Wireless channel path loss prediction method and device based on multi-view visual perception
Through multi-view vision perception technology, multi-view perception images in dynamic scenarios are obtained and convolutional neural networks are used to predict the path loss problem, which solves the problem of insufficient accuracy of path loss prediction in high-dynamic environments in the existing technology, and achieves high precision and Support for diversified application scenarios.
Patent Information
- Application Number
- CN202411075402.8
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2024-08-07
- Publication Date
- 2024-11-29
AI Technical Summary
The existing wireless channel path loss prediction method is difficult to achieve accurate prediction of diverse application scenarios in high dynamic environments, especially because a single device perceives that there is insufficient field of view, object occlusion and depth information, which cannot meet the needs of future 6G networks. .
Using a multi-view visual perception method, multi-view perception images in dynamic scenes are obtained, target features are processed and determined, and pre-trained convolutional neural network model is used to predict the path loss, and combined with multi-view selection algorithm to select the optimal perceptual view. , realize high-precision path loss prediction.
This method improves the accuracy of road loss prediction and the richness of application scenarios, solves the disadvantages of limited perception capabilities of a single environment perception device and complex measurement or simulation calculation, and is suitable for the diversity and high dynamic needs of future mobile communication systems.
Smart Images

Figure CN119051786A_ABST
Abstract
Description
Technical Field
[0001] The present application relates to the field of wireless channel prediction, and in particular to a method and device for predicting wireless channel path loss based on multi-view visual perception. Background Art
[0002] The sixth generation of mobile communication networks (6G) presents a development trend of multiple technologies, multiple frequency bands, and multiple scenarios, which puts higher requirements on communication systems. As one of the six modules of the communication system, efficient and accurate path loss prediction of the channel is crucial for the design, deployment, and optimization of 6G networks. Environmental perception provides a new dimension for path loss prediction, which can support the prediction needs of more diverse application scenarios.
[0003] In recent years, with the integration of sensor devices into the network and the development of artificial intelligence technology, channel prediction methods based on environmental information have received widespread attention. Although there are many attempts to use environmental data for channel prediction, existing solutions still have difficulty in dealing with highly dynamic environments. In addition, single device perception has defects such as limited field of view, object occlusion, and insufficient depth information.
[0004] There are two types of traditional path loss modeling methods in related technologies: deterministic modeling and statistical modeling. Statistical modeling methods, such as free space path loss model, Okumura-Hata model, logarithmic distance path loss model, ABG (Alpha-Beta-Gamma) model and other path loss models, are statistical modeling methods that first obtain path loss models for typical scenarios in turn, and then perform network design through offline simulation. Deterministic modeling methods, such as ray tracing (RT) methods using a detailed three-dimensional model of the environment, can provide more accurate prediction results, but due to the high computational complexity, the corresponding implementation cost also increases. The common point of the above two methods is that they require complex measurements or simulations, have long iteration cycles, and are difficult to achieve accurate path loss calculations in a variety of application scenarios.
[0005] There is another related method, a path loss prediction method based on environmental information has been proposed, such as using satellite two-dimensional images, city map information, lidar point cloud, environmental bird's-eye view and height information and other environmental information to achieve path loss prediction. However, the above methods all focus on static scenes and lack real-time perception of dynamic environments; and single device perception has defects such as limited field of view, object occlusion and insufficient depth information, making it difficult to achieve comprehensive environmental perception. Faced with the development trend of high-dynamic communication scenarios of future 6G networks, the shortcomings of the above methods have caused them to be unable to meet the needs. Summary of the invention
[0006] The purpose of this application is to provide a method and device for predicting wireless channel path loss based on multi-view visual perception, so as to solve the problem that the prior art cannot realize path loss prediction for diverse application scenarios.
[0007] To achieve the above object, an embodiment of the present application provides a method for predicting wireless channel path loss based on multi-view visual perception, including:
[0008] Acquire multi-view perception images in dynamic scenes;
[0009] Processing the perceived image to determine target features for road loss prediction;
[0010] Inputting the target features into a pre-trained convolutional neural network model to obtain output results corresponding to different perspectives;
[0011] According to the plurality of output results, a target path loss corresponding to the optimal perception angle is determined; the target path loss is used to represent the path loss prediction result of the multi-perspective perception image. Here, the target path loss may also represent the path loss prediction corresponding to the optimal perception angle in the multi-perspective perception image.
[0012] Optionally, obtaining multi-view perception images in a dynamic scene includes:
[0013] Determine the perception device deployment parameters corresponding to the target scenario;
[0014] Using the image acquisition device corresponding to the perception device deployment parameters, acquiring multi-perspective perception images in dynamic scenes;
[0015] Among them, the perception device deployment parameters include: the horizontal viewing angle of the image acquisition device is greater than a first value, and the first height of the image acquisition device is less than or equal to the second height of the house corresponding to the target scene, and the vertical viewing angle of the image acquisition device has a second value, and the second value is determined by the first height, the length of the house, and the width of the house.
[0016] Optionally, processing the perceived image to determine target features for road loss prediction includes:
[0017] Preprocessing the perceived image to determine scatterers that affect channel variation as effective image features;
[0018] According to the transmitter position and the receiver position in the dynamic scene, the distance characteristics between the transmitter and the receiver are obtained;
[0019] Inputting the effective image features and the perceived image into a preset residual network, and outputting the sight path existence features;
[0020] The effective image feature, the distance feature and the sight path existence feature are determined as the target feature.
