Information processing methods, location fingerprint enhancement model training methods, and communication equipment

Abstract

This application provides an information processing method, a location fingerprint enhancement model training method, and a communication device, relating to the field of data processing technology. The method includes: acquiring a random noise information matrix of L target reference points, wherein the random noise information matrix has a dimension of L rows and M columns, and each row of the random noise information matrix corresponds to a target reference point, where L and M are both positive integers; inputting the random noise information matrix into a trained target location fingerprint enhancement model to obtain a signal strength information enhancement matrix generated by the target location fingerprint enhancement model based on the random noise information matrix, wherein the dimension of the signal strength information enhancement matrix is ​​the same as the dimension of the random noise information matrix; and storing the L target reference points and the signal strength information enhancement matrix in a preset fingerprint database to improve the information enhancement effect.

Description

Technical Field

[0001] This application relates to the field of data processing technology, and in particular to an information processing method, a method for training a location fingerprint enhancement model, and a communication device. Background Technology

[0002] The construction of a fingerprint database (which may include the received signal strength of several collection points) is the key to the positioning accuracy of a fingerprint positioning system. However, in actual indoor positioning scenarios, due to the instability of the signal source of the beacon (mainly the device used to transmit signals in the positioning scenario) and the complex and changeable indoor environment, the actual received signal strength (RSSI) of the positioning terminal fluctuates significantly.

[0003] In indoor positioning scenarios, the common approach to address RSSI signal strength fluctuations is to enhance (or expand) the information samples, thereby increasing the signal strength information to reduce the impact of signal fluctuations. However, the current enhancement (expansion) effect is relatively poor. Summary of the Invention

[0004] This application provides an information processing method, a location fingerprint enhancement model training method, and a communication device to solve the problem of poor information enhancement effects in existing systems.

[0005] To solve the above-mentioned technical problems, this application is implemented as follows:

[0006] In a first aspect, embodiments of this application provide an information processing method, the method comprising:

[0007] Obtain a random noise information matrix for L target reference points. The random noise information matrix has a dimension of L rows and M columns. Each row of the random noise information matrix corresponds to a target reference point. L and M are both positive integers.

[0008] The random noise information matrix is ​​input into the trained target localization fingerprint enhancement model to obtain the signal strength information enhancement matrix generated by the target localization fingerprint enhancement model based on the random noise information matrix. The dimension of the signal strength information enhancement matrix is ​​the same as that of the random noise information matrix.

[0009] The L target reference points are associated with the signal strength information enhancement matrix and stored in a preset fingerprint database.

[0010] Secondly, embodiments of this application provide a method for locating an enhanced fingerprint model, the method comprising:

[0011] Obtain the first signal strength information matrix, which has a dimension of N rows and M columns. Each row of the first signal strength information matrix includes the received signal strength of one of the N reference points for the signals transmitted by M signal transmitting devices. N and M are both positive integers.

[0012] The first signal strength information matrix is ​​transposed to obtain the second signal strength information matrix;

[0013] Based on the first signal strength information matrix and the second signal strength information matrix, the initial localization fingerprint enhancement model is iteratively trained to obtain the target localization fingerprint enhancement model. The generator of the target localization fingerprint enhancement model is used to generate the signal strength information enhancement matrix.

[0014] Thirdly, embodiments of this application provide an information processing apparatus, the apparatus comprising:

[0015] The noise information acquisition module is used to acquire a random noise information matrix of L target reference points. The random noise information matrix has a dimension of L rows and M columns. Each row of the random noise information matrix corresponds to a target reference point. L and M are both positive integers.

[0016] An enhancement matrix acquisition module is used to input the random noise information matrix into a trained target localization fingerprint enhancement model to obtain a signal strength information enhancement matrix generated by the target localization fingerprint enhancement model based on the random noise information matrix. The dimension of the signal strength information enhancement matrix is ​​the same as the dimension of the random noise information matrix.

[0017] The storage module is used to associate and store the L target reference points with the signal strength information enhancement matrix in a preset fingerprint database.

[0018] Fourthly, embodiments of this application provide a fingerprint enhancement model positioning device, the device comprising:

[0019] The first acquisition module is used to acquire a first signal strength information matrix. The first signal strength information matrix has a dimension of N rows and M columns. Each row of the first signal strength information matrix includes the received signal strength of one of the N reference points for the signals transmitted by M signal transmitting devices. N and M are both positive integers.

[0020] The transpose processing module is used to transpose the first signal strength information matrix to obtain the second signal strength information matrix.

[0021] The training module is used to iteratively train the initial localization fingerprint enhancement model based on the first signal strength information matrix and the second signal strength information matrix to obtain the target localization fingerprint enhancement model.

[0022] Fifthly, embodiments of this application provide a communication device, including: a transceiver and a processor.

[0023] The processor is used for:

[0024] Obtain a random noise information matrix for L target reference points. The random noise information matrix has a dimension of L rows and M columns. Each row of the random noise information matrix corresponds to a target reference point. L and M are both positive integers.

[0025] The random noise information matrix is ​​input into the trained target localization fingerprint enhancement model to obtain the signal strength information enhancement matrix generated by the target localization fingerprint enhancement model based on the random noise information matrix. The dimension of the signal strength information enhancement matrix is ​​the same as that of the random noise information matrix.

[0026] The L target reference points are associated with the signal strength information enhancement matrix and stored in a preset fingerprint database.

[0027] Sixthly, embodiments of this application provide a communication device, including: a transceiver and a processor.

[0028] The processor is used for:

[0029] Obtain the first signal strength information matrix, which has a dimension of N rows and M columns. Each row of the first signal strength information matrix includes the received signal strength of one of the N reference points for the signals transmitted by M signal transmitting devices. N and M are both positive integers.

[0030] The first signal strength information matrix is ​​transposed to obtain the second signal strength information matrix;

[0031] Based on the first signal strength information matrix and the second signal strength information matrix, the initial localization fingerprint enhancement model is iteratively trained to obtain the target localization fingerprint enhancement model. The generator of the target localization fingerprint enhancement model is used to generate the signal strength information enhancement matrix.

[0032] In a seventh aspect, embodiments of this application provide a communication device, including: a processor, a memory, and a program stored in the memory and executable on the processor. When the program is executed by the processor, it implements the steps of the information processing method described in the first aspect, or the steps of the location fingerprint enhancement model training method described in the third aspect.

[0033] Eighthly, embodiments of this application provide a computer-readable storage medium storing a computer program, which, when executed by a processor, implements the steps of the information processing method described in the first aspect, or the steps of the location fingerprint enhancement model training method described in the third aspect.

[0034] In the information processing method of this embodiment, a random noise information matrix of L target reference points can be obtained first. A trained target localization fingerprint enhancement model is used to generate a signal strength information enhancement matrix of L target reference points based on the random noise information matrix, thereby enhancing the signal strength information of the target reference points. The L target reference points and the signal strength information enhancement matrix are associated and stored in a preset fingerprint database, which can improve the information enhancement effect. Attached Figure Description

[0035] To more clearly illustrate the technical solutions of the embodiments of this application, the drawings used in the description of the embodiments of this application will be briefly introduced below. Obviously, the drawings described below are only some embodiments of this application. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0036] Figure 1 This is a flowchart of an information processing method provided in an embodiment of this application;

[0037] Figure 2 This is one of the flowcharts of a method for training a localization fingerprint enhancement model provided in the embodiments of this application;

[0038] Figure 3 This is a schematic diagram of an information processing method provided in an embodiment of this application;

[0039] Figure 4 This is one of the signal strength distribution maps provided in the embodiments of this application;

[0040] Figure 5 This is a second signal strength distribution map provided in the embodiments of this application;

[0041] Figure 6 This is the third type of signal strength distribution map provided in the embodiments of this application;

[0042] Figure 7 This is the fourth signal strength distribution map provided in the embodiments of this application;

[0043] Figure 8 This is the second flowchart of a method for training a localization fingerprint enhancement model provided in the embodiments of this application;

[0044] Figure 9 This is a schematic diagram of the structure of an information processing device provided in an embodiment of this application;

[0045] Figure 10 This is one of the structural schematic diagrams of a communication device provided in the embodiments of this application;

[0046] Figure 11 This is a schematic diagram of the structure of a positioning fingerprint enhancement model training device provided in an embodiment of this application;

[0047] Figure 12 This is a second schematic diagram of the structure of a communication device provided in an embodiment of this application. Detailed Implementation

[0048] The technical solutions of the embodiments of this application will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of this application, not all embodiments. Based on the embodiments of this application, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of this application.

[0049] See Figure 1 , Figure 1 This is a flowchart illustrating an information processing method provided in an embodiment of this application, executed by a first communication device, which can be a terminal or a network device. Figure 1 As shown, the information processing method provided in this embodiment includes the following steps:

[0050] Step S101: Obtain the random noise information matrix of L target reference points;

[0051] The random noise information matrix has a dimension of L rows and M columns. Each row in the random noise information matrix corresponds to a target reference point, and L and M are both positive integers.

[0052] Step S102: Input the random noise information matrix into the trained target localization fingerprint enhancement model to obtain the signal strength information enhancement matrix generated by the target localization fingerprint enhancement model based on the random noise information matrix;

[0053] The dimension of the signal strength information enhancement matrix is ​​the same as the dimension of the random noise information matrix;

[0054] Step S103: Associate and store the L target reference points with the signal strength information enhancement matrix in the preset fingerprint database.

[0055] It should be noted that the method in this embodiment can be applied to positioning scenarios, and the aforementioned target reference points are points in the actual positioning scenario. For example, a target reference point can be a point in the positioning scenario where signal strength information has not been collected. A preset fingerprint database (also called a preset positioning fingerprint database) can be used for training a positioning model (used for positioning, also called an artificial intelligence (AI) positioning model) in the positioning scenario. It should be noted that the number of target reference points is L, and each row of random noise information in the random noise information matrix corresponds to one target reference point. The random noise information of L target reference points forms the aforementioned random noise information matrix, meaning that L rows in the random noise information matrix correspond one-to-one with L target reference points. There are many types of target localization fingerprint enhancement models, and no specific limitation is required. For example, in one example, the target localization fingerprint enhancement model can be a target generative adversarial network model. The target generative adversarial network model can include a trained generator and a trained discriminator. Then, the above-mentioned inputting the random noise information matrix into the trained target localization fingerprint enhancement model to obtain the signal strength information enhancement matrix generated by the target localization fingerprint enhancement model based on the random noise information matrix can also include inputting the random noise information matrix into the trained target localization fingerprint enhancement model to obtain the signal strength information enhancement matrix generated by the generator in the target localization fingerprint enhancement model based on the random noise information matrix.

