An electronic map-based communication scenario segmentation and identification method

Abstract

The application discloses a communication scene segmentation and identification method based on an electronic map, which can effectively realize scene identification in an actual communication process and specifically comprises the following steps: 1) performing first-stage scene segmentation and calculating physical environment parameters based on the segmentation result; 2) taking the physical environment parameters as identification features, comparing the identification effects of commonly used machine learning classification algorithms, and obtaining a primary identification result; 3) performing second-stage scene segmentation, which further refines the scene into sub-scenes that may appear in a real environment on the basis of identifying the scene categories, and recalculating the physical environment parameters according to the segmentation range; and 4) taking the calculated physical environment parameters and received power as identification features, comparing different machine learning algorithms, and finally achieving a scene identification accuracy of 96%. The method provides scene type information for a wireless communication system and provides an important precondition for channel modeling.

Description

Technical Field

[0001] This invention belongs to the field of wireless channel research, and in particular relates to a communication scene segmentation and recognition method based on electronic maps. Background Technology

[0002] With the ever-increasing demands of mobile communication systems, many countries are conducting research on 6th Generation Wireless Communication Technology (6G). The vision of 6G technology encompasses the entire spectrum, global coverage, and all application scenarios. Therefore, 6G communication involves a variety of communication scenarios. Studying wireless channels is fundamental to designing, optimizing, and evaluating 6G network performance. Traditional standardized channel models such as 3GPP TR 38.901, WINNER II, MiWEBA, and COST 2100 are widely used to characterize wireless communication channels. However, existing standardized channel models are relatively coarse in their scenario classification. For example, 3GPP TR 38.901 classifies scenarios as "urban macro," "urban micro," "rural macro," "indoor hotspot," and "indoor office." In wireless channel research, different channel models are used to capture the unique channel characteristics of each communication scenario. Traditional standardized channel models struggle to accurately describe the channel characteristics of the numerous 6G communication scenarios. Therefore, in practical 6G communication environments, these scenarios must be further refined into more specific categories. In other words, existing standardized channel models cannot meet the development requirements of the 6G vision.

[0003] To address the above challenges, researchers have extended their approaches to 6G communication scenarios. Many machine learning algorithms have been used to solve scene recognition problems. However, existing scene recognition methods do not consider scene segmentation. These studies are based on directly given scene categories, then using channel measurements and simulations to obtain data for scene identification. These methods still have a certain gap from practical applications. To fill this gap, this invention proposes a new scene segmentation and recognition method from a more practical perspective. Summary of the Invention

[0004] The purpose of this invention is to provide a communication scene segmentation and identification method based on electronic maps, which is a prerequisite for scene-based channel modeling, in order to solve the problem that the current standardized channel model is difficult to accurately describe the channel characteristics under many 6G scenarios.

[0005] To solve the above-mentioned technical problems, the specific technical solution of the present invention is as follows:

[0006] A communication scene segmentation and recognition method based on electronic maps includes the following steps:

[0007] Step 1: Determine the format, accuracy, and available physical environment parameters of the electronic map to be used;

[0008] Step 2: Perform initial scene segmentation of the target urban area, mainly according to four major scene categories: residential scene, commercial scene, office scene, and industrial scene.

[0009] Step 3: Calculate the required physical environment parameters based on the segmentation results;

[0010] Step 4: Preprocess the physical environment parameters, i.e., classify and label them, mainly based on the scenario in which the physical environment parameters were obtained;

[0011] Step 5: Using the labeled data as identification features, perform initial identification using the kernel Naive Bayes algorithm;

[0012] Step 6: After initially identifying the major scene categories, determine the possible sub-scene types that the scene may contain based on the scene type;

[0013] Step 7: Based on the determined sub-scene types, perform secondary segmentation on the identified scene categories;

[0014] Step 8: Calculate the required physical environment parameters again based on the segmentation results, and simulate the received power based on the existing sub-scenario channel model;

[0015] Step 9: Using physical environment parameters and received power together as identification features, the Gaussian Naive Bayes algorithm is used to identify sub-scenes.

