Indoor wireless positioning method and device, electronic equipment and storage medium

Abstract

The invention provides an indoor wireless positioning method and device, electronic equipment and a computer readable storage medium, and relates to the technical field of wireless communication. The indoor wireless positioning method comprises the following steps: acquiring indoor wireless signal data and indoor environment characteristic data to obtain indoor positioning characteristic data; evaluating the complexity of each indoor wireless positioning task based on a supervised learning algorithm and the indoor positioning feature data; and based on a multi-task learning model and the complexity, performing hierarchical learning on each indoor wireless positioning task to output an indoor wireless positioning result. The problem of low accuracy of indoor wireless positioning in the prior art is at least solved. The method is suitable for indoor wireless positioning scenes.

Description

Technical Field

[0001] The present invention relates to the field of wireless communication technology, and in particular to an indoor wireless positioning method, device, electronic equipment and computer-readable storage medium. Background Art

[0002] With the rapid development of mobile Internet and Internet of Things, more and more terminal devices need to achieve accurate indoor positioning in order to provide better services and user experience. In the wireless communication positioning of terminal devices, the existing technology mainly uses machine learning methods to achieve indoor positioning.

[0003] However, most of the existing technologies only focus on one of the three tasks of indoor wireless positioning (i.e., building positioning, floor positioning, and building coordinate positioning). The few models that focus on multi-task learning pay equal attention to different indoor wireless positioning tasks, resulting in poor convergence effect of highly complex indoor wireless positioning tasks in the modeling process, which in turn leads to a decrease in the accuracy of indoor wireless positioning. Summary of the invention

[0004] The technical problem to be solved by the present invention is to provide an indoor wireless positioning method, device, electronic device and computer-readable storage medium in response to the above-mentioned deficiencies in the prior art. The method can realize intelligent and accurate indoor wireless positioning, achieve reasonable attention to building positioning tasks, floor positioning tasks and building coordinate positioning tasks, solve the problem of difficulty in collaborative optimization learning in building positioning tasks, floor positioning tasks and building coordinate positioning tasks, and thereby improve the optimization learning ability of the multi-task model and the indoor wireless positioning accuracy.

[0005] In a first aspect, the present invention provides an indoor wireless positioning method, comprising: collecting indoor wireless signal data and indoor environment feature data to obtain indoor positioning feature data; based on a supervised learning algorithm and the indoor positioning feature data, evaluating the complexity of each indoor wireless positioning task; based on a multi-task learning model and the complexity, performing hierarchical learning on each indoor wireless positioning task to output an indoor wireless positioning result.

[0006] Preferably, the complexity of each indoor wireless positioning task is evaluated based on the supervised learning algorithm and the indoor positioning feature data, specifically including: based on the supervised learning algorithm, classifying and regressing the indoor positioning feature data to obtain classification and regression results, wherein the supervised learning algorithm includes any one of the following: support vector machine, K nearest neighbor algorithm, decision tree; based on the classification and regression results, the complexity of each indoor wireless positioning task is evaluated, and the classification and regression results refer to the supervised learning results of each indoor wireless positioning task.

[0007] Preferably, the indoor wireless positioning task includes a building positioning task, a floor positioning task and a mid-building coordinate positioning task, and the complexity of each indoor wireless positioning task is evaluated based on the classification result and the regression result, specifically including: judging whether the classification and regression results only meet one of the preset targets of the mid-building coordinate positioning task and the preset targets of the floor positioning task; in response to the classification and regression results both meeting the preset targets of the mid-building coordinate positioning task and the preset targets of the floor positioning task, or both not meeting the preset targets of the mid-building coordinate positioning task and the preset targets of the floor positioning task, determining that the complexity of the building positioning task is low, the complexity of the floor positioning task is medium, and the complexity of the mid-building coordinate positioning task is high; in response to the classification and regression results only meeting the preset targets of the mid-building coordinate positioning task, determining that the complexity of the mid-building coordinate positioning task is low, the complexity of the floor positioning task is high, and the complexity of the building positioning task is medium; in response to the classification and regression results only meeting the preset targets of the floor positioning task, determining that the complexity of the floor positioning task is low, the complexity of the mid-building coordinate positioning task is high, and the complexity of the building positioning task is medium.

[0008] Preferably, before performing hierarchical learning on each indoor wireless positioning task based on the multi-task learning model and the complexity to output the indoor wireless positioning result, the indoor wireless positioning method also includes: constructing a multi-task learning model based on a multi-level neural network, wherein the multi-level neural network includes any one of the following: convolutional neural network CNN, recurrent neural network RNN, multi-layer perceptron MLP.

[0009] Preferably, the multi-task learning model includes a shared layer and multiple output layers, and the shared layer h = f(x; θ h ), with the output layer y t =g t (h; θ t ), where h represents the shared layer of the multi-task learning model, f(·) represents the shared feature extraction function, x represents the input data of the multi-task learning model, and θ h Represents the parameters of the shared layer, y t represents the output layer of the tth task of the multi-task learning model, θ t represents the specific parameters of the tth task, g t (·) represents the hierarchical function of the multi-task learning model for the tth task.

[0010] Preferably, based on the multi-task learning model and the complexity, each indoor wireless positioning task is hierarchically learned to output the indoor wireless positioning result, specifically including: based on the complexity and the shared feature extraction function of the multi-task learning model, extracting the building positioning task data set, the floor positioning task data set and the building coordinate positioning task data set from the indoor positioning feature data; based on the complexity and the preset mapping relationship, inputting the building positioning task data set, the floor positioning task data set and the building coordinate positioning task data set into different hierarchical functions of the multi-task learning model respectively, wherein the preset mapping relationship refers to the mapping relationship between the complexity and the hierarchical function; based on the different hierarchical functions of the multi-task learning model, hierarchical learning is performed on the building positioning task data set, the floor positioning task data set and the building coordinate positioning task data set in turn to output the indoor wireless positioning result.

[0011] Preferably, the shared feature extraction function includes a first shared feature extraction function, a second shared feature extraction function and a third shared feature extraction function, and the shared feature extraction function based on the complexity and the multi-task learning model extracts a building positioning task data set, a floor positioning task data set and a mid-building coordinate positioning task data set from the indoor positioning feature data, specifically including: judging whether the complexity satisfies a first sorting condition, wherein the first sorting condition refers to that the complexity of the building positioning task is greater than the complexity of the floor positioning task, and the complexity of the floor positioning task is greater than the complexity of the mid-building coordinate positioning task; in response to the complexity satisfying the first sorting condition, based on the first shared feature extraction function, the second shared feature extraction function and the third shared feature extraction function of the multi-task learning model, respectively extract the building positioning task data set, the floor positioning task data set and the mid-building coordinate positioning task data set from the indoor positioning feature data, wherein the first shared feature extraction function refers to a preset shared feature extraction function, the second shared feature extraction function refers to the optimized first shared feature extraction function, and the third shared feature extraction function refers to the optimized second shared feature extraction function.