[0056] In the information processing method of this embodiment, a random noise information matrix of L target reference points can be obtained first. A trained target localization fingerprint enhancement model is used to generate a signal strength information enhancement matrix of L target reference points based on the random noise information matrix, thereby enhancing the signal strength information of the target reference points. The L target reference points and the signal strength information enhancement matrix are associated and stored in a preset fingerprint database, which can improve the information enhancement effect.

[0057] In one embodiment, the target localization fingerprint enhancement model is obtained through the following training method:

[0058] Obtain the first signal strength information matrix. The first signal strength information matrix has a dimension of N rows and M columns. Each row of the first signal strength information matrix includes the received signal strength of one of the N reference points for the signals transmitted by M signal transmitting devices. N and M are both positive integers.

[0059] The first signal strength information matrix is ​​transposed to obtain the second signal strength information matrix;

[0060] Based on the first signal strength information matrix and the second signal strength information matrix, the localization fingerprint enhancement model is iteratively trained to obtain the target localization fingerprint enhancement model; the generator of the target localization fingerprint enhancement model is used to generate the signal strength information enhancement matrix.

[0061] It should be noted that the N reference points can be pre-selected reference points, such as N reference points pre-selected in the positioning scenario. M signal transmitting devices can be used to transmit signals. For each of the N reference points, the received signal strength at that reference point relative to the signals transmitted by each of the M signal transmitting devices can be collected. That is, each reference point corresponds to M received signal strengths, one-to-one with the M signal transmitting devices. To a certain extent, the closer the reference point is to the signal transmitting device, the stronger the received signal strength at that reference point relative to the signals transmitted by the signal transmitting device. Each row in the first signal strength information matrix corresponds to the M received signal strengths of a reference point, and each row corresponds to a different reference point. The signal transmitting device can be any device capable of transmitting signals; this embodiment does not limit its specific type. For example, the signal transmitting device can be, but is not limited to, a device with Bluetooth functionality (e.g., an iBeacon device).

[0062] Any row of the first signal strength information matrix can characterize the distribution pattern of the received signal strength of M different signal transmitting devices at a reference point. The transposed second signal strength information matrix is ​​a matrix of size M rows and N columns, where any row of the second signal strength information matrix can characterize the distribution pattern of the received signal strength of N different reference points for the same signal transmitting device.

[0063] By iteratively training the initial localization fingerprint enhancement model using the first and second signal strength information matrices, the network can learn not only the distribution patterns of received signal strength at each reference point, but also the distribution patterns of received signal strength at each reference point of each signal transmitting device, enabling the network to learn the correlation between received signal strengths. Thus, based on the target localization fingerprint enhancement model trained using the first and second signal strength information matrices, a generator for the target localization fingerprint enhancement model can generate a signal strength information enhancement matrix that more closely matches the actual distribution (which may include at least one row of enhanced received signal strength information, i.e., at least M enhanced received signal strength information). This achieves signal strength information enhancement (expansion) while improving the signal strength information enhancement effect.

[0064] In the information processing method of this embodiment, a second signal strength information matrix can be obtained by transposing the first signal strength information matrix. The row information of the first signal strength information matrix can represent the distribution pattern of the received signal strength of each reference point, and the row information of the second signal strength information matrix can represent the distribution pattern of the received signal strength of each reference point of the signal transmitting device. By combining the first signal strength information matrix and the second signal strength information matrix, the initial positioning fingerprint enhancement model is iteratively trained to obtain the target positioning fingerprint enhancement model. That is, the network can be trained using signal strength information matrices that represent different distribution patterns to enhance the information of the trained network and improve the performance of the trained target positioning fingerprint enhancement model. Subsequently, the signal strength information enhancement matrix can be generated by the generator of the target positioning fingerprint enhancement model to achieve information enhancement and improve the information enhancement effect.

[0065] In one embodiment, the initial localization fingerprint enhancement model includes a generator and a discriminator. Based on a first signal strength information matrix and a second signal strength information matrix, the initial localization fingerprint enhancement model is iteratively trained to obtain a target localization fingerprint enhancement model, including:

[0066] The first signal strength information matrix and the second signal strength information matrix are input into the localization fingerprint enhancement model. The generator and the discriminator are cross-trained to obtain the target localization fingerprint enhancement model.

[0067] The generator is used to generate a first generated information matrix based on a first preset information matrix and a second generated information matrix based on a second preset information matrix. The dimension of the first preset information matrix is ​​the same as the dimension of the first signal strength information matrix, and the second preset information matrix is ​​the transpose of the first preset information matrix. The discriminator is used to output the probability that the first signal strength information matrix is ​​the true signal strength information, the probability that the second signal strength information matrix is ​​the true signal strength information, the probability that the first generated information matrix is ​​the true signal strength information, and the probability that the second generated information matrix is ​​the true signal strength information.

[0068] The aforementioned first preset information matrix can be a randomly generated information matrix with the same dimension as the first signal strength information matrix. This matrix may include randomly generated signal strengths, or it can be understood as noise signal strengths, while the signal strengths in the first signal strength information are the actual collected signal strengths. The aforementioned second preset information matrix is ​​the transpose of the first preset information matrix. The first preset information matrix, the second preset information matrix, the first signal strength information matrix, and the second signal strength information matrix can be input into the initial localization fingerprint enhancement model to train the initial localization fingerprint enhancement model. That is, the initial localization fingerprint enhancement model can be trained using the first preset information matrix, the second preset information matrix, the first signal strength information matrix, and the second signal strength information matrix.

[0069] During network training, the generator can generate a first generated information matrix based on a first preset information matrix and a second generated information matrix based on a second preset information matrix. For example, the generator can generate a corresponding row of generated information for each row of preset information in the first preset information matrix, and can generate a corresponding row of generated information for each row of preset information in the second preset information matrix. The dimensions of the first generated information matrix and the first preset information matrix can be the same, and the dimensions of the second generated information matrix and the second preset information matrix can also be the same.

[0070] The first and second generated information matrices generated by the generator can be input into the discriminator. Additionally, the first and second signal strength information matrices can also be input into the discriminator. During network training, the discriminator can output the probabilities that the first signal strength information matrix represents true signal strength information, the second signal strength information matrix represents true signal strength information, the first generated information matrix represents true signal strength information, and the second generated information matrix represents true signal strength information. It should be noted that the probability that the first signal strength information matrix represents true signal strength information includes the probability that each row of the first signal strength information matrix represents true signal strength information, i.e., it includes N probabilities. The probability that the second signal strength information matrix represents true signal strength information includes the probability that each row of the second generated information matrix represents true signal strength information, i.e., it includes M probabilities. The probability that the first generated information matrix represents true signal strength information includes the probability that each row of the first generated information matrix represents true signal strength information. The probability that the second generated information matrix represents true signal strength information includes the probability that each row of the second generated information matrix represents true signal strength information.

[0071] The initial localization fingerprint enhancement model consists of a generator and a discriminator. During network training, the generator and discriminator are cross-trained. That is, the generator is trained once (with its parameters adjustable), then the discriminator is trained once (with its parameters adjustable), then the generator is trained again, and so on, in a cyclical manner, until the training stops, completing the network training and obtaining the target localization fingerprint enhancement model. The sum of the number of iterations for the generator and the discriminator is the total number of iterations for the initial localization fingerprint enhancement model. In essence, the training process of the initial localization fingerprint enhancement model is a cross-training process of the generator and discriminator. Through their game-like learning, they mutually improve their performance, reaching a balance point. By training the generator, it is made capable of generating samples that the discriminator struggles to distinguish as genuine or fake; that is, the generator optimizes its parameters to make its generated samples as likely as possible to be identified as genuine by the discriminator. By training the discriminator, it learns how to distinguish between true and false information, thus improving its ability to differentiate between them. The generator and the discriminator learn from each other through a game until an equilibrium is reached.

[0072] In this embodiment, a signal strength information matrix representing different distribution patterns can be used to cross-train the generator and discriminator of the initial localization fingerprint enhancement model, thereby enhancing the information of the training network and improving the performance of the trained target localization fingerprint enhancement model.

[0073] In one embodiment, the discriminator is trained in the kth iteration by using the kth first target loss, which includes the kth first loss and the kth second loss.

[0074] The k-th first loss is determined at least by the k-th first output and the k-th second output. The k-th first output is the probability of the discriminator outputting based on the k-th first generated information submatrix during the k-th iteration training process. The k-th second output is the probability of the discriminator outputting based on the k-th first signal strength information submatrix during the k-th iteration training process. The k-th first generated information submatrix is ​​the information matrix generated by the generator based on the k-th first information sample matrix during the k-th iteration training process. The k-th first information sample matrix is ​​the submatrix input to the generator in the k-th iteration training of the first preset information matrix. The k-th first signal strength information submatrix is ​​the submatrix input to the discriminator in the k-th iteration training of the first signal strength information matrix. The number of rows in the k-th first information sample matrix and the number of rows in the k-th first signal strength information submatrix are both S, and the number of columns are both M, where S is a positive integer less than or equal to N.

[0075] The k-th second loss is determined at least by the k-th third output and the k-th fourth output. The k-th third output is the probability of the discriminator outputting based on the k-th second generated information sub-matrix during the k-th iteration training process. The k-th fourth output is the probability of the discriminator outputting based on the k-th second signal strength information sub-matrix during the k-th iteration training process. The k-th second generated information sub-matrix is ​​the information matrix generated by the generator based on the k-th second information sample matrix during the k-th iteration training process. The k-th second information sample matrix is ​​the sub-matrix input to the generator in the k-th iteration training of the second preset information matrix. The k-th second signal strength information sub-matrix is ​​the sub-matrix input to the discriminator in the k-th iteration training of the second signal strength information matrix. The number of columns in the k-th second information sample matrix and the number of rows in the k-th second signal strength information sub-matrix are both M, and the number of columns is both S. k is a positive integer less than or equal to the total number of iterations.

[0076] The S mentioned above can be the batch size. In this embodiment, it can be understood as batch processing and training of the training data. For example, in each training process of the discriminator, S rows of information are selected from the first preset information matrix as the first information sample matrix, which includes S sample information row vectors (each sample information row vector has a dimension of M). S rows of information are selected from the first signal strength information matrix as the first signal strength information submatrix, which includes S signal strength information row vectors (each signal strength information row vector has a dimension of M). S columns of information are selected from the second preset information matrix as the second information sample matrix, which includes M sample information row vectors (each sample information row vector has a dimension of S). S columns of information are selected from the second signal strength information matrix as the second signal strength information submatrix, which includes M signal strength information row vectors (each signal strength information row vector has a dimension of S). S can be preset. It should be noted that the k-th iteration training mentioned above is the k-th training of the discriminator.