[0016] Furthermore, step 1 includes the following steps:

[0017] Step 1.1: Determine the electronic map format as "planet";

[0018] Step 1.2: Determine the accuracy of the electronic map to be 5 meters;

[0019] Step 1.3: Determine the physical environment parameters that the electronic map can extract:

[0020] (1) Average building height:

[0021]

[0022] in, and The first The height and area of ​​each building.

[0023] (2) Building height variation:

[0024]

[0025] in, This is a function for taking the standard deviation.

[0026] (3) Average building spacing:

[0027]

[0028] in, For the first The perimeter of the building, This represents the area of ​​the base station sector.

[0029] (4) Regional Loss Ratio: This refers to the percentage of LOS grid cells within a 5m*5m rasterization of the target calculation area.

[0030]

[0031] in, The number of Los grid cells, This represents the total number of grid cells.

[0032] (5) Building proportions:

[0033]

[0034] Furthermore, step 4 specifically includes the following steps:

[0035] Step 4.1: Classify the calculated physical environment parameters according to their respective scene types;

[0036] Step 4.2: The sample data volume is 210, including 84 residential scenarios, 71 commercial scenarios, 14 industrial scenarios, and 41 office scenarios. Residential, commercial, industrial, and office scenarios are labeled as 0, 1, 2, and 3, respectively.

[0037] Furthermore, step 5 specifically includes the following steps:

[0038] Step 5.1: Feature space mapping. Use the Gaussian kernel function to map the original feature space to a higher-dimensional space to handle nonlinear relationships.

[0039] Step 5.2, parameter estimation: In the high-dimensional space, for each category, the kernel density estimation method is used to estimate the conditional probability distribution;

[0040] Step 5.3: Calculate the prior probability of each category based on the training data and the conditional probability distribution estimation;

[0041] After step 5.4, which involves estimating the conditional probability distribution and the prior probabilities of the categories, Bayes' theorem is used for classification. For a given test sample, its posterior probability under each category is calculated, and then the category with the highest posterior probability is selected as the final classification result.

[0042] Step 5.5, Evaluation and Tuning, requires evaluating the algorithm's performance and possibly tuning parameters to ensure good classification results. Cross-validation or other evaluation metrics such as accuracy, precision, and recall are typically used.

[0043] Furthermore, step 6 specifically includes the following steps:

[0044] Step 6.1: Determine if the scene category is one of the following: residential scene, commercial scene, office scene, or industrial scene;

[0045] Step 6.2: Based on the existing scene list, residential scenes include high-rise residential buildings, low-rise residential buildings, intersections, and city streets; commercial scenes include department stores, small markets, intersections, and city streets; office scenes include office buildings, government offices, schools, intersections, and city streets; and industrial scenes include large factories, small factories, intersections, and city streets. At this point, the scene types for secondary segmentation and recognition are determined based on this list. Taking residential scenes as an example, the scenes for secondary segmentation and recognition can be identified as high-rise residential buildings, low-rise residential buildings, intersections, and city streets.

[0046] Furthermore, step 8 specifically includes the following steps:

[0047] Step 8.1 Scene segmentation: According to the characteristics of the actual scene, different scenes are segmented into different ranges. High-rise residential buildings and low-rise residential buildings are segmented into squares with a side length of 300 meters. Similarly, the side length of city streets is 200 meters, and the side length of intersections is 100 meters.

[0048] Step 8.2: The physical environment parameters that can be extracted from the map are all calculated within a base station sector. However, the segmentation range is different for different scenarios. Therefore, the physical environment parameters of a segmented scenario need to be summarized based on the segmentation results.

[0049] Step 8.3: Adjust the parameters of the existing 6G universal channel model for specific scenarios to obtain a channel model applicable to the sub-scenario. Then, simulate the channel model of several sub-scenarios to obtain the corresponding received power (the model outputs the channel characteristics of the corresponding scenario type, i.e., the received power).

[0050] Furthermore, step 9 specifically includes the following steps:

[0051] Step 9.1: Prepare the training dataset. The sample data includes physical environment parameters and received power, and label the sample data with corresponding category labels. The sample data size is 138, and the ratio of the training set to the test set is set to 7:3.

[0052] Step 9.2: For each category, calculate the mean (μ) and standard deviation (σ) of each feature to estimate the parameters of the Gaussian distribution;

[0053] Step 9.3: For each category, estimate the prior probability by counting the number of samples in each category in the training data, that is, the frequency of each category in the training data;

[0054] Step 9.4: Use Bayes' theorem for classification. For a given test sample, calculate its posterior probability in each category, and then select the category with the highest posterior probability as the final classification result.