[0021] Optionally, before inputting the target feature into a pre-trained convolutional neural network model to obtain output results corresponding to different perspectives, the method further includes:
[0022] Construct a target deep learning model; the target deep learning model has four stages, each stage contains two basic blocks; each basic block contains two 3*3 convolutional layers and one skip connection;
[0023] Determine a data set corresponding to the perception image and the road loss parameters; the data set includes a training data set and a test data set;
[0024] The target deep learning model is trained and tested using the training data set and the test data set to determine the convolutional neural network model.
[0025] Optionally, determining a data set corresponding to the perception image and the path loss parameter includes:
[0026] Acquiring physical information of a propagation environment corresponding to the perceived image;
[0027] Constructing a three-dimensional environment model according to the physical information;
[0028] Performing channel simulation on the three-dimensional environment model to generate corresponding path loss parameters for each environment sample;
[0029] According to the perceived image, a data set corresponding to the perceived image and the path loss parameter is determined.
[0030] Optionally, the target features are input into a pre-trained convolutional neural network model to obtain output results corresponding to different perspectives, including:
[0031] The target features are input into a pre-trained convolutional neural network model, and the image features, distance features, and sight path existence features in the target features are fused using the convolutional neural network model to obtain fused perceived environment features;
[0032] The linearity and nonlinearity in the convolutional neural network model are used to obtain output results corresponding to different perspectives; wherein the linearity is implemented according to the linear layer module in the convolutional neural network model; the linear layer module includes two hidden layers, each having a first preset value and a second preset value of unit numbers; the first preset value is greater than the second preset value; the nonlinearity is implemented according to the linear rectification function in the convolutional neural network model.
[0033] Optionally, determining a target path loss corresponding to the optimal perception perspective according to the plurality of output results includes:
[0034] Determine the proportion of the black part of the image in different output results according to the plurality of output results; the proportion of the black part of the image is the ratio of the number of black pixels in the image at the current viewing angle to the total number of pixels in the image at the current viewing angle; each of the output results has a corresponding relationship with a perceived image at a viewing angle;
[0035] Determine an average value of the proportion of black portions of a plurality of the images;
[0036] If the proportion of the black part of multiple output results is greater than or less than the average value, the result with a larger proportion among the multiple output results is selected as the target path loss; if only one output result of the black part of multiple output results is greater than the average value, the result greater than the average value is selected as the target path loss.
[0037] To achieve the above object, an embodiment of the present application provides a wireless channel path loss prediction device based on multi-view visual perception, comprising:
[0038] A first acquisition module is used to acquire multi-view perception images in dynamic scenes;
[0039] A first determination module, used for processing the perception image to determine target features for road loss prediction;
[0040] A first processing module is used to input the target features into a pre-trained convolutional neural network model to obtain output results corresponding to different viewing angles;
[0041] The second determination module is used to determine the target path loss corresponding to the optimal perception perspective according to the multiple output results; the target path loss is used to represent the path loss prediction result of the perception image of the multiple perspectives.
[0042] To achieve the above objectives, an embodiment of the present application provides a readable storage medium on which a program or instruction is stored. When the program or instruction is executed by a processor, the steps in the method described above are implemented.
[0043] To achieve the above objective, an embodiment of the present application provides a computer program product, including computer instructions, which implement the steps of the above method when executed by a processor.
[0044] The beneficial effects of the above technical solution of the present application are as follows:
[0045] In an embodiment of the present application, multi-perspective perception images in a dynamic scene are obtained, and the perception images are processed to determine target features for road loss prediction; the target features are input into a pre-trained convolutional neural network model to obtain output results corresponding to different perspectives; based on multiple output results, a target road loss corresponding to the optimal perception perspective for representing the road loss prediction results of the multi-perspective perception images is determined. The present application perceives the actual dynamic scene based on a multi-perspective camera, uses the perception images and the convolutional neural network model for prediction, and selects the optimal perception perspective for the prediction results, thereby realizing high-precision road loss prediction under multi-perspective perception, enriching the application scenarios of road loss prediction, and improving the accuracy of road loss prediction, solving the problems of using complex measurements or simulations to calculate path loss in related technologies and solving the problem of limited perception capabilities of a single environmental perception device. BRIEF DESCRIPTION OF THE DRAWINGS
[0046] Figure 1 One of the flowcharts of the wireless channel path loss prediction method based on multi-view visual perception provided in an embodiment of the present application;
[0047] Figure 2 The second flowchart of the wireless channel path loss prediction method based on multi-view visual perception provided in the embodiment of the present application;
[0048] Figure 3 A cumulative distribution function diagram of prediction errors generated by the convolutional neural network model provided in an embodiment of the present application under different inputs from a single perspective;
[0049] Figure 4 A schematic diagram of the relationship between multi-view perception, single-view perception and Rx position movement provided in an embodiment of the present application;
[0050] Figure 5 A structural diagram of a wireless channel path loss prediction device based on multi-view visual perception provided in an embodiment of the present application. DETAILED DESCRIPTION
[0051] In order to make the technical problems, technical solutions and advantages to be solved by the present application clearer, a detailed description will be given below with reference to the accompanying drawings and specific embodiments.
[0052] It should be understood that the references to "one embodiment" or "an embodiment" throughout the specification mean that the specific features, structures, or characteristics associated with the embodiment are included in at least one embodiment of the present application. Therefore, the references to "in one embodiment" or "in an embodiment" appearing throughout the specification do not necessarily refer to the same embodiment. In addition, these specific features, structures, or characteristics may be combined in any suitable manner in one or more embodiments.
