Urban functional area detection method based on mobile phone signaling data

Abstract

The present invention relates to a method for detecting urban functional areas based on mobile phone signaling data, involving fields such as mobile phone signaling big data, artificial intelligence technology, and image recognition. The method addresses the problems of existing methods for detecting urban functional areas based on mobile phone signaling data, such as being susceptible to interference from noise and boundary effects, requiring accurate and complete geographic data, and having limitations due to the need for large amounts of training data and feature engineering. The present invention obtains mobile phone signaling data and a POI dataset and creates a POI one-dimensional feature data table. The image dataset and the POI one-dimensional feature data table are then input into an improved YOLOv8 deep learning network model for detection to obtain regional detection results. The present invention utilizes mobile phone signaling data and a POI dataset, combining deep learning computer vision technology with mobile phone signaling data to detect urban functional areas, which is a significant advancement for urban construction and development.

Description

Technical Field

[0001] The present invention relates to the fields of mobile phone signaling big data, artificial intelligence technology, and image recognition, and specifically to a method for detecting urban areas based on mobile phone signaling data, which is used to solve urban planning problems by applying deep learning computer vision technology to mobile phone signaling data. Background Art

[0002] The identification of urban functional zones is crucial for urban planning. First, it provides fundamental data and information for land use planning, enabling planners to determine the optimal uses for different areas, thereby improving land use efficiency and avoiding resource waste and irrational land allocation. Second, by understanding the functional characteristics and advantages of different areas, planners can formulate appropriate development strategies to promote industrial development, employment opportunities, and economic growth. Furthermore, the identification of urban functional zones plays a crucial role in urban transportation planning, ecological protection, and environmental planning, as well as in promoting social equity and improving residents' quality of life.

[0003] Existing methods for detecting urban areas based on mobile phone signaling data include density clustering, geographic information system (GIS)-based spatial interpolation, and machine learning-based classification methods. Density clustering distinguishes high-density areas from low-density areas through algorithms (such as DBSCAN), but is susceptible to noise and boundary effects and requires pre-setting parameters. GIS-based spatial interpolation methods use geographic data to associate mobile phone signaling data points to infer the activity level of the entire urban area, but require accurate and complete geographic data. Machine learning-based classification methods use algorithms (such as decision trees, support vector machines, or neural networks) to classify mobile phone signaling data and divide it into different areas, but require a large amount of training data and feature engineering. These methods require comprehensive consideration of their applicability and limitations. Therefore, based on the above considerations, the present invention proposes a method for detecting urban area functions based on a combination of mobile phone signaling data and deep learning computer vision technology.

[0004] Mobile phone signaling data plays a key role in the monitoring and planning of urban functional areas, providing planners with real-time, accurate, and comprehensive information to help formulate more effective development strategies and policies, and promote sustainable and balanced urban development. This integrated application of data and planning methods helps improve urban efficiency, sustainability, and the quality of life of residents, ultimately achieving the city's sustainable development goals.

[0005] Mobile phone signaling data plays a crucial role in urban planning. By analyzing this data, city planners can gain insights into population distribution and mobility, traffic conditions, facility needs, and disaster response. This data can be used to optimize urban infrastructure, transportation networks, and public service layouts, improve traffic flow and reduce congestion, rationally plan commercial areas and public facilities, and enhance urban quality of life and emergency response efficiency.

[0006] The use of mobile phone signaling data in urban planning presents several challenges. Chief among these is the protection of personal privacy, as this data includes user location information and communication behavior. Without adequate protection and anonymization, it could lead to privacy leaks and abuse. Furthermore, data accuracy and representativeness, the complexity of data acquisition and processing, and restrictions on data use also pose challenges to the application of mobile phone signaling data. When using this data, compliance with relevant laws and regulations is crucial to ensure data legitimacy and privacy protection, and appropriate verification and validation must be performed to ensure data accuracy and reliability.

[0007] The application of mobile phone signaling data in urban planning is becoming an important research and practical direction. With the widespread adoption of smartphones and the development of mobile networks, the acquisition and availability of mobile phone signaling data are gradually increasing. Simultaneously, advances in data analysis technology are continuously enhancing the ability to process and analyze mobile phone signaling data. In urban planning, mobile phone signaling data is being applied to traffic planning and optimization. By analyzing crowd movement patterns and traffic flows, planners can optimize traffic signal timing and road planning, thereby improving traffic efficiency. Furthermore, mobile phone signaling data provides guidance for facility layout and public services. By analyzing crowd movement patterns and needs, the layout of commercial areas and public facilities can be optimized, improving the quality of life in cities. Overall, the application of mobile phone signaling data in urban planning holds broad prospects, providing more accurate and real-time data support for urban development.