[0012] In a second aspect, the present invention also provides an indoor wireless positioning device, including: an acquisition module, an evaluation module and a positioning module, the acquisition module is used to collect indoor wireless signal data and indoor environment feature data to obtain indoor positioning feature data, the evaluation module is connected to the acquisition module, and is used to evaluate the complexity of each indoor wireless positioning task based on a supervised learning algorithm and the indoor positioning feature data, and the positioning module is connected to the evaluation module and is used to perform hierarchical learning on each indoor wireless positioning task based on a multi-task learning model and the complexity to output the indoor wireless positioning result.

[0013] In a third aspect, the present invention further provides an electronic device, comprising a memory and a processor, wherein the memory stores a computer program, and the processor is configured to run the computer program to implement the indoor wireless positioning method provided in the first aspect.

[0014] In a fourth aspect, the present invention further provides a computer-readable storage medium having a computer program stored thereon, and when the computer program is executed by a processor, the indoor wireless positioning method provided in the first aspect is implemented.

[0015] The present invention provides an indoor wireless positioning method, device, electronic device and computer-readable storage medium, which obtains indoor positioning feature data, evaluates the complexity of performing various indoor wireless positioning tasks based on the supervised learning algorithm, and performs various indoor wireless positioning tasks at different levels based on the multi-task learning model and complexity, and each indoor wireless positioning task can be collaboratively optimized and learned to output indoor wireless positioning results. Therefore, the present invention can realize intelligent and accurate indoor wireless positioning, realize reasonable attention to each indoor wireless positioning task, solve the problem that it is difficult to collaboratively optimize learning of each indoor wireless positioning task, and improve the optimization learning ability of the multi-task model and the indoor wireless positioning accuracy. BRIEF DESCRIPTION OF THE DRAWINGS

[0016] Figure 1 This is a flow chart of an indoor wireless positioning method according to Embodiment 1 of the present invention;

[0017] Figure 2 This is a flowchart of hierarchical learning of various indoor wireless positioning tasks based on a multi-task learning model in Example 1 of the present invention;

[0018] Figure 3 This is a flow chart of an indoor wireless positioning method according to Embodiment 2 of the present invention;

[0019] Figure 4 This is a schematic diagram of the structure of an indoor wireless positioning device in Example 3 of the present invention. DETAILED DESCRIPTION

[0020] In order to enable those skilled in the art to better understand the technical solution of the present invention, the embodiments of the present invention will be further described in detail below with reference to the accompanying drawings.

[0021] It should be understood that the specific embodiments and drawings described herein are only used to explain the present invention rather than to limit the present invention.

[0022] It can be understood that, in the absence of conflict, the various embodiments of the present invention and the various features in the embodiments can be combined with each other.

[0023] It can be understood that, for the convenience of description, the drawings of the present invention only show the parts related to the present invention, while the parts irrelevant to the present invention are not shown in the drawings.

[0024] It can be understood that each unit and module involved in the embodiments of the present invention may correspond to only one physical structure, or may be composed of multiple physical structures, or multiple units and modules may be integrated into one physical structure.

[0025] It can be understood that, without conflict, the functions and steps marked in the flowcharts and block diagrams of the present invention may occur in an order different from that marked in the drawings.

[0026] It is understood that the flowcharts and block diagrams of the present invention illustrate the possible architectures, functions, and operations of the systems, devices, equipment, and methods according to the various embodiments of the present invention. Each box in the flowchart or block diagram may represent a unit, module, program segment, or code, which contains executable instructions for implementing the specified functions. Moreover, each box or combination of boxes in the block diagram and flowchart may be implemented by a hardware-based system that implements the specified functions, or may be implemented by a combination of hardware and computer instructions.

[0027] It can be understood that the units and modules involved in the embodiments of the present invention can be implemented by software or hardware. For example, the units and modules can be located in a processor.

[0028] Embodiment 1:

[0029] like Figure 1 As shown, this embodiment provides an indoor wireless positioning method. The indoor wireless positioning method includes:

[0030] S101, collecting indoor wireless signal data and indoor environment feature data to obtain indoor positioning feature data.

[0031] In this embodiment, wireless signal strength information and signal interference information are collected from the indoor environment by a signal analyzer, a wireless access point (AP) or a mobile terminal itself to obtain indoor wireless signal data. Obstacle information and spatial distribution information are collected from the indoor environment by an environmental sensor, a laser rangefinder or an ultrasonic sensor, a camera, a computer vision device or a 3D scanner to obtain indoor environment feature data, wherein the wireless signal includes but is not limited to Wi-Fi signals, Bluetooth signals, RFID (Radio Frequency Identification) signals, and the spatial distribution information includes but is not limited to wall distribution information. The indoor environment feature data is transmitted to the indoor wireless positioning system for indoor wireless positioning. In this embodiment, by collecting indoor environment feature data and indoor wireless signal data as indoor positioning feature data for subsequent indoor wireless positioning, the dependence on signal strength data is reduced. In a complex indoor environment, the indoor environment feature data can help indoor wireless positioning better adapt to signal interference and changes, improve the robustness of indoor wireless positioning, and by analyzing the indoor environment feature data, the multipath effect can be more effectively processed, the signal reception quality can be improved, and the positioning accuracy can be improved.

[0032] Specifically, indoor wireless positioning tasks include building positioning tasks, floor positioning tasks and intra-building coordinate positioning tasks.

[0033] S102, evaluating the complexity of each indoor wireless positioning task based on the supervised learning algorithm and the indoor positioning feature data.

[0034] In this embodiment, the complexity of the indoor wireless positioning task execution is affected by the indoor positioning feature data. For example, in an indoor positioning environment with a higher floor and more wireless network sensors deployed, the complexity of the building positioning task is usually higher than the complexity of the floor positioning task; while in an indoor positioning environment with a lower floor and fewer wireless network sensors deployed, the complexity of the building positioning task may be lower than the complexity of the floor positioning task. Therefore, if the complexity of each indoor wireless positioning task is uniformly evaluated based on the conventional task granularity, that is, the complexity of the building positioning task < the complexity of the floor positioning task < the complexity of the coordinate positioning in the building, errors will occur in the complexity of the evaluation. This embodiment evaluates the complexity of each indoor wireless positioning task by performing supervised learning on the indoor positioning feature data, combining the supervised learning results and the task granularity, thereby improving the accuracy of the complexity evaluation and providing more accurate and effective guidance for the reasonable allocation of subsequent computing resources.

