CN116797936A - Urban function area detection method based on mobile phone signaling data - Google Patents
Urban function area detection method based on mobile phone signaling data Download PDFInfo
- Publication number
- CN116797936A CN116797936A CN202311006651.7A CN202311006651A CN116797936A CN 116797936 A CN116797936 A CN 116797936A CN 202311006651 A CN202311006651 A CN 202311006651A CN 116797936 A CN116797936 A CN 116797936A
- Authority
- CN
- China
- Prior art keywords
- data
- mobile phone
- phone signaling
- signaling data
- poi
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Pending
Links
- 230000011664 signaling Effects 0.000 title claims abstract description 122
- 238000001514 detection method Methods 0.000 title claims abstract description 71
- 230000006870 function Effects 0.000 claims abstract description 74
- 238000013135 deep learning Methods 0.000 claims abstract description 35
- 238000000034 method Methods 0.000 claims description 66
- 230000008569 process Effects 0.000 claims description 23
- 238000012545 processing Methods 0.000 claims description 22
- 238000009826 distribution Methods 0.000 claims description 17
- 238000004140 cleaning Methods 0.000 claims description 16
- 238000002372 labelling Methods 0.000 claims description 12
- 238000007781 pre-processing Methods 0.000 claims description 11
- 230000002159 abnormal effect Effects 0.000 claims description 4
- 238000012937 correction Methods 0.000 claims description 4
- 230000004927 fusion Effects 0.000 claims description 3
- 238000013507 mapping Methods 0.000 claims description 3
- 238000010606 normalization Methods 0.000 claims description 3
- 238000011895 specific detection Methods 0.000 claims description 2
- 230000002708 enhancing effect Effects 0.000 claims 1
- 238000011161 development Methods 0.000 abstract description 19
- 238000005516 engineering process Methods 0.000 abstract description 17
- 238000012549 training Methods 0.000 abstract description 14
- 230000000694 effects Effects 0.000 abstract description 7
- 238000010276 construction Methods 0.000 abstract description 3
- 238000013473 artificial intelligence Methods 0.000 abstract description 2
- 230000018109 developmental process Effects 0.000 description 18
- 239000008186 active pharmaceutical agent Substances 0.000 description 15
- 238000004458 analytical method Methods 0.000 description 9
- 238000012360 testing method Methods 0.000 description 8
- 238000012544 monitoring process Methods 0.000 description 7
- 238000005457 optimization Methods 0.000 description 7
- 238000007405 data analysis Methods 0.000 description 5
- 238000012795 verification Methods 0.000 description 5
- 238000011156 evaluation Methods 0.000 description 4
- 238000004519 manufacturing process Methods 0.000 description 4
- 230000011218 segmentation Effects 0.000 description 4
- 230000004913 activation Effects 0.000 description 3
- 230000008901 benefit Effects 0.000 description 3
- 238000002474 experimental method Methods 0.000 description 3
- 230000006872 improvement Effects 0.000 description 3
- 238000010801 machine learning Methods 0.000 description 3
- 238000005192 partition Methods 0.000 description 3
- 238000013528 artificial neural network Methods 0.000 description 2
- 230000006399 behavior Effects 0.000 description 2
- 230000009286 beneficial effect Effects 0.000 description 2
- 238000004364 calculation method Methods 0.000 description 2
- 238000004891 communication Methods 0.000 description 2
- 238000010835 comparative analysis Methods 0.000 description 2
- 238000010586 diagram Methods 0.000 description 2
- 238000000605 extraction Methods 0.000 description 2
- 230000033001 locomotion Effects 0.000 description 2
- 230000004044 response Effects 0.000 description 2
- 238000010200 validation analysis Methods 0.000 description 2
- 230000002776 aggregation Effects 0.000 description 1
- 238000004220 aggregation Methods 0.000 description 1
- 230000008859 change Effects 0.000 description 1
- 238000013524 data verification Methods 0.000 description 1
- 238000003066 decision tree Methods 0.000 description 1
- 230000007812 deficiency Effects 0.000 description 1
- 230000007613 environmental effect Effects 0.000 description 1
- 238000001914 filtration Methods 0.000 description 1
- 238000009499 grossing Methods 0.000 description 1
- 230000012010 growth Effects 0.000 description 1
- 230000010354 integration Effects 0.000 description 1
- 230000007774 longterm Effects 0.000 description 1
- 238000013508 migration Methods 0.000 description 1
- 230000005012 migration Effects 0.000 description 1
- 238000012986 modification Methods 0.000 description 1
- 230000004048 modification Effects 0.000 description 1
- 230000008447 perception Effects 0.000 description 1
- 230000001737 promoting effect Effects 0.000 description 1
- 238000011160 research Methods 0.000 description 1
- 238000012216 screening Methods 0.000 description 1
- 238000012706 support-vector machine Methods 0.000 description 1
- 239000002699 waste material Substances 0.000 description 1
Landscapes
- Image Analysis (AREA)
Abstract
The invention discloses a city function area detection method based on mobile phone signaling data, relates to the fields of mobile phone signaling big data, artificial intelligence technology, image recognition and the like, solves the problems that the existing city function area detection method based on mobile phone signaling data is easy to be interfered by noise and boundary effect, accurate and complete geographic data is required, and the use is limited due to the fact that a large amount of training data and characteristic engineering are required. The invention utilizes the mobile phone signaling data and the POI data set, and utilizes the combination of the deep learning computer vision technology and the mobile phone signaling data to detect the urban functional area, thereby being a great progress for the construction and development of cities.
Description
Technical Field
The invention relates to the fields of mobile phone signaling big data, artificial intelligence technology, image recognition and the like, in particular to a city area detection method based on mobile phone signaling data, which is used for solving the city planning problem by applying a deep learning computer vision technology in the mobile phone signaling data.
