WO2022242018A1 - 一种基于改进的cnn模型的室内目标定位方法 - Google Patents

一种基于改进的cnn模型的室内目标定位方法 Download PDF

Info

Publication number
WO2022242018A1
WO2022242018A1 PCT/CN2021/123586 CN2021123586W WO2022242018A1 WO 2022242018 A1 WO2022242018 A1 WO 2022242018A1 CN 2021123586 W CN2021123586 W CN 2021123586W WO 2022242018 A1 WO2022242018 A1 WO 2022242018A1
Authority
WO
WIPO (PCT)
Prior art keywords
data
indoor
csi
csi data
cnn model
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.)
Ceased
Application number
PCT/CN2021/123586
Other languages
English (en)
French (fr)
Inventor
张登银
徐业鹏
赵远鹏
杨妍
齐城慧
Current Assignee (The listed assignees may be inaccurate. Google has not performed a legal analysis and makes no representation or warranty as to the accuracy of the list.)
Nanjing University of Posts and Telecommunications
Original Assignee
Nanjing University of Posts and Telecommunications
Priority date (The priority date is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the date listed.)
Filing date
Publication date
Application filed by Nanjing University of Posts and Telecommunications filed Critical Nanjing University of Posts and Telecommunications
Priority to US17/542,470 priority Critical patent/US11943735B2/en
Publication of WO2022242018A1 publication Critical patent/WO2022242018A1/zh
Anticipated expiration legal-status Critical
Ceased legal-status Critical Current

Links

Images

Classifications

    • H—ELECTRICITY
    • H04—ELECTRIC COMMUNICATION TECHNIQUE
    • H04W—WIRELESS COMMUNICATION NETWORKS
    • H04W4/00—Services specially adapted for wireless communication networks; Facilities therefor
    • H04W4/02—Services making use of location information
    • H04W4/021—Services related to particular areas, e.g. point of interest [POI] services, venue services or geofences
    • H—ELECTRICITY
    • H04—ELECTRIC COMMUNICATION TECHNIQUE
    • H04W—WIRELESS COMMUNICATION NETWORKS
    • H04W64/00—Locating users or terminals or network equipment for network management purposes, e.g. mobility management
    • G—PHYSICS
    • G01—MEASURING; TESTING
    • G01S—RADIO DIRECTION-FINDING; RADIO NAVIGATION; DETERMINING DISTANCE OR VELOCITY BY USE OF RADIO WAVES; LOCATING OR PRESENCE-DETECTING BY USE OF THE REFLECTION OR RERADIATION OF RADIO WAVES; ANALOGOUS ARRANGEMENTS USING OTHER WAVES
    • G01S5/00—Position-fixing by co-ordinating two or more direction or position line determinations; Position-fixing by co-ordinating two or more distance determinations
    • G01S5/02—Position-fixing by co-ordinating two or more direction or position line determinations; Position-fixing by co-ordinating two or more distance determinations using radio waves
    • G01S5/0252—Radio frequency fingerprinting
    • G01S5/02521—Radio frequency fingerprinting using a radio-map
    • G01S5/02522—The radio-map containing measured values of non-radio values
    • G—PHYSICS
    • G01—MEASURING; TESTING
    • G01S—RADIO DIRECTION-FINDING; RADIO NAVIGATION; DETERMINING DISTANCE OR VELOCITY BY USE OF RADIO WAVES; LOCATING OR PRESENCE-DETECTING BY USE OF THE REFLECTION OR RERADIATION OF RADIO WAVES; ANALOGOUS ARRANGEMENTS USING OTHER WAVES
    • G01S5/00—Position-fixing by co-ordinating two or more direction or position line determinations; Position-fixing by co-ordinating two or more distance determinations
    • G01S5/02—Position-fixing by co-ordinating two or more direction or position line determinations; Position-fixing by co-ordinating two or more distance determinations using radio waves
    • G01S5/0252—Radio frequency fingerprinting
    • G01S5/02521—Radio frequency fingerprinting using a radio-map
    • G01S5/02524—Creating or updating the radio-map
    • G—PHYSICS
    • G06—COMPUTING OR CALCULATING; COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F18/00—Pattern recognition
    • G06F18/20—Analysing
    • G06F18/21—Design or setup of recognition systems or techniques; Extraction of features in feature space; Blind source separation
    • G06F18/213—Feature extraction, e.g. by transforming the feature space; Summarisation; Mappings, e.g. subspace methods
    • G06F18/2135—Feature extraction, e.g. by transforming the feature space; Summarisation; Mappings, e.g. subspace methods based on approximation criteria, e.g. principal component analysis
    • G—PHYSICS
    • G06—COMPUTING OR CALCULATING; COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F18/00—Pattern recognition
    • G06F18/20—Analysing
    • G06F18/23—Clustering techniques
    • G06F18/232—Non-hierarchical techniques
    • G06F18/2321—Non-hierarchical techniques using statistics or function optimisation, e.g. modelling of probability density functions
    • G06F18/23213—Non-hierarchical techniques using statistics or function optimisation, e.g. modelling of probability density functions with fixed number of clusters, e.g. K-means clustering
    • G—PHYSICS
    • G06—COMPUTING OR CALCULATING; COUNTING
    • G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
    • G06N20/00—Machine learning
    • G—PHYSICS
    • G06—COMPUTING OR CALCULATING; COUNTING
    • G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
    • G06N3/00—Computing arrangements based on biological models
    • G06N3/02—Neural networks
    • G06N3/04—Architecture, e.g. interconnection topology
    • G06N3/045—Combinations of networks
    • G—PHYSICS
    • G06—COMPUTING OR CALCULATING; COUNTING
    • G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
    • G06N3/00—Computing arrangements based on biological models
    • G06N3/02—Neural networks
    • G06N3/04—Architecture, e.g. interconnection topology
    • G06N3/0464—Convolutional networks [CNN, ConvNet]
    • G—PHYSICS
    • G06—COMPUTING OR CALCULATING; COUNTING
    • G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
    • G06N3/00—Computing arrangements based on biological models
    • G06N3/02—Neural networks
    • G06N3/08—Learning methods
    • G—PHYSICS
    • G06—COMPUTING OR CALCULATING; COUNTING
    • G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
    • G06N3/00—Computing arrangements based on biological models
    • G06N3/02—Neural networks
    • G06N3/08—Learning methods
    • G06N3/0895—Weakly supervised learning, e.g. semi-supervised or self-supervised learning
    • G—PHYSICS
    • G06—COMPUTING OR CALCULATING; COUNTING
    • G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
    • G06N3/00—Computing arrangements based on biological models
    • G06N3/02—Neural networks
    • G06N3/08—Learning methods
    • G06N3/09—Supervised learning
    • H—ELECTRICITY
    • H04—ELECTRIC COMMUNICATION TECHNIQUE
    • H04W—WIRELESS COMMUNICATION NETWORKS
    • H04W4/00—Services specially adapted for wireless communication networks; Facilities therefor
    • H04W4/02—Services making use of location information
    • H04W4/023—Services making use of location information using mutual or relative location information between multiple location based services [LBS] targets or of distance thresholds
    • H—ELECTRICITY
    • H04—ELECTRIC COMMUNICATION TECHNIQUE
    • H04W—WIRELESS COMMUNICATION NETWORKS
    • H04W4/00—Services specially adapted for wireless communication networks; Facilities therefor
    • H04W4/02—Services making use of location information
    • H04W4/025—Services making use of location information using location based information parameters
    • H—ELECTRICITY
    • H04—ELECTRIC COMMUNICATION TECHNIQUE
    • H04W—WIRELESS COMMUNICATION NETWORKS
    • H04W4/00—Services specially adapted for wireless communication networks; Facilities therefor
    • H04W4/30—Services specially adapted for particular environments, situations or purposes
    • H04W4/33—Services specially adapted for particular environments, situations or purposes for indoor environments, e.g. buildings
    • H—ELECTRICITY
    • H04—ELECTRIC COMMUNICATION TECHNIQUE
    • H04W—WIRELESS COMMUNICATION NETWORKS
    • H04W84/00—Network topologies
    • H04W84/02—Hierarchically pre-organised networks, e.g. paging networks, cellular networks, WLAN [Wireless Local Area Network] or WLL [Wireless Local Loop]
    • H04W84/10—Small scale networks; Flat hierarchical networks
    • H04W84/12—WLAN [Wireless Local Area Networks]
    • Y—GENERAL TAGGING OF NEW TECHNOLOGICAL DEVELOPMENTS; GENERAL TAGGING OF CROSS-SECTIONAL TECHNOLOGIES SPANNING OVER SEVERAL SECTIONS OF THE IPC; TECHNICAL SUBJECTS COVERED BY FORMER USPC CROSS-REFERENCE ART COLLECTIONS [XRACs] AND DIGESTS
    • Y02—TECHNOLOGIES OR APPLICATIONS FOR MITIGATION OR ADAPTATION AGAINST CLIMATE CHANGE
    • Y02D—CLIMATE CHANGE MITIGATION TECHNOLOGIES IN INFORMATION AND COMMUNICATION TECHNOLOGIES [ICT], I.E. INFORMATION AND COMMUNICATION TECHNOLOGIES AIMING AT THE REDUCTION OF THEIR OWN ENERGY USE
    • Y02D30/00—Reducing energy consumption in communication networks
    • Y02D30/70—Reducing energy consumption in communication networks in wireless communication networks