[0035] It should be noted that the building positioning task is usually a classification task, the building coordinate positioning task is usually a regression task, and the floor positioning task can be either a classification task or a regression task. Under normal circumstances, the complexity of the classification task is lower than the complexity of the regression task. If the floor positioning task is a classification task, when faced with the building positioning task and the floor positioning task, the floor positioning task is a more subtle regression task, and the complexity of the floor positioning task is greater than the complexity of the building positioning task; and if the floor positioning task is a regression task, when faced with the floor positioning task and the building coordinate positioning task, the building coordinate positioning task is a more subtle regression task, and the complexity of the building coordinate positioning task is greater than the complexity of the floor positioning task. Therefore, it can be obtained that the conventional task granularity is the complexity of the building positioning task < the complexity of the floor positioning task < the complexity of the building coordinate positioning task.

[0036] Specifically, S102: based on the supervised learning algorithm and the indoor positioning feature data, the complexity of each indoor wireless positioning task is evaluated, including steps S1021-S1022:

[0037] S1021, based on the supervised learning algorithm, classify and regress the indoor positioning feature data to obtain classification and regression results, wherein the supervised learning algorithm includes any one of the following: support vector machine, K nearest neighbor algorithm, decision tree, and the classification and regression results refer to the supervised learning results of each indoor wireless positioning task.

[0038] In this embodiment, taking the supervised learning algorithm as a support vector machine algorithm as an example, the support vector machine algorithm is used to learn each indoor wireless positioning task, analyze the indoor positioning feature data, and obtain the performance of each indoor wireless positioning task, that is, the classification and regression results.

[0039] It should be noted that the K nearest neighbor algorithm in the prior art is mainly used for classification, while the support vector machine and decision tree can be used for classification and regression tasks at the same time. If the K nearest neighbor algorithm is used to classify and regress indoor positioning feature data, when evaluating each indoor positioning task based on the classification and regression results, the complexity of the floor positioning task and the building positioning task is accurate, and the complexity of the building coordinate positioning task has errors. Therefore, this embodiment usually uses the support vector machine and the decision tree as the supervised learning algorithm, and when the indoor wireless positioning task is a single building positioning task or a single floor positioning task, the K nearest neighbor algorithm is used as the supervised learning algorithm.

[0040] S1022: Based on the classification and regression results, evaluate the complexity of each indoor wireless positioning task.

[0041] Specifically, S1022: based on the classification and regression results, the complexity of each indoor wireless positioning task is evaluated, including: judging whether the classification and regression results only meet one of the preset goals of the mid-building coordinate positioning task and the preset goals of the floor positioning task; in response to the classification and regression results both meeting the preset goals of the mid-building coordinate positioning task and the preset goals of the floor positioning task, or both not meeting the preset goals of the mid-building coordinate positioning task and the preset goals of the floor positioning task, determining that the complexity of the building positioning task is low, the complexity of the floor positioning task is medium, and the complexity of the mid-building coordinate positioning task is high; in response to the classification and regression results only meeting the preset goals of the mid-building coordinate positioning task, determining that the complexity of the mid-building coordinate positioning task is low, the complexity of the floor positioning task is high, and the complexity of the building positioning task is medium; in response to the classification and regression results only meeting the preset goals of the floor positioning task, determining that the complexity of the floor positioning task is low, the complexity of the mid-building coordinate positioning task is high, and the complexity of the building positioning task is medium.

[0042] In this embodiment, when the same indoor wireless positioning task is performed based on the same supervised learning algorithm and different indoor positioning feature data, the classification and regression results are better, which means low complexity, and the classification and regression results are worse, which means high complexity. Similarly, when different indoor wireless positioning tasks are performed based on the same supervised learning algorithm and the same indoor positioning feature data, the classification and regression results are better, which means low complexity, and the classification and regression results are worse, which means high complexity.

[0043] This embodiment judges the quality of the classification results and regression results by the preset goals of each indoor wireless positioning task. If the classification and regression results meet the preset goals of the building positioning task, the complexity of the building positioning task is determined to be low. Similarly, if the classification and regression results also meet the preset goals of the floor positioning task and the mid-building coordinate positioning task, then the complexity of the floor positioning task and the mid-building coordinate positioning task are both low. However, considering the task granularity of the building positioning task, the floor positioning task and the mid-building coordinate positioning task, that is, the positioning accuracy, when the complexity of the building positioning task, the floor positioning task and the mid-building coordinate positioning task are all low, the complexity of the building positioning task is less than the complexity of the floor positioning task and the complexity of the mid-building coordinate positioning task, that is, the floor positioning task and the mid-building coordinate positioning task are respectively promoted to medium and high. If the classification and regression results do not meet the preset goals of all indoor wireless positioning tasks, the complexity of each indoor positioning task is directly determined according to the task granularity, that is, the complexity of the building positioning task is low, the complexity of the floor positioning task is medium, and the complexity of the mid-building coordinate positioning task is high. If the classification and regression results meet the preset goals of a single indoor wireless positioning task, such as the classification and regression results meet the preset goals of the mid-building coordinate positioning task, the complexity of the mid-building coordinate positioning task is determined to be low, and the complexity of the building positioning task and the floor positioning task is determined based on the task granularity.

[0044] It should be noted that the classification and regression results include at least one of the following: accuracy, precision, recall, F1-score, confusion matrix, mean square error, root mean square error, mean absolute error, R2 (coefficient of determination), ROC (receiver operating characteristic curve, also known as sensitivity curve, which is the core indicator for evaluating the performance discrimination of medical diagnostic tests and prediction models) curve and AUC (Area Under the ROC curve). UnderCurve, the area under the ROC curve and the coordinate axis) value, where accuracy refers to the proportion of correctly classified samples to the total samples, precision refers to the proportion of samples correctly predicted to be positive to all samples predicted to be positive, recall refers to the proportion of samples correctly predicted to be positive to all samples actually positive, F1-score refers to the harmonic mean of precision and recall, confusion matrix refers to the matrix showing the classification results, including the number of true positive examples, false positive examples, true negative examples, and false negative examples, ROC curve and AUC value refer to the relationship between true positive rate and false positive rate, mean square error refers to the average of the square of the difference between the predicted value and the actual value, root mean square error refers to the square root of the mean square error, mean absolute error refers to the average of the absolute value of the difference between the predicted value and the actual value, and R2 (coefficient of determination) refers to the model's ability to explain data variation.

[0045] Optionally, before S103: performing hierarchical learning on each indoor wireless positioning task based on the multi-task learning model and complexity to output the indoor wireless positioning result, the indoor wireless positioning method further includes:

[0046] S104, constructing a multi-task learning model based on a multi-level neural network, wherein the multi-level neural network includes any one of the following: a convolutional neural network CNN, a recurrent neural network RNN, and a multi-layer perceptron MLP.

