CN-122002336-A - Indoor information source credibility assessment method and system for multichannel double-channel deep learning

CN122002336ACN 122002336 ACN122002336 ACN 122002336ACN-122002336-A

Abstract

The application discloses an indoor information source credibility assessment method and system for multichannel double-channel deep learning, which realize accurate assessment of information source credibility. The method comprises the steps of extracting multidimensional source characteristics based on the positions of N sources and the current estimated positions of positioning targets, wherein the multidimensional source characteristics comprise propagation path characteristics of each source, spatial structure characteristics of N sources, spatial probability characteristics of each source, spatial probability intersection characteristics of N sources and nominal accuracy of each source, the multidimensional source characteristics are input into a deep learning model to obtain the credibility of each source, the deep learning model extracts specific characteristics corresponding to global characteristics and M source types from the multidimensional source characteristics respectively, fusion characteristics are obtained by fusing the global characteristics and the specific characteristics corresponding to the M source types, N-dimensional vectors are obtained by mapping the fusion characteristics, and each element in the N-dimensional vectors represents the credibility of one source.

Inventors

ZHANG WEI
WANG YANKUN
FAN YONG
ZHANG XING
WANG JINGZHE
ZHANG HAIGANG

Assignees

深圳职业技术大学

Dates

Publication Date: 20260508
Application Date: 20260107

Claims (10)

