算效率的问题。

Description

路径损耗的确定方法和装置、存储介质及电子装置
本公开要求于2024年2月22日提交中国专利局、申请号为202410199012.5、发明名称为“路径损耗的确定方法和装置、存储介质及电子装置”的中国专利申请的优先权，其全部内容通过引用结合在本公开中。
技术领域
本公开实施例涉及通信领域，具体而言，涉及一种路径损耗的确定方法和装置、存储介质及电子装置。
背景技术
路径损耗估计(Path Loss Estimation)是在给定基站的位置信息、工程参数、城市场景中的建筑物分布信息、用户接收信号位置等条件下，对无线信号传输的路径损耗进行估计的过程。
针对路径损耗，按照现有解决方案的驱动模式划分，有知识驱动与数据驱动两种驱动模式。
其中，通过数据驱动的方法估计无线信号传输的路径损耗。即采用人工神经网络、卷积神经网络、图神经网络、生成对抗网络等深度学习模型对路径损耗进行估计的方法。依赖训练好的模型，可以快速批量地产生结果，同时由于神经网络拟合复杂非线性关系的强大能力，兼顾了准确性与高效性。然而，数据驱动的方法过分依赖于训练时使用的数据，在泛化性上存在欠缺。
通过知识驱动的方法估计无线信号传输的路径损耗。例如：3GPP标准(3rd Generation Partnership Project)下的各种信道模型，从电磁波传输的本质出发，通过在不同尺度上求解麦克斯韦方程组，计算得到电磁场分布，并通过统计信号的能量衰减得到路径损耗。这种方法充分考虑了城市场景中的建筑物信息、电磁传播规律，从而提升路径损耗估计的准确性与泛化性。但是在求解过程中，往往需要对场景进行细粒度的网格划分，并需要在每个网格中进行复杂计算，增加了运算量，降低了计算效率。
也就是说，现有技术中进行路径损耗估计的方法无法兼顾准确性与计算效率的问题。
因此，有必要对相关技术予以改良以克服相关技术中的所述缺陷。
发明内容
本公开实施例提供了一种路径损耗的确定方法和装置、存储介质及电子装置，以至少解决现有技术中进行路径损耗估计的方法无法兼顾准确性与计算效率的问题。
根据本公开的一个实施例，提供了一种路径损耗的确定方法，包括：根据第一图像、第二图像和目标网格场景的入射场图像确定所述目标网格场景的叠加场图像，其中，所述第一图像用于指示第一对象在所述目标网格场景中的位置，所述第二图像用于指示第二对象在所述目标网格场景中的位置；根据所述入射场图像和所述叠加场图像确定第一损失函数；以及根据所述叠加场图像和所述目标网格场景的路测数据确定第二损失函数；根据所述第一损失函数和所述第二损失函数确定目标损失函数；根据所述目标损失函数训练神经网络模型，以 通过训练后的神经网络模型确定所述目标网格场景的路径损耗。
在一个示例性实施例中，在根据第一图像、第二图像和目标网格场景的入射场图像确定所述目标网格场景的叠加场图像之前，所述方法还包括：根据所述第一图像确定所述第一对象在所述目标网格场景的第一位置，以及根据所述第二图像确定所述第二对象在所述目标网格场景中的第二位置；根据所述第一位置和所述第二位置确定所述目标网格场景中每个网格的入射场；根据所述每个网格的入射场确定所述目标网格场景的入射场图像。
在一个示例性实施例中，根据第一图像、第二图像和目标网格场景的入射场图像确定所述目标网格场景的叠加场图像，包括：确定所述入射场图像对应的复数形式的第一实部部分和所述入射场图像对应的复数形式的第一虚部部分；根据所述第一实部部分确定第一入射场图像，以及根据所述第一虚部部分确定第二入射场图像；根据所述第一图像、所述第二图像、所述第一入射场图像和所述第二入射场图像确定所述叠加场图像。
在一个示例性实施例中，根据所述第一图像、所述第二图像、所述第一入射场图像和所述第二入射场图像确定所述叠加场图像，包括：分别提取所述第一图像的第一图像特征、所述第二图像的第二图像特征、所述第一入射场图像的第三图像特征和所述第二入射场图像的第四图像特征；根据所述第一图像特征、所述第二图像特征、所述第三图像特征和所述第四图像特征确定所述第一图像、所述第二图像、所述第一入射场图像和所述第二入射场图像的关联图像特征；根据所述关联图像特征和体积分方程确定所述叠加场图像。
在一个示例性实施例中，根据所述叠加场图像和所述目标网格场景的路测数据确定第二损失函数，包括：将所述叠加场图像输入至路损地图转换器中，以指示所述路损地图转换器将所述叠加场图像转换为所述目标网格场景对应的路径损耗地图；根据所述路测数据和所述叠加场图像对所述路径损耗地图进行修正，以确定修正后的路径损耗地图；根据所述修正后的目标路径损耗地图和所述路测数据确定所述第二损失函数。
在一个示例性实施例中，根据所述路测数据和所述叠加场图像对所述路径损耗地图进行修正，以确定修正后的目标路径损耗地图，包括：将所述叠加场图像输入至深度学习网络中，以确定所述叠加场图像对应的复数形式的第二实部部分和所述叠加场图像对应的复数形式的第二虚部部分；根据所述第二实部部分和所述第二虚部部分计算所述路径损耗地图的路径损耗数据；根据所述路径损耗数据与所述路测数据之间的偏差对所述路径损耗地图进行修正，以确定所述目标路径损耗地图。
在一个示例性实施例中，根据所述目标损失函数训练神经网络模型，包括：计算步骤：计算所述目标损失函数的梯度；调整步骤：根据预设优化算法调整所述梯度，以根据调整后的梯度更新所述神经网络模型；确定步骤：根据更新后的神经网络模型确定所述目标损失函数的计算结果；循环执行所述调整步骤和所述确定步骤，直至最终得到的计算结果小于或等于预设阈值。
在一个示例性实施例中，根据所述第一损失函数和所述第二损失函数确定目标损失函数，包括：确定所述第一损失函数在所述神经网络模型中的第一权重，或确定所述第二损失函数在所述神经网络模型中的第二权重；在已确定所述第一权重的情况下，确定所述第一权重与所述第一损失函数的第一乘积，将所述第一乘积和所述第二损失函数的和确定为所述目标损失函数；在已确定所述第二权重的情况下，确定所述第二权重与所述第二损失函数的第二乘积；将所述第二乘积和所述第一损失函数的和确定为所述目标损失函数。