Definitions

  • the invention relates to an indoor target positioning method based on an improved CNN model, and belongs to the technical field of indoor positioning.
  • Wi-Fi signals have inherent advantages when applied to indoor positioning environments.
  • the signal is blocked by obstacles in the indoor environment during propagation, such as: walls, ceilings, closed doors and windows, etc.
  • non-line-of-sight errors will occur (direct light is blocked and only reflected light);
  • the receiver will receive the light components generated by reflection, scattering, and diffraction of direct light together, resulting in multipath phenomenon.
  • Non-line-of-sight errors and multipath phenomena will cause great interference to the accuracy of indoor positioning and affect the indoor positioning effect. Therefore, it is necessary to study effective technologies to eliminate interference, so as to obtain indoor positioning equations with higher accuracy.
  • the present invention proposes an indoor target positioning method based on an improved CNN model, which can solve the problem of noise interference in indoor positioning data. Minimize the impact of multipath effects and improve the accuracy of indoor positioning.
  • the present invention adopts the following technical means:
  • the present invention proposes a kind of indoor target localization method based on improved CNN model, comprises the following steps:
  • the generation method of described location fingerprint storehouse comprises:
  • the preprocessed indoor CSI data is divided into multiple data subsets
  • the location fingerprint library is generated by using the trained improved CNN model and the preprocessed indoor CSI data.
  • the SDR platform includes a mobile node, a plurality of signal base stations and a data processing unit, the mobile node is used to periodically broadcast a detection request frame and automatically send a positioning request; the signal base station is used to receive mobile The positioning request of the node, and obtain the WiFi signal of the mobile node; the data processing unit is used to process the WiFi signal obtained by the signal base station, and obtain the indoor CSI data corresponding to the WiFi signal.
  • data preprocessing methods include:
  • PCA is used to perform dimensionality reduction and denoising processing on the corrected CSI data to obtain the preprocessed CSI data.
  • the method of dividing the preprocessed indoor CSI data into multiple data subsets based on the clustering algorithm includes:
  • each data subset is calculated based on the k-means algorithm, and the calculation formula is as follows:
  • ⁇ i represents the center of the i-th data subset
  • C i represents the i-th data subset
  • x represents indoor CSI data
  • i 1,...,k
  • k is the number of data subsets
  • E represents the square error of all data subsets
  • the distance between the preprocessed indoor CSI data and the center of each data subset is calculated separately, and the data subsets are re-divided according to the distance until the square error is minimized to obtain the final data subset.
  • the improved CNN model includes 5 convolutional layers and 2 fully connected layers connected in sequence, wherein the convolution kernels of the 5 convolutional layers are 16, 32, 64, 64 and 128 respectively; the first The first fully connected layer is used to flatten the output of the fifth convolutional layer, and the second fully connected layer is used to output the classification results of the CSI data.
  • the method of using the data subset to train the improved CNN model includes:
  • x p represents the pth indoor CSI data after normalization
  • var p represents the pth indoor CSI data in the data subset
  • mean() is the average function
  • total() is the classification summary function
  • the loss value of the improved CNN model is calculated by using the loss function
  • the step of performing batch processing on the features of the indoor CSI data after convolution processing includes:
  • U B represents the average value of the features of the indoor CSI data after convolution processing
  • M is the number of features of each indoor CSI data after convolution processing
  • s q represents the indoor CSI data after the qth convolution processing
  • y p represents the coordinate information corresponding to the pth indoor CSI data in the data subset
  • ⁇ and ⁇ are the model parameters of the improved CNN model, respectively, Indicates the characteristics of the pth normalized indoor CSI data in the data subset.
  • L represents the loss value of the improved CNN model
  • y′ p represents the real coordinate label of the p-th indoor CSI data in the data subset
  • y p represents the coordinate information corresponding to the p-th indoor CSI data in the data subset.
  • the positioning fingerprint library includes CSI data and corresponding coordinate information.
  • the present invention proposes an indoor target positioning method based on the improved CNN model, which preprocesses the collected CSI data to remove noise interference, and then divides the indoor CSI data into multiple data subsets through the method of data set division. It is beneficial to improve the performance of position estimation, reduce the amount of calculation in the process of position estimation, and then improve the accuracy of indoor positioning data. Finally, the improved CNN module processes the indoor CSI data in the data subset to extract multipath and non-line-of-sight effects , using the fitting curve of the neural network to better eliminate the effects of multipath and non-line-of-sight effects, obtain accurate indoor positioning coordinate information, and effectively improve the accuracy of indoor positioning.
  • the present invention utilizes the Wi-Fi signal of a large indoor place to perform positioning, does not need to set up a new positioning instrument, and has lower cost and great convenience.
  • PCA for data preprocessing can remove most of the background noise and reduce the dimension of the data, so that CSI signals can also be applied to indoor positioning in multipath and non-line-of-sight environments.
  • the noise can be effectively reduced and the redundancy of each data subset can be dealt with.
  • the final error can be eliminated and the progress of positioning can be improved.
  • the improved CNN model is used to learn the features of the input data, which not only solves the problem of large amount of calculation for multi-dimensional data, but also can accurately extract coordinate information.
  • Fig. 1 is a kind of flow chart of the steps of the indoor target localization method based on the improved CNN model of the present invention
  • Fig. 2 is the structural representation of the improved CNN model in the embodiment of the present invention.
  • Fig. 3 is a flow chart of the steps of fingerprint matching in the embodiment of the present invention.
  • the present invention proposes a method for indoor target positioning based on an improved CNN model.
  • the method of the present invention can be used in an offline stage and an online stage, wherein the offline stage mainly includes model training and fingerprint database construction.
  • the fingerprint library built in the offline stage is used for real-time indoor target positioning.
  • Step 1 Obtain the target CSI data of the target to be located, and perform data preprocessing
  • Step 2 Use the preprocessed target CSI data to perform fingerprint matching with the pre-generated positioning fingerprint database to obtain the coordinate information of the target to be positioned.
  • the generation method of positioning fingerprint library of the present invention comprises:
  • Step A using the SDR platform to collect indoor WiFi signals, obtain indoor CSI data corresponding to the WiFi signals, and perform data preprocessing;
  • Step B dividing the preprocessed indoor CSI data into multiple data subsets based on a clustering algorithm
  • Step C using the data subset to train the improved CNN model to obtain the trained improved CNN model
  • Step D using the trained improved CNN model and the preprocessed indoor CSI data to generate a location fingerprint library.
  • the present invention uses the SDR platform for data collection, the SDR platform can collect WiFi signals, and use the WiFi signals to obtain CSI data, the location information contained in CSI is more than that contained in RFID and RSSI, and it is easier to extract the multipath It is beneficial to improve the accuracy of indoor positioning.
  • the SDR platform includes a mobile node, multiple signal base stations and a data processing unit. Both the mobile node and the signal base station are defined by the SDR equipment.
  • the mobile node can periodically broadcast the detection request frame and automatically send the positioning request;
  • the signal base station uses To receive the positioning request of the mobile node and obtain the WiFi signal of the mobile node, a dedicated baseband receiver is used in the base station after the radio frequency front end to reduce the carrier frequency offset and improve the receiving signal-to-noise ratio;
  • the data processing unit communicates with the signal through the coaxial cable
  • the base station is connected to process the WiFi signal obtained by the signal base station, and obtain indoor CSI data corresponding to the WiFi signal.
  • the detection request frame is selected as the request signal, which belongs to the category of the management frame and strictly abides by the ieee802.11b protocol.
  • step 1 and step A data preprocessing can be performed by PCA, which solves the complexity problem associated with very high-dimensional CSI data in complex environments by extracting a set of linearly uncorrelated variables from a large amount of CSI data and data transmission issues; the specific operations of data preprocessing are as follows:
  • PCA essentially takes the direction with the largest variance as the main feature, and "decorrelates" the data in each orthogonal direction, that is, Making them uncorrelated in different directions minimizes the effect of noise on the experimental data.
  • Var(x) represents the variance of the corrected CSI data
  • m is the number of corrected CSI data