[0053] In the embodiments of the present application, the term "and / or" describes the association relationship of the associated objects, indicating that three relationships may exist. For example, A and / or B may represent: A exists alone, A and B exist at the same time, and B exists alone. The character " / " generally indicates that the associated objects before and after are in an "or" relationship.
[0054] In the embodiments of the present application, the term "plurality" refers to two or more than two, and other quantifiers are similar.
[0055] The following will be combined with the drawings in the embodiments of the present application to clearly and completely describe the technical solutions in the embodiments of the present application. Obviously, the described embodiments are only part of the embodiments of the present application, not all of the embodiments. Based on the embodiments in the present application, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of this application.
[0056] The embodiment of the present application provides a relay link connection control method and device. The method and device are based on the same application concept. Since the method and device solve the problem in a similar principle, the implementation of the device and the method can refer to each other, and the repeated parts will not be repeated.
[0057] like Figure 1 As shown, a method for predicting wireless channel path loss based on multi-view visual perception is provided in an embodiment of the present application, including:
[0058] Step 11, obtaining multi-perspective perceived images in a dynamic scene. Here, the multi-perspective perceived images include perceived images of at least two perspectives.
[0059] In order to obtain multi-perspective perception images in dynamic scenes, this application requires the deployment of image perception devices in advance. The image perception devices can be selected as multi-perspective cameras. The deployment of multi-perspective cameras needs to fully consider factors such as light interference in the real world, the perception field of view, and the practicality of the device deployment location. Use the deployed image perception devices to obtain multi-perspective perception images in dynamic scenes.
[0060] Step 12: Process the perceived image to determine target features for road loss prediction.
[0061] The present application can use an image processing algorithm to process the original perceived image. The target features obtained after processing for path loss prediction include the transmitter (Transmitter, Tx) position, the receiver (Receiver, Rx) position, the distance between the transceiver ends, that is, the distance between Tx and Rx, the existence of the line-of-sight (LOS) path, etc.
[0062] Step 13: input the target features into a pre-trained convolutional neural network model to obtain output results corresponding to different perspectives.
[0063] Optionally, the convolutional neural network (CNN) model of the present application is a CNN model improved based on the ResNet-18 model of the present application, which is a prediction model used for road loss prediction in the present application.
[0064] Step 14: Determine a target path loss corresponding to an optimal perception perspective based on the plurality of output results; the target path loss is used to represent a path loss prediction result of the perception image of the multiple perspectives.
[0065] This application compares multiple output results with a preset threshold. If at least two of them are greater than the preset threshold, the one with a larger pixel ratio is selected as the target road loss; if all are less than the preset threshold, the one with a larger pixel ratio is selected as the target road loss. After determining the target road loss, this application can evaluate the accuracy of the LOS path existence prediction and road loss prediction, and evaluate the effectiveness of the multi-view selection algorithm. This application introduces multi-view perception of images in dynamic scenes, and proposes a method for implementing road loss prediction based on visual perception, which improves prediction accuracy while achieving comprehensive scene perception, and solves the problem of limited perception capabilities of a single environmental perception device.
[0066] Reference Figure 2 As shown, in a specific implementation, the present application provides a high-precision road loss prediction method based on multi-view visual perception. Perform differential analysis on the characteristics of the actual scene, and deploy multi-view cameras at appropriate positions and angles to perceive the physical environment. Secondly, design a LOS path occlusion prediction network based on perceived images, and integrate several features such as image features, transceiver distance, and LOS path occlusion conditions. Finally, use a multi-view selection algorithm to output the optimal road loss prediction result. The specific steps include: Step 1, deploy multi-view cameras to perceive the environment in the actual scene; Step 2, construct a perceived image-road loss data set; Step 3, perform perceived image processing to extract effective features for road loss prediction; Step 4, improve the CNN network, and perform road loss prediction based on perceived images. Step 5, select the optimal perceived perspective for the prediction model. Step 6, evaluate the accuracy of the proposed model.
[0067] Optionally, step 11 of the present application includes:
[0068] Determine the perception device deployment parameters corresponding to the target scenario;
[0069] Using the image acquisition device corresponding to the perception device deployment parameters, acquiring multi-perspective perception images in dynamic scenes;
[0070] Among them, the perception device deployment parameters include: the horizontal viewing angle of the image acquisition device is greater than a first value, and the first height of the image acquisition device is less than or equal to the second height of the house corresponding to the target scene, and the vertical viewing angle of the image acquisition device has a second value, and the second value is determined by the first height, the length of the house, and the width of the house.
[0071] In the embodiment of the present application, it is necessary to comprehensively consider factors such as light interference, perception field of view, and practicality of device deployment location in the real world, and deploy multi-view cameras in actual scenes to obtain environmental images. First, it is necessary to determine the perception device deployment parameters corresponding to the target scene. In this step, in the actual scene, the appropriate camera field of view and deployment height are selected according to the scene size to collect environmental image samples under the target scene. Provide perception image data for subsequent steps. The specific steps are as follows:
[0072] In order to select the appropriate field of view and deployment height of the image acquisition device (such as a camera) to achieve full coverage of the scene range, this application analyzes the target scene and uses the size of the room for testing. The requirements met are:
[0073]
[0074] Among them, θ H and θ V Respectively represent the horizontal and vertical viewing angles of the image acquisition device, that is, the horizontal and vertical viewing angles of the camera. c is the first height of the image acquisition device, i.e., the camera height. h represents the second height of the house corresponding to the target scene, i.e., the height of the house. l and w are the length and width of the room, respectively.