根据本公开的另一个实施例，提供了一种路径损耗的确定装置，包括：第一确定模块，设置为根据第一图像、第二图像和目标网格场景的入射场图像确定所述目标网格场景的叠加场图像，其中，所述第一图像用于指示第一对象在所述目标网格场景中的位置，所述第二图像用于指示第二对象在所述目标网格场景中的位置；第二确定模块，设置为根据所述入射场图像和所述叠加场图像确定第一损失函数；以及根据所述叠加场图像和所述目标网格场景的路测数据确定第二损失函数；第三确定模块，设置为根据所述第一损失函数和所述第二损失函数确定目标损失函数；训练模块，设置为根据所述目标损失函数训练神经网络模型，以通过训练后的神经网络模型确定所述目标网格场景的路径损耗。
根据本公开的又一个实施例，还提供了一种计算机可读存储介质，所述计算机可读存储介质中存储有计算机程序，其中，所述计算机程序被设置为运行时执行上述任一项方法实施例中的步骤。
根据本公开的又一个实施例，还提供了一种电子装置，包括存储器和处理器，所述存储器中存储有计算机程序，所述处理器被设置为运行所述计算机程序以执行上述任一项方法实施例中的步骤。
根据本公开的又一个实施例，还提供了一种计算机程序产品，包括计算机指令，所述计算机指令被处理器执行时实现本公开各个实施例中所述方法的步骤。
通过本公开，由于根据用于指示第一对象在目标网格场景中的位置的第一图像、用于指示第二对象在目标网格场景中的位置的第二图像和目标网格图像的入射场图像确定目标网格场景的叠加场图像。根据入射场图像和叠加场图像确定第一损失函数，并根据叠加场图像和路测数据确定第二损失函数。根据第一损失函数和第二损失函数确定用于训练神经网络模型的目标损失函数，进而通过训练后的神经网络模型确定目标网格场景的路径损耗。也就是说，本公开实施例限定了通过叠加场图像和入射场图像确定第一损失函数，以及路测数据和叠加场图像确定第二损失函数，进而根据第一损失函数和第二损失函数确定用于对神经网络模型进行训练的目标损失函数，以通过训练后的神经网络模型计算路径损耗的技术方案。在本公开中，既将描述入射场与叠加场的物理公式引入目标损失函数，又将路测数据引入目标损失函数。解决了现有技术中，进行路径损耗估计的方法无法兼顾准确性与计算效率的问题，进而神经网络模型在通过目标损失函数进行训练后，在确定路径损耗的过程中，既提高了路径损耗估计的准确性，又提高了路径损耗估计的计算效率。
附图说明
图1是本公开实施例的一种路径损耗的确定方法的计算机终端的硬件结构框图；
图2是根据本公开实施例的路径损耗的确定方法的流程图；
图3是根据本公开可选实施例的一种路径损耗估计模型的示意图；
图4是根据本公开可选实施例的二维无线通信场景示意图；
图5是根据本公开可选实施例的知识驱动模块UNet网络结构图；
图6是根据本公开可选实施例的数据驱动模块UNet网络结构图；
图7是根据本公开实施例的路径损耗的确定装置的结构框图。
具体实施方式
下文中将参考附图并结合实施例来详细说明本公开的实施例。
需要说明的是，本公开的说明书和权利要求书及上述附图中的术语“第一”、“第二”等 是用于区别类似的对象，而不必用于描述特定的顺序或先后次序。
本公开实施例中所提供的方法实施例可以在计算机终端或者类似的运算装置中执行。以运行在计算机终端上为例，图1是本公开实施例的一种路径损耗的确定方法的计算机终端的硬件结构框图。如图1所示，计算机终端可以包括一个或多个(图1中仅示出一个)处理器102(处理器102可以包括但不限于微处理器(Central Processing Unit，MCU)或可编程逻辑器件(Field Programmable Gate Array，FPGA)等的处理装置)和用于存储数据的存储器104，其中，上述计算机终端还可以包括用于通信功能的传输设备106以及输入输出设备108。本领域普通技术人员可以理解，图1所示的结构仅为示意，其并不对上述计算机终端的结构造成限定。例如，计算机终端还可包括比图1中所示更多或者更少的组件，或者具有与图1所示不同的配置。
存储器104可用于存储计算机程序，例如，应用软件的软件程序以及模块，如本公开实施例中的路径损耗的确定方法对应的计算机程序，处理器102通过运行存储在存储器104内的计算机程序，从而执行各种功能应用以及数据处理，即实现上述的方法。存储器104可包括高速随机存储器，还可包括非易失性存储器，如一个或者多个磁性存储装置、闪存、或者其他非易失性固态存储器。在一些实例中，存储器104可进一步包括相对于处理器102远程设置的存储器，这些远程存储器可以通过网络连接至计算机终端。上述网络的实例包括但不限于互联网、企业内部网、局域网、移动通信网及其组合。
传输设备106用于经由一个网络接收或者发送数据。上述的网络具体实例可包括计算机终端的通信供应商提供的无线网络。在一个实例中，传输设备106包括一个网络适配器(Network Interface Controller，简称为NIC)，其可通过基站与其他网络设备相连从而可与互联网进行通讯。在一个实例中，传输设备106可以为射频(Radio Frequency，简称为RF)模块，其用于通过无线方式与互联网进行通讯。
在本实施例中提供了一种路径损耗的确定方法，运行于图1中的计算机终端，图2是根据本公开实施例的路径损耗的确定方法的流程图，如图2所示，该流程包括如下步骤：
步骤S202，根据第一图像、第二图像和目标网格场景的入射场图像确定所述目标网格场景的叠加场图像，其中，所述第一图像用于指示第一对象在所述目标网格场景中的位置，所述第二图像用于指示第二对象在所述目标网格场景中的位置；
可以理解的是，上述第一对象可以为基站，进而上述第一图像可以为用于指示基站在目标网格场景中的位置的图像。上述第二对象可以为建筑物，进而上述第二图像可以为用于指示建筑物在目标网格场景中的位置的图像。上述目标网格场景包括：基站、建筑物、道路等。