  • Indicates the square of the value of the jth corrected CSI data, j 1,...,m.
  • the CSI data contains three-dimensional values (x, y, z axes in the three-dimensional space coordinate system), and the three-dimensional values are in the three-dimensional space
  • the direction of mapping the present invention utilizes the gradient ascending law to map to the CSI eigenvector after w direction is maximum:
  • the maximum variance of the corrected CSI data is obtained through the formula (10), and the corrected CSI data is processed according to the maximum variance to obtain the CSI data after dimensionality reduction and denoising, that is, the preprocessed CSI data.
  • the present invention provides a method of data set division, which divides the fingerprint database set of each indoor small space into several subsets, and utilizes k-means clustering
  • the algorithm can get the center of each data subset. This enables PCA to effectively reduce noise and deal with the redundancy of each data subset, and extract multipath and non-line-of-sight effects on each data subset, which can improve the learning performance of position estimation and can reach the final error Eliminate progress for boost positioning.
  • step B is as follows:
  • Step B01. Randomly divide the preprocessed indoor CSI data into multiple data subsets according to geographic locations, and each data subset includes multiple indoor CSI data. For large indoor places, the random division of data subsets can be carried out according to the layout of shops, floors, etc. in the indoor place, or according to the preset area size (such as 50 square meters).
  • Step B02 using the data subset as a cluster in the clustering algorithm, and calculating the center of each data subset based on the k-means algorithm, the specific calculation formula is as follows:
  • ⁇ i represents the center of the i-th data subset (that is, the mean component of the cluster, also called the centroid)
  • C i represents the i-th data subset
  • x represents the indoor CSI data
  • Step B03 in order to make the points in the cluster closely connected together and make the distance between the clusters as large as possible, the present invention calculates the square error of all data subsets according to the center of the data subset, with the goal of minimizing the square error to iterate.
  • the formula for calculating the square error is as follows:
  • E represents the squared error of all data subsets.
  • Step B04. Calculate the distance between the preprocessed indoor CSI data and the center of each data subset, re-divide the data subsets according to the distance, and repeat steps B02-B03 until the square error is minimized to obtain the final data subset.
  • the improved CNN model is used in step C of the embodiment of the present invention.
  • the improved CNN model includes 5 convolutional layers and 2 fully connected layers connected in sequence, wherein the volume of 5 convolutional layers
  • the accumulation kernels are 16, 32, 64, 64 and 128 respectively.
  • For the first convolutional layer convolution operations are performed to track and extract information from the results in 16 cores, followed by batch normalization, max pooling, and activation; for the second convolutional layer, 32 cores
  • the height of each kernel is 16, which is equivalent to the number of kernels in the first convolutional layer; for the data processing stage in the third, fourth, and fifth convolutional layers and the second convolutional layer layers are the same.
  • the first fully connected layer is used to flatten the output of the 5th convolutional layer.
  • the second fully connected layer is used to output the classification result of CSI data.
  • the present invention divides multi-warp and non-line-of-sight effects into 5 windows: (1) monitoring window does not have non-line-of-sight and multi-warp effects; (2) monitoring window occurs Non-line-of-sight effect; (3) The non-line-of-sight effect of the monitoring window disappears; (4) The multi-path effect appears in the monitoring window; (5) The multi-path effect of the monitoring window disappears, and the five neural nodes in the second fully connected layer represent 5 windows defined by the present invention are defined.
  • step C Using the CNN model to train the data subset can obtain a regression model for position estimation and a classification model for eliminating noise interference.
  • the specific operation of step C is as follows:
  • Step C01 in order to reduce the influence of the correlation value error of the data and improve the speed of network convergence, the method of the present invention standardizes each indoor CSI data in each data subset:
  • x p represents the pth indoor CSI data after normalization
  • var p represents the pth indoor CSI data in the data subset
  • mean() is the average function
  • total() is the classification summary function
  • p 1, ..., n
  • n is the number of indoor CSI data in the data subset.
  • Step C02 initialize the model parameters of the improved CNN model, input the standardized indoor CSI data into the convolution layer of the improved CNN model in units of data subsets, obtain the indoor CSI data after convolution processing, and extract the convolution processing After the characteristics of the indoor CSI data.
  • the indoor CSI data in each data subset is input into the convolutional layer in turn, and the features of the tracking results (indoor CSI data after convolution processing) are extracted through a series of kernel functions.
  • a kernel generates a tracking result feature map, and the elements in the kernel are determined by the network training stage.
  • the kernel is a three-dimensional matrix whose rows and columns are artificially set hyperparameters.
  • Step C03 perform batching, max pooling and activation processing on the features of the indoor CSI data after convolution processing, and obtain a mapping of the indoor CSI data after convolution processing.
  • Batching speeds up network training, allows for higher learning rates, makes initial weighting easier, and simplifies the creation of deep networks.
  • the batch operation in the embodiment of the present invention is as follows:
  • U B represents the average value of the features of the indoor CSI data after convolution processing
  • M is the number of features of each indoor CSI data after convolution processing
  • Pooling is an important concept in convolutional layers, which is a form of non-linear downsampling.
  • the present invention divides the tracking result into a group of non-overlapping rectangles through maximum pooling processing, and outputs the maximum value in each sub-region. On the premise of maintaining the main characteristics, the amount of calculation is reduced by reducing the number of parameters.
  • the method of the present invention utilizes Leaky ReLU as an activation function for activation processing, and the activation function can make the CNN neural network exhibit nonlinear characteristics.
  • Step C04 using the fully connected layer of the improved CNN model to classify the mapping of the indoor CSI data after convolution processing, and obtain the coordinate information corresponding to each indoor CSI data in each data subset.
  • the fully connected layer is a column vector including multiple neurons, and the calculation formula of the output coordinate information is as follows:
  • y p represents the coordinate information corresponding to the pth indoor CSI data in the data subset
  • ⁇ and ⁇ are the model parameters of the improved CNN model, respectively, Indicates the characteristics of the pth normalized indoor CSI data in the data subset.
  • Step C05 according to the preset real coordinate label (that is, the real three-dimensional coordinate information corresponding to the CSI data) and the coordinate information (y p ) output by the improved CNN model, use the loss function to calculate the loss value of the improved CNN model, and the calculation formula is as follows :
  • L represents the loss value of the improved CNN model
  • y′ p represents the real coordinate label of the p-th indoor CSI data in the data subset
  • y p represents the coordinate information corresponding to the p-th indoor CSI data in the data subset.
  • Step C06 update the model parameters of the improved CNN model according to the loss value, and use the updated and improved CNN model to process the standardized indoor CSI data, repeat steps C02-C06, observe the loss convergence through the loss function graph, when the loss value Converge to obtain a trained and improved CNN model.
  • step D of the present invention a large amount of indoor CSI data is collected, and the collected data is processed by using the trained and improved CNN model to obtain coordinate information corresponding to the data, and a positioning fingerprint database is generated by using the CSI data and the corresponding coordinate information.
  • the positioning fingerprint database can be used to directly perform indoor positioning in the online stage, and the target CSI data of the target to be positioned is collected through the SDR platform, and in step 2, the target CSI data is compared with the data in the positioning fingerprint database Yes, if the target CSI data is the same as the CSI data in the positioning fingerprint database, the coordinate information can be directly extracted from the fingerprint database.
  • the present invention can also perform fingerprint matching according to the operation shown in Figure 3, and perform multi-dimensional data operations according to the characteristics of the root node and the split point and matching, select the coordinate information with the highest matching degree.
  • the present invention processes the collected CSI data through PCA and the method of dividing data subsets to achieve the effect of removing noise interference and initially improving the positioning accuracy, and then puts the CSI data in the data subset into the improved CNN model for Training can achieve the effect of suppressing multipath components, and further improve the accuracy of indoor positioning.
  • the method of the invention can achieve a good effect of removing noise and multipath, improves the accuracy of positioning, and realizes its application value in large indoor places.