[0077] The discriminator outputs the probability based on the k-th first generated information sub-matrix, which represents the probability that the k-th first generated information sub-matrix contains true signal strength information. The first generated information sub-matrix is ​​an S-row, M-column matrix, and for each row, a corresponding probability can be output. In other words, the discriminator can output the probability that each row of information in the k-th first generated information sub-matrix contains true signal strength information based on the k-th first generated information sub-matrix. Similarly, the discriminator outputs the probability based on the k-th second generated information sub-matrix, which represents the probability that the k-th second generated information sub-matrix contains true signal strength information. The second generated information sub-matrix is ​​an M-row, S-column matrix, and for each row, a corresponding probability can be output. In other words, the discriminator can output the probability that each row of information in the k-th first generated information sub-matrix contains true signal strength information based on the k-th second generated information sub-matrix.

[0078] The probability output by the discriminator based on the k-th first signal strength information submatrix represents the probability that the k-th first signal strength information submatrix is ​​the true signal strength information. The k-th first signal strength information submatrix is ​​an S-row, M-column matrix, and for each row, a corresponding probability can be output. In other words, the discriminator can output the probability that each row of information in the k-th first signal strength information submatrix is ​​the true signal strength information. Similarly, the probability output by the discriminator based on the k-th second signal strength information submatrix represents the probability that the k-th second signal strength information submatrix is ​​the true signal strength information. The k-th second signal strength information submatrix is ​​an M-row, S-column matrix, and for each row, a corresponding probability can be output. In other words, the discriminator can output the probability that each row of information in the k-th first signal strength information submatrix is ​​the true signal strength information.

[0079] During the training process, the training data is processed in batches, that is, the first preset information matrix, the second preset information matrix, the first signal strength information matrix and the second signal strength information matrix are processed in batches respectively. Each batch is trained using the first information sample matrix, the second information sample matrix, the first signal strength information sub-matrix and the second signal strength information sub-matrix. In each training process, the discriminator is trained using its own first target loss function, that is, the parameters of the discriminator are adjusted using its own first target loss function.

[0080] In this embodiment, during the discriminator training process, the first target loss used in each training iteration includes two parts of loss for this training process: the first loss and the second loss. In one example, the k-th first target loss can be determined based on the k-th first loss and the k-th second loss, and the k-th first target loss is positively correlated with both the k-th first loss and the k-th second loss. In another example, the k-th first target loss can be a weighted sum of the k-th first loss and the k-th second loss. The weights of the first loss and the second loss can be preset based on actual application conditions or historical experience. For example, if the first loss and the second loss are of equal importance, their corresponding weights can be equal, for example, both can be set to 1.

[0081] In this embodiment, the first loss is a loss determined at least by the probability of the discriminator outputting the first generated information submatrix in the kth iteration and the probability of the discriminator outputting the first signal strength information submatrix in the kth iteration. The second loss is a loss determined at least by the discriminator outputting the second generated information submatrix in the kth iteration and the probability of the discriminator outputting the second signal strength information submatrix in the kth iteration. In each training process, the discriminator uses a first target loss that includes both the first loss and the second loss. The discriminator is trained using the first target loss to improve its performance.

[0082] In one embodiment, the initial localization fingerprint enhancement model is a bulldozer distance generative adversarial network WGAN-GP with gradient penalty;

[0083] The k-th first loss includes the k-th first sub-loss determined by the k-th first output and the k-th second output, and the k-th first gradient penalty term. The k-th first sub-loss is the result of subtracting the k-th second expectation from the k-th first expectation. The k-th first output includes S k-th first probabilities, each representing the probability that a row in the k-th first information sample matrix represents the true signal strength information, where S is a positive integer less than or equal to N. The k-th second output includes S k-th second probabilities, each representing the probability that a row in the k-th first signal strength information sub-matrix represents the true signal strength information. The k-th first expectation is the average of the S k-th first probabilities, and the k-th second expectation is the average of the S k-th first gradients. The mean of the two probabilities, the first gradient penalty term of the kth iteration is the mean of H first sub-penalty terms of the kth iteration, the first sub-penalty term of the kth iteration is the first penalty factor multiplied by the square of the first parameter value of the kth iteration, the first parameter value of the kth iteration is the result of taking the norm after the gradient operation of the third probability of the kth iteration minus one, the third probability of the kth iteration is the probability output by the discriminator based on the first sampled signal strength information of the kth iteration during the kth iteration training process, the first sampled signal strength information of the kth iteration is the interpolated sample obtained by interpolating the first information sample matrix and the first signal strength information sub-matrix of the kth iteration during the kth iteration training process, and H is the total number of interpolated samples of the first information sample matrix and the first signal strength information sub-matrix of the kth iteration.

[0084] It can be understood that the S k-th first probabilities include the probability that each row of information in the k-th first information sample matrix is ​​true signal strength information, and the S k-th second probabilities include the probability that each row of information in the k-th first signal strength information sub-matrix is ​​true signal strength information. In this embodiment, H interpolation samplings can be performed to obtain H interpolation sampling samples. Any one of the H interpolation sampling samples can be determined by randomly selecting a row of information in the k-th first information sample matrix, randomly selecting a row of information in the k-th first signal strength information sub-matrix, and interpolation parameters (for example, random interpolation parameters). Thus, during the k-th training process, H interpolation samplings can be performed using the above method to obtain H interpolation sampling samples, that is, H k-th first sampled signal strength information, where H can be an integer greater than 1.

[0085] In one example, the first loss in the kth iteration Determined by the following formula:

[0086]

[0087] in, This represents the information in the a-th row of the k-th first information sample matrix (i.e., the a-th sample in the k-th first information sample matrix). This represents the information in the b-th row of the k-th first signal strength information sub-matrix (i.e., the b-th sample in the k-th first signal strength information sub-matrix). express The probability of the true signal strength information can also be understood as... The corresponding first probability, express The probability of the true signal strength information can also be understood as... The corresponding second probability, This represents the c-th first-sampled signal strength information (c-th interpolated sample) among the H k-th first-sampled signal strength information (i.e., H k-th interpolated sampled signals). The corresponding first sub-penalty term of the kth time, γ1 represents the first penalty factor, express The probability of the true signal strength information can also be understood as... The corresponding third probability, Indicates gradient operation, Indicates to The result of performing gradient operations is the L2 norm.

[0088] In one embodiment, the k-th second loss includes the k-th second sub-loss determined by the k-th third and fourth outputs, and the k-th second gradient penalty term. The k-th second sub-loss is the result of subtracting the k-th fourth expectation from the k-th third expectation. The k-th third output includes M k-th fourth probabilities, each representing the probability that a row in the k-th second information sample matrix represents true signal strength information. The k-th fourth output includes M k-th fifth probabilities, each representing the probability that a row in the k-th second signal strength information sub-matrix represents true signal strength information. The k-th third expectation is the average of the M k-th fourth probabilities, and the k-th fourth expectation is the sum of the M k-th fifth probabilities. The mean of the k-th second gradient penalty term is the mean of h k-th second sub-penalty terms. The k-th second sub-penalty term is the second penalty factor multiplied by the square of the k-th second parameter value. The k-th second parameter value is the result of taking the norm after performing gradient operation on the k-th sixth probability minus one. The k-th sixth probability is the probability output by the discriminator based on the k-th second sampled signal strength information during the k-th iteration training process. The k-th second sampled signal strength information is an interpolated sample obtained by performing one interpolation sampling based on the k-th second information sample matrix and the k-th second signal strength information sub-matrix during the k-th iteration training process. h is the total number of interpolation samplings performed on the k-th second information sample matrix and the k-th second signal strength information sub-matrix.

[0089] In one example, the second loss in the kth iteration Determined by the following formula:

[0090]

[0091] in, This represents the information in the a-th row of the k-th second information sample matrix (i.e., the a-th sample in the k-th second information sample matrix). This represents the information in the b-th row of the k-th second signal strength information sub-matrix (i.e., the b-th sample in the k-th second signal strength information sub-matrix). express The probability of the true signal strength information can also be understood as... The corresponding fourth probability, express The probability of the true signal strength information can also be understood as... The corresponding fifth probability, This represents the c-th second-sampled signal strength information (c-th interpolated sample) out of h k-th second-sampled signal strength information (i.e., h k-th interpolated sampled samples). The corresponding second sub-penalty term in the kth instance, γ2 represents the second penalty factor. express The probability of the true signal strength information can also be understood as... The corresponding sixth probability, Indicates gradient operation, Indicates to The result of the gradient operation is taken as the L2 norm. For example, the first sub-penalty term and the second sub-penalty term can be the same.

[0092] In one example, the interpolated sample can be obtained using the following formula: x^ represents the interpolated sample. This represents a row of information randomly selected from the first information sample matrix / the second information sample matrix (i.e., a sample from the first information sample matrix / a sample from the second information sample matrix). ε represents a random selection of one row of information from the first signal strength information submatrix / the second signal strength information submatrix (i.e., a sample from the first signal strength information submatrix / the second signal strength information submatrix), where ε is the interpolation parameter.

[0093] In one embodiment, during the I-th iteration of generator training, the generator is trained using the I-th second objective loss.

[0094] The I-th second target loss includes the I-th third loss and the I-th fourth loss. The I-th third loss is determined at least through the I-th fifth output, which is the probability of the discriminator outputting based on the I-th first generated information sub-matrix during the I-th iteration training process. The I-th fourth loss is determined at least through the I-th sixth output, which is the probability of the discriminator outputting based on the I-th second generated information sub-matrix during the I-th iteration training process. The I-th first generated information sub-matrix is ​​the information matrix generated by the generator based on the I-th first information sample matrix during the I-th iteration training process. The I-th first information sample matrix is ​​the first preset information matrix... The submatrix input to the generator in the I-th iteration training is: the information matrix generated by the generator based on the second information submatrix in the I-th iteration training. The second information submatrix in the I-th iteration training is: the information matrix generated by the generator based on the second information sample matrix in the I-th iteration training. The second information sample matrix in the I-th iteration training is: the submatrix input to the generator in the I-th iteration training from the second preset information matrix. The first information sample matrix in the I-th iteration has S rows and M columns, where S is a positive integer less than or equal to N. The second information sample matrix in the I-th iteration has M rows and S columns, where I is a positive integer less than or equal to the total number of iterations.

[0095] The aforementioned I-th iteration of training is the I-th training of the generator. The discriminator outputs the probability based on the I-th first generated information submatrix, which represents the probability that the I-th first generated information submatrix contains true signal strength information. For each row of the I-th first generated information submatrix, a corresponding probability can be output. In other words, the discriminator can output the probability that each row of information in the I-th first generated information submatrix contains true signal strength information based on the I-th first generated information submatrix. The discriminator outputs the probability based on the I-th second generated information submatrix, which represents the probability that the I-th second generated information submatrix contains true signal strength information. For each row of the I-th second generated information submatrix, a corresponding probability can be output. In other words, the discriminator can output the probability that each row of information in the I-th first generated information submatrix contains true signal strength information based on the I-th second generated information submatrix.