Background
The detection of urban functional areas is of great importance in urban planning. Firstly, the method provides basic data and information for land utilization planning, so that a planner can determine the optimal use of different areas, thereby improving the land utilization efficiency and avoiding resource waste and unreasonable configuration. Secondly, by knowing the functional characteristics and advantages of different areas, a planner can formulate corresponding development strategies to promote industrial development, employment opportunities and economic growth. In addition, detection of urban functional areas plays an important role in urban traffic planning, ecological protection and environmental planning, and improvement of social fairness and resident life quality.
The existing urban area detection method based on mobile phone signaling data comprises a classification method based on density clustering, spatial interpolation based on a Geographic Information System (GIS) and machine learning. The high density region is distinguished from the low density region by an algorithm (e.g., DBSCAN) based on density clustering, but is easily interfered by noise and boundary effects, and parameters need to be set in advance. The GIS-based spatial interpolation method utilizes geographic data to correlate mobile phone signaling data points to infer the activity level of the whole urban area, but requires accurate and complete geographic data. Machine learning based classification methods use algorithms (such as decision trees, support vector machines, or neural networks) to classify the mobile signaling data, dividing different regions, but require a lot of training data and feature engineering. These methods all require a combination of applicability and limitations. Therefore, the invention proposes a method for detecting urban area functions based on the combination of mobile phone signaling data and deep learning computer vision technology.
The mobile phone signaling data plays a key role in the detection and planning of the urban functional area, provides real-time, accurate and comprehensive information for planners, helps to formulate more effective development strategies and policies, and promotes the sustainable and balanced development of cities. The comprehensive application of the integrated data and planning method is helpful for improving the urban efficiency, the urban sustainability and the resident life quality, and achieves the urban sustainable development goal.
The mobile phone signaling data has important significance in city planning. By analyzing the mobile phone signaling data, the city planner can know information on population distribution, flow, traffic conditions, facility requirements, disaster response and the like. The data can be used for optimizing urban infrastructure, traffic network and public service layout, improving traffic flow and reducing congestion, reasonably planning business areas and public facilities, and improving the life quality and emergency rescue efficiency of cities.
The handset signaling data presents some problems in city planning. The most important of these is the protection of personal privacy, as these data relate to the location information and communication behaviour of the user, which if not adequately protected and anonymously handled may lead to disclosure and abuse of personal privacy. In addition, data accuracy and representativeness, complexity of data acquisition and processing, and data usage limitations are also challenges facing cell phone signaling data applications. When using these data, the relevant laws and regulations must be complied with, the legitimacy and privacy protection of the data be ensured, and appropriate corrections and verifications be made to ensure the accuracy and reliability of the data.
The application of mobile phone signaling data in the field of urban planning is becoming an important research and practice direction. With the popularization of smart phones and the development of mobile networks, the acquisition and availability of mobile phone signaling data are gradually improved. Meanwhile, the progress of data analysis technology has led to continuous enhancement of processing and analysis capabilities of mobile phone signaling data. In city planning, the mobile phone signaling data are applied to traffic planning and optimization, and by analyzing the movement modes and traffic flow of people, a planner is helped to optimize traffic signal timing and road planning, and traffic efficiency is improved. In addition, the mobile phone signaling data also provides guidance for facility layout and public service, and the layout of business areas and public facilities is optimized by analyzing the crowd activity pattern and the demand, so that the life quality of cities is improved. In general, the application prospect of the mobile phone signaling data in urban planning is wide, and more accurate and real-time data support is provided for urban development.
The deep learning computer vision technology has the advantages of high efficiency, accuracy, multi-source data integration, real-time monitoring and updating and the like in urban functional area detection. The method can rapidly process a large amount of image and video data, and can identify and detect functional areas through algorithms, so that the working efficiency and accuracy are improved. Meanwhile, the deep learning computer vision technology can integrate various data sources, such as satellite images and monitoring cameras, provide comprehensive and diversified data, and is beneficial to more accurately detecting and detecting urban functional areas. In addition, the technology can monitor the change and evolution of the city in real time, discover new functional areas or changed areas in time and update the new functional areas or changed areas. The deep learning computer vision technology has wide application prospect in urban functional area detection, and provides important data support and decision reference for urban planning and development.
In summary, the application of the mobile phone signaling data is mainly focused on population distribution and flow, traffic planning, business positioning and the like. Through analyzing the mobile phone signaling data, population distribution conditions and flowing trends can be known, and urban planners are helped to better understand population aggregation areas and population migration modes, so that urban population distribution and social service facility layout are optimized. However, it is absolutely inexhaustible to use mobile phone signaling data and deep learning computer vision technology to detect urban functional areas in China. Therefore, the invention provides an image data set generated by utilizing mobile phone signaling data, and then utilizes a deep learning computer vision technology to detect urban functional areas.
Disclosure of Invention
The invention provides a city function area detection method based on mobile phone signaling data, which aims to solve the problems that the existing city function area detection method based on mobile phone signaling data is easy to be interfered by noise and boundary effect, accurate and complete geographic data is needed, and the use is limited due to the fact that a large amount of training data, characteristic engineering and the like are needed.
The urban function area detection method based on the mobile phone signaling data carries out urban function area detection on image data generated by the mobile phone signaling data based on an improved YOLOv8 deep learning network model; the specific detection steps are as follows:
step one, acquiring mobile phone signaling data, preprocessing the mobile phone signaling data, and acquiring preprocessed mobile phone signaling data;
step two, making and labeling an image dataset for the preprocessed mobile phone signaling data obtained in the step one;
step three, acquiring a POI data set, and acquiring a POI one-dimensional characteristic data table by combining mobile phone signaling data;
combining the MLP network model with the YOLOv8 deep learning network model to obtain an improved YOLOv8 deep learning network model;
and (3) inputting the image data set in the step one and the POI one-dimensional characteristic data table in the step three into an improved YOLOv8 deep learning network model for detection, and obtaining a region detection result.