例如：在第一对象为基站且第二对象为建筑物的情况下，可以通过矩量法将一个至少包含基站和建筑物的二维空间进行网格化划分，形成目标网格场景。将目标网格场景中的基站位置设置为第一值(可选地，第一值为1)，将目标网格场景中除基站位置之外的其他位置设置为第二值(可选地，第二值为0)，即可生成第一图像。同理，将目标网格场景中的建筑物位置设置为第一值(可选地，第一值为1)，将目标网格场景中除建筑物位置之外的其他位置设置为第二值(可选地，第二值为0)，即可生成第二图像。
步骤S204，根据所述入射场图像和所述叠加场图像确定第一损失函数；以及根据所述叠加场图像和所述目标网格场景的路测数据确定第二损失函数；
步骤S206，根据所述第一损失函数和所述第二损失函数确定目标损失函数；
可以理解的是，目标损失函数是由第一损失函数和第二损失函数确定的，即通过目标损失函数训练的神经网络模型即满足第一损失函数中的物理公式，又满足第二损失函数中的路测数据。使得神经网络模型即能提高路径损耗估计的准确性和泛化性，又提能高路径损耗估计的计算效率。
步骤S208，根据所述目标损失函数训练神经网络模型，以通过训练后的神经网络模型确定所述目标网格场景的路径损耗。
通过上述步骤，根据用于指示第一对象在目标网格场景中的位置的第一图像、用于指示第二对象在目标网格场景中的位置的第二图像和目标网格图像的入射场图像确定目标网格场景的叠加场图像。根据入射场图像和叠加场图像确定第一损失函数，并根据叠加场图像和路测数据确定第二损失函数。根据第一损失函数和第二损失函数确定用于训练神经网络模型的目标损失函数，进而通过训练后的神经网络模型确定目标网格场景的路径损耗。也就是说，本公开实施例限定了通过叠加场图像和入射场图像确定第一损失函数，以及路测数据和叠加场图像确定第二损失函数，进而根据第一损失函数和第二损失函数确定用于对神经网络模型进行训练的目标损失函数，以通过训练后的神经网络模型计算路径损耗的技术方案。在本公开中，既将描述入射场与叠加场的物理公式引入目标损失函数，又将路测数据引入目标损失函数。解决了现有技术中进行路径损耗估计的方法无法兼顾准确性与计算效率的问题，进而神经网络模型在通过目标损失函数进行训练后，在确定路径损耗的过程中，既提高了路径损耗估计的准确性，又提高了路径损耗估计的计算效率。
可选的，在步骤S202的在根据第一图像、第二图像和目标网格场景的入射场图像确定所述目标网格场景的叠加场图像之前，所述方法还包括：根据所述第一图像确定所述第一对象在所述目标网格场景的第一位置，以及根据所述第二图像确定所述第二对象在所述目标网格场景中的第二位置；根据所述第一位置和所述第二位置确定所述目标网格场景中每个网格的入射场；根据所述每个网格的入射场确定所述目标网格场景的入射场图像。
通过上述技术方案可知，在确定叠加场图像之前，需要确定目标网格场景的入射场图像。确定入射场图像的方法有多种，在本公开中可以根据第一图像和第二图像确定入射场图像。具体的：通过公式：确定目标网格场景中第二位置(每一个建筑物)相对于第一位置(基站)的入射场结果，进而根据入射场结果确定入射场图像。其中，上述代表入射场结果，pn代表第二位置，ptx代表第一位置，G(·)代表二维空间中的格林函数。
其中，代表0阶第二形式的汉克尔函数，是贝塞尔函数的一种，k是波数ε0是自由空间中的介电常数，μ0代表真空中的磁导率，通常被称为真空磁导率，是一个物理常数。
可选的，上述步骤S202的根据第一图像、第二图像和目标网格场景的入射场图像确定所述目标网格场景的叠加场图像，包括：确定所述入射场图像对应的复数形式的第一实部部分和所述入射场图像对应的复数形式的第一虚部部分；根据所述第一实部部分确定第一入射场图像，以及根据所述第一虚部部分确定第二入射场图像；根据所述第一图像、所述第二图像、所述第一入射场图像和所述第二入射场图像确定所述叠加场图像。
其中，根据所述第一图像、所述第二图像、所述第一入射场图像和所述第二入射场图像确定所述叠加场图像，包括：分别提取所述第一图像的第一图像特征、所述第二图像的第二 图像特征、所述第一入射场图像的第三图像特征和所述第二入射场图像的第四图像特征；根据所述第一图像特征、所述第二图像特征、所述第三图像特征和所述第四图像特征确定所述第一图像、所述第二图像、所述第一入射场图像和所述第二入射场图像的关联图像特征；根据所述关联图像特征和体积分方程确定所述叠加场图像。
上述过程限定了确定叠加场图像的技术方案，具体的：
1)确定第一入射场图像和第二入射场图像；
入射场图像对应有复数形式，可以将入射场图像分为由第一实部部分确定的实部图像(即本公开的第一入射场图像)和由第一虚部部分确定的虚部图像(即本公开的第二入射场图像)，其中，实部图像可以用于指示光的强度分布，即光的亮度或暗度分布。而虚部图像可以用于指示相位信息，即光波的相位分布。
2)确定第一图像的第一图像特征、第二图像的第二图像特征、第一入射场图像的第三图像特增和第二入射场图像的第四图像特征；
将第一图像、第二图像、第一入射场图像和第二入射场图像输入至联合网络模型(Unified Network Model，即以UNet为核心的卷积神经网络,也称UNet网络模型)中，通过编码器提取上述第一图像特征、第二图像特征、第三图像特征和第四图像特征。其中，上述图像特征包括但不限于图像的颜色、纹理、形状和边缘等特征。
3)确定关联图像特征；
解码器通过第一图像特征、第二图像特征、第三图像特征和第四图像特征确定关联图像特征，具体的：
解码器可以通过特征匹配：确定第一图像特征、第二图像特征、第三图像特征和第四图像特征的关联关系；对齐与配准：通过图像配准算法对第一图像特征、第二图像特征、第三图像特征和第四图像特征进行配准；特征融合：通过加权平均、特征级联等方式进行特征融合等一系列步骤完成关联图像特征的确定。
4)根据所述关联图像特征和体积分方程确定叠加场图像。
通过体积分方程(I+Wχ)Etot＝Einc和关联图像特征确定叠加场图像。其中，Etot代表叠加场图像，Einc代表入射场图像，W代表对格林函数在单位网格内求积分得到的N×N系数矩阵，χ代表对角矩阵，其对角线上的元素是每个网格的对比度。
