RFID indoor positioning method and system based on hybrid neural network model

Abstract

The present invention proposes an RFID indoor positioning method and system based on a hybrid neural network model, wherein the method includes obtaining an original signal and corresponding position information, and performing differential calculation on the original signal to obtain a corresponding differential signal; establishing a hybrid neural network model, wherein the hybrid neural network model includes a bidirectional differential gated recurrent unit, an adaptive sparse slot attention module, and a Transformer encoder-decoder; inputting the original signal, the corresponding position information, and the corresponding differential signal into the hybrid neural network model to train the hybrid neural network model to obtain a trained indoor positioning model; obtaining a tag signal to be located and its corresponding differential signal, and inputting them into the indoor positioning model to obtain the corresponding position information; thereby, by combining the bidirectional differential gated recurrent unit with the Transformer attention mechanism, the temporal dynamic changes and spatial dependencies in the radio frequency identification signal can be captured, thereby improving positioning accuracy.

Description

Technical Field

[0001] The present invention relates to the field of indoor positioning technology, and in particular to an RFID indoor positioning method based on a hybrid neural network model, a computer-readable storage medium, a computer device, and an RFID indoor positioning system based on a hybrid neural network model. Background Art

[0002] Among related technologies, RFID indoor positioning primarily relies on geometric algorithms (such as time-of-arrival, time-delayed access (TDOA), and RSSI) or shallow machine learning models (such as k-NN and SVR). However, signals are susceptible to multipath effects, noise interference, and dynamic layout, resulting in low positioning accuracy. Traditional attention mechanisms have high computational complexity and are difficult to process long sequences of RFID signals. Unidirectional recurrent networks (such as GRUs) cannot capture bidirectional temporal dependencies, and traditional models struggle to balance local fluctuations with global trends. While existing deep learning methods (such as CNNs and BiGRUs) have shown some improvements, convolutional networks focus only on local features and ignore global spatial correlations. Standard self-attention mechanisms are sensitive to noise and waste computational resources. They also fail to incorporate signal differential features, making it difficult to capture instantaneous changes. Summary of the Invention

[0003] The present invention aims to at least partially address one of the technical problems in the aforementioned technologies. To this end, one objective of the present invention is to propose an RFID indoor positioning method based on a hybrid neural network model. This method, by combining a bidirectional differential gated recurrent unit with an enhanced Transformer attention mechanism, can effectively capture the temporal dynamics and spatial dependencies in RFID signals, thereby improving positioning accuracy.

[0004] A second object of the present invention is to provide a computer-readable storage medium.

[0005] A third object of the present invention is to provide a computer device.

[0006] The fourth object of the present invention is to propose an RFID indoor positioning system based on a hybrid neural network model.

[0007] To achieve the above-mentioned purpose, the first embodiment of the present invention proposes an RFID indoor positioning method based on a hybrid neural network model, which includes the following steps: obtaining an original signal and corresponding position information, and performing differential calculation on the original signal to obtain a corresponding differential signal; establishing a hybrid neural network model, wherein the hybrid neural network model includes a bidirectional differential gated recurrent unit, an adaptive sparse slot attention module and a Transformer encoder-decoder; inputting the original signal, the corresponding position information and the corresponding differential signal into the hybrid neural network model to train the hybrid neural network model to obtain a trained indoor positioning model; obtaining a tag signal to be located and its corresponding differential signal; and inputting the tag signal to be located and its corresponding differential signal into the hybrid neural network model. The bidirectional differential gated recurrent unit input into the trained indoor positioning model is used to extract temporal features to obtain spatiotemporal feature representation; the spatiotemporal feature representation is input into the adaptive sparse slot attention module in the trained indoor positioning model to obtain attention weights, and the spatiotemporal feature representation is weighted according to the attention weights to obtain weighted features; the weighted features are input into the Transformer encoder-decoder in the trained indoor positioning model to output the position information corresponding to the tag signal to be located; thus, by combining the bidirectional differential gated recurrent unit with the enhanced Transformer attention mechanism, the temporal dynamic changes and spatial dependencies in the radio frequency identification signal can be effectively captured, thereby improving the positioning accuracy.

[0008] In addition, the RFID indoor positioning method based on the hybrid neural network model proposed in the above embodiment of the present invention may also have the following additional technical features:

[0009] Optionally, obtaining the original signal and the corresponding position information includes: using an RFID reader to move along a predetermined trajectory, so as to obtain multiple RSSI data of the RFID reader and the target tag and coordinate information of the target tag during the movement.

