CN117998581A

CN117998581A - Wi-Fi fingerprint positioning method and device based on antagonistic attack defense

Info

Publication number: CN117998581A
Application number: CN202410022040.XA
Authority: CN
Inventors: 闫青丽; 熊旺; 代文瑞; 高聪
Original assignee: Xian University of Posts and Telecommunications
Current assignee: Xian University of Posts and Telecommunications
Priority date: 2024-01-05
Filing date: 2024-01-05
Publication date: 2024-05-07

Abstract

The invention relates to the technical field of indoor positioning, in particular to a Wi-Fi fingerprint positioning method and device based on resistance attack defense, wherein the method comprises the following steps: performing resistance attack on RSSI sample data according to a preset resistance attack rule, and reconstructing the RSSI sample data subjected to the resistance attack through a generator to obtain reconstructed sample data; performing similarity discrimination on a plurality of groups of RSSI sample data and reconstructed sample data through a discriminator, and updating parameters of a generator and the discriminator according to discrimination results so that the similarity of the sample data and the reconstructed sample data meets a preset threshold; acquiring Wi-Fi receiving signal intensity information acquired by target equipment, and reconstructing the Wi-Fi receiving signal intensity information through a generator to obtain reconstructed Wi-Fi receiving signal intensity information; and comparing the reconstructed Wi-Fi received signal strength information with a sample database to determine the position information of the target device. The method can resist malicious attack and improve the accuracy of indoor positioning, thereby providing reliable indoor positioning service for users.

Description

Wi-Fi fingerprint positioning method and device based on antagonistic attack defense

Technical Field

The invention relates to the technical field of indoor positioning, in particular to a Wi-Fi fingerprint positioning method and device based on resistance attack defense.

Background

Wi-Fi fingerprint positioning technology utilizes Wi-Fi signal strength or other characteristics to determine the position of a device in space, and is widely applied to indoor environments such as markets, airports, museums and the like to provide indoor navigation and positioning services for users. The indoor positioning system based on Wi-Fi fingerprints mainly utilizes two modes of received signal strength information (RECEIVED SIGNAL STRENGTH Indicator, RSSI) and channel state information (CHANNEL STATE Infor mation, CSI) to carry out coordinate positioning, a positioning technology comprises an offline Wi-Fi fingerprint acquisition stage and an online positioning stage, wi-Fi fingerprints at different positions of an indoor environment are acquired in the offline Wi-Fi fingerprint acquisition stage, the Wi-Fi fingerprints correspond to the positions, a group of data formed by the Wi-Fi fingerprints and the corresponding positions is position fingerprints, and a Wi-Fi fingerprint positioning database is formed by a plurality of position fingerprints. In the online positioning stage, a user submits the currently acquired Wi-Fi fingerprint to a positioning server, and the position corresponding to the position fingerprint with the largest similarity is used as the current position of the user through matching with the position fingerprint of the Wi-Fi fingerprint positioning database.

However, in the process of matching and positioning Wi-Fi fingerprints acquired by the device with the Wi-Fi fingerprint positioning database by the server, accurate positioning is difficult to perform by the server due to the fact that the Wi-Fi fingerprints may be subjected to a resistance attack. Accordingly, there is a need to provide a new indoor positioning method to solve the above-mentioned problems in the related art.

Disclosure of Invention

The present invention is directed to a Wi-Fi fingerprint locating method and apparatus based on resistance attack defense, and further, to overcome at least some of the above-mentioned problems due to the limitations and disadvantages of the related art.

According to one aspect of the invention, there is provided a Wi-Fi fingerprint positioning method based on resistance attack defense, comprising the steps of:

performing resistance attack on RSSI sample data according to a preset resistance attack rule, and reconstructing the RSSI sample data subjected to the resistance attack through a Tra nsformer-based generator to obtain reconstructed sample data;

Performing similarity discrimination on a plurality of groups of RSSI sample data and reconstructed sample data through a discriminator based on a transducer, and updating parameters of the generator based on the transducer and the discriminator based on the transducer according to a discrimination result so that the similarity of the sample data and the reconstructed sample data meets a preset threshold;

acquiring Wi-Fi received signal strength information acquired by target equipment, and reconstructing the Wi-Fi received signal strength information through the Transfor mer-based generator to obtain reconstructed Wi-Fi received signal strength information;

and comparing the reconstructed Wi-Fi received signal strength information with a sample database to determine the position information of the target equipment.

The Wi-Fi fingerprint positioning method based on the resistance attack defense in an embodiment further comprises:

acquiring sample data of a target area, and generating pseudo sample data of the target area based on generation of an countermeasure network according to the sample data;

reconstructing the pseudo sample data through a mapping network to obtain reconstructed sample data of the target area;

and fusing the sample data with the reconstructed sample data to generate a sample database of the target area.