[0047] Specifically, the multi-task learning model includes a shared layer and multiple output layers, and the shared layer h = f(x; θ h ), output layer y t =g t (h; θ t ), where h represents the shared layer of the multi-task learning model, f(·) represents the shared feature extraction function, x represents the input data of the multi-task learning model, and θ h Represents the parameters of the shared layer, y t represents the output layer of the tth task of the multi-task learning model, θ t represents the specific parameters of the tth task, g t (·) represents the hierarchical function of the multi-task learning model for the tth task.

[0048] In this embodiment, since the multi-task learning model has the powerful ability to learn multiple related tasks, accelerate training and reduce overfitting, and solve complex problems, the multi-task learning model is applied to indoor wireless positioning, which can significantly improve the accuracy of indoor wireless positioning. Taking the multi-level neural network as CNN (Convolutional Neural Networks, convolutional neural network), the application scenario of the multi-task learning model is indoor wireless positioning as an example.

[0049] Based on CNN, a multi-task learning model is constructed to perform various indoor wireless positioning tasks, such as Figure 2 As shown, Figure 2 The neural network model in is a multi-task learning model. Figure 2 The backbone of the multi-task learning model is the shared layer, which includes the shared feature extraction functions of each indoor wireless positioning task. Since the parameters θ of the shared feature extraction functions of each indoor wireless positioning task are h Different, the hierarchical learning data extracted from the indoor positioning feature data x for each indoor wireless positioning task is different, Figure 2 The low-level task execution structure in the middle is the hierarchical function corresponding to the indoor wireless positioning task with low complexity, the medium-level task execution structure is the hierarchical function corresponding to the indoor wireless positioning task with medium complexity, and the high-level task execution structure is the hierarchical function corresponding to the indoor wireless positioning task with high complexity. This embodiment adopts a multi-level neural network to construct a multi-task learning model to perform hierarchical learning on each indoor wireless positioning task, improve the efficiency of indoor wireless positioning, reduce the overfitting problem of indoor wireless positioning, and thus significantly improve the accuracy of indoor wireless positioning.

[0050] S103, based on the multi-task learning model and complexity, hierarchical learning is performed on each indoor wireless positioning task to output the indoor wireless positioning result.

[0051] Specifically, S103: Based on the multi-task learning model and complexity, hierarchical learning is performed on each indoor wireless positioning task to output the indoor wireless positioning result, including steps S1031 to S1033:

[0052] S1031, based on the shared feature extraction function of the complexity and multi-task learning model, extract the building positioning task data set, the floor positioning task data set and the building coordinate positioning task data set from the indoor positioning feature data.

[0053] Specifically, the shared feature extraction function includes a first shared feature extraction function, a second shared feature extraction function and a third shared feature extraction function.

[0054] Specifically, S1031: based on the shared feature extraction function of the complexity and multi-task learning model, extracting the building positioning task data set, the floor positioning task data set and the mid-building coordinate positioning task data set from the indoor positioning feature data, including: judging whether the complexity satisfies the first sorting situation, wherein the first sorting situation refers to that the complexity of the building positioning task is greater than the complexity of the floor positioning task, and the complexity of the floor positioning task is greater than the complexity of the mid-building coordinate positioning task; in response to the complexity satisfying the first sorting situation, based on the first shared feature extraction function, the second shared feature extraction function and the third shared feature extraction function of the multi-task learning model, respectively extracting the building positioning task data set, the floor positioning task data set and the mid-building coordinate positioning task data set from the indoor positioning feature data, wherein the first shared feature extraction function refers to the preset shared feature extraction function, the second shared feature extraction function refers to the optimized first shared feature extraction function, and the third shared feature extraction function refers to the optimized second shared feature extraction function.

[0055] In this embodiment, Figure 2 As shown, indoor wireless positioning tasks with low complexity can help indoor wireless positioning tasks with medium or high complexity extract shared indoor positioning identification features from indoor positioning feature data, thereby helping to improve the hierarchical learning results of indoor wireless positioning tasks with medium or high complexity. For example, if the complexity of the building positioning task is low, the complexity of the floor positioning task is medium, and the complexity of the coordinate positioning task in the building is high, the hierarchical learning data for the building positioning task, i.e., the building positioning task data set, is directly extracted from the indoor positioning feature data based on the preset shared feature extraction function to perform hierarchical learning of the building positioning task; the parameter θ of the preset shared feature extraction function is optimized based on the hierarchical learning result of the building positioning task. h , obtain the second shared feature extraction function, based on the second shared feature extraction function, extract the hierarchical learning data for the floor positioning task from the indoor positioning feature data, that is, the floor positioning task data set, to perform hierarchical learning of the floor positioning task; similarly, optimize the parameters of the second shared feature extraction function based on the hierarchical learning result of the floor positioning task, obtain the third shared feature weighting function, and extract the mid-building coordinate positioning task data set from the indoor positioning feature data based on the third shared feature extraction function, to perform hierarchical learning of the mid-building coordinate positioning task. This embodiment uses the hierarchical learning results of the indoor wireless positioning tasks with low complexity to help the indoor wireless positioning tasks with medium or high complexity to extract their own hierarchical learning data from the indoor positioning feature data, thereby helping to improve the hierarchical learning results of the indoor wireless positioning tasks with medium or high complexity, which can significantly improve the collaborative optimization learning ability of the multi-task learning model, thereby improving the accuracy and stability of indoor wireless positioning.

[0056] It should be noted that in addition to optimizing the parameters of the shared feature extraction function to help indoor wireless positioning tasks with medium or high complexity extract their own hierarchical learning data, this embodiment can also filter the indoor positioning feature data through the hierarchical learning results of indoor wireless positioning tasks with low complexity to obtain hierarchical learning data of indoor wireless positioning tasks with medium or high complexity, such as: the indoor wireless positioning task with low complexity is a building positioning task, the indoor wireless positioning task with medium complexity is a floor positioning task, the indoor wireless positioning task with high complexity is a building coordinate positioning task, and the hierarchical learning result of the building positioning task is Building A11 of XX Community. The data belonging to Building A11 of XX Community in the indoor positioning feature data is filtered, and the floor positioning task data set is extracted from the filtered indoor positioning feature data based on the shared feature extraction function.

[0057] S1032, based on the complexity and the preset mapping relationship, the building positioning task data set, the floor positioning task data set and the building coordinate positioning task data set are respectively input into different hierarchical functions of the multi-task learning model, wherein the preset mapping relationship refers to the mapping relationship between the complexity and the hierarchical function.

[0058] S1033, based on different hierarchical functions of the multi-task learning model, hierarchical learning is performed on the building positioning task data set, the floor positioning task data set and the building coordinate positioning task data set in turn to output the indoor wireless positioning result.