[0055] Step 9.5: Evaluate the algorithm performance and fine-tune the parameters to ensure good classification results.

[0056] The communication scene segmentation and recognition method based on electronic maps of the present invention has the following advantages:

[0057] This invention, building upon existing machine learning-based scene recognition methods, proposes a two-stage scene recognition method emphasizing scene segmentation. First, the features used for recognition are physical environment parameters extracted from electronic maps and received power obtained through channel model simulation, rather than the measurement data used in common scene recognition methods. This significantly reduces the manpower and resources required for channel measurement. Second, this invention proposes scene segmentation, a crucial consideration in practical scene recognition applications, especially in mobile communication. Finally, the two-stage segmentation approach, after identifying the major scene categories, limits the number of sub-scene types, avoiding the challenges of identifying numerous scene types in real-world environments. In summary, this invention provides an effective scene recognition method, laying the groundwork for accurate channel modeling. Attached Figure Description

[0058] Figure 1 This is a model block diagram of the scene segmentation and recognition method based on electronic maps provided in Embodiment 1 of the present invention;

[0059] Figure 2 This is a schematic diagram of the overall communication scene segmentation and recognition in Embodiment 1 of the present invention;

[0060] Figure 3 This is a schematic diagram of the confusion matrix of scene category recognition results in Embodiment 1 of the present invention;

[0061] Figure 4 This is a schematic diagram of a portion of the electronic map in Embodiment 1 of the present invention;

[0062] Figure 5 This is a schematic diagram of the scene segmentation result based on an electronic map in Embodiment 1 of the present invention;

[0063] Figure 6 This is a schematic diagram of the confusion matrix of the sub-scene recognition results in Example 1. Detailed Implementation

[0064] To make the objectives, technical solutions, and advantages of the embodiments of the present invention clearer, the technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0065] Example 1:

[0066] See Figure 1 This embodiment provides a communication scene segmentation and recognition method based on electronic maps, including the following steps:

[0067] Step 1: Obtain an electronic map of the target recognition environment, determine the format, accuracy, and physical environment parameters that can be extracted from the selected electronic map.

[0068] Specifically, in this embodiment, the electronic map is in "planet" format with a resolution of 5 meters. By analyzing multiple key layers of the electronic map, including but not limited to the Buildings layer, Height layer, and Vector layer, a series of important physical environment parameters are obtained. These parameters are of significant value in the accurate identification of communication scenarios. These physical environment parameters include:

[0069] 1) Average building height: The average building height within the base station sector is calculated by thoroughly analyzing information from the Buildings and Height layers.

[0070] 2) Building height variation: By using data processing and analysis of the Height layer, accurate information on changes in building height can be obtained, which helps to understand the path diversity of communication signals.

[0071] 3) Average building spacing: Through the study of data such as Vector layer, we accurately calculated the average distance between buildings, thus providing key information for the rational deployment of communication equipment and the planning of communication range.

[0072] 4) Area Los Percentage: This parameter reflects the proportion of line of sight (LOS) propagation within a given area, which is crucial for accurately predicting the feasibility of communication signal transmission.

[0073] 5) Building ratio: By analyzing the Buildings layer information, the building ratio within the base station sector was calculated, which helps to assess the degree of urbanization and the characteristics of building distribution.

[0074] The extraction and calculation of these physical environment parameters are based on the rich and detailed building information in the electronic map, ensuring a high degree of accuracy in the distribution of buildings in the real environment.

[0075] Step 2: Perform initial scene segmentation on the target area, dividing it into four main types: residential scene, commercial scene, office scene, and industrial scene.

[0076] Specifically, in this embodiment, the purpose of scene segmentation is to determine the range of the scene to be identified and the elements within that scene, which is a necessary prerequisite for scene recognition. To better perform statistical analysis of physical environment parameters based on base station sectors, a 400-meter-sided bounding box is used to segment the four main scene categories. The goal of this initial scene segmentation is to subdivide a large target area into different types of scenes, in order to better classify sub-scenes and perform accurate identification in the next step.