In one embodiment, the preset attack countermeasure rule includes: a rapid gradient descent method, a projection gradient descent method and/or a momentum iteration method.

In one embodiment, the expression of the maximum and minimum countermeasures for performing similarity discrimination on the plurality of groups of RSSI sample data and reconstructed sample data by using a transducer-based discriminator is as follows:

Wherein, θ _G is the weight parameter of the generator, θ _D is the weight parameter of the discriminator, X _adv is the challenge sample data, Reconstruction of sample data output by the generator,/>To reconstruct the discriminative output of the samples, P _G is the feature space of the generator,/>To obey the feature space of the P _G distribution,/>The desire for the feature space of the generator,To determine the expectation that the real data is true, P _data(x) is a clean data prior distribution.

In one embodiment, the Wi-Fi fingerprint positioning method based on the resistance attack defense further comprises:

generating pseudo-sample data by generating a generator in the countermeasure network with the random variable as input data;

and judging a plurality of groups of training data and pseudo sample data by generating a discriminator in the countermeasure network, and adjusting parameters of the generator and the discriminator according to the judging result so that the judging result meets a preset threshold.

In one embodiment, training data in UJIIndoorLoc data sets is obtained, and a generated countermeasure network is trained according to the training data;

And acquiring verification data in the UJIIndoorLoc dataset, and verifying the generated countermeasure network according to the verification data.

In one embodiment the mapping network comprises 8 fully connected layers and is capable of accepting 522-dimensional noise vectors or sample data as input, by which 522-dimensional potential spatial vectors can be generated.

According to another aspect of the present invention, there is provided a Wi-Fi fingerprint positioning device based on a resistance attack defense, comprising:

the anti-attack module is used for carrying out anti-attack on the RSSI sample data according to a preset anti-attack rule, and reconstructing the anti-attack RSSI sample data through a generator based on a transducer to obtain reconstructed sample data;

The generator training module is used for judging the similarity of a plurality of groups of RSSI sample data and reconstructed sample data through a transducer-based discriminator, and updating parameters of the Tra nsformer-based generator and the transducer-based discriminator according to a judging result so that the similarity of the sample data and the reconstructed sample data meets a preset threshold;

The information reconstruction module is used for acquiring Wi-Fi received signal strength information acquired by the target equipment, and reconstructing the Wi-Fi received signal strength information through the generator based on the transducer to acquire reconstructed Wi-Fi received signal strength information;

And the position determining module is used for comparing the reconstructed Wi-Fi received signal strength information with a sample database so as to determine the position information of the target equipment.

According to another aspect of the present invention, there is provided an electronic apparatus including:

at least one processor; and a memory communicatively coupled to the at least one processor; wherein the memory stores instructions executable by the at least one processor to enable the at least one processor to perform the method described above.

According to another aspect of the present invention, there is provided a computer readable storage medium storing a computer program which, when executed by a processor, implements the above method.

The invention provides a Wi-Fi fingerprint positioning method and device based on resistance attack defense, wherein a generator based on a transducer is introduced to serve as a defense mechanism of resistance attack, and RSSI vectors are reconstructed by the generator based on the transducer to eliminate interference caused by resistance sum. The generator based on the transducer has a self-attention mechanism, and can better process complex and diversified signal characteristics, so that the accuracy of indoor positioning is improved, the defending method for resistance attack enables the system to be more stable, malicious attack can be resisted, and reliable indoor positioning service is provided for users. In addition, the method adopts a mapping network and a generation countermeasure network to generate high-quality Wi-Fi signal fingerprint data, thereby reducing the data acquisition cost in an offline stage and increasing the diversity of data. By the combination of mapping the network and generating the antagonism network, more comprehensive data is provided, which is helpful for improving the accuracy of indoor positioning.

Drawings

Fig. 1 is a flowchart of a Wi-Fi fingerprint positioning method based on defensive against a resistance attack in an exemplary embodiment of the present invention;

FIG. 2 is a schematic diagram of a Wi-Fi fingerprint positioning application scenario in an exemplary embodiment of the invention;

FIG. 3 is a schematic diagram of a generating countermeasure network based on a transducer in an exemplary embodiment of the invention;

FIG. 4 is a schematic view of a location division of a target area in an exemplary embodiment of the invention;

FIG. 5 is a schematic diagram of a structure for generating an countermeasure network in an exemplary embodiment of the invention;

FIG. 6 is a schematic diagram of a mapping network in an exemplary embodiment of the invention;

FIG. 7 is a top plan view of a building of UJIIndoorLoc dataset in an exemplary embodiment of the invention;

FIG. 8 is a positioning error contrast chart in an exemplary embodiment of the invention;

Fig. 9 is a schematic structural diagram of a Wi-Fi fingerprint positioning device based on resistance attack defense in an exemplary embodiment of the present invention.