[0075] And the rotation angle of the image acquisition device (such as a camera) satisfies:
[0076]
[0077] The effective propagation environment collected here is 9×6×2.5m 3 A camera with a horizontal field of view of 110° and a vertical field of view of 75° was used, mounted at a height of 2m in the corner of the room with a rotation angle of 35°.
[0078] Among them, Tx is set in the upper right corner of the room, and Rx is set as a movable human mobile phone terminal. And the target scene contains three scatterers. The training and test data are constructed by randomly placing scatterers with different initial positions for generalization verification of the prediction method. For each time Rx moves to a new position, there are 10 groups of random initial positions corresponding to scatterers. Collect multi-perspective perception images of the target scene. The sample of perception images collected by the deployed cameras in dynamic scenes can be expressed as: in 9×6×2.5m 3 A total of 1851 training data, 232 validation data and 232 test data were collected in the propagation environment.
[0079] Optionally, step 12 of the present application includes:
[0080] Preprocessing the perceived image to determine scatterers that affect channel variation as effective image features;
[0081] According to the transmitter position and the receiver position in the dynamic scene, the distance characteristics between the transmitter and the receiver are obtained;
[0082] Inputting the effective image features and the perceived image into a preset residual network, and outputting the presence features of the sight path;
[0083] The effective image feature, the distance feature and the sight path existence feature are determined as the target feature.
[0084] In an embodiment of the present application, an image processing algorithm is used to preprocess the original perceived image, and scatterers that affect channel changes are determined as effective image features. Effective image features are preprocessed image features. Preprocessing includes foreground and background segmentation, binarization, image denoising, image filtering, etc. Preprocessing is used to retain image features that are effective for path loss prediction, that is, to determine effective image features. Furthermore, the distance feature can obtain the transmitter position and receiver position in a dynamic scene according to the parameters set during channel simulation. For example, the parameters that can be determined during channel simulation include: Tx position features and Rx position features. The distance features between the transmitter and the receiver can be determined based on the Tx position features and the Rx position features. The line-of-sight path existence feature is determined using effective image features and the original perceived image. The target features of the present application are used as input features of a pre-trained convolutional neural network model. The target features include effective image features, transceiver distance features, LOS path existence features, etc.
[0085] Specifically, in this step, the multi-perspective perception images are binarized to extract effective image features. By analyzing the wireless propagation mechanism, the main factors that have a direct impact on the path loss are determined. Provide feature assistance for subsequent prediction models. For example: Perception image processing: Use the maximum inter-class variance algorithm to binarize the image, remove redundant background, and focus on scene elements that have a key role in channel changes, that is, scatterers that have a key impact on channel changes, as effective image features. Perform feature selection: By analyzing the wireless propagation mechanism, the main factors that have a direct impact on the path loss are determined. Here, the distance between Tx and Rx and the existence of the LOS path are used to assist the image in predicting the path loss.
[0086] It should be noted that the LOS path existence feature is determined by learning the LOS path existence condition that has a key impact on the path loss using the CNN network. The ResNet residual neural network is used here. As a mature deep learning network, ResNet can effectively handle nonlinear classification problems. For this specific task, the main choice of CNN is the pre-trained ResNet-18 model. The model is modified on the original network structure of the ResNet-18 model by removing the last layer and replacing it with a fully connected layer of neurons of size 2, changing it to a binary classification network and outputting the LOS path existence condition.
[0087] For example, based on the improved ResNet-18 model, this application uses the preprocessed effective image features and the original red, green, and blue primary color (Red, Green, Blue, RGB) images in the original perceived image as deep learning network input to predict the existence characteristics of the LOS path in a wireless propagation environment.
[0088] Optionally, before step 13, the method of the present application further includes:
[0089] Construct a target deep learning model; the target deep learning model has four stages, each stage contains two basic blocks; each basic block contains two 3*3 convolutional layers and one skip connection;
[0090] Determine a data set corresponding to the perception image and the road loss parameters; the data set includes a training data set and a test data set;
[0091] The target deep learning model is trained and tested using the training data set and the test data set to determine the convolutional neural network model.
[0092] In an embodiment of the present application, before performing road loss prediction, it is necessary to first construct a convolutional neural network model for road loss prediction. The convolutional neural network model of the present application is an improved CNN model. The road loss prediction task is a continuous floating point value, which is essentially a regression task. Here, an improved network based on ResNet is used to predict road loss and output the road loss value corresponding to the perceived image. In the feature extraction stage, the present application adopts the ResNet-18 structure, which includes 4 stages, and each stage contains 2 basic blocks. Each basic block contains two 3x3 convolutional layers and a jump connection; the last fully connected layer of the ResNet-18 model is modified to adjust its output dimension to 256, thereby obtaining a target deep learning model. Using a data set with a corresponding relationship between the perceived image and the road loss parameters determined by the perceived image, feature training and testing are performed on the target deep learning model to determine the convolutional neural network model for road loss prediction of the present application.