The invention has the beneficial effects that:
1. the method of the invention firstly provides the detection of the urban function area based on the mobile phone signaling data. Cleaning, processing and visualizing the mobile phone signaling data, and making and marking an image data set; and meanwhile, cleaning and processing the POI data set, and manufacturing a POI one-dimensional characteristic data table. Then, the detection of the urban functional area is performed. The urban function area detection method based on the combination of the mobile phone signaling data and the deep learning computer vision technology has innovation and advancement.
2. The model framework in the method is a multi-task cascade framework, and the first stage of cascade tasks is to improve a YOLOv8 deep learning network model; the MLP network model is integrated into the YOLOv8 network, meanwhile, the loss function of the improved model is defined, and the layer number, the channel number, the super parameters and the like of the network are set. The second stage of the cascading task is to input the image dataset and the POI one-dimensional feature data table into the modified YOLOv8 model, and the network will output the class, bounding box and confidence information for each detected object. According to the confidence threshold, detection results can be screened, and targets with high confidence can be selected.
3. The method of the invention utilizes the combination of the mobile phone signaling data and the POI data set and the deep learning computer vision technology and the mobile phone signaling data to detect the urban functional area, thereby being a great progress for the construction and development of cities.
Drawings
Fig. 1 is a flowchart of a city function area detection method based on mobile phone signaling data according to the present invention;
FIG. 2 is a network structure diagram of a linear model according to the present invention;
fig. 3 is a block diagram of an MLP network in the urban function area detection method based on mobile phone signaling data according to the present invention.
Detailed Description
In a first embodiment, a city function area detection method based on mobile phone signaling data is described with reference to fig. 1, 2 and 3, where the city function area detection is performed on image data generated by mobile phone signaling data based on an improved YOLOv8 deep learning network model.
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 categories and bounding boxes present in each cell. Compared with the traditional method, the YOLO has the advantages that the detection precision is not lost while the high speed is kept, and the YOLO is widely applied to the field of computer vision, such as automatic driving, video monitoring, object recognition and the like. YOLO v8 is the latest series of YOLO based on object detection model, introduced by Ultralytics company.
According to the detection method, firstly, mobile phone signaling data of a city are obtained, and preprocessing and cleaning operations are carried out on the data, so that poor-quality data in the mobile phone signaling data are removed, and the accuracy of the mobile phone signaling data is improved; secondly, generating an image by using the mobile phone signaling data, manufacturing an image data set, and labeling the image data set. And then acquiring a POI data set, cleaning the POI data set, and combining the cleaned POI data set with mobile phone signaling data to generate a POI one-dimensional characteristic data table. And next, inputting the image data set and the POI one-dimensional characteristic data table into an improved YOLOv8 network model to obtain an image detection result. Finally, after the detection result of the image is obtained, the detection result is subjected to contrast analysis, and a corresponding city functional area is formulated according to the detection result of the image, so that the construction of the city is assisted.
In this embodiment, the POI data refers to point of interest (Point of Interest) data, which includes various entities in geographical locations, such as shops, restaurants, scenic spots, banks, and the like. POI data typically includes location coordinates, category, name, address, telephone number, and the like. POI data plays an important role in Geographic Information Systems (GIS) and location services applications. The method can be used in the fields of map navigation, position search, business analysis, city planning and the like. By collecting and maintaining POI data, accurate location information and rich geographic features can be provided, helping users to understand the surrounding environment, find places of interest, and support business decisions and service optimization. In mobile applications and online map services, POI data is an important component in building user-friendly and personalized experiences.
The detection method of the present embodiment is specifically implemented by the following steps:
s1, acquiring mobile phone signaling data, preprocessing the mobile phone signaling data, and acquiring preprocessed mobile phone signaling data;
in this embodiment, the collection and processing of mobile phone signaling data: and obtaining the original data from the mobile phone signaling data source, and carrying out data processing through preprocessing and cleaning. In preprocessing, the original data is parsed, missing values, outliers, etc. are processed. Data cleansing is then performed, including noise removal, error data correction, repeated recording and outliers processing, etc., to ensure accuracy and integrity of the data. Finally, data verification is performed to check the consistency, integrity and accuracy of the data and to find potential problems or errors. The goal of this step is to obtain high quality, reliable handset signaling data, providing a reliable basis for subsequent data analysis and application. The data were pre-processed as follows:
s11, acquiring signaling data of a mobile phone; first, raw data needs to be acquired from a handset signaling data source, involving negotiating and acquiring data with a network provider, mobile operator, or other data provider.
S12, processing and cleaning mobile phone signaling data; processing is required for ping-pong data and drift data. For ping-pong data, the data can be screened by setting a signal strength threshold value, and the influence caused by frequent switching between devices is eliminated. For drift data, smoothing or filtering techniques can be used to smooth signal strength, remove noise and outliers in the data, and ensure accuracy and integrity of the data.
S13, verifying mobile phone signaling data; to ensure the quality of the cleaned data, the data is checked for consistency, integrity and accuracy and verified to ensure compliance with expected standards and requirements. The results of the verification may help discover potential problems or errors and provide confidence and reliability for use in subsequent experiments.