Detailed Description

In order to make the objects, features and advantages of the present invention more obvious and understandable, the technical solutions of the embodiments and examples of the present invention will be clearly and completely described below with reference to the accompanying drawings. However, the example implementations and embodiments may be embodied in many forms and should not be construed as limited to the examples set forth herein; rather, these embodiments and examples are provided so that this disclosure will be thorough and complete, and will fully convey the concept of the example embodiments and examples to those skilled in the art. The described features, structures, or characteristics of the invention may be combined in any suitable manner in one or more embodiments and examples. In the following description, numerous specific details are provided to give a thorough understanding of embodiments and examples of the invention. One skilled in the relevant art will recognize, however, that the invention may be practiced without one or more of the specific details, or with other methods, components, devices, steps, etc. In other instances, well-known aspects have not been shown or described in detail to avoid obscuring aspects of the invention.

Furthermore, the drawings are merely schematic illustrations of the present invention and are not necessarily drawn to scale. The same reference numerals in the drawings denote the same or similar parts, and thus a repetitive description thereof will be omitted. Although the steps of the methods of the present invention are depicted in the accompanying drawings in a particular order, this is not required to or suggested that the steps must be performed in this particular order or that all of the steps shown be performed in order to achieve desirable results. The flow diagrams depicted in the figures are exemplary only and not necessarily all steps are included. For example, some steps may be decomposed, and some steps may be combined or partially combined, so that the order of actual execution may be changed according to actual situations. Some of the block diagrams shown in the figures are functional entities and do not necessarily correspond to physically or logically separate entities. These functional entities may be implemented in software or in one or more hardware modules or integrated circuits or in different networks and/or processor devices and/or microcontroller devices.

Wi-Fi fingerprint positioning technology utilizes Wi-Fi signal strength or other characteristics to determine the position of a device in space, and is widely applied to indoor environments such as markets, airports, museums and the like to provide indoor navigation and positioning services for users. The indoor positioning system based on Wi-Fi fingerprints mainly utilizes two modes of a received signal strength Indicator (RECEIVED SIGNAL STRENGTH Indicator, RSSI) and channel state information (CHANNEL STATE INF ormation, CSI) to carry out coordinate positioning, a positioning technology comprises an offline Wi-Fi fingerprint acquisition stage and an online positioning stage, wi-Fi fingerprints at different positions of an indoor environment are acquired in the offline Wi-Fi fingerprint acquisition stage, the Wi-Fi fingerprints correspond to the positions, a group of data formed by the Wi-Fi fingerprints and the corresponding positions is position fingerprints, and a Wi-Fi fingerprint positioning database is formed by a plurality of position fingerprints. In the online positioning stage, a user submits the currently acquired Wi-Fi fingerprint to a positioning server, and the position corresponding to the position fingerprint with the largest similarity is used as the current position of the user through matching with the position fingerprint of the Wi-Fi fingerprint positioning database. However, in the process of matching and positioning Wi-Fi fingerprints acquired by the device with the Wi-Fi fingerprint positioning database by the server, the Wi-Fi received signal strength information acquired by the device may suffer from a resistance attack, so that positioning errors are increased, robustness is reduced, and even privacy and security are threatened. A resistance attack in a hospital setting may lead to patient positioning tracking errors, threaten patient safety and medical services, and may jeopardize the safety of surgery and treatment; a resistance attack in a mall may cause goods to be incorrectly positioned, affect inventory management and sales, reduce operational efficiency, and have navigation systems affected, thereby confusing customers and reducing shopping experience; the navigation error in the field of automatic driving automobiles can cause traffic accidents, and threatens the safety of drivers and other road users; a resistance attack in a military system may cause the navigation system to fail, increasing the risk of the military being in the field, and even causing the military operation to fail. Therefore, ensuring the robustness of these systems and implementing countermeasure defenses are particularly important, and how to eliminate the countermeasure disturbance to achieve high positioning accuracy is a problem that needs to be solved in indoor positioning technology. In addition, because the positioning accuracy of the indoor positioning method based on wireless fidelity is greatly influenced by the quality of the position fingerprints of the Wi-Fi fingerprint positioning database, the Wi-Fi fingerprints in the position fingerprints are poorer in timeliness, for example, the indoor environment changes such as indoor pattern changes, wi-Fi hot spot position changes and the like can cause Wi-Fi fingerprints to change, so that the positioning accuracy is reduced, the position fingerprints in the Wi-Fi fingerprint positioning database need to be periodically updated, and the updating cost and the updating efficiency are high in a manual mode. Wi-Fi fingerprinting requires the collection of RSSI fingerprint data at different locations in space during the off-line phase to build a fingerprint database, but large-scale RSSI fingerprint data collection typically requires a lot of time and human resources, and requires measurements at different locations, which can result in high data acquisition costs.