[0093] Furthermore, step 13 in the present application includes:
[0094] The target features are input into a pre-trained convolutional neural network model, and the image features, distance features, and sight path existence features in the target features are fused using the convolutional neural network model to obtain fused perceived environment features;
[0095] The linearity and nonlinearity in the convolutional neural network model are used to obtain output results corresponding to different perspectives; wherein the linearity is implemented according to the linear layer module in the convolutional neural network model; the linear layer module includes two hidden layers, each having a first preset value and a second preset value of unit numbers; the first preset value is greater than the second preset value; the nonlinearity is implemented according to the linear rectification function in the convolutional neural network model. Here, the number of units of the first preset value can be selected as 128 units; the number of units of the second preset value can be selected as 64 units.
[0096] In the embodiment of the present application, the target feature is input into the convolutional neural network model to obtain the image feature F ing ∈R 256*1 , distance feature F dist ∈R 1*1 、LOS path has feature F LOS ∈R 1*1 , and the image feature F ing , distance feature F dist ∈R 1*1 、LOS path has feature F LOS ∈R 1*1 Perform feature fusion to obtain the fused perceived environment feature F fusion ∈R 258*1The present application also introduces an additional linear layer module into the convolutional neural network model, which includes two hidden layers with 128 and 64 units respectively, and increases nonlinearity through an activation function (such as a linear rectification function, such as the activation function ReLU), that is, utilizing the linear characteristics of the linear layer module and the nonlinear characteristics of the activation function, and completing the path loss prediction output of the network through a linear layer with 1 unit, and obtaining output results corresponding to different perspectives.
[0097] Optionally, in the present application, determining a data set corresponding to a perception image and a path loss parameter includes:
[0098] Acquiring physical information of a propagation environment corresponding to the perceived image;
[0099] Constructing a three-dimensional environment model according to the physical information;
[0100] Performing channel simulation on the three-dimensional environment model to generate corresponding path loss parameters for each environment sample;
[0101] According to the perceived image, a data set corresponding to the perceived image and the path loss parameter is determined.
[0102] In an embodiment of the present application, determining the data set corresponding to the perceived image and the path loss parameters includes the following steps: environmental information collection, obtaining the physical information of the propagation environment corresponding to the perceived image by a certain method, including actual measurement, map-assisted acquisition, automatic acquisition using computer vision related technology, etc. Obtain the overall layout of the original environment and scatterer information, such as scene size, scatterer position, scatterer size information, Tx position, Rx position, transceiver distance, etc. Based on the collected environmental information, a three-dimensional (3D) modeling tool is used to construct a three-dimensional (3D) environmental model in the same coordinate system as the actual scene. A channel simulation tool is used to generate the corresponding channel parameters for each environmental sample, and the channel parameters can be path loss parameters. And based on the path loss parameters and the perceived image, a data set with a corresponding relationship between the perceived image and the path loss parameters is determined, and the data set includes a training data set and a test data set.
[0103] Specifically, in this step, the physical information of the scatterer is obtained through actual measurement, so as to use the 3D modeling tool to construct the environmental 3D model sample. The corresponding channel of each environmental sample is generated by using the channel simulation tool, and the path loss information is extracted. A data set of perception image-path loss is constructed to provide data support for the subsequent path loss prediction model. Specifically, it includes: environmental information collection, obtaining the overall layout and scatterer information of the original environment corresponding to the perception image through actual measurement, including scene size, scatterer position, and scatterer size information. A 3D environmental model is constructed in the same coordinate system as the actual scene using a 3D modeling tool. The environmental samples are input into the channel simulation tool one by one, and the corresponding path loss data is obtained by performing channel simulation by setting the same communication parameter configuration, Tx, and Rx positions; the simulation parameters are set here as sixth-order reflection and first-order diffraction. A data set of perception image-path loss is constructed. Here, a 28GHz channel corresponding to the perception image is generated, the path loss parameters are extracted, and a training data set for prediction model training and a test data set for prediction model testing are constructed.
[0104] Optionally, step 14 of the present application includes:
[0105] Determine the proportion of the black part of the image in different output results according to the plurality of output results; the proportion of the black part of the image is the ratio of the number of black pixels in the image at the current viewing angle to the total number of pixels in the image at the current viewing angle; each of the output results has a corresponding relationship with a perceived image at a viewing angle;
[0106] Determine an average value of the proportion of black portions of a plurality of the images;
[0107] If the proportion of the black part of multiple output results is greater than or less than the average value, the result with a larger proportion among the multiple output results is selected as the target path loss; if only one output result of the black part of multiple output results is greater than the average value, the result greater than the average value is selected as the target path loss.
[0108] In an embodiment of the present application, by performing image analysis on the perceived images of multiple perspectives and designing a multi-perspective selection algorithm, high-precision road loss prediction under multi-perspective perception is achieved. The multi-perspective perceived images of the present application include perceived images of at least two perspectives; a perceived image of one perspective corresponds to an output result, and therefore, the present application analyzes and compares at least two output results. First, the pixel ratio in the perceived image of at least two output results is calculated to provide a basis for subsequent perspective selection. For example, the pixel ratio in the perceived image of each output result of the multiple perspectives is calculated, that is, the ratio of the black part in the binary image corresponding to each output result is calculated. i .
[0109]
[0110] Among them, N black,i N represents the number of black pixels in the i-th viewing angle image (the viewing angle image corresponding to the i-th output result); total,i Represents the total number of pixels in the i-th view image.