[0008] Deep learning computer vision technology offers advantages in detecting urban functional areas, including high efficiency, accuracy, multi-source data integration, and real-time monitoring and updating. It can rapidly process large amounts of image and video data, using algorithms to identify and detect functional areas, improving efficiency and accuracy. Furthermore, deep learning computer vision technology can integrate multiple data sources, such as satellite imagery and surveillance cameras, providing comprehensive and diverse data that facilitates more accurate detection and monitoring of urban functional areas. Furthermore, this technology can monitor urban changes and evolution in real time, promptly identifying and updating new or changing functional areas. Deep learning computer vision technology has broad application prospects in detecting urban functional areas, providing important data support and decision-making reference for urban planning and development.

[0009] In summary, the applications of mobile phone signaling data primarily focus on population distribution and mobility, transportation planning, and commercial location. By analyzing mobile phone signaling data, we can understand population distribution and mobility trends, helping urban planners better understand population concentration areas and migration patterns, thereby optimizing a city's population distribution and the layout of social service facilities. However, the use of mobile phone signaling data and deep learning computer vision technology to detect urban functional areas is unprecedented in China. Therefore, this paper proposes an image dataset generated using mobile phone signaling data, which is then used to detect urban functional areas using deep learning computer vision technology. Summary of the Invention

[0010] The present invention provides a method for detecting urban functional areas based on mobile phone signaling data to solve the problems that the existing method for detecting urban functional areas based on mobile phone signaling data is susceptible to interference from noise and boundary effects, requires accurate and complete geographic data, and has usage limitations due to the need for a large amount of training data and feature engineering.

[0011] A method for detecting urban functional areas based on mobile phone signaling data is proposed. This method uses an improved YOLOv8 deep learning network model to detect urban functional areas in image data generated by mobile phone signaling data. The specific detection steps are as follows:

[0012] Step 1: obtaining mobile phone signaling data, preprocessing the mobile phone signaling data, and obtaining preprocessed mobile phone signaling data;

[0013] Step 2: Create and label an image dataset based on the pre-processed mobile phone signaling data obtained in step 1;

[0014] Step 3: Obtain the POI dataset and combine it with the mobile phone signaling data to obtain the POI one-dimensional feature data table;

[0015] Step 4: Combine the MLP network model with the YOLOv8 deep learning network model to obtain an improved YOLOv8 deep learning network model;

[0016] The image dataset described in step 1 and the POI one-dimensional feature data table described in step 3 are input into the improved YOLOv8 deep learning network model for detection to obtain the region detection results.

[0017] Beneficial effects of the present invention:

[0018] 1. The method described in this invention is the first to detect urban functional areas based on mobile phone signaling data. The mobile phone signaling data is cleaned, processed, and visualized to create and annotate an image dataset. Simultaneously, the point of interest (POI) dataset is cleaned and processed to create a one-dimensional feature data table for the POIs. Urban functional areas are then detected. This method, combining mobile phone signaling data with deep learning computer vision technology, is innovative and advanced.

[0019] Second, the model framework used in the method described in this invention is a multi-task cascade framework. The first stage of the cascade task involves improving the YOLOv8 deep learning network model; integrating the MLP network model into the YOLOv8 network, defining the improved model's loss function, and setting the network's number of layers, channels, and hyperparameters. The second stage of the cascade task involves inputting the image dataset and the POI one-dimensional feature data table into the improved YOLOv8 model. The network then outputs the category, bounding box, and confidence level of each detected object. Detection results can be filtered based on the confidence threshold to select objects with high confidence.

[0020] 3. The method of the present invention detects urban functional areas by utilizing mobile phone signaling data and POI data sets, and combining deep learning computer vision technology with mobile phone signaling data, which is a great improvement for urban construction and development. BRIEF DESCRIPTION OF THE DRAWINGS

[0021] Figure 1 This is a flow chart of the method for detecting urban functional areas based on mobile phone signaling data according to the present invention;

[0022] Figure 2 A network structure diagram of the linear model of the present invention;

[0023] Figure 3 This is a diagram of the MLP network structure in the urban functional area detection method based on mobile phone signaling data described in the present invention. DETAILED DESCRIPTION

[0024] Specific implementation method 1. Combination Figure 1 、 Figure 2 and Figure 3 This embodiment describes a method for detecting urban functional areas based on mobile phone signaling data. The method performs urban functional area detection on image data generated by mobile phone signaling data based on an improved YOLOv8 deep learning network model.

[0025] YOLO (You Only Look Once) is a deep learning-based object detection algorithm that achieves fast and accurate object detection by dividing an image into grid cells and predicting the object category and bounding box within each cell. Compared to traditional methods, YOLO maintains high speed without sacrificing detection accuracy. It is widely used in computer vision fields such as autonomous driving, video surveillance, and object recognition. YOLOv8 is the latest in the YOLO family of object detection models released by Ultralytics.