[0096] The probability output by the discriminator based on the I-th first signal strength information submatrix represents the probability that the I-th first signal strength information submatrix is ​​the true signal strength information. For each row of the I-th first signal strength information submatrix, a corresponding probability can be output. In other words, the discriminator can output the probability that each row of information in the I-th first signal strength information submatrix is ​​the true signal strength information. Similarly, the probability output by the discriminator based on the I-th second signal strength information submatrix represents the probability that the I-th second signal strength information submatrix is ​​the true signal strength information. For each row of the I-th second signal strength information submatrix, a corresponding probability can be output. In other words, the discriminator can output the probability that each row of information in the I-th first signal strength information submatrix is ​​the true signal strength information.

[0097] During the training process, the training data is processed in batches, that is, the first preset information matrix, the second preset information matrix, the first signal strength information matrix and the second signal strength information matrix are processed in batches respectively. Each batch is trained using the first information sample matrix, the second information sample matrix, the first signal strength information sub-matrix and the second signal strength information sub-matrix. In each training process, the discriminator is trained using its own first target loss function, that is, the parameters of the discriminator are adjusted using its own first target loss function.

[0098] In this embodiment, during generator training, the second objective loss used in each training iteration includes two parts of loss for this training process: the third loss and the fourth loss. In one example, the I-th second objective loss can be determined based on the I-th third loss and the I-th fourth loss, and the I-th second objective loss is positively correlated with both the I-th third and I-th fourth losses. In another example, the I-th second objective loss can be a weighted sum of the I-th third and I-th fourth losses. The weights of the third and fourth losses can be preset based on actual application conditions or historical experience. For example, if the third and fourth losses are of equal importance, their corresponding weights can be equal, for example, both can be set to 1.

[0099] In this embodiment, the third loss is determined by the probability of the discriminator outputting the first generated information submatrix in the Ith generation, and the fourth loss is determined by the probability of the discriminator outputting the second generated information submatrix in the Ith generation. In each training process, the discriminator uses the first target loss, which includes the first and second losses mentioned above. The discriminator is trained using the first target loss to improve its performance.

[0100] In one embodiment, the I-th third loss includes the I-th third sub-loss determined by the I-th fifth output, the I-th third sub-loss is the negative of the I-th fifth expectation, the I-th fifth output includes S I-th seventh probabilities, one I-th seventh probability is used to represent the probability that a row of information in the I-th first generated information sub-matrix is ​​the true signal strength information, and the I-th fifth expectation is the mean of the S I-th seventh probabilities.

[0101] The I-th fourth loss includes the I-th fourth sub-loss determined by the I-th sixth output. The I-th fourth sub-loss is the negative of the I-th sixth expectation. The I-th sixth output includes M I-th eighth probabilities. One I-th eighth probability is used to represent the probability that a row of information in the I-th second generated information sub-matrix is ​​the true signal strength information. The I-th sixth expectation is the mean of the M I-th eighth probabilities.

[0102] For example, in one example, the I-th third sub-loss Determined by the following formula:

[0103]

[0104] in, This represents the information in the a-th row of the first generated information submatrix in the I-th generation (i.e., the a-th sample in the first generated information submatrix in the I-th generation). express The probability of the true signal strength information can also be understood as... The corresponding seventh probability.

[0105] For example, in one example, the fourth loss of the Ith time. Determined by the following formula:

[0106]

[0107] in, This represents the information in the b-th row of the i-th second-generation information sub-matrix (i.e., the b-th sample in the i-th second-generation information sub-matrix). express The probability of the true signal strength information can also be understood as... The corresponding eighth probability.

[0108] In one embodiment, the sum of the number of iterations of the generator and the number of iterations of the discriminator is the total number of iterations of the initial localization fingerprint enhancement model.

[0109] In the Tth iteration training of the initial localization fingerprint enhancement model, if the sum of errors in the Tth iteration is less than the preset error, the iteration stops. The total number of iterations for the initial localization fingerprint enhancement model is T, and the sum of errors in the Tth iteration is the sum between the sum of S first sub-errors in the Tth iteration and the sum of M second sub-errors in the Tth iteration.

[0110] Here, a first sub-error of the Tth generation is the distance between a row of information of a reference point in the first generated information sub-matrix of the Tth generation and a row of information of a reference point in the first signal strength information sub-matrix of the Tth generation. The first generated information sub-matrix of the Tth generation is the information matrix generated by the generator based on the first information sample matrix of the Tth generation during the Tth iteration training process. The first information sample matrix of the Tth generation is the sub-matrix input to the generator in the first preset information matrix during the Tth iteration training process. A second sub-error of the Tth generation is the distance between a row of information corresponding to a signal transmitting device in the second generated information sub-matrix of the Tth generation and a row of information corresponding to a signal transmitting device in the second signal strength information sub-matrix of the Tth generation. The second generated information sub-matrix of the Tth generation is the information matrix generated by the generator based on the second information sample matrix of the Tth generation during the Tth iteration training process. The second information sample matrix of the Tth generation is the sub-matrix input to the generator in the second preset information matrix during the Tth iteration training process.

[0111] It should be noted that the above-mentioned Tth iteration training refers to the Tth iteration training of the localization fingerprint enhancement model. For example, if the generator has been trained once and the discriminator has been trained once, then the second training of the generator indicates that it is the third iteration training of the localization fingerprint enhancement model.

[0112] like Figure 2 As shown in the embodiments of this application, a method for training a fingerprint enhancement model is also provided, characterized in that the method includes:

[0113] Step 201: Obtain the first signal strength information matrix;

[0114] The first signal strength information matrix has a dimension of N rows and M columns. Each row of the first signal strength information matrix includes the received signal strength of one of the N reference points for the signals transmitted by M signal transmitting devices. N and M are both positive integers.

[0115] Step 202: Transpose the first signal strength information matrix to obtain the second signal strength information matrix;

[0116] Step 203: Based on the first signal strength information matrix and the second signal strength information matrix, the generative adversarial network is iteratively trained to obtain the target generative adversarial network. The generator of the target generative adversarial network is used to generate the signal strength information enhancement matrix.

[0117] It should be noted that the process of training the fingerprint enhancement model in this embodiment corresponds to the process of training the fingerprint enhancement model in the above information processing method, and will not be repeated here.

[0118] The above method process will be specifically described below with specific embodiments, taking a Bluetooth beacon as an example.

[0119] 1. The overall process for indoor fingerprint database location is as follows:

[0120] The construction of a pre-set fingerprint database is crucial to the positioning accuracy of a fingerprint positioning system. However, in actual indoor positioning scenarios, due to factors such as the instability of the Bluetooth beacon signal source and the complex and variable indoor environment, the RSSI signal strength actually received by the positioning terminal exhibits significant fluctuations. A field test in an indoor environment showed that the range of Bluetooth signal RSSI strength values ​​(maximum value - minimum value) can reach 40+ dB, which will greatly affect positioning accuracy.

[0121] Figure 3 This paper presents an overall process for indoor fingerprint database positioning, including offline and online stages. Currently, the commonly used approach to address RSSI signal strength fluctuations is Solution 1, which enriches the fingerprint database by sampling multiple times at the same fingerprint acquisition reference point to reduce the impact of Bluetooth signal fluctuations. However, this method is costly and time-consuming, especially impractical in large positioning areas, making fingerprint acquisition a major bottleneck hindering the large-scale deployment of indoor fingerprint positioning systems. To address this issue, this application proposes Solution 2, based on an improved WGAN-GP positioning fingerprint enhancement model. This model requires only a single traversal sampling of the indoor positioning scenario to generate diverse fingerprint data, simulating signal strength fluctuations, enriching the fingerprint database, saving fingerprint acquisition time, and improving positioning accuracy.

[0122] 2. Fingerprint data acquisition and preprocessing

[0123] A Bluetooth fingerprint information acquisition tool was used to perform traversal line sampling of the location scene. The sample fingerprint database generated after preprocessing the collected Bluetooth fingerprint information is shown in Table 1. This database contains N fingerprints and M iBeacon Bluetooth beacons. Each row represents the Cartesian coordinates and fingerprint information corresponding to the fingerprint ID, RSSI. ji This represents the average value of all received signal strengths from the i-th iBeacon Bluetooth beacon received at the j-th reference point, where uncollected Bluetooth beacon signal strength values ​​are uniformly filled with any specific value less than -120 (e.g., -1000).

[0124] Table 1 Bluetooth Fingerprint Database

[0125] Fingerprint ID Rectangular coordinates iBeacon1 iBeacon2 iBeacon3 … iBeaconM <![CDATA[ID1]]> <![CDATA[x1,y1]]> <![CDATA[RSSI 11 ]]> <![CDATA[RSSI 12 ]]> <![CDATA[RSSI 13 ]]> … <![CDATA[RSSI 1M <!-- 12 -->]]> <![CDATA[ID2]]> <![CDATA[x2,y2]]> <![CDATA[RSSI 21 ]]> <![CDATA[RSSI 22 ]]> <![CDATA[RSSI 23 ]]> … <![CDATA[RSSI 2M ]]> … … … … … … … <![CDATA[ID N ]]> <![CDATA[x N ,and N ]]> <![CDATA[RSSI N1 ]]> <![CDATA[RSSI N2 ]]> <![CDATA[RSSI N3 ]]> … <![CDATA[RSSI NM ]]>

[0126] Table 1 shows that each row in the fingerprint database represents one sample data point. Therefore, the input to the WGAN-GP fingerprint enhancement model is an N×M matrix. However, compared to other types of data, the fingerprint database differs in that the column data (iBeacon1, iBeacon2, ..., iBeaconM) represented by each iBeacon Bluetooth beacon also exhibits a certain distribution pattern. Figure 4-7 The paper presents heatmaps of signal strength for different Bluetooth beacons. Depending on the beacon's location, the heatmaps exhibit distinct distribution patterns: sample points closer to the source show higher RSSI signal strength values, while those farther from the source show lower values. This indicates that the signal strength of each beacon at each fingerprint acquisition reference point follows a specific distribution pattern, rather than being random. Therefore, this paper improves the WGAN-GP fingerprint enhancement model by using the transpose of the original fingerprint database matrix (M×N matrix) as input. Through training, the model learns not only the signal strength distribution patterns of each beacon at each fingerprint acquisition reference point but also the signal strength distribution patterns of each fingerprint acquisition reference point within each beacon. This allows the model to fully grasp the correlation between the signal strength data in the fingerprint database, thereby generating simulated fingerprint data that more closely reflects the real-world distribution.