[0111] Further multi-view selection includes: setting a pixel threshold, selecting the predicted value corresponding to the view whose pixel ratio is greater than the pixel threshold as the algorithm output. Specifically, the average value of the ratio is selected as the pixel threshold T i . If the proportion of the black part of the i-th perspective is greater than the pixel threshold, it is considered that there is less redundant information in the image at this time, and the perception image of this perspective is selected for path loss calculation. If the output results corresponding to at least two perspective images are all greater than or all less than the pixel threshold, directly compare the proportion of the black part of the two perspective images, and select the output result of the perspective image with a larger proportion as the target path loss. If only one of the output results corresponding to at least two perspective images is greater than the pixel threshold, directly use the output result greater than the pixel threshold as the target path loss.
[0112] Of course, if two or more of the output results corresponding to at least two perspective images are greater than the pixel threshold, then among the two or more results greater than the pixel threshold, the output result of the perspective image with a larger proportion is selected as the target path loss.
[0113] After determining the target path loss, the method of the present application also includes: evaluating the accuracy of the proposed model, for example, comparing with the existing model to evaluate the accuracy of the model; comparing with the single-view perception method to evaluate the accuracy of the multi-view selection algorithm.
[0114] Among them, the accuracy of the proposed model is evaluated. In this step, the accuracy of the LOS path existence prediction and the path loss prediction is evaluated respectively, and the effectiveness of the multi-view selection algorithm is evaluated. Specifically, the accuracy of the model is evaluated by comparing the root mean square error (RMSE) of the model. The network is trained using the hyperparameters in Table 1.
[0115] Table 1:
[0116] parameter Road damage prediction LOS path existence prediction Batch Size 128 128 Learning Rate 0.001 0.001 Optimizer Adam Adam Epoch 60 20
[0117] Among them, Batch Size is an important hyperparameter in deep learning, which determines the number of samples processed simultaneously in each model training iteration. Adam optimizer is an adaptive algorithm that can adjust the learning rate according to historical gradient information. One epoch means training once with all samples in the training set, which is equivalent to Batch Size equal to the number of samples in the training set. The accuracy of the LOS path existence prediction model is shown in Table 2. The cumulative distribution function (CDF) of the prediction error generated by the path loss prediction model under different inputs from a single perspective is shown in Table 2. Figure 3 shown.
[0118] Table 2:
[0119]
[0120] It can be concluded that for single-view prediction, road loss prediction focuses more on the contours and edges of objects, and binary images can better highlight these features and effectively eliminate redundant image information. However, LOS path existence prediction needs to consider the relative position and color information between objects, so using the original RGB image can provide richer information.
[0121] Binarization of the original image and adding feature assistance can improve the accuracy of road loss prediction. There is a lot of redundant information in RGB images, and adding feature assistance can effectively supplement the environmental information that affects road loss. In the binary image, the relevant structural features have been condensed and can include most of the environmental features. The impact of adding feature assistance is small but still exists. The RMSE of the algorithm that adds feature assistance to the binary image can reach 4.02dB, further improving the accuracy.
[0122] Among them, in evaluating the accuracy of the multi-view selection algorithm: after the multi-view selection algorithm, the average RMSE of the test set can reach 3.66dB, which is 9% higher than the prediction based on single-view images. Figure 4 The relationship between multi-view perception, single-view perception and Rx position movement after traversing all data is given in Figure 1. It can be concluded that multi-view camera perception can achieve a lower RMSE at almost all positions, realize comprehensive perception of the scene, and effectively improve the accuracy of road loss prediction.
[0123] In summary, compared with the prior art, the present application has the following advantages:
[0124] (1) Directly mapping the perceived image of the actual environment with the path loss can alleviate the dependence on the propagation scenario type and support specific predictions for specific environments.
[0125] (2) This method uses multi-view cameras to perceive the actual scene, which can effectively solve the limitations of a single device, obtain comprehensive environmental information, and provide rich environmental information for channel prediction.
[0126] (3) This method is more suitable for the future mobile communication system's requirements for a wide variety of application scenarios and high dynamics. It is more scenario-universal and can perform efficient and accurate path loss prediction for specific actual application scenarios.
[0127] Reference Figure 5 As shown, the embodiment of the present application also provides a wireless channel path loss prediction device based on multi-view visual perception, including:
[0128] A first acquisition module 51 is used to acquire multi-view perception images in a dynamic scene;
[0129] A first determination module 52, configured to process the perception image and determine target features for road loss prediction;
[0130] A first processing module 53 is used to input the target feature into a pre-trained convolutional neural network model to obtain output results corresponding to different viewing angles;
[0131] The second determination module 54 is used to determine the target path loss corresponding to the optimal perception perspective according to the multiple output results; the target path loss is used to represent the path loss prediction result of the perception image of the multiple perspectives.
[0132] Optionally, in the present application, the first acquisition module 51 includes:
[0133] A first determination unit, used to determine the perception device deployment parameters corresponding to the target scene;
[0134] A first acquisition unit, configured to acquire multi-perspective perception images in a dynamic scene by using an image acquisition device corresponding to the perception device deployment parameter;
[0135] Among them, the perception device deployment parameters include: the horizontal viewing angle of the image acquisition device is greater than a first value, and the first height of the image acquisition device is less than or equal to the second height of the house corresponding to the target scene, and the vertical viewing angle of the image acquisition device has a second value, and the second value is determined by the first height, the length of the house, and the width of the house.
[0136] Optionally, in the present application, the first determining module 52 includes:
[0137] A second determination unit is used to preprocess the perceived image and determine a scatterer that affects the channel change as a valid image feature;
[0138] A second acquisition unit, used to acquire a distance feature between the transmitter and the receiver according to the transmitter position and the receiver position in the dynamic scene;
[0139] A first processing unit, configured to input the effective image features and the perceived image into a preset residual network, and output a sight path existence feature;
[0140] The third determining unit is used to determine the effective image feature, the distance feature and the sight path existence feature as the target feature.