Landscapes

  • Engineering & Computer Science (AREA)
  • Physics & Mathematics (AREA)
  • Theoretical Computer Science (AREA)
  • Data Mining & Analysis (AREA)
  • General Physics & Mathematics (AREA)
  • Artificial Intelligence (AREA)
  • Evolutionary Computation (AREA)
  • General Engineering & Computer Science (AREA)
  • Life Sciences & Earth Sciences (AREA)
  • Software Systems (AREA)
  • Mathematical Physics (AREA)
  • Computing Systems (AREA)
  • Health & Medical Sciences (AREA)
  • Biomedical Technology (AREA)
  • Biophysics (AREA)
  • Computational Linguistics (AREA)
  • General Health & Medical Sciences (AREA)
  • Molecular Biology (AREA)
  • Computer Networks & Wireless Communication (AREA)
  • Signal Processing (AREA)
  • Computer Vision & Pattern Recognition (AREA)
  • Bioinformatics & Cheminformatics (AREA)
  • Evolutionary Biology (AREA)
  • Bioinformatics & Computational Biology (AREA)
  • Probability & Statistics with Applications (AREA)
  • Radar, Positioning & Navigation (AREA)
  • Remote Sensing (AREA)
  • Medical Informatics (AREA)
  • Position Fixing By Use Of Radio Waves (AREA)
  • Image Analysis (AREA)

Abstract

本发明公开了一种基于改进的CNN模型的室内目标定位方法，其包括：利用待定位目标的目标CSI数据与预先生成的定位指纹库进行指纹匹配，获得待定位目标的坐标信息；定位指纹库的生成方法为：利用SDR平台采集室内CSI数据，基于聚类算法将室内CSI数据划分为多个数据子集，并利用数据子集训练改进的CNN模型，获得训练好的改进的CNN模型，最后利用训练好的改进的CNN模型生成定位指纹库。本发明能够解决室内定位数据的噪声干扰问题，最大限度的减少多径效应的影响，提高室内定位的精度。