1. The indoor information source credibility assessment method for multichannel double-channel deep learning is characterized by comprising the following steps of: Extracting multidimensional information source characteristics based on the positions of N information sources and the current estimated positions of positioning targets, wherein the multidimensional information source characteristics comprise propagation path characteristics of each information source, spatial structure characteristics of the N information sources, spatial probability characteristics of each information source, spatial probability intersection characteristics of the N information sources and nominal precision of each information source, and N is a positive integer; The method comprises the steps of inputting the multi-dimensional information source characteristics into a deep learning model to obtain the credibility of each information source, wherein the deep learning model extracts global characteristics and specific characteristics corresponding to M information source types from the multi-dimensional information source characteristics respectively, fuses the global characteristics and the specific characteristics corresponding to M information source types to obtain fusion characteristics, and maps the fusion characteristics to obtain N-dimensional vectors, wherein each element in the N-dimensional vectors represents the credibility of one information source.
2. The method of claim 1, wherein inputting the multi-dimensional source features into a deep learning model yields a confidence level for each source, comprising: Generating a first scene structure feature matrix based on the propagation path feature of each information source and the space structure features between every two N information sources, wherein in the first scene structure feature matrix, the ith row and the ith column are propagation path features of the ith information source, the ith row and the jth column are space structure features between the ith information source and the jth information source, j is not equal to i, and i and j are positive integers less than or equal to N; Generating a first environmental noise feature matrix based on the spatial probability feature of each source and the spatial probability intersection feature between every two N sources, wherein in the first environmental noise feature matrix, an ith row and an ith column element are the spatial probability features of the ith source, and an ith row and an jth column element are the spatial probability intersection features between the ith source and the jth source; Generating a first nominal precision feature matrix based on the nominal precision of each information source, wherein the first nominal precision feature matrix is a diagonal matrix, and the ith row and ith column elements in the first nominal precision feature matrix are the nominal precision of the ith information source; and inputting the first scene structure feature matrix, the first environmental noise feature matrix and the first nominal precision feature matrix into the deep learning model to obtain the credibility of each information source.
3. The method of claim 2, wherein the deep learning model comprises a generic convolution kernel and M sets of special convolution kernels, the M sets of special convolution kernels corresponding one-to-one to the M source types; The deep learning model performs feature extraction on the first scene structure feature matrix, the first environmental noise feature matrix and the first nominal precision feature matrix through the general convolution kernel to obtain the global feature; And for each information source type in the M information source types, the deep learning model performs feature extraction on the first scene structure feature matrix, the first environment noise feature matrix and the first nominal precision feature matrix through a special convolution kernel corresponding to the information source type to obtain specific features corresponding to the information source type.
4. A method according to claim 3, wherein the specific features for each source type are extracted by: determining the sequence of the N information sources; Performing single-heat coding on the information source types based on the sequence of the N information sources to obtain coding vectors of the information source types, wherein an ith element in the coding vectors represents whether the ith information source belongs to the information source types or not; generating a mask matrix of the information source type based on the coding vector, wherein the element in the kth row and the kth column in the mask matrix is the average value of the element in the coding vector and the element in the kth column, and k is a positive integer less than or equal to N; masking operation is carried out on the first scene structure feature matrix, the first environment noise feature matrix and the first nominal precision feature matrix based on the mask matrix of the information source type, so that a second scene structure feature matrix, a second environment noise feature matrix and a second nominal precision feature matrix are obtained; And extracting features of the second scene structure feature matrix, the second environment noise feature matrix and the second nominal precision feature matrix through special convolution kernels corresponding to the information source types, so as to obtain specific features corresponding to the information source types.
5. The method of claim 1, wherein the propagation path characteristics of each source are extracted by: Determining a propagation path of the source based on the position of the source and the current estimated position of the positioning target; Converting the space structure information of the indoor environment where the information source is currently located into a space plane equation; calculating the propagation path and the space plane equation to obtain the number of obstacles on the propagation path; and determining the number of the obstacles as the propagation path characteristics of the source.
6. The method of claim 1, wherein spatial structural features between any two of the N sources are extracted by: Determining a first spatial vector from the positioning target to a first source based on a location of the first source and a current estimated location of the positioning target; Determining a second spatial vector from the positioning target to a second source based on a location of the second source and a current estimated location of the positioning target; And determining an included angle between the first space vector and the second space vector, and determining the included angle as a space structural feature between the first information source and the second information source, wherein the first information source and the second information source are any two information sources in the N information sources.
7. The method of claim 1, wherein the spatial probability features for each source are extracted by: determining a spherical area corresponding to the information source based on the position of the information source and the distance observation value of the information source, wherein the spherical area takes the position of the information source as a sphere center and the distance observation value of the information source as a radius; determining a vertical plane where the positioning target is located based on the current estimated position of the positioning target, wherein the vertical plane is perpendicular to a connection line between the current estimated position of the positioning target and the information source position and passes through the current estimated position of the positioning target; and determining a first subarea formed by intersecting the vertical plane from the spherical area, and determining the volume of the first subarea as the space probability characteristic of the information source, wherein the volume of the first subarea is smaller than the rest areas of the spherical area.
8. The method of claim 1, wherein the spatial probability intersection feature between any two of the N sources is extracted by: Determining a first spherical area based on the position of a first information source and the distance observation value of the first information source, wherein the first spherical area takes the position of the first information source as a sphere center and takes the distance observation value of the first information source as a radius; determining a second spherical region based on the position of a second information source and the distance observation value of the second information source, wherein the second spherical region takes the position of the second information source as a sphere center and takes the distance observation value of the second information source as a radius; And determining the volume of a second subarea formed by intersecting the first spherical area and the second spherical area, and determining the volume of the second subarea as a space probability intersection characteristic between the first information source and the second information source, wherein the first information source and the second information source are any two information sources in the N information sources.
9. The method according to any one of claims 1 to 8, wherein the current estimated position of the positioning target is estimated based on a preset positioning model and observed data of the N sources for the positioning target; After the multidimensional information source characteristics are input into a deep learning model to obtain the credibility of each information source, the method further comprises the following steps: adjusting parameters of the positioning model based on the credibility of each information source, and re-estimating the position of the positioning target based on the adjusted positioning model and the observation data of the N information sources for the positioning target; if the distance between the re-estimated position and the current estimated position is larger than a preset threshold value, the re-estimated position is used as a new current estimated position, the first scene structure feature matrix and the first environment noise feature matrix are updated, the credibility of each information source is re-estimated, then the information sources are re-positioned and judged, iteration is continued until the requirements are met, and the credibility is not estimated any more.
10. An indoor source credibility evaluation system for multichannel double-channel deep learning, which is characterized by comprising: The system comprises an extraction module, a positioning module and a positioning module, wherein the extraction module is used for extracting multidimensional information source characteristics based on the positions of N information sources and the current estimated positions of positioning targets, wherein the multidimensional information source characteristics comprise propagation path characteristics of each information source, spatial structure characteristics between every two information sources, spatial probability characteristics of each information source, spatial probability intersection characteristics between every two information sources and nominal precision of each information source, and N is a positive integer; The system comprises an evaluation module, a deep learning module and a mapping module, wherein the evaluation module is used for inputting the multidimensional information source characteristics into the deep learning model to obtain the credibility of each information source, the deep learning model is used for respectively extracting global characteristics and specific characteristics corresponding to M information source types from the multidimensional information source characteristics, fusing the global characteristics and the specific characteristics corresponding to M information source types to obtain fusion characteristics, mapping the fusion characteristics to obtain N-dimensional vectors, and each element in the N-dimensional vectors represents the credibility of one information source.