[0026] The detection method described in this embodiment first obtains the mobile phone signaling data of the city, pre-processes and cleans the data, thereby removing the poor quality data in the mobile phone signaling data, and thus improving the accuracy of the mobile phone signaling data; secondly, uses the mobile phone signaling data to generate images, creates an image dataset, and annotates the image dataset. Then, a POI dataset is obtained, and the POI dataset is cleaned at the same time, and the cleaned POI dataset is combined with the mobile phone signaling data to generate a POI one-dimensional feature data table. Next, the image dataset and the POI one-dimensional feature data table are input into the improved YOLOv8 network model to obtain the image detection results. Finally, after obtaining the image detection results, the detection results are compared and analyzed, and the corresponding urban functional areas are formulated according to the image detection results to help the construction of the city.

[0027] In this embodiment, the POI data refers to Point of Interest data, which includes various meaningful entities at geographical locations, such as stores, restaurants, attractions, banks, etc. POI data typically includes location coordinates, categories, names, addresses, phone numbers, and other related information. POI data plays an important role in geographic information systems (GIS) and location-based service applications. It can be used in map navigation, location search, business analysis, urban planning, and other fields. By collecting and maintaining POI data, accurate location information and rich geographic features can be provided to help users understand their surroundings, find places of interest, and support business decisions and service optimization. In mobile applications and online map services, POI data is an important component of building a user-friendly and personalized experience.

[0028] The detection method described in this embodiment is specifically implemented by the following steps:

[0029] S1. Acquire mobile phone signaling data, preprocess the mobile phone signaling data, and obtain preprocessed mobile phone signaling data;

[0030] In this embodiment, the collection and processing of mobile phone signaling data: the raw data is obtained from the mobile phone signaling data source, and the data is processed after preprocessing and cleaning. In the preprocessing, the raw data will be parsed, and missing values ​​and outliers will be processed. Then data cleaning is performed, including removing noise, correcting erroneous data, processing duplicate records and outliers, etc., to ensure the accuracy and completeness of the data. Finally, data verification is performed to check the consistency, completeness and accuracy of the data and to identify potential problems or errors. The goal of this step is to obtain high-quality and reliable mobile phone signaling data to provide a reliable foundation for subsequent data analysis and application. The steps for preprocessing the data are as follows:

[0031] S11. Acquisition of mobile phone signaling data: First, it is necessary to obtain raw data from the mobile phone signaling data source, which involves negotiating and obtaining data with network providers, mobile operators or other data providers.

[0032] S12. Processing and cleaning of mobile phone signaling data requires processing of both ping-pong and drift data. For ping-pong data, signal strength thresholds can be set to filter the data and eliminate the impact of frequent handoffs between devices. For drift data, smoothing or filtering techniques can be used to stabilize signal strength, remove noise and outliers, and ensure data accuracy and integrity.

[0033] S13. Verification of mobile phone signaling data: To ensure the quality of cleaned data, check the consistency, completeness, and accuracy of the data, and verify that it meets the expected standards and requirements. The results of verification can help identify potential problems or errors and provide confidence and reliability for subsequent experimental applications.

[0034] S2. Create an image dataset using the pre-processed mobile phone signaling data. The specific process of creating the image dataset is as follows:

[0035] S21. Observe the distribution of mobile phone signaling data and collect mobile phone signaling data. This data includes mobile phone connection information, signal strength, and other relevant data at different locations. This information is typically collected by base stations and mobile network equipment and used for communication and positioning purposes. This embodiment uses this data to understand the distribution of mobile phone signals at different geographic locations.

[0036] S22. Map the mobile phone signaling data on a map. In this embodiment, AutoNavi Map is selected as the base map for the dataset. AutoNavi Map is a widely used mapping service that provides detailed geographic information and rich map data. After the mobile phone signaling data is mapped on the map, an area where the mobile phone signaling data is displayed is selected and an image of that area is captured.

[0037] S23. Select an appropriate image size and capture a map image. When capturing a map image, ensure that each captured image is 416×416 pixels in size. This pixel size is widely used in image processing tasks in the field of computer vision and has moderate resolution and computational efficiency. Record the mobile phone signaling latitude and longitude points at the center of each captured image, and name the image according to its latitude and longitude. For example, if the latitude and longitude coordinates of the center point of an image are 22.1245° and 114.1245° respectively, then the image is named 22.1245_114.1245.png. The captured map images are combined into an image dataset, which will serve as an important basis for subsequent data processing and geolocation.

[0038] S24. Label the captured image data using Labelme, generating a corresponding label file in XML format. Labelme is a classic labeling tool that supports tasks such as object detection, semantic segmentation, and instance segmentation. The labeling process involves selecting a map image and labeling it according to its urban functional areas. The urban functional area labels designed in the embodiments of the present invention are: transportation equipment, public services, public services_transportation equipment, commercial, commercial_transportation equipment, commercial_public services, commercial_residential, commercial_industrial, residential, residential_transportation equipment, residential_public services, residential_industrial, industrial, industrial_transportation equipment, industrial_public services, and mixed functional areas.