[0077] Step 3: Using the results of the initial scene segmentation, the physical environment parameters obtained after segmentation are used as data features to label the scene types. Based on this, data preprocessing is performed, including data cleaning, outlier removal, data transformation, and normalization. Finally, the entire dataset is divided into training and testing sets for scene recognition model training and performance evaluation.

[0078] Specifically, in this embodiment, the physical environment parameters obtained from different scenarios are first labeled with their types to clearly define the scenario type to which each data point belongs. This helps to build a supervised learning model for subsequent scenario recognition.

[0079] Next, data preprocessing is performed, including the following steps:

[0080] 1) Data cleaning: Inspect the dataset, delete or repair any errors or missing values ​​to ensure data quality and consistency.

[0081] 2) Outlier removal: Any outlier values ​​that do not conform to common sense or may affect model performance will be removed to ensure the accuracy and stability of model training.

[0082] 3) Data transformation: When necessary, we transform the data to ensure that it meets the input requirements of the model or better reflects the characteristics of a specific scenario.

[0083] 4) Normalization: Data normalization helps ensure that data with different features have similar scales, avoiding unnecessary influence of certain features on the model.

[0084] Finally, the dataset is divided into training and test sets in a 7:3 ratio. This step is to evaluate the performance of the scene recognition model; the training set is used for model building and learning, while the test set is used to verify the model's accuracy and generalization ability.

[0085] Step 4: Use the kernel naive Bayes algorithm to identify the scene categories in the preprocessed dataset.

[0086] The specific steps are as follows:

[0087] Step 401, Feature Space Mapping. The original feature space is mapped to a higher-dimensional space using a Gaussian kernel function. This aims to handle potential non-linear relationships and make the data more suitable for Bayesian classification. The Gaussian kernel function is typically expressed as:

[0088]

[0089] in, and These are sample points. It is the bandwidth parameter of the Gaussian kernel function.

[0090] Step 402, Parameter Estimation. In the high-dimensional space, for each scene category, the kernel density estimation method is used to estimate the conditional probability distribution. This means calculating the probability density function for each category to determine the likelihood of a data point belonging to each category. The formula for conditional probability density estimation is as follows:

[0091]

[0092] in, In the category Observed features The conditional probability, It is the sample size. These are sample points in the training set.

[0093] Step 403, Prior Probability Estimation. Based on the training data and the estimated conditional probability distribution, we calculate the prior probability for each class. Prior probability refers to the probability of each class occurring without considering any features. It is typically expressed as... It means that among them Representative category .

[0094] Step 404, Bayesian Classification. Once the conditional probability distribution and prior probability estimates for each class are complete, classification is performed using Bayes' theorem. For a given test sample, its posterior probability under each class is calculated, and the class with the highest posterior probability is selected as the final classification result. The formula for Bayesian classification is as follows:

[0095]

[0096] in, Given features Subcategories The posterior probability, In the category Observed features The conditional probability, It is a category The prior probability, It is a feature The marginal probability.

[0097] Step 405: Evaluation and Optimization. Finally, the algorithm's performance needs to be evaluated, typically using cross-validation or other evaluation metrics (such as accuracy, precision, recall, etc.). In this embodiment, a schematic diagram of the confusion matrix for class scene recognition is shown below. Figure 3 The proposed method achieves 100% accuracy across all four major scene recognition categories, demonstrating its excellent performance in terms of accuracy. Parameter tuning can be performed further to ensure optimal classification results. This scene recognition method combines the advantages of the kernel-based Naive Bayes algorithm, enabling it to handle nonlinear relationships and multi-class problems, providing a powerful tool for intelligent recognition of communication scenes.

[0098] Step 5: After initially identifying the major scene categories, determine the possible sub-scene types that the scene may contain based on the scene type.

[0099] The specific steps are as follows:

[0100] Step 501: Determine the Scene Category. First, the scene category is determined; it can be one of the following: residential scene, commercial scene, office scene, or industrial scene.

[0101] Step 502: Based on the existing scene list, further refine the scene types, for example:

[0102] 1) In residential scenarios, this includes high-rise residential buildings, low-rise residential buildings, intersections, and city streets.

[0103] 2) In commercial settings, this includes department stores, small markets, intersections, and city streets.