Description

一种基于改进的CNN模型的室内目标定位方法 技术领域
本发明涉及一种基于改进的CNN模型的室内目标定位方法，属于室内定位技术领域。
背景技术
随着城市化的建设，大型室内场所由于其面积大、空间布局复杂等原因，人们对于室内场所的定位需求日益迫切，结合物联网技术的飞速发展和硬件技术的成熟，室内定位技术到了人们的广泛关注。对于已知的定位技术例如全球卫星导航系统，其在室外定位中可以发挥非常好的作用，能够准确的定位物体，但是对于室内的物体，由于钢筋混凝土结构对于信号的干扰作用，会导致其定位效果达不到需求，此外，由于部分室内场所面积较大，空间布局结构复杂，通道走廊交叉分布，GPS定位只能在室外接收卫星信号的位置，所以GPS定位等技术难以较快实现室内定位。
随着大型室内环境中Wi-Fi信号的大量运用，使得Wi-Fi信号应用于室内定位的环境下有着先天的优势。但是如果信号在传播过程中被室内环境中的障碍所阻挡，例如：墙壁、天花板、关闭的门窗等，就会产生非视距的误差(直射光被阻挡只有反射光)；此外，信号在传播过程中由于光滑的物体表面对于直射光会产生反射、散射、衍射等现象，接收器对于直射光经由反射、散射、衍射产生的光的分量一起接受就会产生多径的现象。非视距误差和多径现象都会对室内定位的准确度产生很大的干扰，影响室内定位效果，因此需要研究有效的消除干扰的技术，从而获得精度更高的室 内定位方程。
发明内容
为了解决现有技术中室内定位易受多径效应和非视距误差影响的问题，本发明提出了一种基于改进的CNN模型的室内目标定位方法，能够解决室内定位数据的噪声干扰问题，最大限度的减少多径效应的影响，提高室内定位的精度。
为解决上述技术问题，本发明采用了如下技术手段：
本发明提出了一种基于改进的CNN模型的室内目标定位方法，包括如下步骤：
获取待定位目标的目标CSI数据，并进行数据预处理；
利用预处理后的目标CSI数据与预先生成的定位指纹库进行指纹匹配，获得待定位目标的坐标信息；
所述定位指纹库的生成方法包括：
利用SDR平台采集室内WiFi信号，获得WiFi信号对应的室内CSI数据，并进行数据预处理；
基于聚类算法将预处理后的室内CSI数据划分为多个数据子集；
利用数据子集训练改进的CNN模型，获得训练好的改进的CNN模型；
利用训练好的改进的CNN模型和预处理后的室内CSI数据生成定位指纹库。
进一步的，所述SDR平台包括一个移动节点、多个信号基站和一个数据处理单元，所述移动节点用于通过周期性地广播探测请求帧并自动发送 定位请求；所述信号基站用于接收移动节点的定位请求，并获取移动节点的WiFi信号；所述数据处理单元用于处理信号基站获取的WiFi信号，获得WiFi信号对应的室内CSI数据。
进一步的，数据预处理的方法包括：
对CSI数据进行幅值滤波、相位修正和去均值处理，获得修正后的CSI数据；
利用PCA对修正后的CSI数据进行降维处理和去噪处理，获得预处理后的CSI数据。
进一步的，基于聚类算法将预处理后的室内CSI数据划分为多个数据子集的方法包括：
根据地理位置将预处理后的室内CSI数据随机划分为多个数据子集，每个数据子集中包括多个室内CSI数据；
基于k-means算法计算每个数据子集的中心，计算公式如下：
Figure PCTCN2021123586-appb-000001
其中，μ i表示第i个数据子集的中心，C i表示第i个数据子集，x表示室内CSI数据，i＝1,…,k，k为数据子集的数量；
根据数据子集的中心计算所有数据子集的平方误差，计算公式如下：
Figure PCTCN2021123586-appb-000002
其中，E表示所有数据子集的平方误差；
分别计算预处理后的室内CSI数据与每个数据子集的中心的距离，根据距离重新划分数据子集，直到平方误差最小化，获得最终的数据子集。
进一步的，所述改进的CNN模型包括依次连接的5个卷积层和2个全连接层，其中，5个卷积层的卷积核分别为16、32、64、64和128；第一个全连接层用于展平第5个卷积层的输出，第二个全连接层用于输出CSI数据的分类结果。
进一步的，利用数据子集训练改进的CNN模型的方法包括：
对每个数据子集中的每个室内CSI数据进行规范处理：
Figure PCTCN2021123586-appb-000003
其中，x p表示规范后的第p个室内CSI数据，var p表示数据子集中的第p个室内CSI数据，mean()为求平均值函数，total()为分类汇总函数，p＝1,…,n，n为数据子集中室内CSI数据的个数；
初始化改进的CNN模型的模型参数，以数据子集为单位将规范后的室内CSI数据输入改进的CNN模型的卷积层，获得卷积处理后的室内CSI数据，并提取卷积处理后的室内CSI数据的特征；
对卷积处理后的室内CSI数据的特征进行批量化、最大池化和激活处理，获得卷积处理后的室内CSI数据的映射；
利用改进的CNN模型的全连接层对卷积处理后的室内CSI数据的映射进行分类，获得每个数据子集中每个室内CSI数据对应的坐标信息；
根据预设的真实坐标标签和改进的CNN模型输出的坐标信息，利用损失函数计算改进的CNN模型的损失值；
根据损失值更新改进的CNN模型的模型参数，并利用更新后的改进的CNN模型处理规范后的室内CSI数据，直到损失值收敛，获得训练好的改 进的CNN模型。
进一步的，针对每一个室内CSI数据，对卷积处理后的室内CSI数据的特征进行批量化处理的步骤包括：
根据卷积处理后的室内CSI数据的特征进行均值计算：
Figure PCTCN2021123586-appb-000004
其中，U B表示卷积处理后的室内CSI数据的特征的平均值，M为卷积处理后的每个室内CSI数据的特征的个数，s q表示第q个卷积处理后的室内CSI数据的特征，q＝1,…,M；
根据U B计算特征方差，计算公式如下：
Figure PCTCN2021123586-appb-000005
其中，
Figure PCTCN2021123586-appb-000006
表示卷积处理后的室内CSI数据的特征的方差；
根据U B和
Figure PCTCN2021123586-appb-000007
对卷积处理后的室内CSI数据的特征进行标准化处理：
Figure PCTCN2021123586-appb-000008
其中，
Figure PCTCN2021123586-appb-000009
表示标准化处理后的室内CSI数据的特征，ε为预设的标准化数值。
进一步的，每个数据子集中每个室内CSI数据对应的坐标信息的计算公式如下：
Figure PCTCN2021123586-appb-000010
其中，y p表示数据子集中第p个室内CSI数据对应的坐标信息，γ和β分别为改进的CNN模型的模型参数，
Figure PCTCN2021123586-appb-000011
表示数据子集中第p个标准化处理后的室内CSI数据的特征。
进一步的，所述损失函数的表达式如下：
L＝—∑ py′ plog(y p)    (8)
其中，L表示改进的CNN模型的损失值，y′ p表示数据子集中第p个室内CSI数据的真实坐标标签，y p表示数据子集中第p个室内CSI数据对应的坐标信息。
进一步的，所述定位指纹库包括CSI数据及其对应的坐标信息。
采用以上技术手段后可以获得以下优势：
本发明提出了一种基于改进的CNN模型的室内目标定位方法，对采集到的CSI数据进行预处理去除噪声干扰，再通过数据集划分的方法将室内CSI数据划分为多个数据子集，有利于提高位置估计的性能、减少位置估计过程中的计算量，进而提高室内定位数据的准确性，最后通过改进的CNN模块处理数据子集中的室内CSI数据，进行多径和非视距效应的提取，利用神经网络的拟合曲线达到更好消除多径影响和非视距效应的效果，获得准确的室内定位坐标信息，有效的提高了室内定位的精度。
本发明利用大型室内场所具有的Wi-Fi信号进行定位，不需要架设新的定位仪器，成本更低且具有极大的便利性。利用PCA进行数据预处理，能够去除大部分的背景噪声和降低数据的维度，使得在多径和非视距的环境下CSI信号也能应用于室内定位。通过划分数据子集可以有效地减少噪声和处理每个数据子集的冗余，在每一个数据子集上，可以达到最终误差的消除，提升定位的进度。利用改进的CNN模型对输入数据进行特征学习，既解决了多维数据计算量大的问题，又可以准确提取坐标信息。
附图说明
图1为本发明一种基于改进的CNN模型的室内目标定位方法的步骤流程图；
图2为本发明实施例中改进的CNN模型的结构示意图；
图3为本发明实施例中指纹匹配的步骤流程图。
具体实施方式
下面结合附图对本发明的技术方案作进一步说明：
本发明提出了一种基于改进的CNN模型的室内目标定位方法，如图1所示，本发明方法可以离线阶段和在线阶段，其中离线阶段主要包括模型训练和指纹库构建，在现阶段主要利用离线阶段构建的指纹库进行实时的室内目标定位。
本发明方法的步骤具体如下：
步骤1、获取待定位目标的目标CSI数据，并进行数据预处理；
步骤2、利用预处理后的目标CSI数据与预先生成的定位指纹库进行指纹匹配，获得待定位目标的坐标信息。
本发明定位指纹库的生成方法包括：
步骤A、利用SDR平台采集室内WiFi信号，获得WiFi信号对应的室内CSI数据，并进行数据预处理；
步骤B、基于聚类算法将预处理后的室内CSI数据划分为多个数据子集；
步骤C、利用数据子集训练改进的CNN模型，获得训练好的改进的 CNN模型；
步骤D、利用训练好的改进的CNN模型和预处理后的室内CSI数据生成定位指纹库。
本发明利用SDR平台进行数据采集，SDR平台可以采集WiFi信号，并利用WiFi信号获得CSI数据，CSI中所包含的位置信息要比RFID和RSSI所包含的信息要多，更容易提取其中的多径分量，有利于提升室内定位的精度。SDR平台包括一个移动节点、多个信号基站和一个数据处理单元，移动节点和信号基站都由SDR设备定义来实现，移动节点可以通过周期性地广播探测请求帧并自动发送定位请求；信号基站用于接收移动节点的定位请求，并获取移动节点的WiFi信号，基站中在射频前端之后采用专用的基带接收机来减小载波频偏，提高接收信噪比；数据处理单元通过同轴电缆与信号基站相连，用于处理信号基站获取的WiFi信号，获得WiFi信号对应的室内CSI数据。对于移动节点发送定位请求的场景，选择探测请求帧作为请求信号，探测请求帧属于管理帧的范畴，严格遵守ieee802.11b协议。