确定叠加场图像后可以根据入射场图像和叠加场图像确定第一损失函数，将矩阵形式的体积分方程(I+Wχ)Etot＝Einc引入第一损失函数中，以入射场图像作为学习标签，对叠加场图像进行不断的训练，对网络参数进行优化，使得输出的叠加场接近于理论上的数值计算结果。具体的：第一损失函数E'tot代表输出的叠加场。
确定叠加场图像之后，还可以根据所述叠加场图像和所述目标网格场景的路测数据确定第二损失函数，包括：将所述叠加场图像输入至路损地图转换器中，以指示所述路损地图转换器将所述叠加场图像转换为所述目标网格场景对应的路径损耗地图；根据所述路测数据和所述叠加场图像对所述路径损耗地图进行修正，以确定修正后的路径损耗地图；根据所述修正后的目标路径损耗地图和所述路测数据确定所述第二损失函数。
其中，根据所述路测数据和所述叠加场图像对所述路径损耗地图进行修正，以确定修正后的目标路径损耗地图，包括：将所述叠加场图像输入至深度学习网络中，以确定所述叠加场图像对应的复数形式的第二实部部分和所述叠加场图像对应的复数形式的第二虚部部分； 根据所述第二实部部分和所述第二虚部部分计算所述路径损耗地图的路径损耗数据；根据所述路径损耗数据与所述路测数据之间的偏差对所述路径损耗地图进行修正，以确定所述目标路径损耗地图。
可以理解的是，通过路损地图转换器将叠加场图像转换为目标网格场景的路径损耗地图。进一步的，确定叠加场图像的第二实部部分和第二虚部部分。通过公式确定叠加场的幅度。其中，代表叠加场的幅度，代表第二实部部分，代表第二虚部部分。通过叠加场的幅度可以确定电场的能量，进而得到路径损耗的初步结果，即路径损耗数据。上述路测数据为实际路测数据，路径损耗数据和路测数据之间会存在一定的偏差，可以通过UNet网络学习这个偏差，进而对路径损耗地图进行修正，以得到修正后的路径损耗地图。
进一步的，通过修正后的路径损耗地图和路测数据确定第二损失函数其中，pl'代表修正后的路径损耗地图，pldat代表路测数据。
可选的，根据所述目标损失函数训练神经网络模型，包括：计算步骤：计算所述目标损失函数的梯度；调整步骤：根据预设优化算法调整所述梯度，以根据调整后的梯度更新所述神经网络模型；确定步骤：根据更新后的神经网络模型确定所述目标损失函数的计算结果；循环执行所述调整步骤和所述确定步骤，直至最终得到的计算结果小于或等于预设阈值。
目标损失函数的作用为训练神经网络模型，提高神经网络模型估计路径损耗的准确性、泛化性和计算效率。
通过确定目标损失函数的梯度，通过不断的调整目标损失函数的梯度的方式更新神经网络模型，在根据神经网络模型确定目标损失函数的计算结果小于或等于预设阈值的情况下，可以确定神经网络模型训练完成。
可选的，根据所述第一损失函数和所述第二损失函数确定目标损失函数，包括：确定所述第一损失函数在所述神经网络模型中的第一权重，或确定所述第二损失函数在所述神经网络模型中的第二权重；在已确定所述第一权重的情况下，确定所述第一权重与所述第一损失函数的第一乘积，将所述第一乘积和所述第二损失函数的和确定为所述目标损失函数；在已确定所述第二权重的情况下，确定所述第二权重与所述第二损失函数的第二乘积；将所述第二乘积和所述第一损失函数的和确定为所述目标损失函数。
在λ为第一权重的情况下，目标损失函数在λ为第二权重的情况下，目标损失函数其中，代表第一损失函数，代表第二损失函数，
为了更好的理解上述路径损耗的确定方法的过程，以下再结合可选实施例对上述路径损耗的确定方法的实现流程进行说明，但不用于限定本公开实施例的技术方案。
在无线通信中，对信道的建模一般从大尺度衰落特性和小尺度衰落特性两个方面展开。其中，大尺度衰落特性主要考虑路径损耗这一特性，路径损耗包括传播损耗与阴影衰落两部分。传播损耗与发射器接收器之间的距离有关，阴影衰落则是由场景中的大型障碍物(例如：高山、建筑物)遮挡而形成的。小尺度衰落主要由多径效应产生，与无线传输环境的动态变化有关。
针对于路径损耗，现有技术中通过知识驱动或通过数据驱动的方法估计路径损耗都存在一定的问题，例如：基于传统经验公式和协议标准的路损估计方法，在准确性和泛化性方面存在不足；基于电磁计算的路损估计方法，由于需要在数值上求解麦克斯韦方程，占用计算 与存储资源，影响实际的计算效率；基于数据驱动的神经网络方法，由于模型效果取决于训练用数据，在泛化性上有所欠缺。也就是说，现有技术中路径损耗的估计方法无法兼顾计算路径损耗的准确性、泛化性和计算效率。
本公开可选实施例提出了一种知识数据协同的路径损耗估计模型，本公开可选实施例以体积分方程作为电磁计算的物理知识，通过设计与构造神经网络来代替传统的数值求解方法。具体的：
可选实施例(一)
图3是根据本公开可选实施例的一种路径损耗估计模型的示意图，如图3所示，路径损耗估计模型包括：输入部分、知识驱动模块、数据驱动模块。其中：
1)输入部分：用于通过从公开地图数据中获取城市场景中典型建筑物、道路的空间分布以及尺寸信息，构建基站、散射体的数字化表征，也就是依据矩量法将二维场景网格化(即本公开的目标网格场景)。图4是根据本公开可选实施例的二维无线通信场景示意图，如图4所示，可以将城市中的无线通信场景看作电磁传播场景，以基站作为激励源，城市中的建筑物、植被等作为散射体。
二维场景中的位置信息可以用一个元组p＝(x,y)表示，其中x代表横坐标，y代表纵坐标。对于基站位置图(即本公开的第一图像)，将基站位置(即本公开的第一对象)所在网格设为1，其余网格设为0。对于建筑物分布图(即本公开的第二图像)，将建筑物覆盖(即本公开的第二对象)网格设为1，未覆盖网格设为0。同时根据入射场计算公式：确定每个网格的入射场结果。
其中，代表每个网格的入射场结果，ptx代表基站位置，pn代表建筑物覆盖，G(·)表示二维空间中的格林函数：表示0阶第二形式的汉克尔函数，是贝塞尔函数的一种。k是波数，由此确定目标网格场景的电场入射场图像。由于电场入射场图像对应的复数形式具有实部部分和虚部部分两部分，因此，可以将电场入射场图像拆分为入射场实部图像和入射场虚部图像。