[0039] S3. Obtain the POI dataset and perform feature extraction, and obtain a POI one-dimensional feature data table by combining it with the mobile phone signaling data;

[0040] In order to combine mobile phone signaling data with POI (points of interest) datasets to create a one-dimensional feature data table for joint analysis, you first need to register a developer account on the AutoNavi Open Platform and create an application to obtain an API key. By calling the AutoNavi Map's Web service API, specify keywords and parameters to obtain the POI dataset of interest. The acquired data can include information such as POI name, address, longitude and latitude. Then, the POI dataset is cleaned and tested, missing values ​​and outliers are processed, and standardized and normalized. According to the field characteristics, the POI dataset is tested. Finally, the POI dataset is combined with the mobile phone signaling data, and feature extraction is performed to generate a one-dimensional feature data table. The specific implementation process is as follows:

[0041] S31. Call the web service API to obtain an AutoNavi developer account. First, you need to register a developer account on the AutoNavi Open Platform and create an application to obtain a key for accessing the API. Log in to the AutoNavi Open Platform and follow the instructions to register an account and create an application. Using the key, you can access the POI dataset by calling the AutoNavi Map web service API. AutoNavi Map provides a series of APIs for accessing different types of data, such as geocoding, reverse geocoding, and POI search.

[0042] S32. Obtain a city area POI dataset. Acquire the POI dataset of interest by constructing a correct API request and specifying a specific keyword (Key) or other parameters. For example, you can specify the keyword as restaurant and provide the name or latitude and longitude coordinates of the city where it is located. Using the selected programming language and HTTP client library, send the constructed API request to the API endpoint of Amap. Ensure that the key is included in the request for authentication. After receiving the API response, parse the returned JSON data and process and analyze it as needed. You can extract information such as the POI name, address, latitude and longitude coordinates, and obtain the POI data.

[0043] S33. Clean and test the POI dataset. Then, load and preview the POI data to understand its structure and field information. Next, handle missing values ​​and outliers by deleting or filling missing values ​​and correcting or deleting outliers. Then, standardize and normalize the data, and then test the POI data based on field characteristics; for example, leisure and entertainment, transportation facilities, commercial and residential, and corporate enterprises.

[0044] S34. Combine the POI data and mobile phone signaling data to generate a one-dimensional feature data table. Enter the longitude and latitude points at the center of the captured map image recorded in step S2 into a one-dimensional data table. Count the number of mobile phone signaling data points within 12 time zones and within a 24-hour period at this longitude and latitude point, as well as the number of POI data points within a 1-kilometer radius. Count the categories of POI data points and the number of POI data points in each category. Count the distance to the nearest POI data point for each longitude and latitude point. Then, generate a one-dimensional feature data table for the mobile phone signaling and POI data set, and annotate the data based on the POI data points.

[0045] S4. Combine the MLP network model with the YOLOv8 deep learning network model to obtain an improved YOLOv8 deep learning network model;

[0046] The specific process is:

[0047] S42, YOLOv8 model improvement and network configuration: Configure the network according to the requirements of the YOLOv8 network, set the number of layers, channels and hyperparameters of the network, define the loss function, etc.

[0048] In this implementation, YOLOv8 is an object detection model composed of multiple convolutional layers and fully connected layers. This implementation integrates a multi-layer perceptron (MLP) network model into the YOLOv8 network model. Specifically, the MLP network is connected to the YOLOv8 backbone network, and the MLP network output is connected to the YOLOv8 detection branch, achieving a fusion of the two networks. Furthermore, the single-channel input image is modified to accept an image dataset and a one-dimensional feature data table of point of interest.

[0049] S43. Setting a loss function: In this embodiment, an improved YOLOv8 object detection network model is proposed, and an MLP network is integrated into it. The loss function of the improved YOLOv8 network model adopts a category classification loss function and a bounding box regression loss function.

[0050] The category classification loss function ultimately uses the improved cross entropy loss function, which is as follows:

[0051] Loss = Loss1 + Loss2;

[0052] Among them, Loss1 and Loss2 are as follows:

[0053] ;

[0054] ;

[0055] Where, Custom parameters and , C is the total number of categories, each category has a unique label k, indicating its position in the entire category set. N is the total number of samples in the dataset, i is the index of the sample, and for the i-th sample, its true label vector is expressed as , is the element at the kth position of the true label vector of the i-th sample, where only the kth element is 1 and the rest are 0 ( ), and the model's predicted probability distribution vector for this sample is , represents the model’s predicted probability for each category, is the element at the kth position of the predicted probability distribution vector of the i-th sample.