[0010] Optionally, the original signal, the corresponding position information and the corresponding differential signal are input into the hybrid neural network model to train the hybrid neural network model to obtain a trained indoor positioning model, including: inputting the original signal and the corresponding differential signal into the forward branch and the reverse branch of the bidirectional differential gated cyclic unit respectively to extract timing features in the forward and reverse paths to obtain the corresponding first hidden state and second hidden state, and splicing and fusing the first hidden state and the second hidden state to obtain a spatiotemporal feature representation; inputting the spatiotemporal feature representation into the adaptive sparse slot attention module to perform ReLU square operation to filter low correlation signals through sparse square attention to obtain a first attention weight representing sparse square attention, retaining global dependencies through dense Softmax attention, and obtaining a representation of dense The second attention weight of the Softmax attention is set; the first attention weight and the second attention weight are dynamically fused through the weight coefficient to obtain the attention weight, and the spatiotemporal feature representation is weighted according to the attention weight to obtain the weighted feature; the weighted feature is input into the Transformer encoder-decoder, and the global context aggregation is realized through the multi-layer stacked Transformer blocks of the encoder, and relative position encoding is introduced to enhance sequence position perception. The decoder adopts global average pooling to compress the feature dimension, and then maps the high-dimensional features to the two-dimensional coordinate space through the multi-layer perceptron to output the target position information; the loss function value is determined according to the target position information and the corresponding position information, and the hybrid neural network model is optimized according to the loss function value to obtain a trained indoor positioning model.

[0011] To achieve the above-mentioned purpose, the second aspect of the embodiment of the present invention proposes a computer-readable storage medium, on which an RFID indoor positioning program based on a hybrid neural network model is stored. When the RFID indoor positioning program based on the hybrid neural network model is executed by a processor, the RFID indoor positioning method based on the hybrid neural network model as described above is implemented.

[0012] To achieve the above-mentioned objectives, the third aspect of the present invention proposes a computer device including a memory, a processor, and a computer program stored in the memory and runnable on the processor. When the processor executes the computer program, it implements the RFID indoor positioning method based on the hybrid neural network model as described above.

[0013] To achieve the above-mentioned purpose, the fourth embodiment of the present invention proposes an RFID indoor positioning system based on a hybrid neural network model, including: a first acquisition module, used to obtain the original signal and the corresponding position information, and perform differential calculation on the original signal to obtain the corresponding differential signal; a model building module, used to establish a hybrid neural network model, wherein the hybrid neural network model includes a bidirectional differential gated recurrent unit, an adaptive sparse slot attention module and a Transformer encoder-decoder; a training module, used to input the original signal, the corresponding position information and the corresponding differential signal into the hybrid neural network model to train the hybrid neural network model to obtain a trained indoor positioning model. type; a second acquisition module, used to obtain the tag signal to be located and its corresponding differential signal; a positioning module, used to input the tag signal to be located and its corresponding differential signal into the bidirectional differential gated cyclic unit in the trained indoor positioning model for time series feature extraction to obtain spatiotemporal feature representation; input the spatiotemporal feature representation into the adaptive sparse slot attention module in the trained indoor positioning model to obtain attention weights, and perform weighted processing on the spatiotemporal feature representation according to the attention weights to obtain weighted features; input the weighted features into the Transformer encoder-decoder in the trained indoor positioning model to output the position information corresponding to the tag signal to be located.

[0014] In addition, the RFID indoor positioning system based on the hybrid neural network model proposed in the above embodiment of the present invention may also have the following additional technical features:

[0015] Optionally, the first acquisition module is further configured to use an RFID reader to move along a predetermined trajectory, so as to acquire multiple RSSI data of the RFID reader and the target tag and coordinate information of the target tag during the movement.