[0104] 3) In office settings, this includes office buildings, government offices, schools, intersections, and city streets.

[0105] 4) In industrial settings, this includes large factories, small factories, intersections, and city streets.

[0106] For example, if the initial scene type is identified as a residential scene, this list can be used to determine the sub-scene types that require secondary segmentation and identification, such as high-rise residential buildings, low-rise residential buildings, intersections, and city streets. This step allows for a more refined description of the specific scene characteristics within the target area, enabling the scene recognition system to more comprehensively understand and respond to different types of scenes. This hierarchical and refined scene classification method helps improve the accuracy and precision of scene recognition.

[0107] Step 6: Based on the determined sub-scene types, perform secondary segmentation on the identified scene categories, and recalculate the physical environment parameters based on the segmentation results.

[0108] The specific steps are as follows:

[0109] Step 601: Scene Segmentation. Different segmentation ranges are adopted based on the actual characteristics of different scenes. For example, in a residential scene, high-rise and low-rise residential buildings are segmented using a 300-meter side rectangle, city streets use a 200-meter side rectangle, and intersections use a 100-meter side rectangle. These segmentation ranges are determined according to the needs of the scene characteristics. See the scene segmentation diagram for a residential scene. Figure 5 ;

[0110] Step 602: Calculation of physical environment parameters. The physical environment parameters that can be extracted from the map are typically calculated within a single base station sector. However, since the segmentation range varies for different scenarios, the physical environment parameters for each segmented scenario need to be recalculated based on the segmentation results.

[0111] Step 7: Perform simulations based on existing scenario channel models to obtain the received power for the corresponding scenario in the electronic map. Specifically, adjust the parameters of the existing 6G universal channel model for specific scenarios to obtain channel models applicable to sub-scenarios, and then simulate the corresponding received power based on the channel models of several sub-scenarios.

[0112] Step 8: Use physical environment parameters and received power together as identification features for data preprocessing.

[0113] Specifically, in this embodiment, the previously extracted physical environment parameters and received power data are combined to form a complete identification feature set. These feature sets reflect the physical environment characteristics and signal transmission conditions under different scenarios. Next, data preprocessing is performed, including data cleaning, outlier handling, and normalization. This helps ensure data quality and consistency, reducing unnecessary noise and interference. By combining physical environment parameters and received power, we can more comprehensively describe the communication conditions in different scenarios, thereby improving the accuracy and performance of communication scenario identification.

[0114] Step 9: Using physical environment parameters and received power together as identification features, the Gaussian Naive Bayes algorithm is used to identify sub-scenes.

[0115] The specific steps are as follows:

[0116] Step 901: Prepare the training dataset. Each sample includes physical environment parameters and received power, and the sample data is labeled with the corresponding sub-scene category label. Split the dataset into a training set and a test set, with the ratio of the training set to the test set set set to 7:3, to ensure sufficient training data for model building and performance evaluation.

[0117] Step 902: For each sub-scene category, calculate the mean of each feature (including physical environment parameters and received power). and standard deviation Estimate the parameters of the Gaussian distribution. These parameters describe the probability distribution of each feature across different sub-scene categories. The probability density function of the Gaussian distribution is typically expressed as:

[0118]

[0119] in, In the category Observed features The conditional probability, and They are categories The mean and standard deviation are given.

[0120] Step 903: For each sub-scene category, estimate the prior probability by counting the number of samples of each category in the training data, that is, the frequency of occurrence of each sub-scene category in the training data.

[0121] Step 904: Classify using Bayes' theorem. For a given test sample, we calculate its posterior probability under each sub-scene category, and then select the category with the highest posterior probability as the final classification result.

[0122] Step 905: Evaluate the algorithm performance and fine-tune the parameters to ensure good classification results. See the diagram of the sub-scene recognition confusion matrix in this example. Figure 6 The recognition accuracy rate for urban street scenes was 96.4%, while the recognition accuracy rate for all other sub-scenes was 100%.

[0123] In summary, the communication scene segmentation and recognition method based on electronic maps established in this invention effectively recognizes communication scenes by segmenting and recognizing electronic maps, thus providing an important prerequisite for wireless channel modeling.

[0124] Any aspects of this invention not described in detail are well-known to those skilled in the art.