因此，输入部分的输入图像分别为：基站位置图、建筑物分布图、入射场实部图像和入射场虚部图像。
将上述四个图像输入至以UNet为核心的卷积神经网络中。
2)知识驱动模块：用于通过UNet捕捉基站位置、建筑物分布、电场入射场与目标电场叠加场之间的复杂非线性关系。
将矩阵形式的体积分方程(I+Wχ)Etot＝Einc引入损失函数中，其中，Einc代表电场入射场，其中I是N×N单位矩阵，W是对格林函数在单位网格内求积分得到的N×N系数矩阵，χ是对角矩阵，其对角线上的元素是每个网格的对比度。Etot代表目标电场叠加场。
图5是根据本公开可选实施例的知识驱动模块UNet网络结构图，如图5所示，在知识驱动模块中，以入射场Einc作为学习标签，通过不断地训练对网络参数进行优化，使输出的叠加场E'tot接近于理论上的数值计算结果。
进一步的，根据入射场和训练后的叠加场确定第一损失函数
3)数据驱动模块：通过路损地图转换器将电磁计算的电场强度转换为路径损耗地图。在知识驱动模块中，通过UNet网络输出两个通道的结果，分别是估算叠加场的实部和虚部 通过公示计算叠加场的幅度:进而求得电场的能量。通过与发射源的传输功率比较，可以确定路径损耗的初步分布结果。
上述通过电磁计算的路径损耗的初步分布结果与实际路测数据存在一定的偏差。图6是根据本公开可选实施例的数据驱动模块UNet网络结构图，在数据驱动模块，本公开具可选施例采用如图6所示的UNet网络结构学习路径损耗的初步分布结果与实际路测数据存在的偏差，对中间的输出量加以纠正得到修正的路损地图pl'(即本公开的修正后的路径损耗地图)。在训练过程中，以路损地图数据pldat(即本公开的路测数据)作为真值标签，与修正后的路损地图pl′进行比较，通过不断优化网络参数缩小它们之间的差距，使得估计结果更接近真实数据的分布。
进一步的，确定第二损失函数：
在确定第一损失函数与第二损失函数之后，可以将两个损失函数结合在一起，构成整个网络的损失函数。具体的：可以确定第一损失函数对应的相对权重或第二损失函数的相对权重，进而确定目标损失函数。即
在确定目标损失函数后，需要对神经网络模型进行训练，通过计算目标损失函数的梯度并进行梯度反传，同时不断降低目标损失函数的结果，达到优化网络参数的目的，使得神经网络模型的输出结果逐渐接近路测数据真值。
最后，通过训练好的神经网络模型对路径损耗进行估计，具体的：将目标场景(即本公开的目标网格场景)按照输入维度预处理得到基站位置图、建筑物分布图和电场入射场分布图。再将这些特征输入到训练好的神经网络模型中，以输出的修正路损地图作为模型的路径损耗估计结果。
可选实施例(二)
使用无线电地图数据集(RadioMapSeer)当作数据集，其中包括世界上6个不同城市的700个区域，每个区域有80个不同的基站位置。针对每个场景，数据集中包含基站位置图(即本公开的第一图像)、建筑物分布图(即本公开的第二图像)和路损地图。
设定建筑物的相对介电常数，空气的相对介电常数，根据基站的位置信息计算得到入射场的实部图像(即本公开的第一入射场图像)和虚部图像(即本公开的第二入射场图像)。随机抽取700个不同区域中的500个作为训练集，100个作为验证集，剩下的100个作为测试集。这样划分可以使得模型在训练和测试中面对不同的建筑物分布特点，在准确评估模型性能的同时，也可以保证对模型泛化能力的验证。通过设定好训练相关参数(epoch number、batch size、learning rate、optimizer等)，不断从训练集中采样输入到模型中，对模型的参数进行更新，最终得到训练好的模型(即本公开的训练后的神经网络模型)。
进一步的，对于测试场景，按照训练数据的像素与维度进行网格划分，并转化成相应的基站位置图、建筑物分布图、入射场实部图和虚部图，将这四个图像输入到模型中，将修正的路损地图作为模型输出的路径损耗估计结果。
本公开可选实施例融合物理公式的损失函数与基于路测数据的损失函数，使得输出结果既满足物理知识的体积分方程，又符合实际路测数据的分布规律，提高了路损估计的准确性、效率性和泛化性。
通过以上的实施方式的描述，本领域的技术人员可以清楚地了解到根据上述实施例的方法可借助软件加必需的通用硬件平台的方式来实现，当然也可以通过硬件，但很多情况下前 者是更佳的实施方式。基于这样的理解，本公开的技术方案本质上或者说对现有技术做出贡献的部分可以以软件产品的形式体现出来，该计算机软件产品存储在一个存储介质(如只读存储器/随机存取存储器(Read-Only Memory/Random Access Memory，ROM/RAM)、磁碟、光盘)中，包括若干指令用以使得一台终端设备(可以是手机，计算机，服务器，或者网络设备等)执行本公开各个实施例所述的方法。
在本实施例中还提供了一种路径损耗的确定装置，该装置用于实现上述实施例及优选实施方式，已经进行过说明的不再赘述。如以下所使用的，术语“模块”可以实现预定功能的软件和/或硬件的组合。尽管以下实施例所描述的装置较佳地以软件来实现，但是硬件，或者软件和硬件的组合的实现也是可能并被构想的。
图7是根据本公开实施例的路径损耗的确定装置的结构框图，如图7所示，该装置包括：
第一确定模块72，设置为根据第一图像、第二图像和目标网格场景的入射场图像确定所述目标网格场景的叠加场图像，其中，所述第一图像用于指示第一对象在所述目标网格场景中的位置，所述第二图像用于指示第二对象在所述目标网格场景中的位置；
第二确定模块74，设置为根据所述入射场图像和所述叠加场图像确定第一损失函数；以及根据所述叠加场图像和所述目标网格场景的路测数据确定第二损失函数；
第三确定模块76，设置为根据所述第一损失函数和所述第二损失函数确定目标损失函数；
训练模块78，设置为根据所述目标损失函数训练神经网络模型，以通过训练后的神经网络模型确定所述目标网格场景的路径损耗。