[0056] The border regression loss function includes an improved DFL loss function and CIOU Loss.

[0057] The DFL loss function is as follows:

[0058] ;

[0059] Where, is a quality label in the range of 0 to 1, is the quality label of the i+1th sample, For quality label 、 probability;

[0060] DFL is improved based on Quality Focal Loss (QFL). In QFL, its loss function formula is as follows:

[0061] ;

[0062] in is a quality label from 0 to 1, is the predicted value. The global minimum solution of QFL is = In this way, the cross entropy part becomes the complete cross entropy, and the adjustment factor becomes a power function of the absolute value of the distance. It is an adjustable exponential parameter used to adjust the weight of difficult and easy samples. It is generally set to a positive value, such as 2.

[0063] On top of QFL, we add an additional loss, hoping that the network can quickly focus on the values ​​near the labeled position, making them as probable as possible. This is called Distribution Focal Loss (DFL), which is the DFL in the above formula.

[0064] In DFL, its form is very similar to the right half of QFL. The two positions closest to label y (the quality label 、 ), allowing the network to quickly focus on the distribution of the neighboring area of ​​the target location.

[0065] The CIoU Loss loss function is as follows:

[0066] CIoU Loss ;

[0067] Where, is the area of ​​the intersection of two bounding boxes (or segmentation masks) divided by their union area, is the Euclidean distance between two rectangular boxes, are the center points of the two rectangular boxes, is the weight coefficient, is the distance between the diagonals of the enclosed area of ​​the two rectangular boxes, Used to measure the consistency of the relative proportions of two rectangular boxes.

[0068] This implementation also includes using the gradient descent optimization algorithm to optimize model parameters by minimizing the total loss function to improve the performance of the integrated model on object detection tasks. Integrating the MLP network into the YOLOv8 model enhances the model's feature learning and expression capabilities, thereby achieving better object detection performance.

[0069] S5. Input the image dataset described in step S1 and the POI one-dimensional feature data table described in step S2 into the improved YOLOv8 deep learning network model for detection to obtain region detection results, bounding box results, and confidence results.

[0070] The specific process is:

[0071] S51. Preprocess the image dataset. Preprocess the map images in the acquired image dataset. This includes scaling, cropping, rotating, or other enhancement techniques to ensure that the images are of uniform size and in a format suitable for model input. Also, standardize and normalize the images to ensure that the image data meets the input requirements of the improved YOLOv8 deep learning network model. The image dataset and the POI one-dimensional feature data table are then divided into a training set, a validation set, and a test set.

[0072] S52. The training set is used as input to train the improved YOLOv8 deep learning network model. During this process, the image data from the training set is first extracted through the network feature extraction layer, while the POI one-dimensional feature data table is then processed through the MLP network layer to obtain classification results. The classification results are then combined with the extracted image features for the next training step. During training, the network's weights and biases are updated using a backpropagation algorithm, enabling it to gradually learn the patterns and features of object detection.

[0073] S52. Use the validation set for parameter adjustment and optimization. The validation set is a dataset used during the training of the improved YOLOv8 deep learning network model to adjust the model's hyperparameters. Hyperparameters refer to parameters that need to be set before model training, such as the learning rate and regularization coefficient. By using the validation set, you can experiment with different hyperparameter combinations on the model and select the model configuration that performs best on the test set. This is done to prevent the model from overfitting on the test set, and the validation set plays a "confidential" role during the model parameter adjustment process, preventing the model from being over-optimized for the test set.

[0074] S53. After training is complete, the trained YOLOv8 deep learning network model is used to perform image detection using the test set. The test set is used to evaluate the performance of the machine learning model after training. Once the model is trained on the training set, the samples in the test set are used to evaluate the model's generalization ability, that is, the model's prediction performance on unseen data. First, the test set to be tested is input into the network model to obtain the category, bounding box, and confidence information of each detected object. Second, it is filtered according to the confidence threshold to select objects with high confidence. Finally, the network output is parsed to obtain the region detection results, bounding box results, and confidence results, and the results are visualized.

[0075] This implementation also includes analyzing and evaluating the obtained detection results: this includes counting the number, proportion, and distribution of each functional area category, and understanding their geographical and spatial characteristics. These results are then compared and analyzed with existing urban planning, functional zoning, or planning indicators to determine whether there are areas with missing functions or requiring adjustment. Based on the comparative analysis data, a functional evaluation is conducted, considering the contribution of different types of POI datasets and mobile phone signaling data to urban functions, and evaluating factors such as the service scope, capacity, and convenience of each area.

[0076] Based on the analysis and evaluation results, appropriate urban planning strategies are formulated, along with corresponding urban functional steps. These steps include adjusting planning, improving regional functions, adding or optimizing urban facilities, adjusting transportation layouts, and promoting economic development. Subsequently, the effectiveness of these steps is regularly monitored, assessing urban development achievements and issues, and making necessary adjustments and optimizations based on the monitoring results, in order to continuously improve the city's functional layout and development direction.

