Indoor positioning perception optimization method and system based on star flash technology

Abstract

The invention provides an indoor positioning perception optimization method and system based on a satellite flash technology, and relates to the technical field of wireless positioning, and the method comprises the steps: transmitting and receiving a UWB signal through a satellite flash device, collecting a multi-node measurement data set according to the UWB signal, tDOA measurement data, AOA measurement data, one-way position measurement data and two-way position measurement data are preprocessed, a multi-source fusion positioning algorithm is constructed to collect a multi-source measurement data set, deep learning optimization is carried out, an indoor position estimation result is generated, and dynamic scheduling is carried out in combination with indoor complex environment parameters. And performing cooperative sensing on the indoor area according to the scheduling communication task, and determining an indoor positioning result. The technical problem that in the prior art, an indoor positioning method only depends on single signal measurement data, and consequently the indoor positioning precision and stability are low is solved. The technical effect of improving the indoor positioning precision and stability through fusion of multi-source data and combination with deep learning optimization is achieved.

Description

Technical Field

[0001] This invention relates to the field of wireless positioning technology, and specifically to an indoor positioning and sensing optimization method and system based on star-flash technology. Background Technology

[0002] Indoor positioning refers to the technology of accurately determining the location of people or objects within a building environment using wireless signals, sensors, or geomagnetism. Its core function is to provide real-time location services for various fields such as smart warehousing, medical navigation, shopping mall guidance, and security management, thereby improving operational efficiency and user experience. However, in indoor positioning scenarios, positioning accuracy and reliability are greatly affected by indoor environmental factors. Indoor environments are complex and changeable, with numerous obstacles making signal propagation paths complex and unstable. Simultaneously, frequent movement of people indoors and human obstruction can interfere with signals. Traditional indoor positioning technologies, such as those based on Wi-Fi and Bluetooth, typically rely on only a single type of signal measurement data, such as using signal strength indicators to estimate location. A single signal source is significantly affected by complex environmental factors, making it difficult to guarantee positioning accuracy. Furthermore, when facing dynamically changing indoor environments, the accuracy and stability of indoor positioning are poor, making it difficult to meet the growing demand for high-precision indoor positioning.

[0003] In summary, existing technologies suffer from the problem that indoor positioning methods rely solely on single signal measurement data, resulting in low accuracy and stability in indoor positioning. Summary of the Invention

[0004] The purpose of this application is to provide an indoor positioning perception optimization method and system based on star flash technology, which solves the technical problem that existing indoor positioning methods rely solely on single signal measurement data, resulting in low indoor positioning accuracy and stability.

[0005] In view of the above problems, this application provides an indoor positioning and sensing optimization method and system based on star flash technology.

[0006] The first aspect of this application provides an indoor positioning and sensing optimization method based on star-flash technology. The method includes: transmitting and receiving UWB signals via a star-flash device; collecting a multi-node measurement dataset according to the UWB signals, wherein the measurement dataset includes TDOA measurement data, AOA measurement data, unidirectional position measurement data, and bidirectional position measurement data; preprocessing the TDOA measurement data, AOA measurement data, unidirectional position measurement data, and bidirectional position measurement data to construct a multi-source fusion positioning algorithm; collecting a multi-source measurement dataset using the multi-source fusion positioning algorithm; performing deep learning optimization based on the multi-source measurement dataset to generate an indoor position estimation result; dynamically scheduling the indoor position estimation result in conjunction with indoor complex environmental parameters; and performing collaborative sensing of the indoor area according to the scheduled communication task to determine the indoor positioning result.

[0007] Optionally, the UWB signal is periodically broadcast using a star-flash device according to a preset frame structure; the arrival time parameters of the UWB signal are recorded based on multiple anchor nodes according to a clock synchronization mechanism; the arrival time parameters are mapped to the multiple anchor nodes to calculate the time difference, thereby obtaining the TDOA measurement data; signal phase analysis is performed based on the UWB signal to obtain the signal phase difference, and the signal direction angle is calculated to obtain the AOA measurement data; signal transmission and reception analysis is performed by traversing the multiple anchor nodes to obtain the signal propagation time, and the signal propagation speed is retrieved according to the signal propagation time to calculate the one-way distance, thereby obtaining the one-way position measurement data; a target node is set, and bidirectional ranging interaction is performed between the target node and the multiple anchor nodes to obtain the bidirectional propagation time for clock deviation correction, thereby obtaining the bidirectional position measurement data.

[0008] Optionally, the TDOA measurement data, AOA measurement data, unidirectional position measurement data, and bidirectional position measurement data are cleaned, and the data cleaning results are time-synchronized to generate a preprocessed dataset. Feature analysis is performed on the preprocessed dataset, and quality assessment is conducted based on multiple data features to generate standardized feature vectors. Weights are assigned based on the standardized feature vectors to generate multiple weight coefficients. The multi-source fusion positioning algorithm is used to jointly solve the standardized feature vectors according to the multiple weight coefficients to obtain a multi-source fusion result. The multi-source fusion result is analyzed frame-by-frame according to a time series to obtain the multi-source measurement dataset.

[0009] Optionally, the preprocessed dataset includes TDOA measurement preprocessed data, AOA measurement preprocessed data, unidirectional position measurement preprocessed data, and bidirectional position measurement preprocessed data; the time delay difference of the TDOA measurement preprocessed data is extracted for feature analysis, and the reliability score of the TDOA measurement preprocessed data is performed based on the error distribution characteristics to construct a TDOA feature vector; the signal incident angle of the AOA measurement preprocessed data is extracted for feature analysis, and the spatial reliability score of the AOA measurement preprocessed data is performed based on the angle incident characteristics to construct an AOA feature vector; feature analysis is performed based on the unidirectional position measurement preprocessed data to obtain distance feature values ​​and signal strength features, and feature analysis is performed based on the bidirectional position measurement preprocessed data to obtain round-trip time features and interaction state features; the consistency of the distance feature values, the signal strength features, the round-trip time features, and the interaction state features is verified to construct a ranging feature vector.

[0010] Optionally, a weighted fusion algorithm is used to jointly solve the standardized feature vector according to the multiple weight coefficients to generate an intermediate fusion result; the intermediate fusion result is matched and verified with the measurement dataset of multiple nodes to obtain a data consistency score; the intermediate fusion result is then verified in reverse according to the data consistency score to generate the multi-source fusion result.

[0011] Optionally, a deep learning network model is constructed, and the multi-source measurement dataset is synchronized to the deep learning network model for deep learning: S1: Based on the time dimension, the information content of the multi-source measurement dataset is analyzed to obtain multiple information data. The multiple information data are used as an index to retrieve the measurement time and determine the target measurement time information; S2: Based on the spatial dimension, the positioning contribution of the multi-source measurement dataset is analyzed to obtain multiple positioning contribution degrees; S3: The multiple positioning contribution degrees are sorted in descending order, and the first-order positioning contribution degree is extracted to retrieve the multiple anchor nodes and determine the target anchor node; S4: A spatiotemporal attention mechanism is integrated to analyze the target measurement time information in combination with the target anchor node to generate the indoor location estimation result.