在一个示例性实施例中，所述第一确定模块72，还设置为根据所述第一图像确定所述第一对象在所述目标网格场景的第一位置，以及根据所述第二图像确定所述第二对象在所述目标网格场景中的第二位置；根据所述第一位置和所述第二位置确定所述目标网格场景中每个网格的入射场；根据所述每个网格的入射场确定所述目标网格场景的入射场图像。
在一个示例性实施例中，所述第一确定模块72，还设置为确定所述入射场图像对应的复数形式的第一实部部分和所述入射场图像对应的复数形式的第一虚部部分；根据所述第一实部部分确定第一入射场图像，以及根据所述第一虚部部分确定第二入射场图像；根据所述第一图像、所述第二图像、所述第一入射场图像和所述第二入射场图像确定所述叠加场图像。
在一个示例性实施例中，所述第一确定模块72，还设置为分别提取所述第一图像的第一图像特征、所述第二图像的第二图像特征、所述第一入射场图像的第三图像特征和所述第二入射场图像的第四图像特征；根据所述第一图像特征、所述第二图像特征、所述第三图像特征和所述第四图像特征确定所述第一图像、所述第二图像、所述第一入射场图像和所述第二入射场图像的关联图像特征；根据所述关联图像特征和体积分方程确定所述叠加场图像。
在一个示例性实施例中，所述第二确定模块74，还设置为将所述叠加场图像输入至路损地图转换器中，以指示所述路损地图转换器将所述叠加场图像转换为所述目标网格场景对应的路径损耗地图；根据所述路测数据和所述叠加场图像对所述路径损耗地图进行修正，以确定修正后的路径损耗地图；根据所述修正后的目标路径损耗地图和所述路测数据确定所述第二损失函数。
在一个示例性实施例中，所述第二确定模块74，还设置为将所述叠加场图像输入至深度学习网络中，以确定所述叠加场图像对应的复数形式的第二实部部分和所述叠加场图像对应的复数形式的第二虚部部分；根据所述第二实部部分和所述第二虚部部分计算所述路径损耗 地图的路径损耗数据；根据所述路径损耗数据与所述路测数据之间的偏差对所述路径损耗地图进行修正，以确定所述目标路径损耗地图。
在一个示例性实施例中，所述训练模块78，还设置为计算步骤：计算所述目标损失函数的梯度；调整步骤：根据预设优化算法调整所述梯度，以根据调整后的梯度更新所述神经网络模型；确定步骤：根据更新后的神经网络模型确定所述目标损失函数的计算结果；循环执行所述调整步骤和所述确定步骤，直至最终得到的计算结果小于或等于预设阈值。
在一个示例性实施例中，所述第三确定模块76，还设置为确定所述第一损失函数在所述神经网络模型中的第一权重，或确定所述第二损失函数在所述神经网络模型中的第二权重；在已确定所述第一权重的情况下，确定所述第一权重与所述第一损失函数的第一乘积，将所述第一乘积和所述第二损失函数的和确定为所述目标损失函数；在已确定所述第二权重的情况下，确定所述第二权重与所述第二损失函数的第二乘积；将所述第二乘积和所述第一损失函数的和确定为所述目标损失函数。
通过上述装置，由于根据用于指示第一对象在目标网格场景中的位置的第一图像、用于指示第二对象在目标网格场景中的位置的第二图像和目标网格图像的入射场图像确定目标网格场景的叠加场图像。根据入射场图像和叠加场图像确定第一损失函数，并根据叠加场图像和路测数据确定第二损失函数。根据第一损失函数和第二损失函数确定用于训练神经网络模型的目标损失函数，进而通过训练后的神经网络模型确定目标网格场景的路径损耗。也就是说，本公开实施例限定了通过叠加场图像和入射场图像确定第一损失函数，以及路测数据和叠加场图像确定第二损失函数，进而根据第一损失函数和第二损失函数确定用于对神经网络模型进行训练的目标损失函数，以通过训练后的神经网络模型计算路径损耗的技术方案。在本公开中，既将描述入射场与叠加场的物理公式引入目标损失函数，又将路测数据引入目标损失函数。解决了现有技术中，进行路径损耗估计的方法无法兼顾准确性与计算效率的问题，进而神经网络模型在通过目标损失函数进行训练后，在确定路径损耗的过程中，既提高了路径损耗估计的准确性，又提高了路径损耗估计的计算效率。
需要说明的是，上述各个模块是可以通过软件或硬件来实现的，对于后者，可以通过以下方式实现，但不限于此：上述模块均位于同一处理器中；或者，上述各个模块以任意组合的形式分别位于不同的处理器中。
本公开的实施例还提供了一种计算机可读存储介质，该计算机可读存储介质中存储有计算机程序，其中，该计算机程序被设置为运行时执行上述任一项方法实施例中的步骤。
在一个示例性实施例中，上述计算机可读存储介质可以包括但不限于：U盘、只读存储器(Read-Only Memory，简称为ROM)、随机存取存储器(Random Access Memory，简称为RAM)、移动硬盘、磁碟或者光盘等各种可以存储计算机程序的介质。
本公开的实施例还提供了一种电子装置，包括存储器和处理器，该存储器中存储有计算机程序，该处理器被设置为运行计算机程序以执行上述任一项方法实施例中的步骤。
在一个示例性实施例中，上述电子装置还可以包括传输设备以及输入输出设备，其中，该传输设备和上述处理器连接，该输入输出设备和上述处理器连接。
在一个示例性实施例中，上述计算机程序产品还可以包括计算机指令，所述计算机指令被处理器执行时实现本公开各个实施例中所述方法的步骤。
本实施例中的具体示例可以参考上述实施例及示例性实施方式中所描述的示例，本实施 例在此不再赘述。
显然，本领域的技术人员应该明白，上述的本公开的各模块或各步骤可以用通用的计算装置来实现，它们可以集中在单个的计算装置上，或者分布在多个计算装置所组成的网络上，它们可以用计算装置可执行的程序代码来实现，从而，可以将它们存储在存储装置中由计算装置来执行，并且在某些情况下，可以以不同于此处的顺序执行所示出或描述的步骤，或者将它们分别制作成各个集成电路模块，或者将它们中的多个模块或步骤制作成单个集成电路模块来实现。这样，本公开不限制于任何特定的硬件和软件结合。
以上所述仅为本公开的优选实施例而已，并不用于限制本公开，对于本领域的技术人员来说，本公开可以有各种更改和变化。凡在本公开的原则之内，所作的任何修改、等同替换、改进等，均应包含在本公开的保护范围之内。