[0141] Optionally, the wireless channel path loss prediction device based on multi-view visual perception of the present application further includes:
[0142] A construction module for constructing a target deep learning model; the target deep learning model has four stages, each stage contains two basic blocks; each of the basic blocks contains two 3*3 convolutional layers and one skip connection;
[0143] A third determination module is used to determine a data set corresponding to the perception image and the road loss parameter; the data set includes a training data set and a test data set;
[0144] The fourth determination module is used to train and test the target deep learning model using the training data set and the test data set to determine the convolutional neural network model.
[0145] Optionally, the third determining module includes:
[0146] A third acquisition unit, used to acquire physical information of a propagation environment corresponding to the perceived image;
[0147] A second processing unit, configured to construct a three-dimensional environment model according to the physical information;
[0148] A third processing unit, configured to perform channel simulation on the three-dimensional environment model to generate corresponding path loss parameters for each environment sample;
[0149] The fourth determining unit is used to determine a data set corresponding to the perceived image and the path loss parameter according to the perceived image.
[0150] Optionally, in the present application, the first processing module 53 includes:
[0151] a fourth processing unit, configured to input the target feature into a pre-trained convolutional neural network model, and use the convolutional neural network model to perform feature fusion on the image feature, the distance feature, and the sight path existence feature in the target feature to obtain a fused perceived environment feature;
[0152] The fifth processing unit is used to utilize the linearity and nonlinearity in the convolutional neural network model to obtain output results corresponding to different perspectives; wherein the linearity is implemented according to the linear layer module in the convolutional neural network model; the linear layer module includes two hidden layers, each having a first preset value and a second preset value of unit numbers; the first preset value is greater than the second preset value; the nonlinearity is implemented according to the linear rectification function in the convolutional neural network model.
[0153] Optionally, in the present application, the second determining module 54 includes:
[0154] a fifth determining unit, configured to determine, based on the plurality of output results, a proportion of a black portion of an image in different output results; the proportion of a black portion of an image being a ratio of the number of black pixels in the image at a current viewing angle to the total number of pixels in the image at the current viewing angle; each of the output results having a corresponding relationship with a perceived image at a viewing angle;
[0155] a sixth determining unit, configured to determine an average value of the proportions of the black portions of the plurality of images;
[0156] The sixth processing unit is used to select the result with a larger proportion among the multiple output results as the target path loss if the proportion of the black part of the multiple output results is greater than or less than the average value; if only one output result of the proportion of the black part of the multiple output results is greater than the average value, the result greater than the average value is used as the target path loss.
[0157] Among them, the implementation embodiments of the above-mentioned wireless channel path loss prediction method based on multi-view visual perception are all applicable to the embodiments of the wireless channel path loss prediction device based on multi-view visual perception, and can also achieve the same technical effect.
[0158] A readable storage medium in an embodiment of the present application stores a program or instruction thereon. When the program or instruction is executed by a processor, the steps in the wireless channel path loss prediction method based on multi-perspective visual perception as described above are implemented, and the same technical effect can be achieved. To avoid repetition, it will not be repeated here.
[0159] The processor is a processor in the wireless channel path loss prediction method based on multi-view visual perception described in the above embodiment. The readable storage medium includes a computer readable storage medium, such as a computer read-only memory (ROM), a random access memory (RAM), a disk or an optical disk, etc.
[0160] The present application also provides a computer program product, including computer instructions, which, when executed by a processor, implement the above Figure 1 The various processes of the method embodiment shown can achieve the same technical effect, and will not be described again here to avoid repetition.
[0161] In the embodiment of the present application, module can be implemented with software so that it can be performed by various types of processors. For example, an executable code module of an identification can include one or more physical or logical blocks of computer instructions, for example, it can be constructed as an object, process or function. Nevertheless, the executable code of the identified module does not need to be physically located together, but can include different instructions stored in different locations, and when these instructions are logically combined together, it constitutes a module and realizes the specified purpose of the module.
[0162] In fact, executable code module can be a single instruction or many instructions, and can even be distributed on a plurality of different code segments, distributed among different programs, and distributed across a plurality of memory devices. Similarly, operating data can be identified in the module, and can be implemented and organized in the data structure of any appropriate type according to any appropriate form. The operating data can be collected as a single data set, or can be distributed in different locations (including on different storage devices), and can only be present on a system or network as an electronic signal at least in part.
[0163] When a module can be implemented by software, considering the level of existing hardware technology, a person skilled in the art can build a corresponding hardware circuit to implement the corresponding function of the module that can be implemented by software without considering the cost. The hardware circuit includes a conventional very large scale integration (VLSI) circuit or gate array and existing semiconductors such as logic chips, transistors, or other discrete components. The module can also be implemented by a programmable hardware device, such as a field programmable gate array, a programmable array logic, a programmable logic device, etc.