[0059] In this embodiment, Figure 2 As shown, the hierarchical function of the multi-task learning model sets each indoor wireless positioning task in order from low to high according to the complexity of each indoor wireless positioning task, and performs training and learning. The indoor wireless positioning result is the hierarchical learning result of the indoor wireless positioning task with high complexity, such as: the complexity of the building positioning task is low, the complexity of the floor positioning task is medium, and the complexity of the coordinate positioning task in the building is high. The hierarchical learning result of the building positioning task is Building A11 of XX Community, and the hierarchical learning result of the floor positioning task is Floor 17 of Building A11 of XX Community. Then the indoor wireless positioning result, that is, the hierarchical result of the coordinate positioning task in the building is the longitude x latitude y height z of Household 1703 of Building A11 of XX Community. This embodiment helps to intelligently allocate the resources and priorities of the multi-task learning model by considering the dependencies and complexities between each indoor positioning task at different levels to cope with the requirements of different indoor wireless positioning tasks, and ensure that indoor wireless positioning tasks of different complexities can be properly processed to maximize the overall positioning efficiency.

[0060] It should be noted that in addition to the task sorting method based on complexity, this embodiment can also directly generate a task sorting method according to the order of building positioning tasks, floor positioning tasks and building coordinate positioning tasks, and set the corresponding hierarchical function of the multi-task learning model to determine the execution order of each indoor wireless positioning task.

[0061] The indoor wireless positioning method provided in this embodiment obtains indoor positioning feature data, evaluates the complexity of executing each indoor wireless positioning task based on the supervised learning algorithm, executes each indoor wireless positioning task at different levels based on the multi-task learning model and complexity, and each indoor wireless positioning task can be collaboratively optimized and learned to output indoor wireless positioning results, realize intelligent and accurate indoor wireless positioning, realize reasonable attention to each indoor wireless positioning task, solve the problem that each indoor wireless positioning task is difficult to collaboratively optimize and learn, and improve the optimization learning ability of the multi-task model and the indoor wireless positioning accuracy. In addition, by collecting indoor environment feature data and indoor wireless signal data as indoor positioning feature data for subsequent indoor wireless positioning, the dependence on signal strength data is reduced. In a complex indoor environment, the indoor environment feature data can help indoor wireless positioning better adapt to signal interference and changes, improve the robustness of indoor wireless positioning, and by analyzing the indoor environment feature data, it is possible to more effectively handle multipath effects, improve signal reception quality, and improve positioning accuracy. By conducting supervised learning of each indoor wireless positioning task on the indoor positioning feature data, combined with the supervised learning results and task granularity, the complexity of each indoor wireless positioning task is evaluated, the accuracy of complexity evaluation is improved, and more accurate and effective guidance is provided for the reasonable allocation of subsequent computing resources. By adopting a multi-layered neural network, a multi-task learning model is constructed to conduct hierarchical learning on each indoor wireless positioning task, improve the efficiency of indoor wireless positioning, reduce the overfitting problem of indoor wireless positioning, and significantly improve the accuracy of indoor wireless positioning. Through the hierarchical learning results of indoor wireless positioning tasks with low complexity, it helps indoor wireless positioning tasks with medium or high complexity to extract their own hierarchical learning data from indoor positioning feature data, and then helps improve the hierarchical learning results of indoor wireless positioning tasks with medium or high complexity, which can significantly improve the collaborative optimization learning ability of the multi-task learning model, thereby improving the accuracy and stability of indoor wireless positioning. By considering the dependencies and complexities among various indoor positioning tasks at different levels, it helps to intelligently allocate the resources and priorities of the multi-task learning model to cope with the requirements of different indoor wireless positioning tasks, ensuring that indoor wireless positioning tasks of different complexities can be properly processed to maximize the overall positioning efficiency.

[0062] Embodiment 2:

[0063] like Figure 3 As shown, this embodiment provides an indoor wireless positioning method. The indoor wireless positioning method includes:

[0064] S201, collecting indoor wireless signal data and indoor environment feature data to obtain indoor positioning feature data.

[0065] In this embodiment, Figure 3 The original terminal wireless data in is the indoor wireless signal data.

[0066] Through a signal analyzer, wireless access point (AP) or the mobile terminal itself, wireless signal strength information and signal interference information are collected from the indoor environment to obtain indoor wireless signal data. Through environmental sensors, laser rangefinders or ultrasonic sensors, cameras, computer vision devices or 3D scanners, obstacle information and spatial distribution information are collected from the indoor environment to obtain indoor environment feature data, wherein wireless signals include but are not limited to Wi-Fi signals, Bluetooth signals, RFID (Radio Frequency Identification) signals, and spatial distribution information includes but is not limited to wall distribution information. The indoor environment feature data is transmitted to the indoor wireless positioning system for indoor wireless positioning.

[0067] S202, based on the supervised learning algorithm, classify and regress the indoor positioning feature data to obtain classification and regression results, wherein the classification and regression results refer to the supervised learning results of each indoor wireless positioning task.

[0068] In this embodiment, Figure 3 The single-task learning model in is a model built based on the supervised learning algorithm; Figure 3 Execute each indoor wireless positioning task one by one, that is, classify and regress indoor environment feature data, and obtain classification and regression results; Figure 3 The prediction results in are the classification and regression results.

[0069] The complexity of indoor positioning feature data execution is affected by indoor positioning feature data. For example, in an indoor positioning environment with higher floors and more wireless network sensors deployed, the complexity of building positioning tasks is usually higher than the complexity of floor positioning tasks; while in an indoor positioning environment with lower floors and fewer wireless network sensors deployed, the complexity of building positioning tasks may be lower than the complexity of floor positioning tasks. Therefore, if the complexity of each indoor wireless positioning task is uniformly evaluated at the granularity of conventional tasks, that is, the complexity of building positioning tasks < the complexity of floor positioning tasks < the complexity of coordinate positioning in the building, errors will occur in the complexity of the evaluation.

[0070] It should be noted that the building positioning task is usually a classification task, the building coordinate positioning task is usually a regression task, and the floor positioning task can be either a classification task or a regression task. Under normal circumstances, the complexity of the classification task is lower than the complexity of the regression task. If the floor positioning task is a classification task, when faced with the building positioning task and the floor positioning task, the floor positioning task is a more subtle regression task, and the complexity of the floor positioning task is greater than the complexity of the building positioning task; and if the floor positioning task is a regression task, when faced with the floor positioning task and the building coordinate positioning task, the building coordinate positioning task is a more subtle regression task, and the complexity of the building coordinate positioning task is greater than the complexity of the floor positioning task. Therefore, it can be obtained that the conventional task granularity is the complexity of the building positioning task < the complexity of the floor positioning task < the complexity of the building coordinate positioning task.

[0071] Taking the supervised learning algorithm as the support vector machine algorithm as an example, the support vector machine algorithm is used to learn various indoor wireless positioning tasks, analyze the indoor positioning feature data, and obtain the performance of each indoor wireless positioning task, that is, the classification and regression results.