Claims (12)

  1. 一种路径损耗的确定方法，包括：
    根据第一图像、第二图像和目标网格场景的入射场图像确定所述目标网格场景的叠加场图像，其中，所述第一图像用于指示第一对象在所述目标网格场景中的位置，所述第二图像用于指示第二对象在所述目标网格场景中的位置；
    根据所述入射场图像和所述叠加场图像确定第一损失函数；以及根据所述叠加场图像和所述目标网格场景的路测数据确定第二损失函数；
    根据所述第一损失函数和所述第二损失函数确定目标损失函数；
    根据所述目标损失函数训练神经网络模型，以通过训练后的神经网络模型确定所述目标网格场景的路径损耗。
  2. 根据权利要求1所述的方法，其中，在根据第一图像、第二图像和目标网格场景的入射场图像确定所述目标网格场景的叠加场图像之前，所述方法还包括：
    根据所述第一图像确定所述第一对象在所述目标网格场景的第一位置，以及根据所述第二图像确定所述第二对象在所述目标网格场景中的第二位置；
    根据所述第一位置和所述第二位置确定所述目标网格场景中每个网格的入射场；
    根据所述每个网格的入射场确定所述目标网格场景的入射场图像。
  3. 根据权利要求1所述的方法，其中，根据第一图像、第二图像和目标网格场景的入射场图像确定所述目标网格场景的叠加场图像，包括：
    确定所述入射场图像对应的复数形式的第一实部部分和所述入射场图像对应的复数形式的第一虚部部分；
    根据所述第一实部部分确定第一入射场图像，以及根据所述第一虚部部分确定第二入射场图像：
    根据所述第一图像、所述第二图像、所述第一入射场图像和所述第二入射场图像确定所述叠加场图像。
  4. 根据权利要求3所述的方法，其中，根据所述第一图像、所述第二图像、所述第一入射场图像和所述第二入射场图像确定所述叠加场图像，包括：
    分别提取所述第一图像的第一图像特征、所述第二图像的第二图像特征、所述第一入射场图像的第三图像特征和所述第二入射场图像的第四图像特征；
    根据所述第一图像特征、所述第二图像特征、所述第三图像特征和所述第四图像特征确定所述第一图像、所述第二图像、所述第一入射场图像和所述第二入射场图像的关联图像特征：
    根据所述关联图像特征和体积分方程确定所述叠加场图像。
  5. 根据权利要求1所述的方法，其中，根据所述叠加场图像和所述目标网格场景的路测数据确定第二损失函数，包括：
    将所述叠加场图像输入至路损地图转换器中，以指示所述路损地图转换器将所述叠加场图像转换为所述目标网格场景对应的路径损耗地图；
    根据所述路测数据和所述叠加场图像对所述路径损耗地图进行修正，以确定修正后的路径损耗地图：
    根据所述修正后的目标路径损耗地图和所述路测数据确定所述第二损失函数。
  6. 根据权利要求5所述的方法，其中，根据所述路测数据和所述叠加场图像对所述路径损耗地图进行修正，以确定修正后的目标路径损耗地图，包括：
    将所述叠加场图像输入至深度学习网络中，以确定所述叠加场图像对应的复数形式的第二实部部分和所述叠加场图像对应的复数形式的第二虚部部分；
    根据所述第二实部部分和所述第二虚部部分计算所述路径损耗地图的路径损耗数据；
    根据所述路径损耗数据与所述路测数据之间的偏差对所述路径损耗地图进行修正，以确定所述目标路径损耗地图。
  7. 根据权利要求1所述的方法，其中，根据所述目标损失函数训练神经网络模型，包括：
    计算步骤：计算所述目标损失函数的梯度；
    调整步骤：根据预设优化算法调整所述梯度，以根据调整后的梯度更新所述神经网络模型：
    确定步骤：根据更新后的神经网络模型确定所述目标损失函数的计算结果；
    循环执行所述调整步骤和所述确定步骤，直至最终得到的计算结果小于或等于预设阈值。
  8. 根据权利要求1所述的方法，其中，根据所述第一损失函数和所述第二损失函数确定目标损失函数，包括：
    确定所述第一损失函数在所述神经网络模型中的第一权重，或确定所述第二损失函数在所述神经网络模型中的第二权重；
    在已确定所述第一权重的情况下，确定所述第一权重与所述第一损失函数的第一乘积，将所述第一乘积和所述第二损失函数的和确定为所述目标损失函数；
    在已确定所述第二权重的情况下，确定所述第二权重与所述第二损失函数的第二乘积；将所述第二乘积和所述第一损失函数的和确定为所述目标损失函数。
  9. 一种路径损耗的确定装置，包括：
    第一确定模块，设置为根据第一图像、第二图像和目标网格场景的入射场图像确定所述目标网格场景的叠加场图像，其中，所述第一图像用于指示第一对象在所述目标网格场景中的位置，所述第二图像用于指示第二对象在所述目标网格场景中的位置；
    第二确定模块，设置为根据所述入射场图像和所述叠加场图像确定第一损失函数；以及根据所述叠加场图像和所述目标网格场景的路测数据确定第二损失函数；
    第三确定模块，设置为根据所述第一损失函数和所述第二损失函数确定目标损失函数；
    训练模块，设置为根据所述目标损失函数训练神经网络模型，以通过训练后的神经网络模型确定所述目标网格场景的路径损耗。
  10. 一种计算机可读存储介质，所述计算机可读存储介质中存储有计算机程序，其中，所述计算机程序被处理器执行时实现所述权利要求1至8任一项中所述的方法的步骤。
  11. 一种电子装置，包括存储器、处理器以及存储在所述存储器上并可在所述处理器上运行的计算机程序，所述处理器执行所述计算机程序时实现所述权利要求1至8任一项中所述的方法的步骤。
  12. 一种计算机程序产品，包括计算机指令，其中，所述计算机指令被处理器执行时实现所述权利要求1至8任一项中所述的方法的步骤。
PCT/CN2024/118596 2024-02-22 2024-09-12 路径损耗的确定方法和装置、存储介质及电子装置 Pending WO2025175751A1 (zh)