[0164] The above exemplary embodiments are described with reference to the accompanying drawings, and many different forms and embodiments are feasible without departing from the spirit and teachings of the present application. Therefore, the present application should not be constructed as a limitation of the exemplary embodiments proposed herein. More specifically, these exemplary embodiments are provided so that the present application will be perfect and complete, and the scope of the present application will be conveyed to those who are familiar with the technology. In these figures, the component sizes and relative sizes may be exaggerated for clarity. The terms used here are only based on the purpose of describing specific exemplary embodiments and are not intended to be limiting. As used herein, unless the text clearly indicates otherwise, the singular forms "one", "an" and "the" are intended to include these multiple forms. It will be further understood that the terms "comprising" and / or "including" when used in this specification indicate the presence of the features, integers, steps, operations, components and / or components, but do not exclude the presence or increase of one or more other features, integers, steps, operations, components, components and / or their groups. Unless otherwise indicated, when stated, a range of values includes the upper and lower limits of that range and any subranges therebetween.
[0165] The above is a preferred embodiment of the present application. It should be pointed out that for ordinary technicians in this technical field, several improvements and modifications can be made without departing from the principles described in the present application. These improvements and modifications should also be regarded as the scope of protection of the present application.
Claims
1. A wireless channel path loss prediction method based on multi-view visual perception, characterized in that: include: Acquire multi-view perception images in dynamic scenes; Processing the perceived image to determine target features for road loss prediction; Inputting the target features into a pre-trained convolutional neural network model to obtain output results corresponding to different perspectives; According to the multiple output results, a target path loss corresponding to the optimal perception perspective is determined; the target path loss is used to represent the path loss prediction result of the perception image of the multiple perspectives.
2. The method according to claim 1, characterized in that: Acquire multi-view perception images in dynamic scenes, including: Determine the perception device deployment parameters corresponding to the target scenario; Using the image acquisition device corresponding to the perception device deployment parameters, acquiring multi-perspective perception images in dynamic scenes; Among them, the perception device deployment parameters include: the horizontal viewing angle of the image acquisition device is greater than a first value, and the first height of the image acquisition device is less than or equal to the second height of the house corresponding to the target scene, and the vertical viewing angle of the image acquisition device has a second value, and the second value is determined by the first height, the length of the house, and the width of the house.
3. The method according to claim 1, characterized in that Processing the perceived image to determine target features for road loss prediction includes: Preprocessing the perceived image to determine scatterers that affect channel variation as effective image features; According to the transmitter position and the receiver position in the dynamic scene, the distance characteristics between the transmitter and the receiver are obtained; Inputting the effective image features and the perceived image into a preset residual network, and outputting the sight path existence features; The effective image feature, the distance feature and the sight path existence feature are determined as the target feature.
4. The method according to claim 1, characterized in that: Before inputting the target feature into a pre-trained convolutional neural network model to obtain output results corresponding to different perspectives, the method further includes: Construct a target deep learning model; the target deep learning model has four stages, each stage contains two basic blocks; each basic block contains two 3*3 convolutional layers and one skip connection; Determine a data set corresponding to the perception image and the road loss parameters; the data set includes a training data set and a test data set; The target deep learning model is trained and tested using the training data set and the test data set to determine the convolutional neural network model.
5. The method according to claim 4, characterized in that Determine the data set corresponding to the perception image and the path loss parameters, including: Acquiring physical information of a propagation environment corresponding to the perceived image; Constructing a three-dimensional environment model according to the physical information; Performing channel simulation on the three-dimensional environment model to generate corresponding path loss parameters for each environment sample; According to the perceived image, a data set corresponding to the perceived image and the path loss parameter is determined.
6. The method according to claim 1, characterized in that The target features are input into the pre-trained convolutional neural network model to obtain output results corresponding to different perspectives, including: The target features are input into a pre-trained convolutional neural network model, and the image features, distance features, and sight path existence features in the target features are fused using the convolutional neural network model to obtain fused perceived environment features; The linearity and nonlinearity in the convolutional neural network model are used to obtain output results corresponding to different perspectives; wherein the linearity is implemented according to the linear layer module in the convolutional neural network model; the linear layer module includes two hidden layers, each having a first preset value and a second preset value of unit numbers; the first preset value is greater than the second preset value; the nonlinearity is implemented according to the linear rectification function in the convolutional neural network model.
7. The method according to claim 1, characterized in that Determining a target path loss corresponding to an optimal perception perspective according to the plurality of output results includes: Determine the proportion of the black part of the image in different output results according to the plurality of output results; the proportion of the black part of the image is the ratio of the number of black pixels in the image at the current viewing angle to the total number of pixels in the image at the current viewing angle; each of the output results has a corresponding relationship with a perceived image at a viewing angle; Determine an average value of the proportion of black portions of a plurality of the images; If the proportion of the black part of multiple output results is greater than or less than the average value, the result with a larger proportion among the multiple output results is selected as the target path loss; if only one output result of the black part of multiple output results is greater than the average value, the result greater than the average value is selected as the target path loss.
8. A wireless channel path loss prediction device based on multi-view visual perception, characterized in that: include: A first acquisition module is used to acquire multi-view perception images in dynamic scenes; A first determination module, used for processing the perception image to determine target features for road loss prediction; A first processing module is used to input the target features into a pre-trained convolutional neural network model to obtain output results corresponding to different viewing angles; The second determination module is used to determine the target path loss corresponding to the optimal perception perspective according to the multiple output results; the target path loss is used to represent the path loss prediction result of the perception image of the multiple perspectives.
9. A readable storage medium having a program or instruction stored thereon, characterized in that: When the program or instruction is executed by a processor, the steps in the method according to any one of claims 1 to 7 are implemented.
10. A computer program product, characterized in that The method comprises computer instructions, which, when executed by a processor, implement the steps of the method according to any one of claims 1 to 7.