[0012] Optionally, multiple indoor scene datasets are collected from the indoor area, including CSI data, obstacle distribution data, and personnel activity data; the CSI data, obstacle distribution data, and personnel activity data are labeled to construct a target to be located; a multi-branch deep neural network is constructed based on the standardized feature vector, and the multi-branch deep neural network is trained under supervision based on the multiple anchor nodes and the target to be located as graph structure nodes to construct the deep learning network model.

[0013] Optionally, a convolutional neural network branch is constructed based on the TDOA feature vector, a fully connected network branch is constructed based on the AOA feature vector, and a measurement value processing branch is constructed based on the ranging feature vector; the convolutional neural network branch, the fully connected network branch, and the measurement value processing branch are associated and integrated to construct a dedicated feature extraction branch module; the TDOA feature vector, the AOA feature vector, and the ranging feature vector are mapped to the convolutional neural network branch, the fully connected network branch, and the measurement value processing branch, respectively, and a cross-modal attention mechanism is used to process the TDOA feature vector and the AOA feature vector. The vector and the ranging feature vector are analyzed according to multiple weight coefficients to obtain cross-modal context feature parameters; cross-modal integration is performed based on the cross-modal context feature parameters to construct a multimodal feature fusion module; the TDOA feature vector, the AOA feature vector, and the ranging feature vector are dimensionality reduced according to the cross-modal context feature parameters to obtain a dimensionality reduction parameter set; linear regression is performed based on the dimensionality reduction parameter set to construct a position regression output module; the dedicated feature extraction branch module, the multimodal feature fusion module, and the position regression output module are integrated to construct the multi-branch deep neural network.

[0014] Optionally, environmental remote sensing data is continuously collected and updated in real time to construct a real-time environmental parameter map. The complexity of the indoor area is analyzed by traversing the real-time environmental parameter map to generate indoor complex environmental parameters, which include environmental complexity. The indoor location estimation results are adjusted for regional perception according to the environmental complexity to divide multiple areas to be perceived. A scheduling cycle is set to perform perception analysis on the multiple areas to be perceived to determine multiple sensing nodes. Communication resources are allocated according to the multiple sensing nodes to generate a scheduling communication task, which includes a communication task instruction set. The communication task instruction set is executed to perform collaborative perception of the indoor area to determine the indoor positioning result.

[0015] A second aspect of this application provides an indoor positioning and sensing optimization system based on star-flash technology. The system includes: a data acquisition module for transmitting and receiving UWB signals via a star-flash device, and acquiring a multi-node measurement dataset according to the UWB signals. The measurement dataset includes TDOA measurement data, AOA measurement data, unidirectional position measurement data, and bidirectional position measurement data; a data processing module for preprocessing the TDOA measurement data, AOA measurement data, unidirectional position measurement data, and bidirectional position measurement data to construct a multi-source fusion positioning algorithm, and acquiring a multi-source measurement dataset through the multi-source fusion positioning algorithm; a location generation module for performing deep learning optimization based on the multi-source measurement dataset to generate an indoor location estimation result; and a positioning result determination module for dynamically scheduling the indoor location estimation result in conjunction with indoor complex environmental parameters, performing collaborative sensing of the indoor area according to the scheduled communication task, and determining the indoor positioning result.

[0016] One or more technical solutions provided in this application have at least the following technical effects or advantages: The method provided in this application embodiment transmits and receives UWB signals through a star-flash device, and collects a multi-node measurement dataset according to the UWB signals. The measurement dataset includes TDOA measurement data, AOA measurement data, unidirectional position measurement data, and bidirectional position measurement data. The TDOA measurement data, AOA measurement data, unidirectional position measurement data, and bidirectional position measurement data are preprocessed to construct a multi-source fusion positioning algorithm. This algorithm is then used to collect the multi-source measurement dataset. Deep learning optimization is performed based on the multi-source measurement dataset to generate an indoor position estimation result. The indoor position estimation result is then dynamically scheduled in conjunction with indoor complex environmental parameters. Based on the scheduled communication tasks, collaborative sensing of the indoor area is performed to determine the indoor positioning result. This achieves the technical effect of improving indoor positioning accuracy and stability through the fusion of multi-source data and optimization using deep learning.

[0017] The above description is merely an overview of the technical solution of this application. To enable a clearer understanding of the technical means of this application and to facilitate its implementation according to the description, and to make the above and other objects, features, and advantages of this application more apparent, specific embodiments of this application are described below. It should be understood that the content described in this section is not intended to identify key or important features of the embodiments of this application, nor is it intended to limit the scope of this application. Other features of this application will become readily apparent through the following description. Attached Figure Description

[0018] To more clearly illustrate the technical solutions in this application or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are merely exemplary. For those skilled in the art, other drawings can be obtained based on the provided drawings without creative effort.

[0019] Figure 1 A flowchart illustrating the indoor positioning and sensing optimization method based on star-flash technology provided in this application.

[0020] Figure 2 A schematic diagram of the structure of the indoor positioning and sensing optimization system based on star-flash technology provided in this application.

[0021] Explanation of reference numerals in the attached diagram: Data acquisition module 11, data processing module 12, location generation module 13, and location result determination module 14. Detailed Implementation

[0022] This application provides an indoor positioning perception optimization method and system based on star-flash technology, which addresses the technical problem of low indoor positioning accuracy and stability caused by existing indoor positioning methods relying solely on single signal measurement data. It achieves the technical effect of improving indoor positioning accuracy and stability through the fusion of multi-source data and optimization using deep learning.

[0023] The technical solutions of the present invention will now be clearly and completely described with reference to the accompanying drawings. Obviously, the described embodiments are only a part of the embodiments of the present invention, and not all of them. It should be understood that the present invention is not limited to the exemplary embodiments described herein. All other embodiments obtained by those skilled in the art based on the embodiments of the present invention without creative effort are within the scope of protection of the present invention. It should also be noted that, for ease of description, only the parts related to the present invention are shown in the accompanying drawings, not all of them.

[0024] Example 1, as Figure 1 As shown, this application provides an indoor positioning and sensing optimization method based on star-flash technology, which includes: The device transmits and receives UWB signals through a star flash device, and collects a multi-node measurement dataset according to the UWB signals. The measurement dataset includes TDOA measurement data, AOA measurement data, unidirectional position measurement data, and bidirectional position measurement data.

[0025] Furthermore, the method involves transmitting and receiving UWB signals via a star-flash device, and collecting measurement datasets from multiple nodes based on the UWB signals. This includes: periodically broadcasting the UWB signals using the star-flash device according to a preset frame structure; recording the arrival time parameters of the UWB signals based on multiple anchor nodes using a clock synchronization mechanism; mapping the arrival time parameters to the multiple anchor nodes to calculate the time difference, thereby obtaining the TDOA measurement data; performing signal phase analysis based on the UWB signals to obtain the signal phase difference and calculate the signal direction angle, thereby obtaining the AOA measurement data; traversing the multiple anchor nodes to perform signal transmission and reception analysis to obtain the signal propagation time; retrieving the signal propagation speed based on the signal propagation time to calculate the unidirectional distance, thereby obtaining the unidirectional position measurement data; setting a target node; performing bidirectional ranging interaction between the target node and the multiple anchor nodes to obtain the bidirectional propagation time; performing clock deviation correction, thereby obtaining the bidirectional position measurement data.