在步骤1和步骤A中，可以通过PCA进行数据预处理，PCA通过从大量的CSI数据中提取一组线性不相关的变量，从而解决在复杂环境中与非常高维CSI数据相关的复杂性问题和数据传输问题；数据预处理的具体操作如下：
(1)、对CSI数据进行幅值滤波、相位修正和去均值处理，去均值后CSI数据每个特征维度上的均值都是0，获得修正后的CSI数据。
(2)、利用PCA对修正后的CSI数据进行降维处理和去噪处理，PCA本质上是将方差最大的方向作为主要特征，并且在各个正交方向上将数据“离相关”，也就是让它们在不同的方向上没有相关性，这就使得噪音对实验数据的影响最小化。
利用PCA计算修正后的CSI数据的方差，计算公式如下：
Figure PCTCN2021123586-appb-000012
其中，Var(x)表示修正后的CSI数据的方差，m为修正后的CSI数据的个数，
Figure PCTCN2021123586-appb-000013
表示第j个修正后的CSI数据的值的平方，j＝1,…,m。
(3)、利用修正后的CSI数据的方差计算CSI特征向量；CSI数据中包含三个维度的数值(三维空间坐标系中的x、y、z轴)，三个维度的数值在三维空间中进行映射的方向组成了CSI数据的映射方向w，w＝(w 1,w 2,w 3)，w 1、w 2、w 3分别表示三个维度的数值在x、y、z轴中进行映射的方向，本发明利用梯度上升法令映射到w方向后的CSI特征向量最大：
Figure PCTCN2021123586-appb-000014
其中，
Figure PCTCN2021123586-appb-000015
表示梯度的变化量，X表示CSI特征向量。
(4)、通过公式(10)得到修正后的CSI数据的方差最大值，根据方差最大值处理修正后的CSI数据，得到降维去噪后的CSI数据，即预处理后的CSI数据。
为了能够提高位置估计学习的性能，本发明给出了一种数据集划分的方法，这种数据划分的方法将每个室内小空间的指纹数据库集划分成若干子集，利用k-means聚类算法可以得到每一个数据子集的中心。这使得PCA 能够有效地减少噪声和处理每一个数据子集的冗余，在每一个数据子集上进行多径和非视距效应的提取，能够提高位置估计的学习性能，可以达到最终误差的消除提升定位的进度。
在本发明实施例中，步骤B的具体操作如下：
步骤B01、根据地理位置将预处理后的室内CSI数据随机划分为多个数据子集，每个数据子集中包括多个室内CSI数据。针对大型室内场所，可以室内场所中店铺、楼层等的布局进行数据子集的随机划分，也可以根据预设的面积大小(比如50平方米)进行数据子集的随机划分。
步骤B02、以数据子集作为聚类算法中的簇，基于k-means算法计算每个数据子集的中心，具体计算公式如下：
Figure PCTCN2021123586-appb-000016
其中，μ i表示第i个数据子集的中心(即簇的均值分量，也叫质心)，C i表示第i个数据子集，x表示室内CSI数据，i＝1,…,k，k为数据子集的数量。
步骤B03、为了让簇内的点能够紧密的连接在一起，让簇之间的距离尽量的大，本发明根据数据子集的中心计算所有数据子集的平方误差，以最小化平方误差为目标进行迭代。平方误差的计算公式如下：
Figure PCTCN2021123586-appb-000017
其中，E表示所有数据子集的平方误差。
步骤B04、分别计算预处理后的室内CSI数据与每个数据子集的中心的距离，根据距离重新划分数据子集，重复步骤B02～B03，直到平方误差 最小化，获得最终的数据子集。
本发明实施例的步骤C中运用到了改进的CNN模型，如图2所示，改进的CNN模型包括依次连接的5个卷积层和2个全连接层，其中，5个卷积层的卷积核分别为16、32、64、64和128。对于第一个卷积层，执行卷积运算从16个核中的跟踪结果和提取结果中的信息，然后进行批处理规范化、最大池化和激活；对于第二个卷积层，32个内核用于跟踪结果和提取信息，每个内核的高度为16，相当于第一个卷积层的核数；对于第三、四、五个卷积层中的数据处理阶段与第二个卷积层相同。第一个全连接层用于展平第5个卷积层的输出。第二个全连接层用于输出CSI数据的分类结果，本发明将多经和非视距效应分成5个窗口：(1)监视窗口没有非视距和多经效应；(2)监视窗口出现非视距效应；(3)监视窗口的非视距效应消失；(4)监视窗口出现多经效应；(5)监视窗口的多径效应消失，第二全连接层的5个神经节点分别代表了本发明定义的5个窗口。
利用CNN模型对数据子集训练可以得到用于位置估计的回归模型和用于消除噪声干扰等的分类模型，步骤C的具体操作如下：
步骤C01、为了降低数据的相关值误差的影响，提高网络收敛的速度，本发明方法对每个数据子集中的每个室内CSI数据进行规范处理：
Figure PCTCN2021123586-appb-000018
其中，x p表示规范后的第p个室内CSI数据，var p表示数据子集中的第p个室内CSI数据，mean()为求平均值函数，total()为分类汇总函数，p＝1,…,n，n为数据子集中室内CSI数据的个数。
步骤C02、初始化改进的CNN模型的模型参数，以数据子集为单位将规范后的室内CSI数据输入改进的CNN模型的卷积层，获得卷积处理后的室内CSI数据，并提取卷积处理后的室内CSI数据的特征。
依次将每个数据子集中的室内CSI数据输入卷积层，通过一系列核函数提取跟踪结果(卷积处理后的室内CSI数据)的特征。一个核生成一个跟踪结果特征映射，核中的元素由网络训练阶段确定。核是一个三维矩阵，它的行和列是人为设置的超参数。
步骤C03、对卷积处理后的室内CSI数据的特征进行批量化、最大池化和激活处理，获得卷积处理后的室内CSI数据的映射。
批量化处理可以加速网络的训练、允许更高的学习率，使得初始加权更加容易、简化深层网络的创建。本发明实施例中批量化的操作如下：
(1)根据卷积处理后的室内CSI数据的特征进行均值计算：
Figure PCTCN2021123586-appb-000019
其中，U B表示卷积处理后的室内CSI数据的特征的平均值，M为卷积处理后的每个室内CSI数据的特征的个数，s q表示第q个卷积处理后的室内CSI数据的特征，q＝1,…,M。
(2)根据U B计算特征方差，计算公式如下：
Figure PCTCN2021123586-appb-000020
其中，
Figure PCTCN2021123586-appb-000021
表示卷积处理后的室内CSI数据的特征的方差。
(3)根据U B和
Figure PCTCN2021123586-appb-000022
对卷积处理后的室内CSI数据的特征进行标准化处理：
Figure PCTCN2021123586-appb-000023
其中，
Figure PCTCN2021123586-appb-000024
表示标准化处理后的室内CSI数据的特征，ε为预设的标准化数值。
池化是卷积层的一个重要概念，它是一种非线性下采样形式。本发明通过最大池化处理将追踪的结果划分为一组不重叠的矩形，在每一个子区域输出最大值。在保持主要特征的前提下，通过减少参数个数来降低计算量。
本发明方法利用Leaky ReLU作为激活函数进行激活处理，激活函数能够使CNN神经网络呈现出非线性的特征。
步骤C04、利用改进的CNN模型的全连接层对卷积处理后的室内CSI数据的映射进行分类，获得每个数据子集中每个室内CSI数据对应的坐标信息。
全连接层是一个包括多个神经元的列向量，其输出的坐标信息的计算公式如下：
Figure PCTCN2021123586-appb-000025
其中，y p表示数据子集中第p个室内CSI数据对应的坐标信息，γ和β分别为改进的CNN模型的模型参数，
Figure PCTCN2021123586-appb-000026
表示数据子集中第p个标准化处理后的室内CSI数据的特征。
步骤C05、根据预设的真实坐标标签(即CSI数据对应的真实三维坐标信息)和改进的CNN模型输出的坐标信息(y p)，利用损失函数计算改进的CNN模型的损失值，计算公式如下：
L＝—∑ py′ plog(y p)    (18)
其中，L表示改进的CNN模型的损失值，y′ p表示数据子集中第p个室内CSI数据的真实坐标标签，y p表示数据子集中第p个室内CSI数据对应的坐标信息。
步骤C06、根据损失值更新改进的CNN模型的模型参数，并利用更新后的改进的CNN模型处理规范后的室内CSI数据，重复步骤C02～C06，通过损失函数图观察损失收敛情况，当损失值收敛，获得训练好的改进的CNN模型。
在本发明步骤D中，采集大量的室内CSI数据，利用训练好的改进的CNN模型处理采集到的数据，可以获得数据对应的坐标信息，利用CSI数据及其对应的坐标信息生成定位指纹库。
本发明获取定位指纹库后，在线阶段可以利用定位指纹库直接进行室内定位，通过SDR平台采集待定位目标的目标CSI数据，在步骤2中，将目标CSI数据与定位指纹库中的数据进行比对，如果目标CSI数据与定位指纹库中的CSI数据一样，可以直接从指纹库中提取坐标信息。此外，考虑到目标CSI数据不一定与定位指纹库中的CSI数据完全相同，本发明还可以按照如图3所示的操作进行指纹匹配，根据根节点的特征和分裂点进行多维度的数据运算和匹配，选取匹配度最高的坐标信息。
本发明通过PCA和划分数据子集的方法对采集到的CSI数据进行处理，达到去除噪声干扰和初步的提高定位的精确度的效果，随后将数据子集中的CSI数据放入改进的CNN模型进行训练，达到抑制多径分量的效果，近一步的提高了室内定位的准确度。本发明方法可以达到一个很好的去除噪 声和多径的效果，提高了定位的准确性，实现了其在大型室内场所应用的价值。
以上所述仅是本发明的优选实施方式，应当指出，对于本技术领域的普通技术人员来说，在不脱离本发明技术原理的前提下，还可以做出若干改进和变形，这些改进和变形也应视为本发明的保护范围。