[0077] Specific implementation method 2: Figure 1 、 Figure 2 and Figure 3 As shown, this embodiment is an example of the method for detecting urban functional areas based on mobile phone signaling data described in Specific Implementation 1:

[0078] This embodiment is a model for urban functional area detection based on mobile phone signaling data. The experimental data uses mobile phone signaling data within 24 hours of a certain day in Shenzhen, and the POI dataset selects point of interest data in the Shenzhen area.

[0079] First, obtain Shenzhen mobile phone signaling data. Second, preprocess and clean the mobile phone signaling data. Then, map the mobile phone signaling data set on the map, make it into an image data set, and annotate it. At the same time, obtain the POI data set of Shenzhen city and clean the POI data set. After cleaning the POI data set, combine it with the mobile phone signaling data set to make a POI one-dimensional feature data table. Then, by processing the image data set, input the image data set and the POI one-dimensional feature data table into the improved YOLOv8 network model to obtain the detection results. Then, analyze and evaluate the detection results to formulate corresponding strategies for subsequent urban planning. Specifically, the following steps are included:

[0080] First, obtain mobile phone signaling data, then preprocess and clean it. Obtain the raw data of mobile phone signaling data through consultation with network providers or mobile operators. Table 1 shows the raw data of mobile phone signaling data. The raw data consists of 38,218,717 rows and 4 columns.

[0081] Table 1

[0082] 0 1 2 3 0 55555556 17:39:08 113.809028 22.756181 1 55555556 23:53:07 114.026389 22.626042 2 55555556 13:10:09 114.039792 22.574028 ... ... ... ... ... 32818715 55969825 18:44:15 114.045069 22.525139 32818716 55969826 21:57:41 114.043681 22.535208

[0083] Next, we preprocessed the raw data, including formatting, handling missing values, and outliers. Data cleaning was performed to remove noise, correct erroneous data, and address duplicate records. In this experiment, we retained the first and last occurrences of the ID, longitude, and latitude, and added a new column for total time, formatting the start time to seconds. After cleaning, the raw data had 2,583,202 rows and five columns, as shown in Table 2. Table 2 shows the cleaned mobile phone signaling data.

[0084] Table 2

[0085] ID Start time longitude latitude 0 55555556 17:39:08 113.809028 22.756181 1 55555556 23:53:07 114.026389 22.626042 2 55555556 13:10:09 114.039792 22.574028 ... ... ... ... ... 2583220 55969825 18:44:44 114.045069 22.525139 2583221 55969825 20:09:51 114.042708 22.524583

[0086] Second, map the data on the map and obtain the image dataset.

[0087] First, all collected mobile phone signaling data is displayed on a map. Due to the low precision of mobile phone signaling data, this implementation adds a certain amount of data perturbation processing when displaying it on the map, which can better protect user privacy and display data on the map. At the same time, the 24 hours of a day are divided into 12 time zones, and a different color is selected for each different time zone. Then, a specific area is selected for capture. The size of each captured image is 416×416 pixels, and the mobile phone signaling latitude and longitude point at its center is recorded. Labelme is then used to annotate the image data and generate the corresponding XML format label file.

[0088] 3. Obtain the POI dataset, clean it, and combine it with mobile phone signaling data to create a POI one-dimensional feature data table.

[0089] First, you need to call the AutoNavi Maps web service API. Register an AutoNavi developer account and create an application to obtain an API access key. Build an API request, specifying keywords and parameters, to retrieve the POI data of interest. Use the key to send the API request to the AutoNavi Maps API endpoint and parse the returned JSON data. Clean the POI dataset, including handling missing values ​​and outliers, standardizing and normalizing, and testing the data based on field features. Combined with mobile phone signaling data, the latitude and longitude points at the center of the image are entered into a one-dimensional data table. Count the number of mobile phone signaling data points at this latitude and longitude point within 24 hours and 12 time zones, as well as the surrounding POI data points within a 1-kilometer radius, the types of POI data points, and the number of POI data points in each category. Count the distance to the nearest POI data point for each latitude and longitude point. Then, create a one-dimensional feature data table for the mobile phone signaling-based POI dataset and annotate the data points accordingly.

[0090] At the same time, the urban functional areas selected in this implementation are: transportation equipment, public services, public services_transportation equipment, commerce, commerce_transportation equipment, commerce_public services, commerce_residential, commerce_industry, residential, residential_transportation equipment, residential_public services, residential_industry, industry, industry_transportation equipment, industry_public services, and mixed functional areas.