[0026] Specifically, the StarScan device is a collective term for the hardware devices and related software systems built upon StarScan technology to achieve efficient, stable, and low-latency wireless communication and data interaction between devices. StarScan technology is a new generation of short-range wireless connection technology, which has advantages over Bluetooth and Wi-Fi technologies such as microsecond-level low latency, high transmission rate, low power consumption, high reliability, and strong anti-interference capabilities. By presetting the frame structure of the UWB (Ultra-Wideband) signal, including preamble, synchronization code, data field, and checksum, the StarScan device periodically broadcasts the UWB signal according to the preset frame structure. Periodic broadcasting is used to ensure the stability and coverage of the UWB signal, enabling each anchor node to receive the signal reliably and regularly. The anchor node refers to a reference point fixed in the indoor environment for receiving and processing the UWB signal. It is generally installed in a fixed location in a known location, such as on a ceiling, wall, or pillar, to receive the UWB signal from the StarScan device and record the arrival time of the UWB signal. There are at least three anchor nodes. For example, in a large smart factory workshop, to achieve high-precision positioning and real-time monitoring of equipment such as logistics carts and robots, star-flash devices are installed in various corners of the workshop. Following a preset frame structure, the star-flash devices periodically broadcast UWB signals every 10 milliseconds. Each broadcast UWB signal contains specific identification information and a timestamp, enabling multiple anchor nodes within the workshop to accurately identify and record the signal's arrival time. After receiving the UWB signal, multiple anchor nodes record the arrival time parameters of the UWB signal according to a clock synchronization mechanism. The time parameters include the timestamp of the UWB signal arriving at the anchor node. The clock synchronization mechanism can be achieved through synchronization protocols or hardware synchronization mechanisms, such as atomic clocks, to ensure that the time of each anchor node remains highly consistent. Based on the recorded arrival time parameters, a reference anchor node is selected, and the time difference between other anchor nodes and the reference anchor node is calculated to obtain TDOA measurement data. The TDOA measurement data refers to the time difference of the UWB signal arriving at different anchor nodes. Simultaneously, phase analysis is performed on the UWB signal to calculate the phase difference of the UWB signal on different receiving antennas. Using the relationship between the UWB signal phase difference and the UWB signal direction, an interferometer algorithm is used to calculate the UWB signal direction angle, obtaining AOA measurement data. The AOA measurement data reflects the angle of arrival of the UWB signal. Multiple anchor nodes are traversed, and the propagation time of multiple UWB signals is obtained based on the signal reception and transmission times recorded at these nodes. The average of these propagation times is then calculated to obtain the signal propagation time. The known propagation speed of the UWB signal in air is approximated as the speed of light. Using the formula distance = speed × time, the one-way distance is calculated using the signal propagation time and speed. The one-way position measurement data reflects the distance information between the target node and the anchor nodes.A target node is defined, and bidirectional ranging interaction is performed between the target node and multiple anchor nodes. During bidirectional ranging, the target node and anchor nodes send and receive signals to each other and record the signal propagation time. A symmetrical bilateral ranging algorithm is used to process the bidirectional propagation time and correct for clock deviation, obtaining the deviation-corrected propagation time. Accurate bidirectional position measurement data is calculated using the deviation-corrected propagation time and signal propagation speed. Based on this, a multi-node measurement dataset is collected, including TDOA measurement data, AOA measurement data, unidirectional position measurement data, and bidirectional position measurement data. Using star-flash technology, ranging and angle measurement are simultaneously achieved, including unidirectional and multidirectional position measurement data, reflecting the target node's position information from different angles, thereby improving the accuracy, reliability, and stability of indoor positioning.

[0027] The TDOA measurement data, AOA measurement data, unidirectional position measurement data, and bidirectional position measurement data are preprocessed to construct a multi-source fusion positioning algorithm, and multi-source measurement datasets are collected through the multi-source fusion positioning algorithm.

[0028] Furthermore, the TDOA measurement data, AOA measurement data, unidirectional position measurement data, and bidirectional position measurement data are preprocessed to construct a multi-source fusion positioning algorithm. The algorithm is used to collect a multi-source measurement dataset. The method includes: cleaning the TDOA measurement data, AOA measurement data, unidirectional position measurement data, and bidirectional position measurement data; performing time synchronization processing on the cleaned data to generate a preprocessed dataset; traversing the preprocessed dataset for feature analysis; performing quality assessment based on multiple data features to generate standardized feature vectors; assigning weights based on the standardized feature vectors to generate multiple weight coefficients; using the multi-source fusion positioning algorithm to jointly solve the standardized feature vectors according to the multiple weight coefficients to obtain a multi-source fusion result; and analyzing the multi-source fusion result frame-by-frame according to a time series to obtain the multi-source measurement dataset.

[0029] Specifically, the acquired TDOA, AOA, unidirectional, and bidirectional position measurement data undergo data cleaning. Noise is removed using filtering algorithms such as Kalman filtering. Outliers significantly deviating from the normal range are identified and eliminated. Missing values ​​are filled using mean imputation, median imputation, or interpolation based on adjacent data to ensure data quality. After data cleaning, the timestamps of all measurement data are aligned to the same time base using network time protocols or precise time protocols to ensure alignment across different measurement data in the time dimension, generating a preprocessed dataset. The preprocessed dataset is then traversed, and feature analysis is performed on the TDOA, AOA, unidirectional, and bidirectional position measurement data to obtain multiple data features. Quality assessment is then conducted based on these features, generating standardized feature vectors. Weight coefficients are assigned to each data type based on the confidence scores of multiple features. All weight coefficients are normalized to ensure the sum of the weight coefficients is 1, generating multiple weight coefficients. A multi-source fusion positioning algorithm is adopted. The standardized feature vector is jointly solved according to the weight coefficient. Each feature in the standardized feature vector is multiplied by the corresponding weight coefficient to obtain the weighted multi-source fusion result. Then, the multi-source fusion result is analyzed frame by frame according to the time series to extract the positioning result at each time point. The positioning results in the time series are smoothed to remove jitter and abrupt changes, generating a multi-source measurement dataset. The multi-source measurement dataset contains the target location information after fusion processing, which further improves the accuracy and reliability of indoor positioning.

[0030] Furthermore, feature analysis is performed on the preprocessed dataset, and quality assessment is conducted based on multiple data features to generate standardized feature vectors. The method includes: the preprocessed dataset contains TDOA measurement preprocessed data, AOA measurement preprocessed data, unidirectional position measurement preprocessed data, and bidirectional position measurement preprocessed data; the time delay difference of the TDOA measurement preprocessed data is extracted for feature analysis, and the reliability score of the TDOA measurement preprocessed data is performed based on the error distribution characteristics to construct a TDOA feature vector; the signal incident angle of the AOA measurement preprocessed data is extracted for feature analysis, and the spatial reliability score of the AOA measurement preprocessed data is performed based on the angle incident characteristics to construct an AOA feature vector; feature analysis is performed on the unidirectional position measurement preprocessed data to obtain distance feature values ​​and signal strength features, and feature analysis is performed on the bidirectional position measurement preprocessed data to obtain round-trip time features and interaction state features; the consistency of the distance feature values, the signal strength features, the round-trip time features, and the interaction state features is verified to construct a ranging feature vector.