[0125] The preferred embodiments of the present invention have been described in detail above. It should be understood that those skilled in the art can make numerous modifications and variations based on the concept of the present invention without creative effort. Therefore, all technical solutions that can be obtained by those skilled in the art based on the concept of the present invention through logical analysis, reasoning, or limited experimentation on the basis of existing technology should be within the scope of protection defined by the claims.

Claims

1. A communication scene segmentation and recognition method based on electronic maps, characterized in that, Includes the following steps: Step 1: Perform initial scene segmentation on the target urban area in the electronic map, dividing it into four main scene categories: residential, commercial, office, and industrial. The electronic map is in "planet" format with a resolution of 5 meters. Step 2: Calculate the physical environment parameters for each major scene category; Step 3: Construct a physical environment parameter dataset using the scene categories corresponding to the physical environment parameters as data labels; the physical environment parameters include: average building height, building height variation, average building spacing, regional Los percentage, and building proportion. Step 4: Using the physical environment parameter dataset as recognition features, the kernel naive Bayes algorithm is used to perform initial recognition of the scene categories; Step 5: After initially identifying the major scene categories, determine the sub-scene types that each major scene category may contain; this specifically includes the following steps: Step 5.1: Determine if the scene category is one of the following: residential scene, commercial scene, office scene, or industrial scene; Step 5.2: Determine the scene type for secondary segmentation and recognition based on the existing scene list. The sub-scenes of the residential scene include high-rise residential buildings, low-rise residential buildings, intersections and city streets. The sub-scenes of the commercial scene include department stores, small markets, intersections and city streets. The sub-scenes of the office scene include office buildings, government agencies, schools, intersections and city streets. The sub-scenes of the industrial scene include large factories, small factories, intersections and city streets. Step 6: Based on the determined sub-scene types, perform secondary segmentation on the identified scene categories; Step 7: Based on the segmentation results, recalculate the physical environment parameters required for each sub-scene type, and simulate the received power based on the existing sub-scene channel model; Step 8: Using physical environment parameters and received power together as identification features, the Gaussian Naive Bayes algorithm is used to identify sub-scene types.

2. The communication scene segmentation and recognition method based on electronic maps according to claim 1, characterized in that, Step 4 specifically includes the following steps: Step 4.1: Use the Gaussian kernel function to map the original feature space composed of the physical environment parameter dataset to a higher-dimensional space in order to handle nonlinear relationships; Step 4.2: In the high-dimensional space, for each scene category, the kernel density estimation method is used to estimate the conditional probability distribution. Step 4.3: Calculate the prior probability of each category based on the training data and conditional probability distribution estimation taken from the physical environment parameter dataset; Step 4.4: After estimating the conditional probability distribution and the prior probability of the categories, Bayes' theorem is used for classification. For the test sample taken from the physical environment parameter dataset, its posterior probability under each category is calculated, and then the category with the highest posterior probability is selected as the final classification result.

3. The communication scene segmentation and recognition method based on electronic maps according to claim 1, characterized in that, Step 7 specifically includes the following steps: Step 7.1 Scene segmentation: Based on the characteristics of the actual scene, different sub-scenes are segmented into different ranges; Step 7.2: The physical environment parameters that can be extracted from the map are all calculated within a base station sector. However, the segmentation range is different for different scenarios. At this time, the physical environment parameters of a segmented scenario are summarized according to the segmentation results. Step 7.3: Adjust the parameters of the existing 6G universal channel model for specific scenarios to obtain a channel model applicable to the sub-scenario, and then simulate the corresponding received power based on the channel models of several sub-scenarios.

4. The communication scene segmentation and recognition method based on electronic maps according to claim 1, characterized in that, Step 8 specifically includes the following steps: Step 8.1: Prepare the training dataset. The sample data includes the physical environment parameters and received power calculated in Step 7, and label the sample data with the corresponding sub-scene category labels. Step 8.2: For each category, calculate the mean and standard deviation of each feature to estimate the parameters of the Gaussian distribution; Step 8.3: For each category, estimate the prior probability by counting the number of samples of each category in the training data, that is, the frequency of each category in the training data; Step 8.4: Use Bayes' theorem for classification. For a given test sample, calculate its posterior probability in each category, and then select the category with the highest posterior probability as the final classification result.

Images