[0091] 4. The steps to integrate the MLP network into the YOLOv8 network model to improve the target detection performance are as follows: connect the MLP network to the YOLOv8 backbone network, and connect the output of the MLP network to the detection branch of YOLOv8 to achieve the fusion of the two networks.

[0092] The general formula of the linear model of the MLP network can be expressed as:

[0093] + ;

[0094] Where, represents the predicted output of the model, arrive is the characteristic value of the sample; is the weight coefficient of each eigenvalue, e is the bias parameter, It can be regarded as the weighted sum of all eigenvalues, such as Figure 2 As shown, in Figure 2 In the example, the input features and predicted results are represented by nodes, using the weight coefficients Used to connect these nodes.

[0095] In the MLP model, the algorithm adds hidden layers to the process, then repeats the weighted sum calculation in the hidden layers, and finally uses the results calculated in the hidden layers to generate the final result, such as Figure 3 As shown, the model has many more feature coefficients (weights) to learn. Each input feature has a coefficient associated with a hidden unit, which is the purpose of generating these hidden units. Each hidden unit also has a coefficient associated with the final output. After generating the hidden layer, an activation function is used to nonlinearize the activated units. This nonlinear processing is intended to simplify sample features, enabling neural networks to learn complex, nonlinear datasets.

[0096] 5. Define the loss function of the network model. In this embodiment, the loss function of the improved YOLOv8 network model uses category classification loss and bounding box regression loss. The category classification loss ultimately uses the improved cross entropy loss function, which is as follows:

[0097] + ;

[0098] in , They are as follows:

[0099] ;

[0100] ;

[0101] Where, Custom parameters and , C is the total number of categories, each category has a unique label k, indicating its position in the entire category set. N is the total number of samples in the dataset, i is the index of the sample, and for the i-th sample, its true label vector is expressed as , Represents the element at the kth position of the true label vector of the i-th sample, where only the kth element is 1 and the rest are 0 ( ), and the model's predicted probability distribution vector for this sample is , represents the model’s predicted probability for each category, Represents the element at the kth position of the predicted probability distribution vector for the i-th sample.

[0102] The border regression loss uses the improved DFL and CIOU Loss, and the loss function is as follows:

[0103] The DFL loss function is:

[0104] ;

[0105] Where, is a quality label in the range of 0 to 1, is the quality label of the i+1th sample, For quality label 、 probability;

[0106] The CIoU Loss is:

[0107] CIoU Loss ;

[0108] Where, To calculate the area of ​​the intersection of two bounding boxes (or segmentation masks) divided by their union area, is the Euclidean distance between the two rectangles, are the center points of the two rectangular boxes, is the weight coefficient, is the distance between the diagonals of the enclosed area of ​​the two rectangular boxes, Used to measure the consistency of the relative proportions of two rectangular boxes.

[0109] 6. Obtain the detection results of Shenzhen’s urban functional areas, conduct data analysis and comparative analysis, and then formulate corresponding urban functional steps based on the results, count the number, proportion and distribution of Shenzhen’s urban functional areas, evaluate functional deficiencies and adjustment needs, and consider the contribution of different data to Shenzhen’s urban functions and service scope.

[0110] Data analysis: Conduct data analysis on new test results, including statistics on the number, proportion, and distribution of each functional area category. Understand the geographical location and spatial distribution characteristics of different categories.

[0111] Comparative analysis: Compare and analyze the test results with existing urban planning, urban functional zoning, or urban planning indicators. Compare the consistency of the new test results with urban planning requirements or functional zoning to determine whether there are areas with missing functions or planning adjustments.

[0112] Functional Assessment: Based on the test results and comparative analysis data, the functional characteristics of each area are evaluated. This considers the contribution of different types of POI data and mobile phone signaling data to urban functions such as commerce, transportation, and leisure and entertainment, as well as factors such as the service scope, capacity, and convenience of each area.

[0113] Developing Urban Functional Steps: Based on the results of comparative analysis and functional assessment, corresponding urban functional steps are developed. These involve adjusting urban planning, improving regional functions, adding or optimizing urban facilities, adjusting transportation layout, and promoting economic development. Depending on the specific circumstances, long-term and short-term urban development strategies can be formulated, taking into account regional priorities and phased implementation plans.

[0114] Monitoring and Adjustment: Regularly monitor the implementation of urban functional steps and evaluate the achievements and problems of urban development. Make necessary adjustments and optimizations based on the monitoring results to continuously improve the functional layout and development direction of the city.

[0115] The detection method described in this embodiment combines mobile phone signaling data with deep learning computer vision technology. This combination offers a wealth of application prospects in urban planning. By combining the location and movement information from mobile phone signaling data with deep learning computer vision technology, more accurate detection of urban functional areas can be achieved, providing support for urban spatial perception and planning. This combination provides urban planners with more comprehensive and accurate data support, helping to optimize urban planning decisions and enhance sustainable urban development.