Based on the defects in the related art, the invention provides a Wi-Fi fingerprint positioning method and device based on resistance attack defense, and the Wi-Fi fingerprint positioning method and device based on resistance attack defense are used for eliminating interference caused by the total resistance by introducing a generator based on a transducer as a defense mechanism of the resistance attack and reconstructing an RSSI vector by using the generator based on the transducer. The generator based on the transducer has a self-attention mechanism, and can better process complex and diversified signal characteristics, so that the accuracy of indoor positioning is improved, the defending method for resistance attack enables the system to be more stable, malicious attack can be resisted, and reliable indoor positioning service is provided for users. In addition, the method adopts a mapping network and a generation countermeasure network to generate high-quality Wi-Fi signal fingerprint data, thereby reducing the data acquisition cost in an offline stage and increasing the diversity of data. By the combination of mapping the network and generating the antagonism network, more comprehensive data is provided, which is helpful for improving the accuracy of indoor positioning.

The embodiment of the invention provides a Wi-Fi fingerprint positioning method and device based on resistance attack defense, and fig. 1 is a flow diagram of the Wi-Fi fingerprint positioning method based on resistance attack defense in the embodiment of the invention; as shown in fig. 1, the Wi-Fi fingerprint positioning method based on the resistance attack defense comprises the following steps:

Step S11: performing resistance attack on RSSI sample data according to a preset resistance attack rule, and reconstructing the RSSI sample data subjected to the resistance attack through a generator based on a transducer to obtain reconstructed sample data; step S13: performing similarity discrimination on a plurality of groups of RSSI sample data and reconstructed sample data through a discriminator based on a transducer, and updating parameters of the generator based on the transducer and the discriminator based on the transducer according to a discrimination result so that the similarity of the sample data and the reconstructed sample data meets a preset threshold;

An application scenario of the Wi-Fi fingerprint positioning method based on resistance attack defense is shown in fig. 2, a device to be positioned collects Wi-Fi vectors, which are formed by a plurality of Wi-Fi received signal intensity information, at a certain position in a space and sends the Wi-Fi vectors to a server, and the server determines the position information of the device by comparing the Wi-Fi vectors with a fingerprint database. However, since the Wi-Fi vector acquired by the server may be subjected to a resistance attack of the interference signal in the environment, an attacker changes the original signal by introducing noise into the signal, modifying the signal characteristics, changing the signal strength, and the like, so that an error exists between the Wi-Fi vector acquired by the server and the Wi-Fi vector acquired by the device, and a deviation exists in the positioning of the device by the server. The embodiment performs the antagonism defense training on the transducer-based generator through the RSSI sample data so as to eliminate the deviation caused by the antagonism attack in the online positioning.

After the sample database is built in the offline stage of Wi-Fi fingerprinting, the training of the transducer-based generator can be performed with the aid of the sample database. First, RSSI sample data in a sample database is simulated against attacks, and an exemplary way of against attacks may be white-box attacks and black-box attacks such as a fast gradient descent method (FGS M), a projection gradient descent method (PGD), and/or a Momentum Iteration Method (MIM). White-box attacks are attack processes in which, with knowledge of the structure and parameters of the model, disturbances are added to the input samples to cause the model to output erroneous classification results. The samples added with disturbance and causing the model to output the misclassification result are the generated countermeasure samples. The FGSM performs tiny disturbance on input data by calculating a loss function to generate a countermeasure sample, and performs tiny modification on the input RSSI data according to a gradient direction by defining the loss function and calculating a gradient of the loss function on one-dimensional RSSI data to generate the countermeasure sample; PGD is an improvement over FGSM by performing multiple iterations to generate more powerful challenge samples; MIMs introduce momentum to improve the challenge sample generation process to make the attack more stable, by using momentum to keep a constant perturbation in a certain direction to the RSSI data to better bypass the defense mechanism. In the black box attack which is closer to the actual scene, the attacker lacks internal information on the target model, and the black box attack needs to observe the model output, infer the model behavior and generate an countermeasure sample through more calculation and iteration.

Second, since the challenge sample is obtained by adding a disturbance to the original sample, there is a one-to-one correspondence between the challenge sample and the original sample, and thus the transducer-based generator can be trained with multiple challenge samples and the original sample data set as inputs. The generation countermeasure network (TransGAN) based on the transducer is shown in fig. 3, the structure of which comprises a generator based on the transducer and a discriminator based on Transf ormer, the input of the discriminator comprises original sample data from a sample database and reconstructed sample data generated by the generator, the generator has the function of eliminating the influence of the countermeasure attack by generating a new sample which is highly similar to the original data by reconstructing the countermeasure sample after the countermeasure attack, the weight parameters of the generator are updated to find the optimal mapping, the generated reconstructed sample is highly similar to the clean sample, and the generator generates the reconstructed sample data which is highly similar to the original sample data by the repeated game process of the discriminator and the generator. The discriminator identifies and distinguishes the original sample data and the reconstructed sample data, for example, a preset threshold of the discrimination result can be set as a model training standard, for example, when the accuracy of the discriminator in identifying the original sample data and the reconstructed sample data reaches a certain set value, the reconstructed sample data generated by the generator can be considered to overcome the interference of attack resistance and maintain the characteristics of the original sample data. In one embodiment, the expression defining the maximum and minimum countermeasures performed by the arbiter and the generator is:

Wherein, θ _G is the weight parameter of the generator, θ _D is the weight parameter of the discriminator, X _adv is the challenge sample data, Reconstruction of sample data output by the generator,/>To reconstruct the discriminative output of the samples, P _G is the feature space of the generator,/>To obey the feature space of the P _G distribution,/>The desire for the feature space of the generator,To determine the expectation that the real data is true, P _data(x) is a clean data prior distribution. The decision making is performed by a decision making with the maximum and minimum countermeasures of the generator, while training the generator to mitigate small disturbances on the data set, optimizing the decision making to separate the raw clean data from the reconstructed data obtained from the generator. Wherein the weight parameters θ _G and θ _D have the following relationship:

l_g＝α*l_mse+β*l_adv

l _g is defined as the overall loss function, which is a weighted sum of the different loss functions, α and β are used to regulate the effect of the different loss functions on the generator optimization, l _mse is the loss function used to minimize the content error between the reconstructed data and the original data, by minimizing the mean square error to promote the generated data to be closer in content to the original data, by learning the details and structure of the data by the generator, thereby improving the quality of the generated samples. And l _adv is a loss function calculated based on the probability of all the reconstruction data by the discriminator, and the generated sample is more difficult to be identified as falsification by the discriminator through interaction with the discriminator and optimization of parameters of a generator, so that the robustness of resisting the resistance attack is improved. l _d is defined as the loss function of the arbiter.

Step S15: acquiring Wi-Fi receiving signal intensity information acquired by target equipment, and reconstructing the Wi-Fi receiving signal intensity information through a generator based on a Transformer to obtain reconstructed Wi-Fi receiving signal intensity information; step S17: and comparing the reconstructed Wi-Fi received signal strength information with a sample database to determine the position information of the target equipment.

After training the generator based on the transducer is completed, the method can be used for on-line positioning of target equipment in a scene shown in fig. 2, wherein the target equipment can be terminal equipment such as a smart phone, a tablet computer and the like. After the server receives Wi-Fi vectors sent by the target equipment, positioning reference points with similar data are matched from a sample database according to a certain matching algorithm, then the positions of positioning points are estimated according to the reference points, and the currently existing deterministic positioning algorithms comprise DNN, CNN and the like.

Wi-Fi fingerprint positioning requires acquisition of RSSI fingerprint data of different positions to construct a database in an offline stage, more accurate positioning results can be obtained through the constructed database as the acquisition density is higher, however, the RSSI fingerprint data needs to be measured at different positions, and higher acquisition density also leads to higher data acquisition cost. In the embodiment, by means of the neural network model, only a small amount of RSSI fingerprint data is required to be collected, and generation and expansion are carried out through the neural network model, so that the data density of a database is enhanced, and a more accurate positioning effect is achieved. For example, as shown in fig. 4, the target area of the RSSI fingerprint data is divided into the coordinate distribution of 9*9, and the database needs to be built to collect the RSSI fingerprint data of 100 positions, which is heavy in work and high in cost. In this embodiment, the number of collected data is reduced, for example, only 16 positions of RSSI fingerprint data are collected as sample data in three cells, the data collection workload in the same target area is greatly reduced, and the rest 84 positions of RSSI fingerprint data are generated by generating the countermeasure network based on the collected 16 RSSI fingerprint data. The sample data is RSSI fingerprint data, and each RSSI fingerprint data comprises position information of the position and Wi-Fi receiving signal strength.

Generating pseudo sample data of the target area based on the generation countermeasure network based on the acquired data after a small amount of sample data is acquired; generating a countermeasure network (GAN) as shown in fig. 5, a set of random variables (noise vectors) is first defined, mapped to a data space by a mapping network, generated as a mapping result closer to a real feature space, then the feature vectors generated by the mapping network are input to a generator, the input feature vectors are converted into dummy sample data by the generator so as to be similar to the real data as much as possible, and the real data and the dummy sample data generated by the generator are identified by a discriminator so as to discriminate whether the dummy sample data is identical to the real data and is used as sample data.