[0127] 3. WGAN-GP model network structure

[0128] The original fingerprint database data was used as input data in two ways: N×M(iBeacon_number) and M×N. The input data was processed in batches according to the batch_size setting, with each batch of training data having a size of batch_size×iBeacon_number and iBeacon_number×batch_size, respectively. The loss function for the critic (discriminator) of the improved WGAN-GP model is as follows:

[0129]

[0130]

[0131]

[0132] Where G represents the generator, C represents the critic, x1 is the generator network's noise input data of size batch_size × iBeacon_number (first information sample matrix), x2 is the generator network's noise input data of size iBeacon_number × batch_size (second information sample matrix), real1 is the critic network's real fingerprint database input data of size batch_size × iBeacon_number (first signal strength information sub-matrix), real2 is the critic network's real fingerprint database input data of size iBeacon_number × batch_size (second signal strength information sub-matrix), real_labels are the labels of the real data, set to 1, and fake_labels are the labels of the noise data, set to -1.

[0133] The loss function for the generator is:

[0134]

[0135]

[0136]

[0137] The meanings of G, C, x1, x2, and real_labels are the same as above.

[0138] Based on the input data format, the input-output network structures of the generator and critic in the WGAN-GP model are improved to enable the network to support multi-input multi-output modes. The improved network structures of the generator and critic in the WGAN-GP model are as follows: Figure 8 As shown, it can be understood that the data format of the input model has been improved. Not only is the first signal strength information submatrix collected input, but also its transpose matrix, i.e., the second signal strength information submatrix, input. Thus, the WGAN-GP model can output improved output data according to the improved input data.

[0139] 4. Model Error Evaluation

[0140] In this application, the error between the generated sample and the real sample is calculated using RMSE (Root Mean Square Error), as shown below:

[0141]

[0142] in This represents the signal strength value of the i-th generated sample at the j-th Bluetooth beacon. Let represent the signal strength value of the i-th real sample at the j-th Bluetooth beacon. The total error of the model is defined as the average error of all generated samples. Assuming the model generates N samples in total, the average error is...

[0143] like Figure 9 As shown, Figure 9 This is a schematic diagram of the structure of an information processing device 900 provided in an embodiment of this application, as shown below. Figure 9 As shown, the information processing device 900 includes:

[0144] The noise information acquisition module 901 is used to acquire a random noise information matrix of L target reference points. The random noise information matrix has a dimension of L rows and M columns. Each row in the random noise information matrix corresponds to a target reference point. L and M are both positive integers.

[0145] The enhancement matrix acquisition module 902 is used to input the random noise information matrix into the trained target localization fingerprint enhancement model to obtain the signal strength information enhancement matrix generated by the target localization fingerprint enhancement model based on the random noise information matrix. The dimension of the signal strength information enhancement matrix is ​​the same as that of the random noise information matrix.

[0146] The storage module 903 is used to associate and store L target reference points with the signal strength information enhancement matrix in a preset fingerprint database.

[0147] In one embodiment, the target localization fingerprint enhancement model is obtained through the following training method:

[0148] Obtain the first signal strength information matrix. The first signal strength information matrix has a dimension of N rows and M columns. Each row of the first signal strength information matrix includes the received signal strength of one of the N reference points for the signals transmitted by M signal transmitting devices. N and M are both positive integers.

[0149] The first signal strength information matrix is ​​transposed to obtain the second signal strength information matrix;

[0150] Based on the first signal strength information matrix and the second signal strength information matrix, the initial localization fingerprint enhancement model is iteratively trained to obtain the target localization fingerprint enhancement model.

[0151] In one embodiment, the localization fingerprint enhancement model includes a generator and a discriminator. Based on a first signal strength information matrix and a second signal strength information matrix, the initial localization fingerprint enhancement model is iteratively trained to obtain a target localization fingerprint enhancement model, including:

[0152] The first signal strength information matrix and the second signal strength information matrix are input into the initial localization fingerprint enhancement model. The generator and discriminator are cross-trained to obtain the target localization fingerprint enhancement model.

[0153] The generator is used to generate a first generated information matrix based on a first preset information matrix and a second generated information matrix based on a second preset information matrix. The dimension of the first preset information matrix is ​​the same as the dimension of the first signal strength information matrix, and the second preset information matrix is ​​the transpose of the first preset information matrix. The discriminator is used to output the probability that the first signal strength information matrix is ​​the true signal strength information, the probability that the second signal strength information matrix is ​​the true signal strength information, the probability that the first generated information matrix is ​​the true signal strength information, and the probability that the second generated information matrix is ​​the true signal strength information.

[0154] In one embodiment, during the k-th iteration training of the discriminator, the discriminator is trained using the k-th first target loss, which includes the k-th first loss and the k-th second loss.

[0155] The k-th first loss is determined at least by the k-th first output and the k-th second output. The k-th first output is the probability of the discriminator outputting based on the k-th first generated information submatrix during the k-th iteration training process. The k-th second output is the probability of the discriminator outputting based on the k-th first signal strength information submatrix during the k-th iteration training process. The k-th first generated information submatrix is ​​the information matrix generated by the generator based on the k-th first information sample matrix during the k-th iteration training process. The k-th first information sample matrix is ​​the submatrix input to the generator in the k-th iteration training of the first preset information matrix. The k-th first signal strength information submatrix is ​​the submatrix input to the discriminator in the k-th iteration training of the first signal strength information matrix. The number of rows in the k-th first information sample matrix and the number of rows in the k-th first signal strength information submatrix are both S, and the number of columns are both M, where S is a positive integer less than or equal to N.

[0156] The k-th second loss is determined at least by the k-th third output and the k-th fourth output. The k-th third output is the probability of the discriminator outputting based on the k-th second generated information sub-matrix during the k-th iteration training process. The k-th fourth output is the probability of the discriminator outputting based on the k-th second signal strength information sub-matrix during the k-th iteration training process. The k-th second generated information sub-matrix is ​​the information matrix generated by the generator based on the k-th second information sample matrix during the k-th iteration training process. The k-th second information sample matrix is ​​the sub-matrix input to the generator in the k-th iteration training of the second preset information matrix. The k-th second signal strength information sub-matrix is ​​the sub-matrix input to the discriminator in the k-th iteration training of the second signal strength information matrix. The number of columns in the k-th second information sample matrix and the number of rows in the k-th second signal strength information sub-matrix are both M, and the number of columns is both S. k is a positive integer less than or equal to the total number of iterations.

[0157] In one embodiment, during the I-th iteration of generator training, the generator is trained using the I-th second objective loss.

[0158] The I-th second target loss includes the I-th third loss and the I-th fourth loss. The I-th third loss is determined at least through the I-th fifth output, which is the probability of the discriminator outputting based on the I-th first generated information sub-matrix during the I-th iteration training process. The I-th fourth loss is determined at least through the I-th sixth output, which is the probability of the discriminator outputting based on the I-th second generated information sub-matrix during the I-th iteration training process. The I-th first generated information sub-matrix is ​​the information matrix generated by the generator based on the I-th first information sample matrix during the I-th iteration training process. The I-th first information sample matrix is ​​the first preset information matrix... The submatrix input to the generator in the I-th iteration training is: the information matrix generated by the generator based on the second information submatrix in the I-th iteration training. The second information submatrix in the I-th iteration training is: the information matrix generated by the generator based on the second information sample matrix in the I-th iteration training. The second information sample matrix in the I-th iteration training is: the submatrix input to the generator in the I-th iteration training from the second preset information matrix. The first information sample matrix in the I-th iteration has S rows and M columns, where S is a positive integer less than or equal to N. The second information sample matrix in the I-th iteration has M rows and S columns, where I is a positive integer less than or equal to the total number of iterations.

[0159] In one embodiment, the initial localization fingerprint enhancement model is a bulldozer distance generative adversarial network WGAN-GP with gradient penalty;

[0160] The k-th first loss includes the k-th first sub-loss determined by the k-th first output and the k-th second output, and the k-th first gradient penalty term. The k-th first sub-loss is the result of subtracting the k-th second expectation from the k-th first expectation. The k-th first output includes S k-th first probabilities, each representing the probability that a row in the k-th first information sample matrix represents the true signal strength information, where S is a positive integer less than or equal to N. The k-th second output includes S k-th second probabilities, each representing the probability that a row in the k-th first signal strength information sub-matrix represents the true signal strength information. The k-th first expectation is the average of the S k-th first probabilities, and the k-th second expectation is the average of the S k-th first gradients. The mean of the two probabilities, the first gradient penalty term of the kth iteration is the mean of H first sub-penalty terms of the kth iteration, the first sub-penalty term of the kth iteration is the first penalty factor multiplied by the square of the first parameter value of the kth iteration, the first parameter value of the kth iteration is the result of taking the norm after the gradient operation of the third probability of the kth iteration minus one, the third probability of the kth iteration is the probability output by the discriminator based on the first sampled signal strength information of the kth iteration during the kth iteration training process, the first sampled signal strength information of the kth iteration is the interpolated sample obtained by interpolating the first information sample matrix and the first signal strength information sub-matrix of the kth iteration during the kth iteration training process, and H is the total number of interpolated samples of the first information sample matrix and the first signal strength information sub-matrix of the kth iteration.

[0161] In one embodiment, the k-th second loss includes the k-th second sub-loss determined by the k-th third and fourth outputs, and the k-th second gradient penalty term. The k-th second sub-loss is the result of subtracting the k-th fourth expectation from the k-th third expectation. The k-th third output includes M k-th fourth probabilities, each representing the probability that a row in the k-th second information sample matrix represents true signal strength information. The k-th fourth output includes M k-th fifth probabilities, each representing the probability that a row in the k-th second signal strength information sub-matrix represents true signal strength information. The k-th third expectation is the average of the M k-th fourth probabilities, and the k-th fourth expectation is the sum of the M k-th fifth probabilities. The mean of the k-th second gradient penalty term is the mean of h k-th second sub-penalty terms. The k-th second sub-penalty term is the second penalty factor multiplied by the square of the k-th second parameter value. The k-th second parameter value is the result of taking the norm after performing gradient operation on the k-th sixth probability minus one. The k-th sixth probability is the probability output by the discriminator based on the k-th second sampled signal strength information during the k-th iteration training process. The k-th second sampled signal strength information is an interpolated sample obtained by performing one interpolation sampling based on the k-th second information sample matrix and the k-th second signal strength information sub-matrix during the k-th iteration training process. h is the total number of interpolation samplings performed on the k-th second information sample matrix and the k-th second signal strength information sub-matrix.

[0162] In one embodiment, the sum of the number of iterations of the generator and the number of iterations of the discriminator is the total number of iterations of the initial localization fingerprint enhancement model.

[0163] In the Tth iteration training of the initial localization fingerprint enhancement model, if the sum of errors in the Tth iteration is less than the preset error, the iteration stops. The total number of iterations for the initial localization fingerprint enhancement model is T, and the sum of errors in the Tth iteration is the sum between the sum of S first sub-errors in the Tth iteration and the sum of M second sub-errors in the Tth iteration.