[0031] Specifically, the preprocessed dataset is obtained after data cleaning and time synchronization, including TDOA measurement preprocessed data, AOA measurement preprocessed data, unidirectional position measurement preprocessed data, and bidirectional position measurement preprocessed data. The preprocessed dataset is traversed to extract TDOA measurement preprocessed data. Then, the time delay difference between each anchor node is extracted from the TDOA measurement preprocessed data. This time delay difference refers to the time difference in signal arrival at different receiving points, reflecting the relative positional relationship between the target and each receiving point. Statistical characteristics of the time delay difference, such as mean and standard deviation, are calculated to obtain error distribution characteristics. Based on these error distribution characteristics, the TDOA measurement preprocessed data is given a reliability score. If the mean is close to zero and the standard deviation is small, it indicates that the TDOA measurement values ​​fluctuate little around the true value and have a high degree of concentration, thus receiving a higher score. If the mean deviates significantly from zero or the standard deviation is large, it indicates large measurement errors, data dispersion, and low accuracy, thus receiving a lower score. A TDOA feature vector is then constructed. Similarly, the signal incident angle is extracted from the AOA measurement preprocessing data, and its statistical characteristics, such as mean and standard deviation, are calculated. The spatial distribution characteristics of the incident angle are analyzed, and a spatial reliability score is assigned to the AOA measurement data based on the incident angle characteristics. The spatial reliability score reflects the spatial reliability of the AOA measurement data; a higher score indicates more reliable data. An AOA feature vector is then constructed based on the signal incident angle and the spatial reliability score. The unidirectional position measurement preprocessing data records the time from signal transmission to reception. Using the known speed of light, the distance the signal propagates is calculated, which is the distance feature value. The signal strength feature is obtained by calculating the ratio of the received signal power to the transmitted power. The distance feature value and signal strength feature reflect the physical distance between the target node and the anchor node and the signal attenuation during transmission, respectively. Feature analysis is performed on the preprocessed bidirectional position measurement data to obtain round-trip time and interaction status features. The round-trip time feature is calculated by recording the time difference between the signal being sent from the target node to the anchor node and returning, i.e., the time interval from the signal transmission time to the reception time, reflecting the total propagation time of the signal on the round-trip path. The interaction status features include whether the communication is normal during bidirectional communication and whether the signal is successfully received. For example, if the signal is successfully received and returned, the interaction status is successful; if it fails to be received due to interference or other reasons, it is marked as a failure.Finally, the distance feature value, signal strength feature, round-trip time feature, and interaction status feature are consistent. Based on the known signal propagation speed, the theoretical round-trip distance is calculated using the round-trip time feature. This calculated distance is then compared with the distance feature value directly calculated from unidirectional position measurement data to check if they are within a reasonable error range. Combined with the interaction status feature, if the interaction is successful, the signal strength feature is further verified to ensure it conforms to the strength range of normal communication. If the interaction fails, it is checked whether the failure is due to low signal strength or other abnormal conditions, thus identifying and removing abnormal data. Through consistency verification, outliers and inconsistent data can be effectively removed, constructing a ranging feature vector that matches multiple measurement data points and is logically sound. This generates a standardized feature vector, providing comprehensive, reliable, and accurate data for multi-source fusion positioning algorithms, effectively improving the accuracy and reliability of indoor positioning.

[0032] Furthermore, the multi-source fusion localization algorithm is used to jointly solve the standardized feature vector according to the multiple weight coefficients to obtain the multi-source fusion result. The method includes: using a weighted fusion algorithm to jointly solve the standardized feature vector according to the multiple weight coefficients to generate an intermediate fusion result; matching and verifying the intermediate fusion result with the measurement dataset of multiple nodes to obtain a data consistency score; and performing multi-level verification on the intermediate fusion result according to the data consistency score to generate the multi-source fusion result.

[0033] Specifically, a weighted fusion algorithm is employed, jointly solving standardized feature vectors according to multiple weight coefficients. Through weighted summation, data with different weights are fused according to their importance to generate an intermediate fusion result. This intermediate fusion result is matched with a multi-node measurement dataset, calculating the distance and angle differences between the intermediate fusion result and the measurement data of each node, and evaluating them according to preset thresholds to obtain a data consistency score. Based on the data consistency score, the intermediate fusion result undergoes reverse verification. Reverse verification includes checking the calculation results of each step to ensure there are no calculation errors or data anomalies. A multi-level verification mechanism is used to perform multi-level verification of the intermediate fusion result. Multi-level verification includes checks at different levels. If the data consistency score is high, it indicates that the intermediate fusion result is highly consistent with the data in the measurement dataset, and the fusion result is reliable; only simple verification is performed, such as checking the rationality and completeness of the data. If the data consistency score is low, indicating significant inconsistencies, a relocation mechanism is activated to analyze the source of error, re-collect data, and perform fusion calculation, ultimately generating an accurate and reliable multi-source fusion result, improving the accuracy and reliability of indoor positioning.

[0034] Deep learning optimization is performed based on the multi-source measurement dataset to generate indoor location estimation results.

[0035] Furthermore, deep learning optimization is performed based on the multi-source measurement dataset to generate indoor location estimation results. The method includes: constructing a deep learning network model and synchronizing the multi-source measurement dataset to the deep learning network model for deep learning: S1: performing information content analysis on the multi-source measurement dataset based on the time dimension to obtain multiple information data, and using the multiple information data as an index to retrieve the measurement time to determine the target measurement time information; S2: performing positioning contribution analysis on the multi-source measurement dataset based on the spatial dimension to obtain multiple positioning contribution degrees; S3: sorting the multiple positioning contribution degrees in descending order, extracting the first-order positioning contribution degree to retrieve the multiple anchor nodes to determine the target anchor node; S4: integrating a spatiotemporal attention mechanism to analyze the target measurement time information in conjunction with the target anchor node to generate the indoor location estimation results.

[0036] Specifically, a deep learning network model is constructed to process multi-source measurement datasets and generate indoor location estimation results. The multi-source measurement dataset is used as input data and fed into the deep learning network model for deep learning. The deep learning network model performs the following steps: S1: Based on the time dimension, the information content of the measurement data at each time point in the multi-source measurement dataset is calculated. The calculated information content data is used as an index to retrieve the measurement time and determine the target measurement time information. S2: Based on the spatial dimension, the multi-source measurement dataset is analyzed. The contribution of each anchor node to the target node's positioning is calculated based on the distance between the anchor node and the target node and the signal strength. S3: The calculated positioning contributions are sorted in descending order. The first-order positioning contribution is extracted, and multiple anchor nodes are retrieved to obtain the anchor node corresponding to the first-order positioning contribution, which is used as the target anchor node, i.e., the anchor node with the greatest impact on positioning in the current environment. S4: A spatiotemporal attention mechanism is integrated to perform correlation analysis between the target measurement time information and the target anchor node to generate the indoor location estimation results. By using star-flash technology, ranging and angle measurement can be achieved simultaneously, and multiple measurement data can be fused from multiple sources. By using deep learning technology, multi-source positioning information fusion can be achieved, improving positioning accuracy and reducing interference from multipath effects, thus providing reliable and effective positioning information for dynamic scheduling and collaborative sensing.