S2, manufacturing an image data set from the preprocessed mobile phone signaling data, wherein the specific process for manufacturing the image data set is as follows:
s21, observing the distribution condition of the mobile phone signaling data, and collecting the mobile phone signaling data, wherein the data comprise connection information, signal strength and other related data of the mobile phone at different positions. Such information is typically collected by base stations and mobile network devices and used for communication and positioning purposes. The present embodiment uses these data to understand the distribution of the mobile phone signals at different geographic locations.
S22, mapping the mobile phone signaling data on a map; in this embodiment, a high-level map is selected as a base map of a data set, and is a widely used map service that provides detailed geographic information and rich map data. After the mobile phone signaling data is mapped on the map, selecting an area where the mobile phone signaling data is displayed on the map, and intercepting an image of the area.
S23, selecting a proper size of an image, and intercepting a map image; when the map image is intercepted, the size of each intercepted image is 416 multiplied by 416 pixels, and the pixel size is widely applied to image processing tasks in the field of computer vision, and has moderate resolution and calculation efficiency. The mobile phone signaling latitude and longitude points of each intercepted image center are recorded, the name of the image is named according to the latitude and longitude, for example, the latitude and the longitude of the coordinates of a certain image center point are 22.1245 degrees and 114.1245 degrees respectively, and then the image is named as 22.1245_114.1245.Png. The intercepted map images are formed into an image data set to be used as an important basis for subsequent data processing and geographic positioning.
S24, labeling the intercepted image data, and labeling the image data by utilizing Labelme to generate a corresponding XML-format tag file; labelme is a classical labeling tool and supports tasks such as target detection, semantic segmentation, instance segmentation and the like. The labeling process is to select a map image to label according to the city function area, and the city function area label designed by the embodiment of the invention is as follows: traffic devices, public services, public service_traffic devices, businesses, business_traffic devices, business_public services, business_living, business_industrial, living, living_traffic devices, living_public services, living_industrial, industrial, industrial_traffic devices, industrial_public services, mixed functional areas.
S3, acquiring a POI data set, extracting features, and acquiring a POI one-dimensional feature data table by combining mobile phone signaling data;
in order to combine the mobile phone signaling data and the POI (point of interest) data set into a one-dimensional feature data table for combined analysis, firstly, a developer account is registered in a Hide open platform, and an application program is created to acquire an API key. The POI dataset of interest is obtained by invoking the Web services API of the hadamard map, specifying keywords and parameters. The acquired data may include information such as POI names, addresses, longitude and latitude. And then cleaning and detecting the POI data set, processing the missing value and the abnormal value, and performing standardization and normalization. And detecting the POI data set according to the field characteristics. And finally, combining the POI data set with the mobile phone signaling data, and performing feature extraction to generate a one-dimensional feature data table. The specific implementation process is as follows:
s31, calling a Web service API to acquire an account number of a Hilder developer; first, it is necessary to register a developer account with the open platform of the german and create an application to obtain a Key (Key) to access the API. Logging in the Gaoder open platform, registering account numbers according to the guide and creating application programs. Using the key, the POI dataset may be obtained by calling a Web services API of the golgi map. The Goldmap provides a series of APIs to access different types of data, such as geocoding, inverse geocoding, POI searching, etc.
S32, acquiring a POI data set of the urban area; by constructing the correct API request, a specific Key (Key) or other parameter is specified to obtain the POI dataset of interest. For example, a keyword may be designated as a restaurant and provide the name or latitude and longitude coordinates of the city in which it is located. The constructed API request is sent to the API endpoint of the Goldmap using the selected programming language and the HTTP client library. The key contained in the request is ensured for authentication. After receiving the response of the API, analyzing the returned JSON data, and processing and analyzing according to the need. The name, address, longitude and latitude coordinates and other information of the POI can be extracted to obtain POI data.
S33, cleaning and detecting the POI data set, and then loading and previewing the POI data to know the structure and field information of the data. Then, the missing values and the outliers are processed, and the missing values may be deleted or filled, and the outliers may be corrected or deleted. Then, the data are standardized and normalized, and then POI data are detected according to field characteristics; such as recreational entertainment, transportation facilities, commercial housing, corporate enterprises, and the like.
S34, combining POI data with mobile phone signaling data to generate a one-dimensional characteristic data table; and (3) inputting the longitude and latitude points of the center of the intercepted map image recorded in the step (S2) into a one-dimensional data table. Counting how many mobile phone signaling data points exist in the longitude and latitude points within 24 hours and 12 time zones, and what types of POI data points exist in the surrounding 1 km range, wherein each type of POI data points is in a plurality of types. Counting the distance of the nearest POI data point of each longitude and latitude point; and then the data are made into a one-dimensional characteristic data table of the mobile phone signaling combined POI data set, and labeling is carried out according to POI data points.
S4, combining the MLP network model with the YOLOv8 deep learning network model to obtain an improved YOLOv8 deep learning network model;
the specific process is as follows:
s42, YOLOv8 model improvement and network configuration: and performing network configuration according to the requirement of the YOLOv8 network, setting the layer number, the channel number and the super parameters of the network, defining a loss function and the like.
In this embodiment, YOLOv8 is a target detection model that consists of multiple convolution layers and full-link layers. In this embodiment, a multi-layer perceptron (MLP) network model is integrated into a YOLOv8 network model, namely: and connecting the MLP network with a main network of the YOLOv8, and connecting the output of the MLP network with a detection branch of the YOLOv8 to realize the fusion of the two networks. And improves its single channel input image to a network model that can input an image dataset and a POI one-dimensional feature data table.
S43, setting a loss function: an improved YOLOv8 target detection network model is presented in this embodiment and incorporates an MLP network therein. The loss function of the improved YOLOv8 network model uses a class classification loss function and a frame regression loss function.