[0072] It should be noted that the K nearest neighbor algorithm in the prior art is mainly used for classification, while the support vector machine and decision tree can be used for classification and regression tasks at the same time. If the K nearest neighbor algorithm is used to classify and regress indoor positioning feature data, when evaluating each indoor positioning task based on the classification and regression results, the complexity of the floor positioning task and the building positioning task is accurate, and the complexity of the building coordinate positioning task has errors. Therefore, this embodiment usually uses the support vector machine and the decision tree as the supervised learning algorithm, and when the indoor wireless positioning task is a single building positioning task or a single floor positioning task, the K nearest neighbor algorithm is used as the supervised learning algorithm.

[0073] S203: Based on the classification and regression results, evaluate the complexity of each indoor wireless positioning task.

[0074] In this embodiment, Figure 3 The task in is the indoor wireless positioning task.

[0075] When performing the same indoor wireless positioning task based on the same supervised learning algorithm and different indoor positioning feature data, the classification and regression results are better, which means low complexity, and the classification and regression results are worse, which means high complexity. Similarly, when performing different indoor wireless positioning tasks based on the same supervised learning algorithm and the same indoor positioning feature data, the classification and regression results are better, which means low complexity, and the classification and regression results are worse, which means high complexity.

[0076] This embodiment judges the quality of the classification results and regression results by the preset goals of each indoor wireless positioning task. If the classification and regression results meet the preset goals of the building positioning task, the complexity of the building positioning task is determined to be low. Similarly, if the classification and regression results also meet the preset goals of the floor positioning task and the mid-building coordinate positioning task, then the complexity of the floor positioning task and the mid-building coordinate positioning task are both low. However, considering the task granularity of the building positioning task, the floor positioning task and the mid-building coordinate positioning task, that is, the positioning accuracy, when the complexity of the building positioning task, the floor positioning task and the mid-building coordinate positioning task are all low, the complexity of the building positioning task is less than the complexity of the floor positioning task and the complexity of the mid-building coordinate positioning task, that is, the floor positioning task and the mid-building coordinate positioning task are respectively promoted to medium and high. If the classification and regression results do not meet the preset goals of all indoor wireless positioning tasks, the complexity of each indoor positioning task is directly determined according to the task granularity, that is, the complexity of the building positioning task is low, the complexity of the floor positioning task is medium, and the complexity of the mid-building coordinate positioning task is high. If the classification and regression results meet the preset goals of a single indoor wireless positioning task, such as the classification and regression results meet the preset goals of the mid-building coordinate positioning task, the complexity of the mid-building coordinate positioning task is determined to be low, and the complexity of the building positioning task and the floor positioning task is determined based on the task granularity.

[0077] It should be noted that the classification and regression results include at least one of the following: accuracy, precision, recall, F1-score, confusion matrix, mean square error, root mean square error, mean absolute error, R2 (coefficient of determination), ROC curve and AUC value, among which, accuracy refers to the proportion of correctly classified samples to the total samples, precision refers to the proportion of samples correctly predicted as positive to all samples predicted as positive, recall refers to the proportion of samples correctly predicted as positive to all samples actually positive, F1-score refers to the harmonic mean of precision and recall, confusion matrix refers to a matrix showing the classification results, including the number of true positive examples, false positive examples, true negative examples and false negative examples, ROC curve and AUC value refer to the relationship between true positive rate and false positive rate, mean square error refers to the average of the square of the difference between the predicted value and the actual value, root mean square error refers to the square root of the mean square error, mean absolute error refers to the average of the absolute value of the difference between the predicted value and the actual value, and R2 (coefficient of determination) refers to the model's ability to explain data variation.

[0078] S204: sorting the indoor wireless positioning tasks based on the complexity of the indoor wireless positioning tasks.

[0079] In this embodiment, Figure 3 The complexity of multiple indoor positioning tasks is the complexity of each indoor wireless positioning task.

[0080] S205, based on the multi-task learning model, hierarchical learning is performed on the sorted indoor wireless positioning tasks in sequence to output indoor wireless positioning results.

[0081] In this embodiment, Figure 3 The indoor positioning task in is the indoor wireless positioning task.

[0082] Since the multi-task learning model has the powerful ability to learn multiple related tasks, accelerate training, reduce overfitting, and solve complex problems, the multi-task learning model is applied to indoor wireless positioning, which can significantly improve the accuracy of indoor wireless positioning. Taking the multi-layer neural network as CNN (Convolutional Neural Networks), the application scenario of the multi-task learning model is indoor wireless positioning as an example.

[0083] Based on CNN, a multi-task learning model is constructed to perform various indoor wireless positioning tasks, such as Figure 2 As shown, Figure 2 The neural network model in is a multi-task learning model. Figure 2 The backbone of the multi-task learning model is the shared layer, which includes the shared feature extraction functions of each indoor wireless positioning task. Since the parameters θ of the shared feature extraction functions of each indoor wireless positioning task are h Different, the hierarchical learning data extracted from the indoor positioning feature data x for each indoor wireless positioning task is different, Figure 2 The low-level task execution structure is the hierarchical function corresponding to the indoor wireless positioning task with low complexity, the medium-level task execution structure is the hierarchical function corresponding to the indoor wireless positioning task with medium complexity, and the high-level task execution structure is the hierarchical function corresponding to the indoor wireless positioning task with high complexity.

[0084] like Figure 2 As shown, indoor wireless positioning tasks with low complexity can help indoor wireless positioning tasks with medium or high complexity extract shared indoor positioning identification features from indoor positioning feature data, thereby helping to improve the hierarchical learning results of indoor wireless positioning tasks with medium or high complexity. For example, if the complexity of the building positioning task is low, the complexity of the floor positioning task is medium, and the complexity of the coordinate positioning task in the building is high, the hierarchical learning data for the building positioning task, i.e., the building positioning task data set, is directly extracted from the indoor positioning feature data based on the preset shared feature extraction function to perform hierarchical learning of the building positioning task; the parameter θ of the preset shared feature extraction function is optimized based on the hierarchical learning result of the building positioning task. h, a second shared feature extraction function is obtained, and based on the second shared feature extraction function, hierarchical learning data for the floor positioning task, namely, the floor positioning task data set, is extracted from the indoor positioning feature data to perform hierarchical learning of the floor positioning task; similarly, the parameters of the second shared feature extraction function are optimized based on the hierarchical learning result of the floor positioning task to obtain a third shared feature weighting function, and based on the third shared feature extraction function, a mid-building coordinate positioning task data set is extracted from the indoor positioning feature data to perform hierarchical learning of the mid-building coordinate positioning task.