[0037] Furthermore, the construction process of the deep learning network model includes: collecting multiple indoor scene datasets in an indoor area, wherein the multiple indoor scene datasets include CSI data, obstacle distribution data, and personnel activity data; labeling the CSI data, obstacle distribution data, and personnel activity data to construct a target to be located; constructing a multi-branch deep neural network based on the standardized feature vectors; and performing supervised training on the multi-branch deep neural network based on the multiple anchor nodes and the target to be located as graph structure nodes to construct the deep learning network model.

[0038] Furthermore, a multi-branch deep neural network is constructed based on the standardized feature vectors. The method includes: constructing a convolutional neural network branch based on the TDOA feature vector, a fully connected network branch based on the AOA feature vector, and a measurement value processing branch based on the ranging feature vector; associating and integrating the convolutional neural network branch, the fully connected network branch, and the measurement value processing branch to construct a dedicated feature extraction branch module; and mapping the TDOA feature vector, the AOA feature vector, and the ranging feature vector to the convolutional neural network branch, the fully connected network branch, and the measurement value processing branch using a cross-modal attention mechanism. The OA feature vector, the AOA feature vector, and the ranging feature vector are analyzed according to multiple weight coefficients to obtain cross-modal context feature parameters; cross-modal integration is performed based on the cross-modal context feature parameters to construct a multimodal feature fusion module; the TDOA feature vector, the AOA feature vector, and the ranging feature vector are dimensionality reduced according to the cross-modal context feature parameters to obtain a dimensionality reduction parameter set; linear regression is performed based on the dimensionality reduction parameter set to construct a position regression output module; the dedicated feature extraction branch module, the multimodal feature fusion module, and the position regression output module are integrated to construct the multi-branch deep neural network.

[0039] Specifically, multiple indoor scene datasets are collected within the indoor area. CSI data (Channel State Information), reflecting the amplitude and phase information of the signal on different subcarriers, is collected via a mobile terminal equipped with signal acquisition devices. Obstacle distribution data, including the location, shape, and material of obstacles in the indoor environment, is collected using multiple indoor layout drawings combined with cameras. Simultaneously, personnel activity data, including movement trajectories and stopping positions of people, is collected using cameras or wearable sensors deployed within the indoor scene. This results in multiple indoor scene datasets containing CSI data, obstacle distribution data, and personnel activity data. The CSI data, obstacle distribution data, and personnel activity data in the indoor scene datasets are then labeled, adding explicit tags to multiple data points. For example, tags are used to mark CSI data characteristics corresponding to specific locations, obstacle distribution at those locations, and personnel activity status. A target for localization is constructed; the target for localization refers to the node object whose location needs to be determined.

[0040] A multi-branch deep neural network is constructed based on standardized feature vectors. Specifically, this involves: preprocessing the TDOA feature vectors into a two-dimensional matrix suitable for convolutional neural network processing, such as arranging them in chronological order or combinations of different anchor nodes; selecting an appropriate convolution kernel size, such as 3×3; and performing convolution operations by sliding the kernel across the matrix to automatically extract local temporal difference features from the TDOA features. Feature extraction is progressively deepened by stacking multiple convolutional layers to capture patterns of signal propagation time differences. Max pooling or average pooling is used to downsample the output of the convolutional layers, reducing data dimensionality and computational cost while enhancing the translation invariance of the features. Finally, the features processed by multiple convolutions and pooling are input into a fully connected layer for further integration and feature mapping, constructing a convolutional neural network branch capable of effectively processing TDOA feature vectors. Similarly, a fully connected network branch is constructed based on the AOA feature vectors, and a measurement processing branch is constructed based on the ranging feature vectors. These convolutional neural network branches, fully connected network branches, and measurement processing branches are then integrated to form a dedicated feature extraction branch module. After mapping the TDOA feature vector, AOA feature vector, and ranging feature vector to their corresponding branches, a cross-modal attention mechanism is used to weight these vectors using multiple weighting coefficients. The importance of each modal feature is dynamically adjusted based on contextual feature parameters to obtain cross-modal contextual feature parameters. This cross-modal attention mechanism, a deep learning technique, automatically focuses on the most relevant parts of different feature vectors to localization, assigning different weights based on the importance of each feature vector to localization. This extracts more representative cross-modal contextual feature parameters, which better reflect the correlation and complementarity between different feature vectors. Based on these cross-modal contextual feature parameters, the weighted feature vectors from different modalities, including the TDOA, AOA, and ranging feature vectors, are concatenated and fused to form a comprehensive feature vector. A neural network layer then integrates information from this comprehensive feature vector, eliminating redundant information between different features. This allows features from different modalities to complement and reinforce each other, forming a multimodal feature fusion module. Based on cross-modal contextual feature parameters, dimensionality reduction techniques, such as PCA, are employed to reduce the dimensionality of TDOA feature vectors, AOA feature vectors, and ranging feature vectors, obtaining a set of dimensionality-reduced parameters. Linear regression is then performed based on this parameter set, and a location regression output module is constructed by learning the linear relationship between the dimensionality-reduced parameters and the actual location. Finally, the dedicated feature extraction branch module, multimodal feature fusion module, and location regression output module are integrated to construct a complete multi-branch deep neural network.The multi-branch deep neural network first uses a dedicated feature extraction branch module to accurately extract corresponding feature information from TDOA, AOA, and ranging data. Then, the different modal features output by the dedicated feature extraction branch module are input into a multi-modal feature fusion module. A cross-modal mechanism is used to achieve deep fusion and enhancement of multi-modal features. Finally, the fused features are input into a location regression output module, which processes the data and outputs the indoor location estimation result. Multiple modules in the multi-branch deep neural network are connected in an orderly manner to perform different functions and work collaboratively. After constructing the multi-branch deep neural network, multiple anchor nodes and the target to be located are used as graph structure nodes. Anchor nodes are fixed points in the indoor environment with known precise locations that can transmit or receive signals, while the target to be located is the node object whose location needs to be determined. Features extracted from multi-source data and processed by the multi-branch deep network are used as input, combined with the real location information of the target to be located as supervision labels, and then input into the multi-branch deep neural network. During training, the deep learning network model automatically adjusts the weights and biases of neurons in each layer based on the difference between input features and the true location using the backpropagation algorithm, continuously learning the mapping relationship from input features to target location. As the number of training iterations increases, the deep learning network model gradually optimizes its parameters, making the predicted location increasingly closer to the true location, thus gradually improving positioning accuracy. Training stops when the positioning error on the validation set reaches a preset stable and ideal level, or when the preset maximum number of iterations is reached, resulting in a completed deep learning network model. By using multi-branch deep neural networks and multimodal feature fusion, the advantages of different data sources are fully utilized to improve indoor positioning accuracy. Through supervised training and dynamic weight adjustment, the deep learning network model can adapt to different indoor environments and dynamic scenarios, improving its adaptability. Through graph-structured nodes and supervised training, the deep learning network model can generate reliable positioning results, providing reliable and effective positioning information for dynamic scheduling and collaborative perception.