The class classification loss function finally adopts an improved cross entropy loss function, and the improved cross entropy loss function is as follows:
Loss＝Loss 1 +Loss 2 ；
wherein Loss is 1 ，Loss 2 The method comprises the following steps:
；
；
in the method, in the process of the invention,custom parameters andc is the total number of categories, each category having a unique reference number k indicating its location in the entire set of categories. N is the total number of samples in the dataset, i is the index of the samples, and for the ith sample, its true tag vector is expressed as,The element at the kth position of the true label vector of the ith sample, wherein only the kth element is 1, and the rest elements are 0 #) And the predictive probability distribution vector of the model for the sample isRepresenting the model's predicted probability for each category,is the element of the kth position of the predictive probability distribution vector for the ith sample.
The frame regression Loss function includes a modified DFL Loss function and a CIOU Loss.
The DFL loss function is as follows:
；
in the method, in the process of the invention,a mass label in the range of 0 to 1,for the quality label of the i +1 th sample,is a quality label、Probability of (2);
whereas DFL was developed based on Quality Focal Loss (QFL), in QFL the loss function is formulated as follows:
；
wherein the method comprises the steps ofA mass label of 0 to 1,is a predicted value. QFL the global minimum solution is=. The cross entropy part thus becomes a complete cross entropy, while the adjustment factor becomes a power function of the absolute value of the distance.Is an adjustable index parameter for adjusting the weight of the difficulty sample, and is typically set to a positive value, for example, a value of 2.
With the additional loss of QFL, it is desirable that the network be able to focus quickly on values near the labeling location so that their probability is as high as possible. Based on this, the term Distribution Focal Loss (DFL), i.e., DFL of the above formula.
In the DFL, it is very similar in form to the right half of QFL, in the DFLFor the two positions left and right (mass labels) where the label y is closest、) Thereby allowing the network to quickly focus on the distribution of the vicinity of the target location.
The CIoU Loss function is as follows:
CIoU Loss；
in the method, in the process of the invention,dividing the intersection area of two bounding boxes (or segmentation masks) by their union area，Is the euclidean distance between two rectangular boxes,respectively the center points of the two rectangular frames,as the weight coefficient of the light-emitting diode,is the distance of the diagonal of the closure areas of the two rectangular boxes,for measuring the consistency of the relative proportions of the two rectangular boxes.
In this embodiment, the method further includes using an optimization algorithm to reduce the gradient, and optimizing the model parameters by minimizing the total loss function, so as to improve the performance of the integrated model on the target detection task. The MLP network is integrated into the YOLOv8 model, so that the feature learning and expression capability of the model can be enhanced, and better target detection performance is obtained.
S5, inputting the image dataset described in the step S1 and the POI one-dimensional characteristic data table described in the step S2 into an improved YOLOv8 deep learning network model for detection, and obtaining a region detection result, a boundary box result and a confidence coefficient result.
The specific process is as follows:
s51, preprocessing the image data set, and preprocessing the map image in the acquired image data set. Including scaling, cropping, rotation, or other enhancement techniques of the image to ensure that the image has a uniform size and format suitable for model input. Meanwhile, the image is standardized and normalized, so that the image data is ensured to meet the input requirement of an improved YOLOv8 deep learning network model. Then dividing the image data set and the POI one-dimensional characteristic data table into a training set, a verification set and a test set.
S52, the training set is a data set for inputting the improved YOLOv8 deep learning network model and training the YOLOv8 deep learning network model. In the process, firstly, image data of a training set is subjected to network feature extraction layer image features, and meanwhile, a POI one-dimensional feature data table is subjected to an MLP network layer to obtain a classification result. And secondly, combining the classification result with the extracted image features to perform the next training. During the training process, the weights and biases of the network are updated through a back propagation algorithm so that the network can gradually learn the target detection and the detected modes and features.
S52, parameter adjustment and optimization are carried out by using the verification set. The validation set is a data set used in the improved YOLOv8 deep learning network model training process for adjusting the hyper parameters (hyper parameters) of the model. Super-parameters refer to parameters that need to be set prior to model training, such as learning rate, regularization coefficients, etc. By using the validation set, experiments with different combinations of hyper-parameters can be performed on the models, selecting the model configuration that performs best on the test set. The purpose of this is to avoid the model from being over fitted on the test set, while the verification set plays a role in "confidentiality" in the model parameter tuning process, preventing the model from being over optimized for the test set.
And S53, after training is completed, using a test set to carry out image detection on the trained YOLOv8 deep learning network model. 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 generalization ability of the model, i.e., the predicted behavior of the model to data that has not been seen, is estimated using the samples in the test set. Firstly, inputting a test set to be detected into a network model, and acquiring the category, the bounding box and the confidence information of each detected target. And secondly, screening according to a confidence threshold value, and selecting a target with high confidence. And finally, analyzing the result output by the network to obtain a region detection result, a boundary box result and a confidence coefficient result, and visualizing the result.
In this embodiment, the method further includes performing analysis and evaluation on the obtained detection result: including counting the number, proportion and distribution of the functional area categories and knowing their geographic location and spatial characteristics. These results are then compared with existing city plans, functional partitions or planning metrics to determine if there are areas of missing functionality or need to be adjusted. Based on the data of the comparison analysis, the function evaluation is carried out, the contribution degree of POI data sets and mobile phone signaling data of different categories to the city function is considered, and factors such as the service range, the capacity and the convenience of each area are evaluated.
And (3) formulating a corresponding city planning strategy according to the analysis and evaluation result, and formulating corresponding city function steps, including adjusting planning, improving regional functions, adding or optimizing city facilities, adjusting traffic layout, promoting economic development and the like. And then, periodically monitoring the implementation effect of the urban function step, evaluating urban development results and problems, and carrying out necessary adjustment and optimization according to the monitoring results so as to continuously perfect the functional layout and development direction of the city.