[0164] Here, a first sub-error of the Tth generation is the distance between a row of information of a reference point in the first generated information sub-matrix of the Tth generation and a row of information of a reference point in the first signal strength information sub-matrix of the Tth generation. The first generated information sub-matrix of the Tth generation is the information matrix generated by the generator based on the first information sample matrix of the Tth generation during the Tth iteration training process. The first information sample matrix of the Tth generation is the sub-matrix input to the generator in the first preset information matrix during the Tth iteration training process. A second sub-error of the Tth generation is the distance between a row of information corresponding to a signal transmitting device in the second generated information sub-matrix of the Tth generation and a row of information corresponding to a signal transmitting device in the second signal strength information sub-matrix of the Tth generation. The second generated information sub-matrix of the Tth generation is the information matrix generated by the generator based on the second information sample matrix of the Tth generation during the Tth iteration training process. The second information sample matrix of the Tth generation is the sub-matrix input to the generator in the second preset information matrix during the Tth iteration training process.

[0165] The information processing device 900 provided in this embodiment can implement each process of each embodiment of the above information processing method, with one-to-one correspondence of technical features and can achieve the same technical effect. To avoid repetition, it will not be described again here.

[0166] This application also provides a communication device, including: a processor, a memory, and a program stored in the memory and executable on the processor. When the program is executed by the processor, it implements the various processes of the above-described information processing method embodiments and achieves the same technical effect. To avoid repetition, it will not be described again here.

[0167] For details, see Figure 10 This application also provides a communication device, including a bus 1001, a transceiver 1002, an antenna 1003, a bus interface 1004, a processor 1008, and a memory 1006.

[0168] The processor 1008 is used for:

[0169] Obtain a random noise information matrix for L target reference points. The random noise information matrix has a dimension of L rows and M columns. Each row in the random noise information matrix corresponds to a target reference point. L and M are both positive integers.

[0170] Input the random noise information matrix into the trained target localization fingerprint enhancement model to obtain the signal strength information enhancement matrix generated by the target localization fingerprint enhancement model based on the random noise information matrix. The dimension of the signal strength information enhancement matrix is ​​the same as that of the random noise information matrix.

[0171] The L target reference points are associated with the signal strength information enhancement matrix and stored in a preset fingerprint database.

[0172] In one embodiment, the target localization fingerprint enhancement model is obtained through the following training method:

[0173] Obtain the first signal strength information matrix. The first signal strength information matrix has a dimension of N rows and M columns. Each row of the first signal strength information matrix includes the received signal strength of one of the N reference points for the signals transmitted by M signal transmitting devices. N and M are both positive integers.

[0174] The first signal strength information matrix is ​​transposed to obtain the second signal strength information matrix;

[0175] Based on the first signal strength information matrix and the second signal strength information matrix, the initial localization fingerprint enhancement model is iteratively trained to obtain the target localization fingerprint enhancement model.

[0176] In one embodiment, the initial localization fingerprint enhancement model includes a generator and a discriminator. Based on a first signal strength information matrix and a second signal strength information matrix, the initial localization fingerprint enhancement model is iteratively trained to obtain a target localization fingerprint enhancement model, including:

[0177] The first signal strength information matrix and the second signal strength information matrix are input into the initial localization fingerprint enhancement model. The generator and discriminator are cross-trained to obtain the target localization fingerprint enhancement model.

[0178] The generator is used to generate a first generated information matrix based on a first preset information matrix and a second generated information matrix based on a second preset information matrix. The dimension of the first preset information matrix is ​​the same as the dimension of the first signal strength information matrix, and the second preset information matrix is ​​the transpose of the first preset information matrix. The discriminator is used to output the probability that the first signal strength information matrix is ​​the true signal strength information, the probability that the second signal strength information matrix is ​​the true signal strength information, the probability that the first generated information matrix is ​​the true signal strength information, and the probability that the second generated information matrix is ​​the true signal strength information.

[0179] In one embodiment, during the k-th iteration training of the discriminator, the discriminator is trained using the k-th first target loss, which includes the k-th first loss and the k-th second loss.

[0180] The k-th first loss is determined at least by the k-th first output and the k-th second output. The k-th first output is the probability of the discriminator outputting based on the k-th first generated information submatrix during the k-th iteration training process. The k-th second output is the probability of the discriminator outputting based on the k-th first signal strength information submatrix during the k-th iteration training process. The k-th first generated information submatrix is ​​the information matrix generated by the generator based on the k-th first information sample matrix during the k-th iteration training process. The k-th first information sample matrix is ​​the submatrix input to the generator in the k-th iteration training of the first preset information matrix. The k-th first signal strength information submatrix is ​​the submatrix input to the discriminator in the k-th iteration training of the first signal strength information matrix. The number of rows in the k-th first information sample matrix and the number of rows in the k-th first signal strength information submatrix are both S, and the number of columns are both M, where S is a positive integer less than or equal to N.

[0181] The k-th second loss is determined at least by the k-th third output and the k-th fourth output. The k-th third output is the probability of the discriminator outputting based on the k-th second generated information sub-matrix during the k-th iteration training process. The k-th fourth output is the probability of the discriminator outputting based on the k-th second signal strength information sub-matrix during the k-th iteration training process. The k-th second generated information sub-matrix is ​​the information matrix generated by the generator based on the k-th second information sample matrix during the k-th iteration training process. The k-th second information sample matrix is ​​the sub-matrix input to the generator in the k-th iteration training of the second preset information matrix. The k-th second signal strength information sub-matrix is ​​the sub-matrix input to the discriminator in the k-th iteration training of the second signal strength information matrix. The number of columns in the k-th second information sample matrix and the number of rows in the k-th second signal strength information sub-matrix are both M, and the number of columns is both S. k is a positive integer less than or equal to the total number of iterations.

[0182] In one embodiment, during the I-th iteration of generator training, the generator is trained using the I-th second objective loss.

[0183] The I-th second target loss includes the I-th third loss and the I-th fourth loss. The I-th third loss is determined at least through the I-th fifth output, which is the probability of the discriminator outputting based on the I-th first generated information sub-matrix during the I-th iteration training process. The I-th fourth loss is determined at least through the I-th sixth output, which is the probability of the discriminator outputting based on the I-th second generated information sub-matrix during the I-th iteration training process. The I-th first generated information sub-matrix is ​​the information matrix generated by the generator based on the I-th first information sample matrix during the I-th iteration training process. The I-th first information sample matrix is ​​the first preset information matrix... The submatrix input to the generator in the I-th iteration training is: the information matrix generated by the generator based on the second information submatrix in the I-th iteration training. The second information submatrix in the I-th iteration training is: the information matrix generated by the generator based on the second information sample matrix in the I-th iteration training. The second information sample matrix in the I-th iteration training is: the submatrix input to the generator in the I-th iteration training from the second preset information matrix. The first information sample matrix in the I-th iteration has S rows and M columns, where S is a positive integer less than or equal to N. The second information sample matrix in the I-th iteration has M rows and S columns, where I is a positive integer less than or equal to the total number of iterations.

[0184] In one embodiment, the initial localization fingerprint enhancement model is a bulldozer distance generative adversarial network WGAN-GP with gradient penalty;

[0185] The k-th first loss includes the k-th first sub-loss determined by the k-th first output and the k-th second output, and the k-th first gradient penalty term. The k-th first sub-loss is the result of subtracting the k-th second expectation from the k-th first expectation. The k-th first output includes S k-th first probabilities, each representing the probability that a row in the k-th first information sample matrix represents the true signal strength information, where S is a positive integer less than or equal to N. The k-th second output includes S k-th second probabilities, each representing the probability that a row in the k-th first signal strength information sub-matrix represents the true signal strength information. The k-th first expectation is the average of the S k-th first probabilities, and the k-th second expectation is the average of the S k-th first gradients. The mean of the two probabilities, the first gradient penalty term of the kth iteration is the mean of H first sub-penalty terms of the kth iteration, the first sub-penalty term of the kth iteration is the first penalty factor multiplied by the square of the first parameter value of the kth iteration, the first parameter value of the kth iteration is the result of taking the norm after the gradient operation of the third probability of the kth iteration minus one, the third probability of the kth iteration is the probability output by the discriminator based on the first sampled signal strength information of the kth iteration during the kth iteration training process, the first sampled signal strength information of the kth iteration is the interpolated sample obtained by interpolating the first information sample matrix and the first signal strength information sub-matrix of the kth iteration during the kth iteration training process, and H is the total number of interpolated samples of the first information sample matrix and the first signal strength information sub-matrix of the kth iteration.

[0186] In one embodiment, the k-th second loss includes the k-th second sub-loss determined by the k-th third and fourth outputs, and the k-th second gradient penalty term. The k-th second sub-loss is the result of subtracting the k-th fourth expectation from the k-th third expectation. The k-th third output includes M k-th fourth probabilities, each representing the probability that a row in the k-th second information sample matrix represents true signal strength information. The k-th fourth output includes M k-th fifth probabilities, each representing the probability that a row in the k-th second signal strength information sub-matrix represents true signal strength information. The k-th third expectation is the average of the M k-th fourth probabilities, and the k-th fourth expectation is the sum of the M k-th fifth probabilities. The mean of the k-th second gradient penalty term is the mean of h k-th second sub-penalty terms. The k-th second sub-penalty term is the second penalty factor multiplied by the square of the k-th second parameter value. The k-th second parameter value is the result of taking the norm after performing gradient operation on the k-th sixth probability minus one. The k-th sixth probability is the probability output by the discriminator based on the k-th second sampled signal strength information during the k-th iteration training process. The k-th second sampled signal strength information is an interpolated sample obtained by performing one interpolation sampling based on the k-th second information sample matrix and the k-th second signal strength information sub-matrix during the k-th iteration training process. h is the total number of interpolation samplings performed on the k-th second information sample matrix and the k-th second signal strength information sub-matrix.

[0187] In one embodiment, the sum of the number of iterations of the generator and the number of iterations of the discriminator is the total number of iterations of the initial localization fingerprint enhancement model.

[0188] In the Tth iteration training of the initial localization fingerprint enhancement model, if the sum of errors in the Tth iteration is less than the preset error, the iteration stops. The total number of iterations for the initial localization fingerprint enhancement model is T, and the sum of errors in the Tth iteration is the sum between the sum of S first sub-errors in the Tth iteration and the sum of M second sub-errors in the Tth iteration.