Description

Indoor information source credibility assessment method and system for multichannel double-channel deep learning Technical Field The application relates to the technical field of positioning, in particular to an indoor information source credibility assessment method and system for multichannel double-channel deep learning. Background Ultra Wide-Band (UWB), wireless fidelity (WIRELESS FIDELITY, wi-Fi), bluetooth and other terrestrial radio range positioning technologies have become the dominant schemes of indoor positioning technologies. However, positioning accuracy and reliability are severely limited due to non-line-of-sight propagation, multipath effects, and dynamic interference in indoor environments. The quality of the information source observation is evaluated, and a corresponding information source screening and weight distribution strategy is constructed, so that the method has become a key technology for inhibiting the influence of noise and environmental disturbance of the information source. Disclosure of Invention The embodiment of the application aims to provide an indoor information source credibility assessment method and system for multichannel double-channel deep learning, which can be used for accurately assessing information source credibility. In order to achieve the above object, the embodiment of the present application adopts the following technical scheme: In a first aspect, an embodiment of the present application provides an indoor source trust evaluation method for multichannel dual-path deep learning, including: Extracting multidimensional information source characteristics based on the positions of N information sources and the current estimated positions of positioning targets, wherein the multidimensional information source characteristics comprise propagation path characteristics of each information source, spatial structure characteristics of the N information sources, spatial probability characteristics of each information source, spatial probability intersection characteristics of the N information sources and nominal precision of each information source, and N is a positive integer; The method comprises the steps of inputting the multi-dimensional information source characteristics into a deep learning model to obtain the credibility of each information source, wherein the deep learning model extracts global characteristics and specific characteristics corresponding to M information source types from the multi-dimensional information source characteristics respectively, fuses the global characteristics and the specific characteristics corresponding to M information source types to obtain fusion characteristics, and maps the fusion characteristics to obtain N-dimensional vectors, wherein each element in the N-dimensional vectors represents the credibility of one information source. In a second aspect, an embodiment of the present application provides an indoor source trust evaluation system for multi-channel dual-path deep learning, including: The system comprises an extraction module, a positioning module and a positioning module, wherein the extraction module is used for extracting multidimensional information source characteristics based on the positions of N information sources and the current estimated positions of positioning targets, wherein the multidimensional information source characteristics comprise propagation path characteristics of each information source, spatial structure characteristics between every two information sources, spatial probability characteristics of each information source, spatial probability intersection characteristics between every two information sources and nominal precision of each information source, and N is a positive integer; The system comprises an evaluation module, a deep learning module and a mapping module, wherein the evaluation module is used for inputting the multidimensional information source characteristics into the deep learning model to obtain the credibility of each information source, the deep learning model is used for respectively extracting global characteristics and specific characteristics corresponding to M information source types from the multidimensional information source characteristics, fusing the global characteristics and the specific characteristics corresponding to M information source types to obtain fusion characteristics, mapping the fusion characteristics to obtain N-dimensional vectors, and each element in the N-dimensional vectors represents the credibility of one information source. The above at least one technical scheme adopted by the embodiment of the application can achieve the following beneficial effects: First, the evaluation mechanism is more prospective and intrinsic. By establishing a cross verification paradigm and utilizing the embedded space between the information sources and the probability relation to conduct research and judgment, the inherent defect that the traditional m