As shown in fig. 1, 2 and 3, the present embodiment is an example of the urban function area detection method based on mobile phone signaling data according to the first embodiment.
In this embodiment, the model for urban function area detection based on urban function area detection of mobile phone signaling data is used as experimental data, the mobile phone signaling data within 24 hours in a certain day is used as experimental data, and the POI data set selects interest point data in the Shenzhen urban area.
Firstly, shenzhen mobile phone signaling data are acquired, and secondly, the mobile phone signaling data are preprocessed and cleaned. Then, mapping the mobile phone signaling data set on a map, making the mobile phone signaling data set into an image data set, and labeling; meanwhile, acquiring a POI data set of Shenzhen city, and cleaning the POI data set; after the POI data set is cleaned, the POI one-dimensional characteristic data table is manufactured by combining the mobile phone signaling data set. Then, the image data set and the POI one-dimensional characteristic data table are input into an improved YOLOv8 network model through the processing of the image data set; and obtaining a detection result. And then analyzing and evaluating the detection result to obtain a strategy corresponding to the subsequent city planning degree. The method specifically comprises the following steps:
firstly, acquiring mobile phone signaling data, and then preprocessing and cleaning. The original data of the signaling data of the mobile phone is obtained through negotiation with a network provider or a mobile operator, etc. The original data of the mobile phone signaling is shown in table 1, and table 1 is the original mobile phone signaling data, and the original data has 38218717 rows and 4 columns of data.
TABLE 1
| 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 |
Next, these raw data are preprocessed, including parsing formats, processing missing values, outliers, etc. Data cleaning, noise removal, error correction, repeated processing, etc., are performed, the data of the first and last occurrence of the ID, longitude, latitude are reserved in this experiment, and a list of total times, i.e., the total time of all the start times is formatted into a total time of seconds, is newly added. After the cleaning of the original data is completed, 2583202 rows and 5 columns of data are shown in table 2, and table 2 is the signaling data of the cleaned mobile phone.
TABLE 2
| ID | Start time | Longitude and latitude | Latitude of 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 |
2. Map the data on a map and acquire an image dataset.
First, the collected mobile phone signaling data is displayed on the map. Because of the low precision of the mobile phone signaling data, the embodiment adds a certain data disturbance process to the display of the mobile phone signaling data on the map, so that the privacy of a user can be better protected and the data can be displayed on the map. At the same time, 24 hours a day is divided into 12 time zones, and each different time zone is selected to be different in color. Then, a specific area is selected for interception. Each intercepted image has the size of 416 multiplied by 416 pixels, the longitude and latitude points of the mobile phone signaling in the center of each intercepted image are recorded, and the image data are marked by Labelme to generate a corresponding XML format tag file.
3. And acquiring a POI data set, cleaning and preparing a POI one-dimensional characteristic data table by combining mobile phone signaling data.
First, a Web service API of the hadamard map needs to be called. A german developer account is registered and an application is created to obtain a key to access the API. By constructing an API request, keywords and parameters are specified to obtain POI data of interest. The key is used to send an API request to the hyperde map API endpoint and parse the returned JSON data. And cleaning the POI data set, namely processing the missing value, the abnormal value, the standardization and the normalization, and detecting the data according to the field characteristics. The longitude and latitude points of the recorded picture center are input into a one-dimensional data table by the combined mobile phone signaling data, the number of mobile phone signaling data points in 12 time zones and POI data points in the surrounding 1 km range are counted in 24 hours, and the number of the POI data points is several, and each category is several. Counting the distance of the nearest POI data point of each longitude and latitude point; and then the data are made into a one-dimensional characteristic data table of the mobile phone signaling combined POI data set and marked according to POI data points.
Meanwhile, the city function areas selected in the embodiment are as follows: traffic devices, public services, public service_traffic devices, businesses, business_traffic devices, business_public services, business_living, business_industrial, living, living_traffic devices, living_public services, living_industrial, industrial, industrial_traffic devices, industrial_public services, mixed functional areas.
4. The steps of integrating the MLP network into the YOLOv8 network model to improve the target detection performance are as follows: and connecting the MLP network with a main network of the YOLOv8, and connecting the output of the MLP network with a detection branch of the YOLOv8 to realize the fusion of the two networks.
The general formula for a linear model of an MLP network can be expressed as:
+；
in the method, in the process of the invention,the predicted output of the model is represented,to the point ofIs a characteristic value of the sample;for each characteristic value, a weight coefficient, e is a bias parameter,can be regarded as a weighted sum of all feature values, as shown in fig. 2, in which fig. 2 the input features and the predicted result are represented by nodes, using the weight coefficientsFor connecting the nodes.
In the MLP model, a Hidden layer (Hidden Layers) is added in the process of the algorithm, then the weighted summation calculation is repeatedly performed on the Hidden layer, and finally the result calculated by the Hidden layer is used to generate a final result, as shown in fig. 3, the characteristic coefficient (weight) to be learned by the model is much more. There is a coefficient between each input feature and the hidden units (hidden units), which is also done to generate these hidden units. And there is also a coefficient between each hidden unit and the final result. After the hidden layer is generated, the activation unit is non-linearized using an activation function (activation function) because the non-linear processing is to simplify sample features so that the neural network can learn complex non-linear data sets.