[0041] The indoor location estimation results are combined with indoor complex environment parameters for dynamic scheduling. Based on the scheduling communication tasks, the indoor area is collaboratively perceived to determine the indoor positioning result.

[0042] Furthermore, the indoor location estimation results are dynamically scheduled in conjunction with indoor complex environmental parameters. Based on the scheduled communication tasks, collaborative sensing of the indoor area is performed to determine the indoor positioning result. The method includes: continuously collecting and updating environmental remote sensing data in real time to construct a real-time environmental parameter map; traversing the real-time environmental parameter map to perform complexity analysis on the indoor area, generating indoor complex environmental parameters, which include environmental complexity; adjusting the indoor location estimation results for regional perception according to the environmental complexity, dividing the area into multiple areas to be perceived; setting a scheduling cycle to perform perception analysis on the multiple areas to be perceived, determining multiple sensing nodes; allocating communication resources according to the multiple sensing nodes, generating a scheduled communication task, which includes a communication task instruction set; and executing the communication task instruction set to perform collaborative sensing of the indoor area to determine the indoor positioning result.

[0043] Specifically, multiple sensors are used to continuously collect environmental remote sensing data, including UWB signals and indoor environmental data, to monitor and update the indoor environment in real time. A real-time environmental parameter map is constructed using the collected real-time environmental remote sensing data. This map reflects environmental parameters at various locations within the indoor environment, including signal strength, obstacle distribution, and human activity. By traversing the data in the real-time environmental parameter map, the complexity of the indoor area is analyzed based on factors such as indoor obstacle distribution, signal interference levels, and human activity density. Areas with higher complexity contain more obstacles, stronger signal interference, or higher frequency of human activity. Through complexity analysis, indoor complex environmental parameters are generated, including environmental complexity, which reflects the ease of signal propagation, obstacle distribution, and the density of human activity within the indoor environment. Then, the indoor space is divided into grids according to differences in complexity, resulting in multiple areas to be sensed. For example, a complexity threshold range is set. Areas with environmental complexity exceeding the high threshold are marked as high-complexity areas, those below the low threshold are marked as low-complexity areas, and those in between are marked as medium-complexity areas. For high-complexity areas, due to their significant impact on positioning accuracy, they are divided into small grids of 1-2 square meters, with each grid serving as a sensing area. Medium-complexity areas are divided into grids of 3-5 square meters, and low-complexity areas are divided into grids of 6-10 square meters. A scheduling cycle is pre-set based on actual needs and data change frequency. Within each scheduling cycle, a comprehensive sensing analysis is performed on each sensing area to determine the sensing nodes within each area. These sensing nodes are key location points in the indoor environment used for positioning and sensing. Based on the determined sensing nodes, a dynamic bandwidth allocation algorithm is used to allocate communication resources to each sensing node, generating scheduling communication tasks. Communication resources include bandwidth and power, and scheduling communication tasks include the specific operations that each sensing node needs to perform, such as data acquisition frequency and transmission time, to achieve collaborative sensing. Finally, the communication task instruction set is executed to collect multi-dimensional information including UWB signal arrival time and angle. A multi-source data fusion positioning algorithm is then used to perform collaborative sensing of the indoor area to determine the indoor positioning result. By dynamically scheduling the indoor location estimation results in conjunction with complex indoor environmental parameters, efficient collaborative sensing of the indoor area can be achieved, effectively improving the accuracy, stability, and reliability of indoor positioning results, ensuring accurate and real-time positioning results even in complex and changing indoor environments.

[0044] Example 2, based on the same inventive concept as the indoor positioning and sensing optimization method based on star-flash technology in the previous examples, such as... Figure 2 As shown, this application provides an indoor positioning and sensing optimization system based on star-flash technology, wherein the indoor positioning and sensing optimization system based on star-flash technology includes: The data acquisition module 11 is used to transmit and receive UWB signals through a star-flash device, and to acquire a multi-node measurement dataset according to the UWB signals. The measurement dataset includes TDOA measurement data, AOA measurement data, unidirectional position measurement data, and bidirectional position measurement data. The data processing module 12 is used to preprocess the TDOA measurement data, AOA measurement data, unidirectional position measurement data, and bidirectional position measurement data to construct a multi-source fusion positioning algorithm, and to acquire a multi-source measurement dataset through the multi-source fusion positioning algorithm. The location generation module 13 is used to perform deep learning optimization based on the multi-source measurement dataset to generate an indoor location estimation result. The positioning result determination module 14 is used to dynamically schedule the indoor location estimation result in combination with indoor complex environmental parameters, and to perform collaborative sensing of the indoor area according to the scheduled communication task to determine the indoor positioning result.

[0045] Furthermore, the data acquisition module 11 in the indoor positioning and sensing optimization system based on star-flash technology is also used for: periodically broadcasting the UWB signal according to a preset frame structure using a star-flash device; recording the arrival time parameters of the UWB signal based on multiple anchor nodes according to a clock synchronization mechanism; calculating the time difference by mapping the arrival time parameters to the multiple anchor nodes to obtain the TDOA measurement data; performing signal phase analysis based on the UWB signal to obtain the signal phase difference and calculating the signal direction angle to obtain the AOA measurement data; traversing the multiple anchor nodes to perform signal transmission and reception analysis to obtain the signal propagation time, retrieving the signal propagation speed according to the signal propagation time to calculate the one-way distance and obtain the one-way position measurement data; setting a target node, performing bidirectional ranging interaction between the target node and the multiple anchor nodes to obtain the bidirectional propagation time for clock deviation correction, and obtaining the bidirectional position measurement data.

[0046] Furthermore, the data processing module 12 in the indoor positioning and perception optimization system based on star-flash technology is also used for: cleaning the TDOA measurement data, the AOA measurement data, the unidirectional position measurement data, and the bidirectional position measurement data; performing time synchronization processing on the data cleaning results to generate a preprocessed dataset; traversing the preprocessed dataset for feature analysis; performing quality assessment based on multiple data features to generate standardized feature vectors; assigning weights based on the standardized feature vectors to generate multiple weight coefficients; using the multi-source fusion positioning algorithm to jointly solve the standardized feature vectors according to the multiple weight coefficients to obtain multi-source fusion results; and analyzing the multi-source fusion results frame by frame according to the time series to obtain the multi-source measurement dataset.

[0047] Furthermore, the data processing module 12 in the indoor positioning and perception optimization system based on star-flash technology is also used for: the preprocessing dataset includes TDOA measurement preprocessing data, AOA measurement preprocessing data, unidirectional position measurement preprocessing data, and bidirectional position measurement preprocessing data; extracting the time delay difference of the TDOA measurement preprocessing data for feature analysis, scoring the credibility of the TDOA measurement preprocessing data based on the error distribution characteristics, and constructing a TDOA feature vector; extracting the signal incident angle of the AOA measurement preprocessing data for feature analysis, scoring the spatial credibility of the AOA measurement preprocessing data based on the angle incident characteristics, and constructing an AOA feature vector; performing feature analysis based on the unidirectional position measurement preprocessing data to obtain distance feature values ​​and signal strength features, and performing feature analysis based on the bidirectional position measurement preprocessing data to obtain round-trip time features and interaction state features; and performing consistency verification between the distance feature values, the signal strength features, the round-trip time features, and the interaction state features to construct a ranging feature vector.