Applications Claiming Priority (2)

Application Number Priority Date Filing Date Title
CN202410199012.5A CN120526279A (zh) 2024-02-22 2024-02-22 路径损耗的确定方法和装置、存储介质及电子装置
CN202410199012.5 2024-02-22

Publications (1)

Publication Number Publication Date
WO2025175751A1 true WO2025175751A1 (zh) 2025-08-28

Family

ID=96747758

Family Applications (1)

Application Number Title Priority Date Filing Date
PCT/CN2024/118596 Pending WO2025175751A1 (zh) 2024-02-22 2024-09-12 路径损耗的确定方法和装置、存储介质及电子装置

Country Status (2)

Country Link
CN (1) CN120526279A (zh)
WO (1) WO2025175751A1 (zh)

Citations (4)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN114692929A (zh) * 2020-12-31 2022-07-01 上海华为技术有限公司 一种传播路损的预测方法以及计算设备
CN115175103A (zh) * 2022-07-06 2022-10-11 中国联合网络通信集团有限公司 路径损耗估计方法、装置、设备和介质
CN115866656A (zh) * 2021-09-26 2023-03-28 华为技术有限公司 路径损耗计算方法及装置
WO2023061303A1 (zh) * 2021-10-13 2023-04-20 中兴通讯股份有限公司 大尺度衰落的建模及估计方法、系统、设备和存储介质

Patent Citations (4)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN114692929A (zh) * 2020-12-31 2022-07-01 上海华为技术有限公司 一种传播路损的预测方法以及计算设备
CN115866656A (zh) * 2021-09-26 2023-03-28 华为技术有限公司 路径损耗计算方法及装置
WO2023061303A1 (zh) * 2021-10-13 2023-04-20 中兴通讯股份有限公司 大尺度衰落的建模及估计方法、系统、设备和存储介质
CN115175103A (zh) * 2022-07-06 2022-10-11 中国联合网络通信集团有限公司 路径损耗估计方法、装置、设备和介质

Also Published As

Publication number Publication date
CN120526279A (zh) 2025-08-22

Similar Documents

Publication Publication Date Title
CN105528785A (zh) 一种双目视觉图像立体匹配方法
CN103841602B (zh) 一种邻区配置方法及系统
CN111147163A (zh) 一种基于dnn神经网络的无线通信链路损耗预测方法
CN109635748A (zh) 高分辨率影像中道路特征的提取方法
CN113194493A (zh) 基于图神经网络的无线网络数据缺失属性恢复方法及装置
Gao et al. Localization-oriented digital twinning in 6G: A new indoor-positioning paradigm and proof-of-concept
CN118678377B (zh) 基于异构图神经网络的频谱覆盖地图构建方法及装置
CN112866904A (zh) 基于波束索引地图的免信道训练大维通信波束对齐方法
CN111667409B (zh) 一种基于超分辨率算法的绝缘子影像分辨率增强方法
CN104113849A (zh) 传播模型的校正方法
Li et al. RMTransformer: Accurate radio map construction and coverage prediction
CN116321187A (zh) 一种信号路径损耗预测模型的训练方法及装置
CN110659593A (zh) 基于改进DiracNet的城市雾霾能见度检测方法
WO2025175751A1 (zh) 路径损耗的确定方法和装置、存储介质及电子装置
CN108207005B (zh) 一种lte无线网络的评估方法及服务器
CN119915685B (zh) 一种基于半动态图神经网络的区域多站点pm2.5预测方法
CN113141570A (zh) 地下场景定位方法、装置、计算设备及计算机存储介质
CN118509037B (zh) 基于卫星互联网的无人机电力巡检不间断通信方法及系统
CN114710742A (zh) 一种基于多链插值构建指纹地图的室内定位方法
CN114997047B (zh) 一种基于循环生成对抗网络的电磁频谱信息补全方法
CN104507102A (zh) 一种基于rssi和fmm的空间特性估计方法
Flórez-González et al. Fast indoor radio propagation prediction using deep learning
CN120201513B (zh) 结合通信路径状态监测的无线通信控制方法及系统
Du et al. Path loss map construction based on radioresunet
CN108271183A (zh) 一种基于原始测量报告的邻区优化方法及装置

Legal Events

Date Code Title Description
121 Ep: the epo has been informed by wipo that ep was designated in this application

Ref document number: 24925491

Country of ref document: EP

Kind code of ref document: A1