Claims (10)

  1. 一种基于改进的卷积神经网络(CNN)模型的室内目标定位方法，其特征在于，包括如下步骤：
    获取待定位目标的目标相机串行接口(CSI)数据，并进行数据预处理；
    利用预处理后的目标CSI数据与预先生成的定位指纹库进行指纹匹配，获得待定位目标的坐标信息；
    所述定位指纹库的生成方法包括：
    利用软件无线电(SDR)平台采集室内WiFi信号，获得WiFi信号对应的室内CSI数据，并进行数据预处理；
    基于聚类算法将预处理后的室内CSI数据划分为多个数据子集；
    利用数据子集训练改进的CNN模型，获得训练好的改进的CNN模型；
    利用训练好的改进的CNN模型和预处理后的室内CSI数据生成定位指纹库。
  2. 根据权利要求1所述的一种基于改进的CNN模型的室内目标定位方法，其特征在于，所述SDR平台包括一个移动节点、多个信号基站和一个数据处理单元，所述移动节点用于通过周期性地广播探测请求帧并自动发送定位请求；所述信号基站用于接收移动节点的定位请求，并获取移动节点的WiFi信号；所述数据处理单元用于处理信号基站获取的WiFi信号，获得WiFi信号对应的室内CSI数据。
  3. 根据权利要求1所述的一种基于改进的CNN模型的室内目标定位方法，其特征在于，数据预处理的方法包括：
    对CSI数据进行幅值滤波、相位修正和去均值处理，获得修正后的CSI 数据；
    利用主成分分析(PCA)对修正后的CSI数据进行降维处理和去噪处理，获得预处理后的CSI数据。
  4. 根据权利要求1所述的一种基于改进的CNN模型的室内目标定位方法，其特征在于，基于聚类算法将预处理后的室内CSI数据划分为多个数据子集的方法包括：
    根据地理位置将预处理后的室内CSI数据随机划分为多个数据子集，每个数据子集中包括多个室内CSI数据；
    基于k-means算法计算每个数据子集的中心，计算公式如下：
    Figure PCTCN2021123586-appb-100001
    其中，μ i表示第i个数据子集的中心，C i表示第i个数据子集，x表示室内CSI数据，i＝1,…,k，k为数据子集的数量；
    根据数据子集的中心计算所有数据子集的平方误差，计算公式如下：
    Figure PCTCN2021123586-appb-100002
    其中，E表示所有数据子集的平方误差；
    分别计算预处理后的室内CSI数据与每个数据子集的中心的距离，根据距离重新划分数据子集，直到平方误差最小化，获得最终的数据子集。
  5. 根据权利要求1所述的一种基于改进的CNN模型的室内目标定位方法，其特征在于，所述改进的CNN模型包括依次连接的5个卷积层和2个全连接层，其中，5个卷积层的卷积核分别为16、32、64、64和128；第 一个全连接层用于展平第5个卷积层的输出，第二个全连接层用于输出CSI数据的分类结果。
  6. 根据权利要求1或5所述的一种基于改进的CNN模型的室内目标定位方法，其特征在于，利用数据子集训练改进的CNN模型，方法包括：
    对每个数据子集中的每个室内CSI数据进行规范处理：
    Figure PCTCN2021123586-appb-100003
    其中，x p表示规范后的第p个室内CSI数据，var p表示数据子集中的第p个室内CSI数据，mean( )为求平均值函数，total( )为分类汇总函数，p＝1,…,n，n为数据子集中室内CSI数据的个数；
    初始化改进的CNN模型的模型参数，以数据子集为单位将规范后的室内CSI数据输入改进的CNN模型的卷积层，获得卷积处理后的室内CSI数据，并提取卷积处理后的室内CSI数据的特征；
    对卷积处理后的室内CSI数据的特征进行批量化、最大池化和激活处理，获得卷积处理后的室内CSI数据的映射；
    利用改进的CNN模型的全连接层对卷积处理后的室内CSI数据的映射进行分类，获得每个数据子集中每个室内CSI数据对应的坐标信息；
    根据预设的真实坐标标签和改进的CNN模型输出的坐标信息，利用损失函数计算改进的CNN模型的损失值；
    根据损失值更新改进的CNN模型的模型参数，并利用更新后的改进的CNN模型处理规范后的室内CSI数据，直到损失值收敛，获得训练好的改进的CNN模型。
  7. 根据权利要求6所述的一种基于改进的CNN模型的室内目标定位方法，其特征在于，针对每一个室内CSI数据，对卷积处理后的室内CSI数据的特征进行批量化处理的步骤包括：
    根据卷积处理后的室内CSI数据的特征进行均值计算：
    Figure PCTCN2021123586-appb-100004
    其中，U B表示卷积处理后的室内CSI数据的特征的平均值，M为卷积处理后的每个室内CSI数据的特征的个数，s q表示第q个卷积处理后的室内CSI数据的特征，q＝1,…,M；
    根据U B计算特征方差，计算公式如下：
    Figure PCTCN2021123586-appb-100005
    其中，
    Figure PCTCN2021123586-appb-100006
    表示卷积处理后的室内CSI数据的特征的方差；
    根据U B和
    Figure PCTCN2021123586-appb-100007
    对卷积处理后的室内CSI数据的特征进行标准化处理：
    Figure PCTCN2021123586-appb-100008
    其中，
    Figure PCTCN2021123586-appb-100009
    表示标准化处理后的室内CSI数据的特征，ε为预设的标准化数值。
  8. 根据权利要求6所述的一种基于改进的CNN模型的室内目标定位方法，其特征在于，每个数据子集中每个室内CSI数据对应的坐标信息的计算公式如下：
    Figure PCTCN2021123586-appb-100010
    其中，y p表示数据子集中第p个室内CSI数据对应的坐标信息，γ和β分别为改进的CNN模型的模型参数，
    Figure PCTCN2021123586-appb-100011
    表示数据子集中第p个标准化处理后的室内CSI数据的特征。
  9. 根据权利要求6所述的一种基于改进的CNN模型的室内目标定位方法，其特征在于，所述损失函数的表达式如下：
    Figure PCTCN2021123586-appb-100012
    其中，L表示改进的CNN模型的损失值，y′ p表示数据子集中第p个室内CSI数据的真实坐标标签，y p表示数据子集中第p个室内CSI数据对应的坐标信息。
  10. 根据权利要求1所述的一种基于改进的CNN模型的室内目标定位方法，其特征在于，所述定位指纹库包括CSI数据及其对应的坐标信息。
PCT/CN2021/123586 2021-05-21 2021-10-13 一种基于改进的cnn模型的室内目标定位方法 Ceased WO2022242018A1 (zh)

Priority Applications (1)