[0048] Furthermore, the data processing module 12 in the indoor positioning and perception optimization system based on star-flash technology is also used to: employ a weighted fusion algorithm to jointly solve the standardized feature vector according to the multiple weight coefficients to generate an intermediate fusion result; perform matching verification between the intermediate fusion result and the measurement dataset of multiple nodes to obtain a data consistency score; and perform multi-level verification on the intermediate fusion result according to the data consistency score to generate the multi-source fusion result.

[0049] Furthermore, the location generation module 13 in the indoor positioning and perception optimization system based on star-flash technology is also used for: constructing a deep learning network model and synchronizing the multi-source measurement dataset to the deep learning network model for deep learning: S1: performing information content analysis on the multi-source measurement dataset based on the time dimension to obtain multiple information content data, using the multiple information content data as an index to retrieve the measurement time and determine the target measurement time information; S2: performing positioning contribution analysis on the multi-source measurement dataset based on the spatial dimension to obtain multiple positioning contribution degrees; S3: sorting the multiple positioning contribution degrees in descending order, extracting the first-order positioning contribution degree to retrieve the multiple anchor nodes and determine the target anchor node; S4: integrating a spatiotemporal attention mechanism to analyze the target measurement time information in conjunction with the target anchor node and generate the indoor location estimation result.

[0050] Furthermore, the location generation module 13 in the indoor positioning and perception optimization system based on star-flash technology is also used to: collect multiple indoor scene datasets in the indoor area, the multiple indoor scene datasets including CSI data, obstacle distribution data, and personnel activity data; label the CSI data, obstacle distribution data, and personnel activity data to construct the target to be located; construct a multi-branch deep neural network based on the standardized feature vector, and perform supervised training on the multi-branch deep neural network based on the multiple anchor nodes and the target to be located as graph structure nodes to construct the deep learning network model.

[0051] Furthermore, the location generation module 13 in the indoor positioning and perception optimization system based on star-flash technology is also used for: constructing a convolutional neural network branch based on the TDOA feature vector, constructing a fully connected network branch based on the AOA feature vector, and constructing a measurement value processing branch based on the ranging feature vector; associating and integrating the convolutional neural network branch, the fully connected network branch, and the measurement value processing branch to construct a dedicated feature extraction branch module; and mapping the TDOA feature vector, the AOA feature vector, and the ranging feature vector to the convolutional neural network branch, the fully connected network branch, and the measurement value processing branch using a cross-modal attention mechanism. The TDOA feature vector, the AOA feature vector, and the ranging feature vector are analyzed according to multiple weight coefficients to obtain cross-modal context feature parameters; cross-modal integration is performed based on the cross-modal context feature parameters to construct a multimodal feature fusion module; the TDOA feature vector, the AOA feature vector, and the ranging feature vector are dimensionality reduced according to the cross-modal context feature parameters to obtain a dimensionality reduction parameter set; linear regression is performed based on the dimensionality reduction parameter set to construct a position regression output module; the dedicated feature extraction branch module, the multimodal feature fusion module, and the position regression output module are integrated to construct the multi-branch deep neural network.

[0052] Furthermore, the positioning result determination module 14 in the indoor positioning perception optimization system based on star-flash technology is also used for: continuously collecting environmental remote sensing data for real-time updates, constructing a real-time environmental parameter map, traversing the real-time environmental parameter map to perform complexity analysis on the indoor area, generating indoor complex environmental parameters, the indoor complex environmental parameters including environmental complexity; adjusting the indoor location estimation result according to the environmental complexity, dividing multiple areas to be perceived; setting a scheduling cycle to perform perception analysis on the multiple areas to be perceived, determining multiple perception nodes; allocating communication resources according to the multiple perception nodes, generating a scheduling communication task, the scheduling communication task including a communication task instruction set; executing the communication task instruction set to perform collaborative perception of the indoor area, and determining the indoor positioning result.

[0053] The various embodiments in this specification are described in a progressive manner, with each embodiment focusing on its differences from other embodiments. Figure 1 The indoor positioning perception optimization method and specific examples based on star-flash technology in Example 1 are also applicable to the indoor positioning perception optimization system based on star-flash technology in this example. Through the foregoing detailed description of the indoor positioning perception optimization method based on star-flash technology, those skilled in the art can clearly understand the indoor positioning perception optimization system based on star-flash technology in this example. Therefore, for the sake of brevity, it will not be described in detail here.

[0054] The above description of the disclosed embodiments enables those skilled in the art to make or use this application. Various modifications to these embodiments will be readily apparent to those skilled in the art, and the general principles defined herein may be implemented in other embodiments without departing from the spirit or scope of this application. Therefore, this application is not to be limited to the embodiments shown herein, but is to be accorded the widest scope consistent with the principles and novel features disclosed herein.

[0055] Obviously, those skilled in the art can make several improvements and modifications to this application without departing from the principles of this application, and these improvements and modifications also fall within the protection scope of this application.

Claims

1. An indoor positioning and sensing optimization method based on star-flash technology, characterized in that, The method includes: The device transmits and receives UWB signals through a star flash device, and collects a multi-node measurement dataset according to the UWB signals. The measurement dataset includes TDOA measurement data, AOA measurement data, unidirectional position measurement data, and bidirectional position measurement data. The TDOA measurement data, AOA measurement data, unidirectional position measurement data, and bidirectional position measurement data are preprocessed to construct a multi-source fusion positioning algorithm, and multi-source measurement datasets are collected through the multi-source fusion positioning algorithm. Deep learning optimization is performed based on the multi-source measurement dataset to generate indoor location estimation results; The indoor location estimation results are combined with indoor complex environment parameters for dynamic scheduling. Based on the scheduling communication tasks, the indoor area is collaboratively perceived to determine the indoor positioning result.

2. The indoor positioning and sensing optimization method based on star-flash technology as described in claim 1, characterized in that, The method involves transmitting and receiving UWB signals via a star strobe device, and collecting a multi-node measurement dataset according to the UWB signals. The UWB signal is periodically broadcast using a star-flash device according to a preset frame structure; The arrival time parameters of the UWB signal are recorded based on multiple anchor nodes according to a clock synchronization mechanism. The arrival time parameters are mapped to the multiple anchor nodes to calculate the time difference, thereby obtaining the TDOA measurement data. Based on the UWB signal, signal phase analysis is performed to obtain the signal phase difference, and the signal direction angle is calculated to obtain the AOA measurement data; The signal transmission and reception analysis is performed by traversing the multiple anchor nodes to obtain the signal propagation time. The signal propagation speed is retrieved according to the signal propagation time to calculate the one-way distance and obtain the one-way position measurement data. A target node is set, and bidirectional ranging interaction is performed between the target node and the multiple anchor nodes to obtain the bidirectional propagation time for clock deviation correction, thereby obtaining the bidirectional position measurement data.