[0189] Here, a first sub-error of the Tth generation is the distance between a row of information of a reference point in the first generated information sub-matrix of the Tth generation and a row of information of a reference point in the first signal strength information sub-matrix of the Tth generation. The first generated information sub-matrix of the Tth generation is the information matrix generated by the generator based on the first information sample matrix of the Tth generation during the Tth iteration training process. The first information sample matrix of the Tth generation is the sub-matrix input to the generator in the first preset information matrix during the Tth iteration training process. A second sub-error of the Tth generation is the distance between a row of information corresponding to a signal transmitting device in the second generated information sub-matrix of the Tth generation and a row of information corresponding to a signal transmitting device in the second signal strength information sub-matrix of the Tth generation. The second generated information sub-matrix of the Tth generation is the information matrix generated by the generator based on the second information sample matrix of the Tth generation during the Tth iteration training process. The second information sample matrix of the Tth generation is the sub-matrix input to the generator in the second preset information matrix during the Tth iteration training process.

[0190] In one embodiment, the processor 1008 is further configured to:

[0191] Obtain the random noise information matrix of the target reference point. The random noise information matrix has a dimension of L rows and M columns, where L is a positive integer.

[0192] The random noise information matrix is ​​input into the target localization fingerprint enhancement model to obtain the signal strength information enhancement matrix generated by the generator based on the random noise information in the target localization fingerprint enhancement model. The dimension of the signal strength information enhancement matrix is ​​the same as that of the random noise information matrix.

[0193] The target reference point is associated with the signal strength information enhancement matrix and stored in a preset fingerprint database.

[0194] exist Figure 10 In this document, a bus architecture (represented by bus 1001) is used. Bus 1001 may include any number of interconnected buses and bridges, linking various circuits including one or more processors represented by processor 1008 and memory represented by memory 1006. Bus 1001 may also link various other circuits such as peripheral devices, voltage regulators, and power management circuits, which are well known in the art and therefore will not be described further herein. Bus interface 1004 provides an interface between bus 1001 and transceiver 1002. Transceiver 1002 may be a single element or multiple elements, such as multiple receivers and transmitters, providing a unit for communicating with various other devices over a transmission medium. Data processed by processor 1008 is transmitted over a wireless medium via antenna 1003, which further receives data and transmits it to processor 1008.

[0195] The processor 1008 is responsible for managing the bus 1001 and general processing, and can also provide various functions, including timing, peripheral interfaces, voltage regulation, power management, and other control functions. The memory 1006 can be used to store the data used by the processor 1008 during operation.

[0196] Optionally, the processor 1008 can be a CPU, ASIC, FPGA, or CPLD.

[0197] This application also provides a computer-readable storage medium storing a computer program. When executed by a processor, the computer program implements the various processes of the above-described information processing method embodiments and achieves the same technical effects. To avoid repetition, it will not be described again here. The computer-readable storage medium may be a read-only memory (ROM), a random access memory (RAM), a magnetic disk, or an optical disk.

[0198] See Figure 11 This application also provides a location fingerprint enhancement model training device 1100, comprising:

[0199] The first acquisition module 1101 is used to acquire the first signal strength information matrix. The first signal strength information matrix has a dimension of N rows and M columns. Each row of the first signal strength information matrix includes the received signal strength of one of the N reference points for the signals transmitted by M signal transmitting devices. N and M are both positive integers.

[0200] The transpose processing module 1102 is used to transpose the first signal strength information matrix to obtain the second signal strength information matrix.

[0201] The training module 1103 is used to iteratively train the localization fingerprint enhancement model based on the first signal strength information matrix and the second signal strength information matrix to obtain the target localization fingerprint enhancement model, which is used to generate the signal strength information enhancement matrix.

[0202] It should be noted that the training process of the aforementioned fingerprint enhancement model training device 1100 corresponds to the training process in the aforementioned information processing device 900, and will not be described again here.

[0203] This application also provides a communication device, including: a processor, a memory, and a program stored in the memory and executable on the processor. When the program is executed by the processor, it implements the various processes of the above-described embodiments of the location fingerprint enhancement model training method and achieves the same technical effect. To avoid repetition, it will not be described again here.

[0204] For details, see Figure 12 This application also provides a communication device, including a bus 1201, a transceiver 1202, an antenna 1203, a bus interface 1204, a processor 1208, and a memory 1206.

[0205] The processor 1208 is used for:

[0206] Obtain the first signal strength information matrix. The first signal strength information matrix has a dimension of N rows and M columns. Each row of the first signal strength information matrix includes the received signal strength of one of the N reference points for the signals transmitted by M signal transmitting devices. N and M are both positive integers.

[0207] The first signal strength information matrix is ​​transposed to obtain the second signal strength information matrix;

[0208] Based on the first signal strength information matrix and the second signal strength information matrix, the initial localization fingerprint enhancement model is iteratively trained to obtain the target localization fingerprint enhancement model, which is used to generate the signal strength information enhancement matrix.

[0209] It should be noted that the iterative training process of processor 1208 corresponds to the iterative training process of processor 1008 mentioned above, and will not be described again.

[0210] exist Figure 12 In this document, a bus architecture (represented by bus 1201) is used. Bus 1201 may include any number of interconnected buses and bridges, linking various circuits including one or more processors represented by processor 1208 and memory represented by memory 1206. Bus 1201 may also link various other circuits such as peripheral devices, voltage regulators, and power management circuits, which are well known in the art and therefore will not be described further herein. Bus interface 1204 provides an interface between bus 1201 and transceiver 1202. Transceiver 1202 may be a single element or multiple elements, such as multiple receivers and transmitters, providing a unit for communicating with various other devices over a transmission medium. Data processed by processor 1208 is transmitted over a wireless medium via antenna 1203, which further receives data and transmits it to processor 1208.

[0211] Processor 1208 is responsible for managing bus 1201 and general processing, and can also provide various functions, including timing, peripheral interface, voltage regulation, power management, and other control functions. Memory 1206 can be used to store data used by processor 1208 during operation.

[0212] Optionally, the processor 1208 can be a CPU, ASIC, FPGA, or CPLD.

[0213] This application also provides a computer-readable storage medium storing a computer program. When executed by a processor, this computer program implements the various processes of the above-described fingerprint enhancement model training method embodiments and achieves the same technical effects. To avoid repetition, it will not be described again here. The computer-readable storage medium may be a read-only memory (ROM), a random access memory (RAM), a magnetic disk, or an optical disk.

[0214] It should be noted that, in this document, the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such a process, method, article, or apparatus. Unless otherwise specified, an element defined by the phrase "comprising one..." does not exclude the presence of other identical elements in the process, method, article, or apparatus that includes that element.

[0215] Through the above description of the embodiments, those skilled in the art can clearly understand that the methods of the above embodiments can be implemented by means of software plus necessary general-purpose hardware platforms. Of course, they can also be implemented by hardware, but in many cases the former is a better implementation method. Based on this understanding, the technical solution of this application, in essence, or the part that contributes to the prior art, can be embodied in the form of a software product. This computer software product is stored in a storage medium (such as ROM / RAM, magnetic disk, optical disk) and includes several instructions to cause a terminal (which may be a mobile phone, computer, server, air conditioner, or network device, etc.) to execute the methods of the various embodiments of this application.

[0216] The embodiments of this application have been described above with reference to the accompanying drawings. However, this application is not limited to the specific embodiments described above. The specific embodiments described above are merely illustrative and not restrictive. Those skilled in the art can make many other forms under the guidance of this application without departing from the spirit and scope of the claims, and all of these forms are within the protection scope of this application.

Claims

1. An information processing method, characterized in that, The method includes: Obtain a random noise information matrix for L target reference points. The random noise information matrix has a dimension of L rows and M columns. Each row of the random noise information matrix corresponds to a target reference point. L and M are both positive integers. The random noise information matrix is ​​input into the trained target localization fingerprint enhancement model to obtain the signal strength information enhancement matrix generated by the target localization fingerprint enhancement model based on the random noise information matrix. The dimension of the signal strength information enhancement matrix is ​​the same as that of the random noise information matrix. The L target reference points are associated with the signal strength information enhancement matrix and stored in a preset fingerprint database; The target localization fingerprint enhancement model is obtained through the following training method: Obtain the first signal strength information matrix, which has a dimension of N rows and M columns. Each row of the first signal strength information matrix includes the received signal strength of one of the N reference points for the signals transmitted by M signal transmitting devices, where N is a positive integer. The first signal strength information matrix is ​​transposed to obtain the second signal strength information matrix; Based on the first signal strength information matrix and the second signal strength information matrix, the initial localization fingerprint enhancement model is iteratively trained to obtain the target localization fingerprint enhancement model, which is used to generate the signal strength information enhancement matrix.

2. The method according to claim 1, characterized in that, The initial localization fingerprint enhancement model includes a generator and a discriminator. The step of iteratively training the initial localization fingerprint enhancement model based on the first signal strength information matrix and the second signal strength information matrix to obtain the target localization fingerprint enhancement model includes: The first signal strength information matrix and the second signal strength information matrix are input into the initial localization fingerprint enhancement model, and the generator and the discriminator are cross-trained to obtain the target localization fingerprint enhancement model. The generator is used to generate a first generated information matrix based on a first preset information matrix and a second generated information matrix based on a second preset information matrix. The dimension of the first preset information matrix is ​​the same as the dimension of the first signal strength information matrix. The second preset information matrix is ​​the transpose of the first preset information matrix. The discriminator is used to output the probability that the first signal strength information matrix is ​​true signal strength information, the probability that the second signal strength information matrix is ​​true signal strength information, the probability that the first generated information matrix is ​​true signal strength information, and the probability that the second generated information matrix is ​​true signal strength information.

3. The method according to claim 2, characterized in that, The discriminator is trained through the k-th iteration, using the k-th first target loss, which includes the k-th first loss and the k-th second loss. The k-th first loss is determined at least by the k-th first output and the k-th second output. The k-th first output is the probability that the discriminator outputs based on the k-th first generated information sub-matrix during the k-th iteration training process. The k-th second output is the probability that the discriminator outputs based on the k-th first signal strength information sub-matrix during the k-th iteration training process. The k-th first generated information sub-matrix is: the information matrix generated by the generator based on the k-th first information sample matrix during the k-th iteration training process. The k-th first information sample matrix is ​​a sub-matrix input to the generator in the k-th iteration training of the first preset information matrix. The k-th first signal strength information sub-matrix is: the sub-matrix input to the discriminator in the k-th iteration training of the first signal strength information matrix. The number of rows in the k-th first information sample matrix and the number of columns in the k-th first signal strength information sub-matrix are both S, and the number of columns is both M, where S is a positive integer less than or equal to N. The k-th second loss is determined at least by the k-th third output and the k-th fourth output. The k-th third output is the probability of the discriminator outputting based on the k-th second generated information sub-matrix during the k-th iteration training process. The k-th fourth output is the probability of the discriminator outputting based on the k-th second signal strength information sub-matrix during the k-th iteration training process. The k-th second generated information sub-matrix is: the information matrix generated by the generator based on the k-th second information sample matrix during the k-th iteration training process. The k-th second information sample matrix is: the sub-matrix input to the generator in the k-th iteration training of the second preset information matrix. The k-th second signal strength information sub-matrix is: the sub-matrix input to the discriminator in the k-th iteration training of the second signal strength information matrix. The number of columns in the k-th second information sample matrix and the number of rows in the k-th second signal strength information sub-matrix are both M, and the number of columns is both S. k is a positive integer less than or equal to the total number of iterations.