In one embodiment, the Wi-Fi fingerprint positioning method based on the resistance attack defense further comprises training a generated resistance network, and generating pseudo sample data through a generator in the generated resistance network by taking a random variable as input data; and judging the training data and the pseudo sample data by generating a discriminator in the countermeasure network, and adjusting parameters of the generator and the discriminator according to the judging result so that the judging result meets a preset threshold. Specifically, the generator provides the generated data to the discriminator for evaluation, the discriminator outputs the probability that each sample is a true sample, and the generator updates its own parameters according to the feedback of the discriminator to generate more realistic pseudo samples. Meanwhile, the discriminator updates its own parameters according to the real data and the data generated by the generator to more effectively distinguish the real sample from the generated sample. The optimization process to generate the countermeasure network can be described as a "binary minimum and maximum game" problem, whose mathematical expression is as follows:

Wherein x is real data, z is random noise, G (z) is pseudo sample data output by a generator, D (x) is output by a discriminator, D (G (z)) is a discrimination result of the discriminator on the pseudo sample, In order to determine the expectation that the real data is true,To discriminate the expectation that the dummy sample data is true.

Through collaborative training of the generator and the discriminator, the generator gradually learns the feature space output by the mapping network, generates more and more realistic RSSI data samples, continuously improves the fidelity of the generated data, and simultaneously the discriminator is continuously optimized in the process so as to more accurately discriminate whether the data generated by the generator is real or not, and continuously improves the discrimination capability of the real and the generated data. This iterative training process helps the generator gradually understand the characteristics of the real data and generate data similar to it, thereby improving the quality and quantity of the overall dataset. The generator model parameters and the arbiter model parameters to be used in the present technology are shown in table 1. The ReLU activation function and the batch standardization layer are skillfully applied in the generator, so that necessary nonlinearity is introduced, and the training stability is enhanced. The output layer of the discriminator adopts a Sigmoid activation function, is specially designed for processing two classification tasks, and maps the output range to [0,1], thereby clearly indicating the probability of whether the sample is real or fake. The linear output layer of the generator is selected so that the generated data can have a flexible real value range. Overall, the increase in model depth helps to increase its expressive power, enabling it to learn more effectively complex data representations. Such a model structure may play a key role in tasks such as wireless positioning, and can generate high-quality and realistic data. In the optimization process, an Adam optimizer is selected to update parameters, and the initial learning rate is 0.0001 so as to ensure that the network can perform effective weight adjustment in training.

TABLE 1

In an exemplary embodiment, to further enhance the authenticity of the data, the pseudo-sample data generated based on the generation of the challenge network is reconstructed by the mapping network to ensure that it is closer to the real data in the feature space. Illustratively, the model structure of the mapping network is shown in fig. 6, which includes 8 full connection layers, accepts 522-dimensional noise vector Z or actual RSSI samples (including RSSI vectors and coordinates) as input, and generates 522-dimensional potential spatial vectors through the full connection layers. Alternatively, the loss function employs a Mean Square Error (MSE) in the mapping network, and the optimizer selects Adam, with the batch size set to 128. When the mapping network processes random noise or real RSSI samples, different types of activation functions (including ReLU, linear and leakage ReLU) are skillfully used to successfully map input data to a feature space, so that the quality and fidelity of data generated by a generator are improved, noise in the generated data is reduced, the problem of model overfitting is solved, and the diversity of the generated data is enhanced to be more close to the situation of the real world. The mapping network can reduce the cost of data acquisition and improve the quality of the generated data by generating the synthesized data, thereby providing a more cost-effective data generation solution for the application fields of wireless positioning, tracking and the like, inputting the potential space W generated by the mapping network into the GAN for training, so that the GAN can generate new RSSI data, and further enhancing or expanding the original data set.

In one embodiment, using the disclosed UJIIndoorLoc dataset for RSSI fingerprinting, the UJIInd oorLoc dataset can be used to study and evaluate the development of indoor positioning algorithms covering three buildings of university Jaume I (top view as shown in fig. 7), each building comprising 4 or more floors, with a total area of approximately 110,000 square meters. The data set is intended to support different types of localization tasks, including classification tasks for identification of actual buildings and floors, and regression tasks for estimating actual longitudes and latitudes. UJIIndoorLoc the dataset was created in 2013, engaged by more than 20 different users and 25 Android devices. The dataset comprises two main parts: 19937 training/reference records (trainingdata. Csv file) and 1111 validation/test records (validlationdata. Csv file). These records contain 529 attributes, including WiFi fingerprints, acquisition location coordinates, and other useful information. The WiFi fingerprint is described by the detected wireless access points WAPs and the corresponding received signal strength values RSSI. The received signal strength value is represented by a negative integer ranging from-104 dBm (very bad signal) to 0dBm. If a WAP is not detected, a positive value of 100 is used to indicate this. During the creation of the dataset, a total of 520 different WAPs are detected, so each WiFi fingerprint consists of 520 intensity values.