3. The indoor positioning and sensing optimization method based on star-flash technology as described in claim 2, characterized in that, The TDOA measurement data, AOA measurement data, unidirectional position measurement data, and bidirectional position measurement data are preprocessed to construct a multi-source fusion positioning algorithm. The multi-source measurement dataset is then collected using this algorithm. The method includes: The TDOA measurement data, AOA measurement data, unidirectional position measurement data, and bidirectional position measurement data are cleaned, and the data cleaning results are time-synchronized to generate a preprocessed dataset. The preprocessed dataset is traversed for feature analysis, and quality assessment is performed based on multiple data features to generate standardized feature vectors. Weights are assigned based on the standardized feature vectors to generate multiple weight coefficients. The multi-source fusion localization algorithm is used to jointly solve the standardized feature vector according to the multiple weight coefficients to obtain the multi-source fusion result. The multi-source fusion results are analyzed frame by frame according to the time series to obtain the multi-source measurement dataset.

4. The indoor positioning and sensing optimization method based on star-flash technology as described in claim 3, characterized in that, The method involves iterating through the preprocessed dataset to perform feature analysis, conducting quality assessment based on multiple data features, and generating standardized feature vectors. The preprocessed dataset includes TDOA measurement preprocessed data, AOA measurement preprocessed data, unidirectional position measurement preprocessed data, and bidirectional position measurement preprocessed data. The time delay difference of the TDOA measurement preprocessing data is extracted for feature analysis. The credibility of the TDOA measurement preprocessing data is scored based on the error distribution characteristics, and a TDOA feature vector is constructed. The signal incident angle of the AOA measurement preprocessing data is extracted for feature analysis. Based on the angle incident features, the spatial reliability score of the AOA measurement preprocessing data is performed, and an AOA feature vector is constructed. Based on the unidirectional position measurement preprocessing data, feature analysis is performed to obtain distance feature values ​​and signal strength features. Based on the bidirectional position measurement preprocessing data, feature analysis is performed to obtain round-trip time features and interaction status features. The distance feature value, the signal strength feature, the round-trip time feature, and the interaction state feature are checked for consistency to construct a ranging feature vector.

5. The indoor positioning and sensing optimization method based on star-flash technology as described in claim 3, characterized in that, The multi-source fusion localization algorithm is used to jointly solve the standardized feature vector according to the multiple weight coefficients to obtain the multi-source fusion result. The method includes: A weighted fusion algorithm is used to jointly solve the standardized feature vector according to the multiple weight coefficients to generate an intermediate fusion result. Based on the intermediate fusion result, a data consistency score is obtained by matching and verifying it with the measurement dataset of multiple nodes. The intermediate fusion results are then subjected to multi-level verification based on the data consistency score to generate the multi-source fusion results.

6. The indoor positioning and sensing optimization method based on star-flash technology as described in claim 4, characterized in that, Deep learning optimization is performed based on the multi-source measurement dataset to generate indoor location estimation results. The method includes: Construct a deep learning network model and synchronize the multi-source measurement dataset to the deep learning network model for deep learning: S1: Perform information content analysis on the multi-source measurement dataset based on the time dimension to obtain multiple information content data. Use the multiple information content data as an index to retrieve the measurement time and determine the target measurement time information. S2: Perform location contribution analysis on the multi-source measurement dataset based on the spatial dimension to obtain multiple location contribution degrees; S3: Sort the multiple positioning contribution scores in descending order, extract the positioning contribution score of the first position, and search the multiple anchor nodes to determine the target anchor node; S4: The integrated spatiotemporal attention mechanism analyzes the target measurement time information in conjunction with the target anchor node to generate the indoor location estimation result.

7. The indoor positioning and sensing optimization method based on star-flash technology as described in claim 6, characterized in that, The process and methods for constructing deep learning network models include: Collect multiple indoor scene datasets in the indoor area, which include CSI data, obstacle distribution data, and personnel activity data; The CSI data, obstacle distribution data, and personnel activity data are labeled to construct the target to be located; A multi-branch deep neural network is constructed based on the standardized feature vectors. The multi-branch deep neural network is trained under supervision using the multiple anchor nodes and the target to be located as graph structure nodes, thereby constructing the deep learning network model.

8. The indoor positioning and sensing optimization method based on star-flash technology as described in claim 7, characterized in that, The method for constructing a multi-branch deep neural network based on the standardized feature vectors includes: A convolutional neural network branch is constructed based on the TDOA feature vector, a fully connected network branch is constructed based on the AOA feature vector, and a measurement value processing branch is constructed based on the ranging feature vector. The convolutional neural network branch, the fully connected network branch, and the measurement value processing branch are linked and integrated to construct a dedicated feature extraction branch module; The TDOA feature vector, AOA feature vector, and ranging feature vector are mapped to the convolutional neural network branch, the fully connected network branch, and the measurement value processing branch. A cross-modal attention mechanism is used to analyze the TDOA feature vector, AOA feature vector, and ranging feature vector according to multiple weight coefficients to obtain cross-modal context feature parameters. Cross-modal integration is performed based on the cross-modal contextual feature parameters to construct a multimodal feature fusion module; The TDOA feature vector, AOA feature vector, and ranging feature vector are dimensionality reduced according to the cross-modal context feature parameters to obtain a dimensionality reduction parameter set. Based on the reduced-dimensionality parameter set, linear regression is performed to construct a location regression output module; The dedicated feature extraction branch module, the multimodal feature fusion module, and the location regression output module are integrated to construct the multi-branch deep neural network.

9. The indoor positioning and sensing optimization method based on star-flash technology as described in claim 1, characterized in that, The indoor location estimation results are combined with indoor complex environment parameters for dynamic scheduling. Based on the scheduling communication tasks, collaborative sensing of the indoor area is performed to determine the indoor positioning result. The method includes: Continuously collect environmental remote sensing data and update it in real time to construct a real-time environmental parameter map. Traverse the real-time environmental parameter map to perform complexity analysis on the indoor area and generate indoor complex environmental parameters, which include environmental complexity. The indoor location estimation results are adjusted for regional perception based on the environmental complexity, and multiple areas to be perceived are divided. A scheduling cycle is set to perform perception analysis on the multiple areas to be perceived, and multiple perception nodes are determined. According to the multiple sensing nodes, communication resources are allocated to generate scheduled communication tasks, and the scheduled communication tasks include a set of communication task instructions. The communication task instruction set is executed to perform collaborative perception of the indoor area and determine the indoor positioning result.

10. An indoor positioning and sensing optimization system based on star-flash technology, characterized in that, The steps for implementing the indoor positioning and sensing optimization method based on star-flash technology according to any one of claims 1 to 9 include: The data acquisition module is used to transmit and receive UWB signals through the star flash device, and to acquire a multi-node measurement dataset according to the UWB signal. The measurement dataset includes TDOA measurement data, AOA measurement data, unidirectional position measurement data, and bidirectional position measurement data. The data processing module is used to preprocess the TDOA measurement data, the AOA measurement data, the unidirectional position measurement data, and the bidirectional position measurement data, construct a multi-source fusion positioning algorithm, and collect multi-source measurement datasets through the multi-source fusion positioning algorithm; The location generation module is used to perform deep learning optimization based on the multi-source measurement dataset to generate indoor location estimation results; The positioning result determination module is used to dynamically schedule the indoor location estimation results in combination with indoor complex environment parameters, and to perform collaborative perception of the indoor area according to the scheduling communication task to determine the indoor positioning result.

Images