[0085] It should be noted that in addition to optimizing the parameters of the shared feature extraction function to help indoor wireless positioning tasks with medium or high complexity extract their own hierarchical learning data, this embodiment can also filter the indoor positioning feature data through the hierarchical learning results of indoor wireless positioning tasks with low complexity to obtain hierarchical learning data of indoor wireless positioning tasks with medium or high complexity, such as: the indoor wireless positioning task with low complexity is a building positioning task, the indoor wireless positioning task with medium complexity is a floor positioning task, the indoor wireless positioning task with high complexity is a building coordinate positioning task, and the hierarchical learning result of the building positioning task is Building A11 of XX Community. The data belonging to Building A11 of XX Community in the indoor positioning feature data is filtered, and the floor positioning task data set is extracted from the filtered indoor positioning feature data based on the shared feature extraction function.

[0086] like Figure 2 As shown, the hierarchical function of the multi-task learning model sets each indoor wireless positioning task in sequence from low to high according to the complexity of each indoor wireless positioning task, and performs training and learning. The indoor wireless positioning result is the hierarchical learning result of the indoor wireless positioning task with high complexity, such as: the complexity of the building positioning task is low, the complexity of the floor positioning task is medium, and the complexity of the building coordinate positioning task is high. The hierarchical learning result of the building positioning task is Building A11 of XX Community, and the hierarchical learning result of the floor positioning task is Floor 17 of Building A11 of XX Community. Then the indoor wireless positioning result, that is, the hierarchical result of the building coordinate positioning task is the longitude x, latitude y, and altitude z of Household 1703 of Building A11 of XX Community.

[0087] It should be noted that in addition to the task sorting method based on complexity, this embodiment can also directly generate a task sorting method according to the order of building positioning tasks, floor positioning tasks and building coordinate positioning tasks, and set the corresponding hierarchical function of the multi-task learning model to determine the execution order of each indoor wireless positioning task.

[0088] An indoor wireless positioning method provided in this embodiment obtains indoor positioning feature data, evaluates the complexity of executing each indoor wireless positioning task based on the indoor positioning feature data based on a supervised learning algorithm, executes each indoor wireless positioning task at different levels based on a multi-task learning model and complexity, and each indoor wireless positioning task can be collaboratively optimized and learned to output indoor wireless positioning results, thereby realizing intelligent and accurate indoor wireless positioning, achieving reasonable attention to each indoor wireless positioning task, solving the problem that each indoor wireless positioning task is difficult to collaboratively optimize and learn, and improving the optimization learning capability of the multi-task model and thus the indoor wireless positioning accuracy.

[0089] This embodiment 3:

[0090] like Figure 4 As shown, this embodiment also provides an indoor wireless positioning device, including: an acquisition module 31, an evaluation module 32 and a positioning module 33, the acquisition module 31 is used to collect indoor wireless signal data and indoor environment feature data to obtain indoor positioning feature data, the evaluation module 32 is connected to the acquisition module 31, and is used to evaluate the complexity of each indoor wireless positioning task based on a supervised learning algorithm and the indoor positioning feature data, and the positioning module 33 is connected to the evaluation module 32, and is used to perform hierarchical learning on each indoor wireless positioning task based on a multi-task learning model and the complexity to output an indoor wireless positioning result.

[0091] Specifically, the evaluation module 32 includes: a classification and regression unit 321 and an evaluation unit 322. The classification and regression unit 321 is based on a supervised learning algorithm and is used to classify and regress indoor positioning feature data to obtain classification and regression results, wherein the supervised learning algorithm includes any one of the following: support vector machine, K nearest neighbor algorithm, decision tree, the classification and regression results refer to the supervised learning results of each indoor wireless positioning task, and the evaluation unit 322 is used to evaluate the complexity of each indoor wireless positioning task based on the classification and regression results.

[0092] Specifically, the evaluation unit 322 includes: a first judgment subunit, a first response subunit, a second response subunit and a third response subunit. The first judgment subunit is used to judge whether the classification and regression results only meet one of the preset goals of the mid-building coordinate positioning task and the preset goals of the floor positioning task. The first response subunit is used to determine the building positioning task in response to the classification and regression results both meeting the preset goals of the mid-building coordinate positioning task and the preset goals of the floor positioning task, or both not meeting the preset goals of the mid-building coordinate positioning task and the preset goals of the floor positioning task. The complexity is low, the complexity of the floor positioning task is medium, and the complexity of the coordinate positioning task in the building is high. The second response subunit is used to respond to the classification and regression results that only meet the preset target of the coordinate positioning task in the building, and determine that the complexity of the coordinate positioning task in the building is low, the complexity of the floor positioning task is high, and the complexity of the building positioning task is medium. The third response subunit is used to respond to the classification and regression results that only meet the preset target of the floor positioning task, and determine that the complexity of the floor positioning task is low, the complexity of the coordinate positioning task in the building is high, and the complexity of the building positioning task is medium.

[0093] Optionally, the indoor wireless positioning device also includes: a construction module 34, which is used to construct a multi-task learning model based on a multi-level neural network, wherein the multi-level neural network includes any of the following: a convolutional neural network CNN, a recurrent neural network RNN, and a multi-layer perceptron MLP.

[0094] Specifically, the positioning module 33 includes: an extraction unit 331, an input unit 332 and an output unit 333. The extraction unit 331 is used to extract a building positioning task data set, a floor positioning task data set and a building coordinate positioning task data set from the indoor positioning feature data based on the complexity and the shared feature extraction function of the multi-task learning model. The input unit 332 is used to input the building positioning task data set, the floor positioning task data set and the building coordinate positioning task data set into different hierarchical functions of the multi-task learning model based on the complexity and the preset mapping relationship, wherein the preset mapping relationship refers to the mapping relationship between the complexity and the hierarchical function. The output unit 333 is used to perform hierarchical learning on the building positioning task, the floor positioning task and the building coordinate positioning task in turn based on the different hierarchical functions of the multi-task learning model to output the indoor wireless positioning result.

[0095] Specifically, the extraction unit 331 includes: a second judgment subunit and a fourth response subunit, the second judgment subunit is used to judge whether the complexity satisfies the first sorting situation, wherein the first sorting situation refers to that the complexity of the building positioning task is greater than the complexity of the floor positioning task, and the complexity of the floor positioning task is greater than the complexity of the mid-building coordinate positioning task, and the fourth response subunit is used to respond to the complexity satisfying the first sorting situation, based on the first shared feature extraction function, the second shared feature extraction function and the third shared feature extraction function of the multi-task learning model, respectively extract the building positioning task data set, the floor positioning task data set and the building coordinate positioning task data set from the indoor positioning feature data, wherein the first shared feature extraction function refers to a preset shared feature extraction function, the second shared feature extraction function refers to the optimized first shared feature extraction function, and the third shared feature extraction function refers to the optimized second shared feature extraction function.

[0096] It can be understood that the indoor wireless positioning device provided above executes the indoor wireless positioning method corresponding to the embodiment 1 provided above. Therefore, the beneficial effects that can be achieved can refer to the beneficial effects of the scheme corresponding to the indoor wireless positioning method of the embodiment 1 above, and will not be repeated here.