In a fingerprint localization experiment using UJIIndoorLoc dataset, an experimental result with an average localization error of 7.3 meters was obtained. After the Wi-Fi received signal strength information collected by the device is subjected to the resistance attack by three different white-box attack methods (FGSM rapid gradient descent, PGD projection gradient descent and MIM momentum iteration method), the cumulative distribution of positioning errors of the data after the attack is shown in fig. 8. The impact on the data set is apparent when the deep learning model is subject to a resistance attack. This effect can greatly affect positioning errors, resulting in serious impairment of positioning accuracy.

An exemplary embodiment of the present invention provides a Wi-Fi fingerprint positioning device based on resistance attack defense, and fig. 9 is a schematic structural diagram of a Wi-Fi fingerprint positioning device based on resistance attack defense in an exemplary embodiment of the present invention. Referring to fig. 9, the Wi-Fi fingerprint positioning device based on the resistance attack defense includes:

The attack resistance module 90 is configured to perform an attack resistance on the RSSI sample data according to a preset attack resistance rule, and reconstruct the attack resistance RSSI sample data through a transducer-based generator to obtain reconstructed sample data;

a generator training module 92, configured to perform similarity discrimination on multiple sets of the RSSI sample data and the reconstructed sample data by using a transducer-based discriminator, and update parameters of the transducer-based generator and the transducer-based discriminator according to a discrimination result, so that the similarity between the sample data and the reconstructed sample data meets a preset threshold;

the information reconstruction module 94 is configured to obtain Wi-Fi received signal strength information acquired by the target device, and reconstruct the Wi-Fi received signal strength information through the generator based on the transducer, so as to obtain reconstructed Wi-Fi received signal strength information;

The location determining module 96 is configured to compare the reconstructed Wi-Fi received signal strength information with a sample database to determine location information of the target device.

In the Wi-Fi fingerprint positioning method and device, the comprehensive mapping network and the method for generating the countermeasure network are introduced, so that the acquisition cost is reduced, the accuracy of a model is improved, the quality of generated data is stable, and the performance and the reliability of an indoor positioning system are improved. The use of deep learning models offers various advantages in performing RSSI fingerprinting, but also faces challenges to resistance attacks. The objective of the challenge attack is to intentionally fool the deep learning model, causing it to make false predictions or decisions, thereby affecting the performance and reliability of the positioning system. The indoor positioning resistance attack defense method based on TransGAN aims to solve the problems by taking effective measures under the condition that a dataset is polluted, and has the following beneficial effects: (1) data set quality improvement: the high-quality pseudo tag can be generated through the off-line mapping network and the GAN technology, so that the data acquisition cost is effectively reduced, a more comprehensive and representative data set is constructed, and a better training basis is provided for the model. (2) attack resistance enhancement: transGAN is introduced in the online stage to cope with the attack of the access point, so that the robustness of the model to the attack is enhanced, the reliable performance of the system can be provided even if the data set is polluted in a real-time scene, and the accuracy and stability of indoor positioning are remarkably improved. (3) global information learning: the introduction of the Transformer enables the generator to better capture the global dependency relationship, and compared with the traditional convolution operation, the method can more flexibly process the global correlation and the spatial position information between the objects, improves the learning representation capability of the generating network on the global information, and is particularly important for processing the resistance attack. (4) alleviation of gradient vanishing problem: conventional GAN may face problems such as gradient disappearance during training, and the introduction of a transducer mechanism helps to alleviate these problems, making the network easier to train, and improving the co-training effect of the generator and discriminator. Comprehensively, the technical scheme brings remarkable beneficial effects in the aspects of data processing, attack defense, global information learning and the like by integrating the mapping network, the GAN technology and the transducer, and provides powerful support for improving the performance of the indoor positioning system.

The specific details of each module/unit in the above apparatus are described in the corresponding method section, and are not repeated here. It should be noted that although in the above detailed description several modules or units of a device for action execution are mentioned, such a division is not mandatory. Indeed, the features and functions of two or more modules or units described above may be embodied in one module or unit in accordance with embodiments of the invention. Conversely, the features and functions of one module or unit described above may be further divided into a plurality of modules or units to be embodied.

In addition to the methods and apparatus described above, embodiments of the invention may also be a computer program product comprising computer program instructions which, when executed by a processor, cause the processor to perform steps in a method according to various embodiments of the invention described in the "exemplary methods" section of this specification.

The computer program product may write program code for performing operations of embodiments of the present invention in any combination of one or more programming languages, including an object oriented programming language such as Java, C++ or the like and conventional procedural programming languages, such as the C programming language or similar programming languages. The program code may execute entirely on the user's computing device, partly on the user's device, as a stand-alone software package, partly on the user's computing device, partly on a remote computing device, or entirely on the remote computing device or server.

Another embodiment of the invention provides an electronic device that may be used to perform all or part of the steps of the method described in this example embodiment. The device comprises: at least one processor; and a memory communicatively coupled to the at least one processor; wherein the memory stores instructions executable by the at least one processor to enable the at least one processor to perform steps in a method according to various embodiments of the invention as described in the above "exemplary methods" of the present specification.