4. The method according to claim 2, characterized in that, During the I-th iteration training of the generator, the generator is trained using the I-th second objective loss; The I-th second target loss includes an I-th third loss and an I-th fourth loss. The I-th third loss is determined at least through the I-th fifth output, which is the probability of the discriminator outputting based on the I-th first generated information submatrix during the I-th iteration of training. The I-th fourth loss is determined at least through the I-th sixth output, which is the probability of the discriminator outputting based on the I-th second generated information submatrix during the I-th iteration of training. The first generated information sub-matrix of the I-th iteration is: the information matrix generated by the generator based on the first generated information sample matrix during the I-th iteration training process. The first generated information sample matrix is: the sub-matrix input to the generator in the I-th iteration training within the first preset information matrix. The second generated information sub-matrix of the I-th iteration is: the information matrix generated by the generator based on the second generated information sub-matrix during the I-th iteration training process. The second generated information sub-matrix of the I-th iteration is: the information matrix generated by the generator based on the second generated information sample matrix during the I-th iteration training process. The second generated information sample matrix is: the sub-matrix input to the generator in the I-th iteration training within the second preset information matrix. The first generated information sample matrix has S rows and M columns, where S is a positive integer less than or equal to N. The second generated information sample matrix has M rows and S columns, where I is a positive integer less than or equal to the total number of iterations.

5. The method according to claim 3, characterized in that, The initial localization fingerprint enhancement model is a bulldozer distance generative adversarial network WGAN-GP with gradient penalty; The k-th first loss includes the k-th first sub-loss determined by the k-th first output and the k-th second output, and the k-th first gradient penalty term. The k-th first sub-loss is the result of subtracting the k-th second expectation from the k-th first expectation. The k-th first output includes S k-th first probabilities, each representing the probability that a row in the k-th first information sample matrix is ​​true signal strength information, where S is a positive integer less than or equal to N. The k-th second output includes S k-th second probabilities, each representing the probability that a row in the k-th first signal strength information sub-matrix is ​​true signal strength information. The k-th first expectation is the mean of the S k-th first probabilities, and the k-th second expectation is the mean of the S k-th second... The mean of the probabilities, the k-th first gradient penalty term is the mean of H k-th first sub-penalty terms, the k-th first sub-penalty term is the first penalty factor multiplied by the square of the k-th first parameter value, the k-th first parameter value is the result of taking the norm after performing gradient operation on the k-th third probability minus one, the k-th third probability is the probability output by the discriminator based on the k-th first sampled signal strength information during the k-th iteration training process, the k-th first sampled signal strength information is an interpolated sample obtained by performing one interpolation sampling based on the k-th first information sample matrix and the k-th first signal strength information sub-matrix during the k-th iteration training process, and H is the total number of interpolation samplings performed on the k-th first information sample matrix and the k-th first signal strength information sub-matrix.

6. The method according to claim 3 or 5, characterized in that, The k-th second loss includes the k-th second sub-loss determined by the k-th third output and the k-th fourth output, and the k-th second gradient penalty term. The k-th second sub-loss is the result of subtracting the k-th fourth expectation from the k-th third expectation. The k-th third output includes M k-th fourth probabilities, each representing the probability that a row in the k-th second information sample matrix represents true signal strength information. The k-th fourth output includes M k-th fifth probabilities, each representing the probability that a row in the k-th second signal strength information sub-matrix represents true signal strength information. The k-th third expectation is the mean of the M k-th fourth probabilities, and the k-th fourth expectation is the mean of the M k-th fifth probabilities. The k-th second gradient penalty term is the mean of h k-th second sub-penalty terms. The k-th second sub-penalty term is the second penalty factor multiplied by the square of the k-th second parameter value. The k-th second parameter value is the result of taking the norm after performing gradient operation on the k-th sixth probability minus one. The k-th sixth probability is the probability output by the discriminator based on the k-th second sampled signal strength information during the k-th iteration training process. The k-th second sampled signal strength information is an interpolated sample obtained by performing one interpolation sampling based on the k-th second information sample matrix and the k-th second signal strength information sub-matrix during the k-th iteration training process. h is the total number of interpolation samplings performed on the k-th second information sample matrix and the k-th second signal strength information sub-matrix.

7. The method according to claim 2, characterized in that, The sum of the number of iterations for training the generator and the number of iterations for training the discriminator is the total number of iterations for training the initial localization fingerprint enhancement model; In the Tth iteration training of the initial localization fingerprint enhancement model, if the sum of errors in the Tth iteration is less than a preset error, the iteration stops. The total number of iterations for the initial localization fingerprint enhancement model is T. The sum of errors in the Tth iteration is the sum between the sum of S first sub-errors in the Tth iteration and the sum of M second sub-errors in the Tth iteration. Wherein, a first sub-error of the Tth generation is the distance between a row of information of a reference point in the first generated information sub-matrix of the Tth generation and a row of information of the same reference point in the first signal strength information sub-matrix of the Tth generation. The first generated information sub-matrix of the Tth generation is an information matrix generated by the generator based on the first information sample matrix of the Tth generation during the Tth iteration training process. The first information sample matrix of the Tth generation is a sub-matrix of the first preset information matrix input to the generator in the Tth iteration training process. A second sub-error of the Tth generation is the distance between a row of information corresponding to a signal transmitting device in the second generated information sub-matrix of the Tth generation and a row of information corresponding to the same signal transmitting device in the second signal strength information sub-matrix of the Tth generation. The second generated information sub-matrix of the Tth generation is an information matrix generated by the generator based on the second information sample matrix of the Tth generation during the Tth iteration training process. The second information sample matrix of the Tth generation is a sub-matrix of the second preset information matrix input to the generator in the Tth iteration training process.

8. A method for training a fingerprint enhancement model, characterized in that, The method includes: Obtain the first signal strength information matrix, which has a dimension of N rows and M columns. Each row of the first signal strength information matrix includes the received signal strength of one of the N reference points for the signals transmitted by M signal transmitting devices. N and M are both positive integers. The first signal strength information matrix is ​​transposed to obtain the second signal strength information matrix; Based on the first signal strength information matrix and the second signal strength information matrix, the generative adversarial network is iteratively trained to obtain the target generative adversarial network. The generator of the target generative adversarial network is used to generate the signal strength information enhancement matrix.

9. An information processing device, characterized in that, The device includes: The noise information acquisition module is used to acquire a random noise information matrix of L target reference points. The random noise information matrix has a dimension of L rows and M columns. Each row of the random noise information matrix corresponds to a target reference point. L and M are both positive integers. An enhancement matrix acquisition module is used to input the random noise information matrix into a trained target localization fingerprint enhancement model to obtain a signal strength information enhancement matrix generated by the target localization fingerprint enhancement model based on the random noise information matrix. The dimension of the signal strength information enhancement matrix is ​​the same as the dimension of the random noise information matrix. The storage module is used to associate and store the L target reference points with the signal strength information enhancement matrix in a preset fingerprint database; The target localization fingerprint enhancement model is obtained through the following training method: Obtain the first signal strength information matrix, which has a dimension of N rows and M columns. Each row of the first signal strength information matrix includes the received signal strength of one of the N reference points for the signals transmitted by M signal transmitting devices, where N is a positive integer. The first signal strength information matrix is ​​transposed to obtain the second signal strength information matrix; Based on the first signal strength information matrix and the second signal strength information matrix, the initial localization fingerprint enhancement model is iteratively trained to obtain the target localization fingerprint enhancement model, which is used to generate the signal strength information enhancement matrix.

10. A training device for a localization fingerprint enhancement model, characterized in that, The device includes: The first acquisition module is used to acquire a first signal strength information matrix. The first signal strength information matrix has a dimension of N rows and M columns. Each row of the first signal strength information matrix includes the received signal strength of one of the N reference points for the signals transmitted by M signal transmitting devices. N and M are both positive integers. The transpose processing module is used to transpose the first signal strength information matrix to obtain the second signal strength information matrix. The training module is used to iteratively train the initial localization fingerprint enhancement model based on the first signal strength information matrix and the second signal strength information matrix to obtain the target localization fingerprint enhancement model, which is used to generate the signal strength information enhancement matrix.

11. A communication device, characterized in that, Including transceivers and processors, The processor is used for: Obtain a random noise information matrix for L target reference points. The random noise information matrix has a dimension of L rows and M columns. Each row of the random noise information matrix corresponds to a target reference point. L and M are both positive integers. The random noise information matrix is ​​input into the trained target localization fingerprint enhancement model to obtain the signal strength information enhancement matrix generated by the target localization fingerprint enhancement model based on the random noise information matrix. The dimension of the signal strength information enhancement matrix is ​​the same as that of the random noise information matrix. The L target reference points are associated with the signal strength information enhancement matrix and stored in a preset fingerprint database; The target localization fingerprint enhancement model is obtained through the following training method: Obtain the first signal strength information matrix, which has a dimension of N rows and M columns. Each row of the first signal strength information matrix includes the received signal strength of one of the N reference points for the signals transmitted by M signal transmitting devices, where N is a positive integer. The first signal strength information matrix is ​​transposed to obtain the second signal strength information matrix; Based on the first signal strength information matrix and the second signal strength information matrix, the initial localization fingerprint enhancement model is iteratively trained to obtain the target localization fingerprint enhancement model, which is used to generate the signal strength information enhancement matrix.

12. A communication device, characterized in that, Including transceivers and processors, The processor is used for: Obtain the first signal strength information matrix, which has a dimension of N rows and M columns. Each row of the first signal strength information matrix includes the received signal strength of one of the N reference points for the signals transmitted by M signal transmitting devices. N and M are both positive integers. The first signal strength information matrix is ​​transposed to obtain the second signal strength information matrix; Based on the first signal strength information matrix and the second signal strength information matrix, the initial localization fingerprint enhancement model is iteratively trained to obtain the target localization fingerprint enhancement model, which is used to generate the signal strength information enhancement matrix.

13. A communication device, characterized in that, include: A processor, a memory, and a program stored in the memory and executable on the processor, wherein the program, when executed by the processor, implements the steps of the method as described in any one of claims 1 to 8.

14. A computer-readable storage medium having a computer program stored thereon, the computer program being executed by a processor to implement the steps of the method of any one of claims 1-8.

Images