[0097] Embodiment 4:

[0098] This embodiment further provides an electronic device, including a memory and a processor, wherein the memory stores a computer program, and the processor is configured to run the computer program to implement the indoor wireless positioning method for business data in the above-mentioned embodiment 1.

[0099] Embodiment 5:

[0100] This embodiment further provides a computer-readable storage medium on which a computer program is stored. When the computer program is executed by a processor, the indoor wireless positioning method in the above-mentioned embodiment 1 is implemented.

[0101] It is to be understood that the above embodiments are merely exemplary embodiments used to illustrate the principles of the present invention, but the present invention is not limited thereto. For those of ordinary skill in the art, various modifications and improvements can be made without departing from the spirit and essence of the present invention, and these modifications and improvements are also considered to be within the scope of protection of the present invention.

Claims

1. An indoor wireless positioning method, characterized in that: include: Collect indoor wireless signal data and indoor environment feature data to obtain indoor positioning feature data; Based on supervised learning algorithms and indoor positioning feature data, the complexity of each indoor wireless positioning task is evaluated; Based on the multi-task learning model and the complexity, each indoor wireless positioning task is learned hierarchically to output the indoor wireless positioning result.

2. The indoor wireless positioning method according to claim 1, characterized in that: The complexity of each indoor wireless positioning task is evaluated based on the supervised learning algorithm and indoor positioning feature data, specifically including: Based on the supervised learning algorithm, the indoor positioning feature data is classified and regressed to obtain the classification and regression results, wherein the supervised learning algorithm includes any one of the following: support vector machine, K nearest neighbor algorithm, decision tree, and the classification and regression results refer to the supervised learning results of each indoor wireless positioning task; Based on the classification and regression results, the complexity of each indoor wireless positioning task is evaluated.

3. The indoor wireless positioning method according to claim 2, characterized in that: The indoor wireless positioning tasks include building positioning tasks, floor positioning tasks and building coordinate positioning tasks. Based on the classification results and regression results, the complexity of each indoor wireless positioning task is evaluated, specifically including: Determine whether the classification and regression results meet only one of the preset goals of the building coordinate positioning task and the preset goals of the floor positioning task; In response to the classification and regression results both satisfying the preset target of the mid-building coordinate positioning task and the preset target of the floor positioning task, or both not satisfying the preset target of the mid-building coordinate positioning task and the preset target of the floor positioning task, determining that the complexity of the building positioning task is low, the complexity of the floor positioning task is medium, and the complexity of the building coordinate positioning task is high; In response to the classification and regression results only satisfying the preset goal of the mid-building coordinate positioning task, determining that the complexity of the mid-building coordinate positioning task is low, the complexity of the floor positioning task is high, and the complexity of the building positioning task is medium; In response to the classification and regression results only satisfying the preset target of the floor positioning task, the complexity of the floor positioning task is determined to be low, the complexity of the mid-building coordinate positioning task is determined to be high, and the complexity of the building positioning task is determined to be medium.

4. The indoor wireless positioning method according to claim 1, characterized in that: Before performing hierarchical learning on each indoor wireless positioning task based on the multi-task learning model and the complexity to output the indoor wireless positioning result, the method further includes: A multi-task learning model is constructed based on a multi-level neural network, wherein the multi-level neural network includes any of the following: convolutional neural network CNN, recurrent neural network RNN, and multi-layer perceptron MLP.

5. The indoor wireless positioning method according to claim 4, characterized in that: The multi-task learning model includes a shared layer and multiple output layers. Shared layer h = f(x; θ h ), Output layer y t =g t (h; θ t ), where h represents the shared layer of the multi-task learning model, f(·) represents the shared feature extraction function, x represents the input data of the multi-task learning model, and θ h Represents the parameters of the shared layer, y t represents the output layer of the tth task of the multi-task learning model, θ t represents the specific parameters of the tth task, g t (·) represents the hierarchical function of the multi-task learning model for the tth task.

6. The indoor wireless positioning method according to claim 5, characterized in that: The hierarchical learning of each indoor wireless positioning task based on the multi-task learning model and the complexity to output the indoor wireless positioning result specifically includes: Based on the shared feature extraction function of the complexity and multi-task learning model, extracting a building positioning task data set, a floor positioning task data set and a building coordinate positioning task data set from the indoor positioning feature data; Based on the complexity and the preset mapping relationship, the building positioning task data set, the floor positioning task data set and the building coordinate positioning task data set are respectively input into different hierarchical functions of the multi-task learning model, wherein the preset mapping relationship refers to the mapping relationship between the complexity and the hierarchical function; Based on the different hierarchical functions of the multi-task learning model, the building positioning task dataset, the floor positioning task dataset and the building coordinate positioning task dataset are hierarchically learned in turn to output the indoor wireless positioning results.

7. The indoor wireless positioning method according to claim 6, characterized in that: The shared feature extraction function includes a first shared feature extraction function, a second shared feature extraction function and a third shared feature extraction function, The shared feature extraction function based on the complexity and multi-task learning model extracts a building positioning task data set, a floor positioning task data set and a building coordinate positioning task data set from the indoor positioning feature data, specifically including: Determine whether the complexity satisfies a first sorting condition, wherein the first sorting condition means that the complexity of the building positioning task is greater than the complexity of the floor positioning task, and the complexity of the floor positioning task is greater than the complexity of the building coordinate positioning task; In response to the complexity satisfying the first sorting condition, based on the first shared feature extraction function, the second shared feature extraction function and the third shared feature extraction function of the multi-task learning model, respectively extract a building positioning task data set, a floor positioning task data set and a building coordinate positioning task data set from the indoor positioning feature data, Among them, the first shared feature extraction function refers to a preset shared feature extraction function, the second shared feature extraction function refers to the optimized first shared feature extraction function, and the third shared feature extraction function refers to the optimized second shared feature extraction function.

8. An indoor wireless positioning device, characterized in that: include: Acquisition module, evaluation module and positioning module, The acquisition module is used to collect indoor wireless signal data and indoor environment feature data to obtain indoor positioning feature data. The evaluation module is connected to the acquisition module and is used to evaluate the complexity of each indoor wireless positioning task based on the supervised learning algorithm and indoor positioning feature data. The positioning module is connected to the evaluation module and is used to perform hierarchical learning on each indoor wireless positioning task based on the multi-task learning model and the complexity to output the indoor wireless positioning result.

9. An electronic device, characterized in that: It comprises a memory and a processor, wherein the memory stores a computer program, and the processor is configured to run the computer program to implement an indoor wireless positioning method as described in any one of claims 1 to 7.

10. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by a processor, an indoor wireless positioning method as described in any one of claims 1 to 7 is implemented.

Images