[0116] The technical features of the above-mentioned embodiments can be combined arbitrarily. In order to make the description concise, not all possible combinations of the technical features in the above-mentioned embodiments are described. However, as long as there is no contradiction in the combination of these technical features, they should be considered to be within the scope of this specification.

[0117] The above-described embodiments merely represent several implementation methods of the present invention. While the descriptions are relatively specific and detailed, they should not be construed as limiting the scope of the patent. It should be noted that a person skilled in the art could make various modifications and improvements without departing from the spirit of the present invention, all of which fall within the scope of protection of the present invention. Therefore, the scope of protection of the patent for this invention shall be determined by the appended claims.

Claims

1. A method for detecting urban functional areas based on mobile phone signaling data, characterized by: This method uses an improved YOLOv8 deep learning network model to detect urban functional areas in images generated by mobile phone signaling data. The specific detection steps are as follows: Step 1: obtaining mobile phone signaling data, preprocessing the mobile phone signaling data, and obtaining preprocessed mobile phone signaling data; Step 2: Create and label an image dataset based on the pre-processed mobile phone signaling data obtained in step 1; Step 3: Obtain the POI dataset and combine it with the mobile phone signaling data to obtain the POI one-dimensional feature data table; Step 4: Combine the MLP network model with the YOLOv8 deep learning network model to obtain an improved YOLOv8 deep learning network model; The image dataset described in step 1 and the POI one-dimensional feature data table described in step 3 are input into the improved YOLOv8 deep learning network model for detection to obtain the region detection results.

2. The method for detecting urban functional areas based on mobile phone signaling data according to claim 1, characterized in that: In step 1, the mobile phone signaling data is pre-processed, specifically: The raw data is obtained from the mobile phone signaling data source, and the raw data is parsed, and missing values ​​and outliers are processed; then the data is cleaned, including removing noise, correcting erroneous data, and processing duplicate records and outliers; finally, the consistency, completeness and accuracy of the data are verified.

3. The method for detecting urban functional areas based on mobile phone signaling data according to claim 1, characterized in that: In step 2, the process of creating an image dataset from the pre-processed mobile phone signaling data is as follows: Map the mobile phone signaling data on the map, select Amap as the base map, and take a screenshot of the corresponding area on the map where the mobile phone signaling data is mapped. This will obtain an image dataset consisting of map images. The image data will then be annotated to generate a corresponding label file in XML format.

4. The method for detecting urban functional areas based on mobile phone signaling data according to claim 1, characterized in that: In step 3, the acquired POI dataset is cleaned, and then combined with the mobile phone signaling data, the mobile phone signaling data points within 24 hours and 12 time zones of the corresponding latitude and longitude points and the POI data points within a 1 km range are counted; and the distance to the nearest POI data point of each latitude and longitude point is counted; The statistical data is then made into a one-dimensional feature data table of the mobile phone signaling data combined with the POI data set, and annotated according to the POI data points.

5. The method for detecting urban functional areas based on mobile phone signaling data according to claim 1, characterized in that: In step 4, the improved YOLOv8 deep learning network model is connected by connecting the MLP network with the YOLOv8 backbone network, and connecting the output of the MLP network with the detection branch of YOLOv8 to achieve the fusion of the two network models.

6. The method for detecting urban functional areas based on mobile phone signaling data according to claim 5, characterized in that: Improve the loss functions of the YOLOv8 network model, including category classification loss function and bounding box regression loss function; The classification loss function It can be expressed as follows: + ;in , They are as follows: ； ； Where, Custom parameters and , C is the total number of categories, each category has a unique label k, indicating its position in the entire category set; N is the total number of samples in the data set, is the k-th element of the true label vector of the i-th sample, is the element at the kth position of the predicted probability distribution vector of the i-th sample; The border regression loss includes improved DFL and CIOU Loss, and the loss function is as follows: The DFL loss function is: )； Where, is a quality label in the range of 0 to 1, is the quality label of the i+1th sample, For quality label 、 probability; The CIoU Loss is: CIoU Loss ; Where, To calculate the area of ​​the intersection of two bounding boxes divided by the area of ​​the union of two bounding boxes, is the Euclidean distance between two rectangular boxes, are the center points of the two rectangular boxes, is the weight coefficient, is the distance between the diagonals of the enclosed area of ​​the two rectangular boxes, Used to measure the consistency of the relative proportions of two rectangular boxes.

7. The method for detecting urban functional areas based on mobile phone signaling data according to claim 1, characterized in that: Step four also includes preprocessing the map images in the image dataset input into the improved YOLOv8 deep learning network model; including scaling, cropping, rotating or enhancing the map images so that the map images have a uniform size and a format suitable for model input; at the same time, standardizing and normalizing the map images so that the map images meet the input requirements of the improved YOLOv8 deep learning network model.

Images