5. A loss function of the network model is defined. In this embodiment, the improved YOLOv8 network model loss function uses class classification loss and frame regression loss. The class classification loss finally adopts an improved cross entropy loss function, and the class loss function is as follows:
+；
wherein the method comprises the steps of，The method comprises the following steps:
；
；
in the method, in the process of the invention,custom parameters andc is the total number of categories, each category having a unique reference number k indicating its location in the entire set of categories. N is the total number of samples in the dataset, i is the index of the samples, and for the ith sample, its true tag vector is expressed as,The element representing the kth position of the true label vector of the ith sample, wherein only the kth element is 1 and the remaining elements are 0 #) And the predictive probability distribution vector of the model for the sample isRepresenting the model's predicted probability for each category,an element representing the kth position of the predictive probability distribution vector for the ith sample.
The frame regression Loss adopts improved DFL and CIOU Loss, and the Loss function is as follows:
wherein the DFL loss function is:
；
in the method, in the process of the invention,a mass label in the range of 0 to 1,quality tag for the (i+1) th sample，Is a quality label、Probability of (2);
wherein CIoU Loss is:
CIoU Loss；
in the method, in the process of the invention,to divide the intersection area of two bounding boxes (or segmentation masks) by their union area,is the Euclidean distance between two rectangular frames,Respectively the center points of the two rectangular frames,as the weight coefficient of the light-emitting diode,is the distance of the diagonal of the closure areas of the two rectangular boxes,for measuring the consistency of the relative proportions of the two rectangular boxes.
6. And obtaining a detection result of the Shenzhen urban function region, making corresponding urban function steps according to the result through data analysis and comparison analysis, counting the number, proportion and distribution conditions of the Shenzhen urban function region, evaluating the function deficiency and adjustment requirements, and considering the contribution degree and service range of different data to the Shenzhen urban function.
Data analysis: and carrying out data analysis on the new detection result, including counting the number, proportion and distribution of each functional area category. The geographical location and spatial distribution characteristics of the different categories are known.
Comparison analysis: and comparing and analyzing the detection result with the existing city planning, city functional partition or city planning index. And comparing the new detection result with the city planning requirement or the consistency of the functional partitions, and determining whether the areas with the functions missing or to be adjusted are planned.
Functional assessment: and evaluating the functional characteristics of each region according to the detection result and the data of the comparative analysis. The contribution degree of POI data and mobile phone signaling data of different categories to urban functions, such as business, traffic, leisure entertainment and the like, and factors such as service range, capacity and convenience of each area are considered.
The city function making step: and according to the results of the comparative analysis and the function evaluation, corresponding city function steps are formulated. The method relates to the adjustment of urban planning, the improvement of regional functions, the addition or optimization of urban facilities, the adjustment of traffic layout, the promotion of economic development and the like. According to specific situations, a long-term city development strategy and a short-term city development strategy can be formulated, and priorities and staged implementation plans of different areas are considered.
Monitoring and adjusting: the implementation effect of the urban function steps is monitored regularly, and urban development achievements and problems are evaluated. And carrying out necessary adjustment and optimization according to the monitoring result so as to continuously perfect the functional layout and development direction of the city.
The detection method of the embodiment combines the mobile phone signaling data with the deep learning computer vision technology. The mobile phone signaling data and the deep learning computer vision technology are combined, so that a rich application prospect is brought in urban planning. By combining the position and movement information of the mobile phone signaling data and the deep learning computer vision technology, more accurate urban function area detection can be realized, and urban space perception and planning support can be provided. The combination provides more comprehensive and accurate data support for city planners, and is helpful for optimizing city planning decisions and improving the sustainable development of cities.
The technical features of the above-described embodiments may be arbitrarily combined, and all possible combinations of the technical features in the above-described embodiments are not described for brevity of description, however, as long as there is no contradiction between the combinations of the technical features, they should be considered as the scope of the description.
The above examples illustrate only a few embodiments of the invention, which are described in detail and are not to be construed as limiting the scope of the invention. It should be noted that it is possible for a person skilled in the art to make several variants and modifications without departing from the inventive concept, which fall within the scope of protection of the present invention, which is therefore subject to the appended claims.
Claims (7)
1. The city function area detection method based on the mobile phone signaling data is characterized in that: the method is based on an improved YOLOv8 deep learning network model to detect urban functional areas of images generated by mobile phone signaling data; the specific detection steps are as follows:
step one, acquiring mobile phone signaling data, preprocessing the mobile phone signaling data, and acquiring preprocessed mobile phone signaling data;
step two, making and labeling an image dataset for the preprocessed mobile phone signaling data obtained in the step one;
step three, acquiring a POI data set, and acquiring a POI one-dimensional characteristic data table by combining mobile phone signaling data;
combining the MLP network model with the YOLOv8 deep learning network model to obtain an improved YOLOv8 deep learning network model;
and (3) inputting the image data set in the step one and the POI one-dimensional characteristic data table in the step three into an improved YOLOv8 deep learning network model for detection, and obtaining a region detection result.
2. The urban function area detection method based on mobile phone signaling data according to claim 1, wherein the method comprises the following steps: in the first step, preprocessing the signaling data of the mobile phone, specifically:
acquiring original data from a mobile phone signaling data source, and processing the original data by analyzing, missing values and abnormal values; then, data cleaning is carried out, including noise removal, error data correction, repeated record processing and abnormal value processing; and finally, verifying the consistency, the integrity and the accuracy of the data.
3. The urban function area detection method based on mobile phone signaling data according to claim 1, wherein the method comprises the following steps: in the second step, the process of making the image data set for the preprocessed mobile phone signaling data is as follows:
mapping the mobile phone signaling data on a map, selecting a Goldmap as a base map, selecting a corresponding region of the mobile phone signaling data mapped on the map for screenshot, obtaining an image data set composed of map images, and then labeling the image data to generate a corresponding XML format tag file.