Another embodiment of the present invention provides a computer-readable storage medium having stored thereon computer program instructions which, when executed by a processor, cause the processor to perform steps in a method according to various embodiments of the present invention described in the "exemplary method" above in this specification.

The computer readable storage medium may employ any combination of one or more readable media. The readable medium may be a readable signal medium or a readable storage medium. The readable storage medium may include, for example, but is not limited to, an electronic, magnetic, optical, electromagnetic, infrared, or semiconductor system, apparatus, or device, or a combination of any of the foregoing. More specific examples (a non-exhaustive list) of the readable storage medium would include the following: an electrical connection having one or more wires, a portable disk, a hard disk, random Access Memory (RAM), read-only memory (ROM), erasable programmable read-only memory (EPROM or flash memory), optical fiber, portable compact disk read-only memory (CD-ROM), an optical storage device, a magnetic storage device, or any suitable combination of the foregoing.

The basic principles of the present invention have been described above in connection with specific embodiments, but it should be noted that the advantages, benefits, effects, etc. mentioned in the present invention are merely examples and not intended to be limiting, and these advantages, benefits, effects, etc. are not to be construed as necessarily possessed by the various embodiments of the invention. Furthermore, the specific details disclosed herein are for purposes of illustration and understanding only, and are not intended to be limiting, as the invention is not necessarily limited to practice with the above described specific details.

The block diagrams of the devices, apparatuses, devices, systems referred to in the present invention are only illustrative examples and are not intended to require or imply that the connections, arrangements, configurations must be made in the manner shown in the block diagrams. As will be appreciated by one of skill in the art, the devices, apparatuses, devices, systems may be connected, arranged, configured in any manner. Words such as "including," "comprising," "having," and the like are words of openness and mean "including but not limited to," and are used interchangeably therewith. The terms "or" and "as used herein refer to and are used interchangeably with the term" and/or "unless the context clearly indicates otherwise. The term "such as" as used herein refers to, and is used interchangeably with, the phrase "such as, but not limited to.

Other embodiments of the invention will be apparent to those skilled in the art from consideration of the specification and practice of the invention disclosed herein. This invention is intended to cover any variations, uses, or adaptations of the invention following, in general, the principles of the invention and including such departures from the present disclosure as come within known or customary practice within the art to which the invention pertains. It is intended that the specification and examples be considered as exemplary only, with a true scope and spirit of the invention being indicated by the following claims.

It is to be understood that the invention is not limited to the precise arrangements and instrumentalities shown in the drawings, which have been described above, and that various modifications and changes may be effected without departing from the scope thereof. The scope of the invention is limited only by the appended claims.

Claims

1. A Wi-Fi fingerprint positioning method based on resistance attack defense is characterized by comprising the following steps:

2. The Wi-Fi fingerprint positioning method based on resistive attack defense of claim 1, further comprising:

3. The Wi-Fi fingerprint positioning method based on resistive attack defense of claim 1, wherein the preset resistive attack rules comprise: a rapid gradient descent method, a projection gradient descent method and/or a momentum iteration method.

4. The Wi-Fi fingerprint positioning method of claim 1, wherein the expression of the maximum and minimum countermeasures for similarity discrimination of the plurality of sets of the RSSI sample data and the reconstructed sample data by a transducer-based discriminator is:

Wherein, θ _G is the weight parameter of the generator, θ _D is the weight parameter of the discriminator, X _adv is the challenge sample data, Reconstruction of sample data output by the generator,/>To reconstruct the discriminative output of the samples, P _G is the feature space of the generator,/>To obey the feature space of the P _G distribution,/>Desired feature space for generator,/>To determine the expectation that the real data is true, P _data(x) is a clean data prior distribution.

5. The Wi-Fi fingerprint positioning method based on resistive attack defense of claim 2, further comprising:

6. The Wi-Fi fingerprint positioning method based on resistive attack defense of claim 5, wherein training data in a UJIIndoorLoc dataset is obtained, and a generated countermeasure network is trained according to the training data;

7. Wi-Fi fingerprint positioning method based on resistive attack defense according to claim 2, characterized in that the mapping network comprises 8 fully connected layers and can accept 522-dimensional noise vectors or sample data as input, by which 522-dimensional potential space vectors can be generated.

8. Wi-Fi fingerprint locating device based on resistance attack defense, characterized by comprising:

9. An electronic device, comprising:

At least one processor; and a memory communicatively coupled to the at least one processor; wherein the memory stores instructions executable by the at least one processor to enable the at least one processor to perform the Wi-Fi fingerprint locating method based on resistance attack defense of any of claims 1-7.

10. A computer readable storage medium storing a computer program, wherein the computer program when executed by a processor implements the Wi-Fi fingerprint positioning method based on resistive attack defense of any of claims 1-7.