Application Number Priority Date Filing Date Title
US17/542,470 US11943735B2 (en) 2021-05-21 2021-12-05 Indoor target positioning method based on improved convolutional neural network model

Applications Claiming Priority (2)

Application Number Priority Date Filing Date Title
CN202110556627.5A CN113271539B (zh) 2021-05-21 2021-05-21 一种基于改进的cnn模型的室内目标定位方法
CN202110556627.5 2021-05-21

Related Child Applications (1)

Application Number Title Priority Date Filing Date
US17/542,470 Continuation-In-Part US11943735B2 (en) 2021-05-21 2021-12-05 Indoor target positioning method based on improved convolutional neural network model

Publications (1)

Publication Number Publication Date
WO2022242018A1 true WO2022242018A1 (zh) 2022-11-24

Family

ID=77232520

Family Applications (1)

Application Number Title Priority Date Filing Date
PCT/CN2021/123586 Ceased WO2022242018A1 (zh) 2021-05-21 2021-10-13 一种基于改进的cnn模型的室内目标定位方法

Country Status (3)

Country Link
US (1) US11943735B2 (zh)
CN (1) CN113271539B (zh)
WO (1) WO2022242018A1 (zh)

Cited By (3)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN116047410A (zh) * 2023-01-16 2023-05-02 华东师范大学 一种基于图卷积神经网络的uwb室内行人定位方法
CN116234005A (zh) * 2022-12-20 2023-06-06 同济大学 一种多建筑物室内定位方法
WO2024197655A1 (zh) * 2023-03-29 2024-10-03 中国科学院深圳先进技术研究院 基于去噪自编码器和卷积神经网络的WiFi指纹室内定位方法

Families Citing this family (6)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN113271539B (zh) * 2021-05-21 2024-06-25 南京邮电大学 一种基于改进的cnn模型的室内目标定位方法
CN116567541B (zh) * 2023-04-17 2025-07-01 天津大学 基于无线Wi-Fi的智能货架交互行为识别方法
CN116347598B (zh) * 2023-05-30 2023-08-15 华南师范大学 一种基于Wi-Fi的室内定位方法和装置
CN118803638A (zh) * 2023-07-14 2024-10-18 中国移动通信有限公司研究院 信息处理方法、定位指纹增强模型训练方法及通信设备
CN118283524A (zh) * 2024-03-28 2024-07-02 河南理工大学 一种基于双向时间卷积网络的uwb煤矿井下定位方法
CN119494947A (zh) * 2025-01-20 2025-02-21 成都理工大学 一种面向智能工厂的自适应定位方法及系统

Citations (4)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN112153736A (zh) * 2020-09-14 2020-12-29 南京邮电大学 一种基于信道状态信息的人员动作识别和位置估计方法
US20210092611A1 (en) * 2019-09-19 2021-03-25 Colorado State University Research Foundation Security-enhanced Deep Learning Fingerprint-Based Indoor Localization
CN112712557A (zh) * 2020-12-17 2021-04-27 上海交通大学 一种基于卷积神经网络的超分辨率cir室内指纹定位方法
CN113271539A (zh) * 2021-05-21 2021-08-17 南京邮电大学 一种基于改进的cnn模型的室内目标定位方法

Family Cites Families (3)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
US11363416B2 (en) * 2019-10-04 2022-06-14 Samsung Electronics Co., Ltd. System and method for WiFi-based indoor localization via unsupervised domain adaptation
CN111212379B (zh) * 2020-01-06 2021-11-23 浙江工易云服科技有限公司 一种基于卷积神经网络的新型csi室内定位方法
CN112333642B (zh) * 2020-11-06 2022-08-30 南京邮电大学 基于信道状态信息的室内定位方法

Patent Citations (4)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
US20210092611A1 (en) * 2019-09-19 2021-03-25 Colorado State University Research Foundation Security-enhanced Deep Learning Fingerprint-Based Indoor Localization
CN112153736A (zh) * 2020-09-14 2020-12-29 南京邮电大学 一种基于信道状态信息的人员动作识别和位置估计方法
CN112712557A (zh) * 2020-12-17 2021-04-27 上海交通大学 一种基于卷积神经网络的超分辨率cir室内指纹定位方法
CN113271539A (zh) * 2021-05-21 2021-08-17 南京邮电大学 一种基于改进的cnn模型的室内目标定位方法

Non-Patent Citations (1)

* Cited by examiner, † Cited by third party
Title
曹建荣等 (CAO, JIANRONG ET AL.): "结合CNN 和WiFi 指纹库的室内定位算法 (Indoor Location Algorithm Combining CNN and WiFi Fingerprint Database)", 计算机系统应用 (COMPUTER SYSTEMS & APPLICATIONS), 15 July 2020 (2020-07-15) *

Cited By (3)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN116234005A (zh) * 2022-12-20 2023-06-06 同济大学 一种多建筑物室内定位方法
CN116047410A (zh) * 2023-01-16 2023-05-02 华东师范大学 一种基于图卷积神经网络的uwb室内行人定位方法
WO2024197655A1 (zh) * 2023-03-29 2024-10-03 中国科学院深圳先进技术研究院 基于去噪自编码器和卷积神经网络的WiFi指纹室内定位方法

Also Published As

Publication number Publication date
US20220386264A1 (en) 2022-12-01
US11943735B2 (en) 2024-03-26
CN113271539B (zh) 2024-06-25
CN113271539A (zh) 2021-08-17

Similar Documents

Publication Publication Date Title
CN113271539B (zh) 一种基于改进的cnn模型的室内目标定位方法
Zhu et al. BLS-location: A wireless fingerprint localization algorithm based on broad learning
CN112748397B (zh) 非视距条件下基于自适应bp神经网络的uwb定位方法
CN112101278B (zh) 基于k近邻特征提取和深度学习的宅基地点云分类方法
CN107818571B (zh) 基于深度学习网络和均值漂移的船只自动跟踪方法及系统
CN106912105B (zh) 基于pso_bp神经网络的三维定位方法
CN115220007B (zh) 一种针对姿态识别的雷达点云数据增强方法
Chidlovskii et al. Semi-supervised variational autoencoder for WiFi indoor localization
CN113947636B (zh) 一种基于深度学习的激光slam定位系统及方法
CN107064913A (zh) 一种基于深度学习的无线定位方法及系统
CN110536245A (zh) 一种基于深度学习的室内无线定位方法及系统
CN111263295B (zh) 一种wlan室内定位方法和装置
CN107529222A (zh) 一种基于深度学习的WiFi室内定位系统
CN109814066B (zh) 基于神经网络学习的rssi室内定位测距方法、室内定位平台
CN105120479B (zh) 终端间Wi-Fi信号的信号强度差异修正方法
CN113805142B (zh) 一种基于联邦学习的建筑物楼层室内定位方法
CN111461251A (zh) 基于随机森林与自编码器的WiFi指纹的室内定位方法
CN114943766A (zh) 重定位方法、装置、电子设备及计算机可读存储介质
CN118746794A (zh) 一种多无人机协同目标定位方法、介质及装置
Wanqing et al. Improved PSO-extreme learning machine algorithm for indoor localization
CN105866732A (zh) 一种改进mk模型和wknn算法相结合的混合室内定位方法
WO2024197655A1 (zh) 基于去噪自编码器和卷积神经网络的WiFi指纹室内定位方法
CN113724325B (zh) 一种基于图卷积网络的多场景单目相机位姿回归方法
CN121462981A (zh) 一种基于生成式对抗网络的毫米波csi室内定位方法及系统
CN113008226B (zh) 基于门控循环神经网络和粒子滤波的地磁室内定位方法

Legal Events

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

Ref document number: 21940464

Country of ref document: EP

Kind code of ref document: A1

NENP Non-entry into the national phase

Ref country code: DE

122 Ep: pct application non-entry in european phase

Ref document number: 21940464

Country of ref document: EP

Kind code of ref document: A1