4. The urban function area detection method based on mobile phone signaling data according to claim 1, wherein the method comprises the following steps: step three, cleaning the acquired POI data set, and then combining the mobile phone signaling number and counting mobile phone signaling data points in 12 time zones and POI data points in a range of 1 km for 24 hours; and counting the distance between nearest POI data points of each longitude and latitude point; and then the counted data is made into a one-dimensional characteristic data table of the mobile phone signaling data combined with the POI data set, and labeling is carried out according to the POI data points.
5. The urban function area detection method based on mobile phone signaling data according to claim 1, wherein the method comprises the following steps: and step four, connecting the MLP network with a main network of the Yolov8, and connecting the output of the MLP network with a detection branch of the Yolov8 to realize the fusion of the two network models for the improved Yolov8 deep learning network model.
6. The urban function area detection method based on mobile phone signaling data according to claim 5, wherein the method comprises the following steps: the loss functions of the improved YOLOv8 network model comprise category classification loss functions and frame regression loss functions;
the classification loss functionExpressed by the following formula: />+/>The method comprises the steps of carrying out a first treatment on the surface of the Wherein->，/>The method comprises the following steps:
；
；
in the method, in the process of the invention,custom parameters and->C is the total number of categories, each category has a unique index k, which indicates the position in the whole category set; n is the total number of samples in the dataset, +.>Element of the kth position of the true tag vector for the ith sample, +.>An element of a kth position of the predictive probability distribution vector for an ith sample;
the frame regression Loss includes the improved DFL and CIOUs Loss, the Loss function is as follows:
wherein the DFL loss function is:
)；
in the method, in the process of the invention,mass label in the range of 0 to 1, < >>Mass label for the i+1th sample, +.>For quality label->、/>Probability of (2);
wherein CIoU Loss is:
CIoU Loss；
in the method, in the process of the invention,to divide the intersection area of two bounding boxes by the union area of two bounding boxes, +.>Is the Euclidean distance between two rectangular frames,>respectively the center points of two rectangular frames, < + >>Is a weight coefficient>Distance of diagonal of closure area for two rectangular frames, < >>For measuring the consistency of the relative proportions of the two rectangular boxes.
7. The urban function area detection method based on mobile phone signaling data according to claim 1, wherein the method comprises the following steps: step four, preprocessing a map image in an image dataset input with the improved YOLOv8 deep learning network model; the method comprises the steps of scaling, clipping, rotating or enhancing the map image to ensure that the map image has uniform size and a format suitable for model input; and meanwhile, the map image is subjected to standardization and normalization processing, so that the map image meets the input requirement of the improved YOLOv8 deep learning network model.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202311006651.7A CN116797936A (en) | 2023-08-11 | 2023-08-11 | Urban function area detection method based on mobile phone signaling data |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202311006651.7A CN116797936A (en) | 2023-08-11 | 2023-08-11 | Urban function area detection method based on mobile phone signaling data |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN116797936A true CN116797936A (en) | 2023-09-22 |
Family
ID=88048254
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202311006651.7A Pending CN116797936A (en) | 2023-08-11 | 2023-08-11 | Urban function area detection method based on mobile phone signaling data |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN116797936A (en) |
-
2023
- 2023-08-11 CN CN202311006651.7A patent/CN116797936A/en active Pending
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| Hong et al. | Measuring inequality in community resilience to natural disasters using large-scale mobility data | |
| Zheng et al. | Detecting collective anomalies from multiple spatio-temporal datasets across different domains | |
| Zhao et al. | Understanding the bias of call detail records in human mobility research | |
| Zhong et al. | Detecting the dynamics of urban structure through spatial network analysis | |
| US8620624B2 (en) | Event identification in sensor analytics | |
| US20240265038A1 (en) | Method and apparatus for dividing region, storage medium, and electronic device | |
| Huang et al. | Predicting human mobility with activity changes | |
| KR20190139130A (en) | Analysis method of fluidized population information capable of providing real-time fluidized population data by pcell algorithm | |
| CN110191416A (en) | For analyzing the devices, systems, and methods of the movement of target entity | |
| Wismans et al. | Improving a priori demand estimates transport models using mobile phone data: a Rotterdam-region case | |
| Mei et al. | Population spatialization with pixel-level attribute grading by considering scale mismatch issue in regression modeling | |
| Mor et al. | Who is a tourist? Classifying international urban tourists using machine learning | |
| Sinnott et al. | Accident black spot identification and verification through social media | |
| Hu et al. | Prioritizing disaster mapping tasks for online volunteers based on information value theory | |
| CN107291784B (en) | Method and device for acquiring geo-fence categories and business equipment | |
| Siripirote et al. | Updating of travel behavior model parameters and estimation of vehicle trip chain based on plate scanning | |
| Ramsahai et al. | Crime prediction in Trinidad and Tobago using big data analytics: Predictive policing in developing countries | |
| KR101935161B1 (en) | Prediction system and method based on combination of sns and public opinion poll | |
| CN113408867A (en) | City burglary crime risk assessment method based on mobile phone user and POI data | |
| Lu et al. | A two-stage stochastic programming approach for dynamic OD estimation using LBSN data | |
| Mohammadi et al. | Probabilistic prediction of trip travel time and its variability using hierarchical Bayesian learning | |
| CN116797936A (en) | Urban function area detection method based on mobile phone signaling data | |
| Encalada et al. | Mining big data for tourist hot spots: Geographical patterns of online footprints | |
| Liu et al. | Estimation of travel flux between urban blocks by combining spatio-temporal and purpose correlation | |
| CN116233759A (en) | Resident travel track investigation method and system |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| PB01 | Publication | ||
| PB01 | Publication | ||
| SE01 | Entry into force of request for substantive examination | ||
| SE01 | Entry into force of request for substantive examination |