[0016] Optionally, the training module is further used to input the original signal and the corresponding differential signal into the forward branch and the reverse branch of the bidirectional differential gated cyclic unit respectively to extract timing features in the forward and reverse paths to obtain the corresponding first hidden state and second hidden state, and concatenate and fuse the first hidden state and the second hidden state to obtain a spatiotemporal feature representation; input the spatiotemporal feature representation into the adaptive sparse slot attention module to perform ReLU square operation to filter low-correlation signals through sparse square attention to obtain a first attention weight representing sparse square attention, retain global dependencies through dense Softmax attention to obtain a second attention weight representing dense Softmax attention; combine the first attention weight and the second Attention weights are dynamically fused through weight coefficients to obtain attention weights, and the spatiotemporal feature representations are weighted according to the attention weights to obtain weighted features; the weighted features are input into the Transformer encoder-decoder, global context aggregation is achieved through the encoder's multi-layer stacked Transformer blocks, and relative position encoding is introduced to enhance sequence position perception. The decoder uses global average pooling to compress feature dimensions, and then maps high-dimensional features to a two-dimensional coordinate space through a multi-layer perceptron to output target position information; the loss function value is determined based on the target position information and the corresponding position information, and the hybrid neural network model is optimized based on the loss function value to obtain a trained indoor positioning model. BRIEF DESCRIPTION OF THE DRAWINGS

[0017] Figure 1 Schematic diagram of the process of an RFID indoor positioning method based on a hybrid neural network model according to an embodiment of the present invention;

[0018] Figure 2 A schematic diagram of the structure of a hybrid neural network model according to an embodiment of the present invention;

[0019] Figure 3 2. A schematic diagram of the structure of a dual-branch attention fusion mechanism of an adaptive sparse slot attention module according to an embodiment of the present invention;

[0020] Figure 4 2 is a schematic diagram of the structure of a Transformer encoder-decoder according to an embodiment of the present invention;

[0021] Figure 5 4 is a block diagram of an RFID indoor positioning system based on a hybrid neural network model according to an embodiment of the present invention. DETAILED DESCRIPTION

[0022] The following describes embodiments of the present invention in detail, examples of which are shown in the accompanying drawings, wherein the same or similar reference numerals throughout represent the same or similar elements or elements having the same or similar functions. The embodiments described below with reference to the accompanying drawings are exemplary and are intended to be used to explain the present invention, and are not to be construed as limiting the present invention.

[0023] To better understand the above technical solutions, exemplary embodiments of the present invention will be described in more detail below with reference to the accompanying drawings. Although exemplary embodiments of the present invention are shown in the accompanying drawings, it should be understood that the present invention can be implemented in various forms and should not be limited by the embodiments described herein. Instead, these embodiments are provided to enable a more thorough understanding of the present invention and to fully convey the scope of the present invention to those skilled in the art.

[0024] In order to better understand the above technical solution, the above technical solution will be described in detail below with reference to the accompanying drawings and specific implementation methods.

[0025] Figure 1 FIG. 1 is a flow chart of an RFID indoor positioning method based on a hybrid neural network model according to an embodiment of the present invention. Figure 1 As shown, the RFID indoor positioning method based on the hybrid neural network model of the embodiment of the present invention includes the following steps:

[0026] S101, obtaining an original signal and corresponding position information, and performing a differential calculation on the original signal to obtain a corresponding differential signal.

[0027] It should be noted that RFID readers and RFID tags can transmit data through antennas. The RFID reader sends a broadcast query request through the antenna, and the RFID tag responds to the reader's request and returns data. After the reader receives the feedback signal, the read-write module hands it over to the RFID middleware system for processing.

[0028] As an embodiment, obtaining the original signal and the corresponding location information includes: using an RFID reader to move along a predetermined trajectory, so as to obtain multiple RSSI data of the RFID reader and the target tag and coordinate information of the target tag during the movement.

[0029] In other words, the RFID reader can continuously send multiple query requests along the set trajectory. The RFID middleware system saves the RSSI data returned by these multiple queries and uses them as the feature values ​​X used for training. The tag position coordinates are used as the tag y used for training to form a data set and saved in the database for model training and testing.

[0030] S102, establishing a hybrid neural network model, wherein the hybrid neural network model includes a bidirectional differential gated recurrent unit, an adaptive sparse slot attention module and a Transformer encoder-decoder.

[0031] S103: Input the original signal, the corresponding position information and the corresponding differential signal into the hybrid neural network model to train the hybrid neural network model to obtain a trained indoor positioning model.

[0032] As an embodiment, the original signal, the corresponding position information and the corresponding differential signal are input into a hybrid neural network model to train the hybrid neural network model to obtain a trained indoor positioning model, including: inputting the original signal and the corresponding differential signal into the forward branch and the reverse branch of the bidirectional differential gated cyclic unit respectively to extract the timing features in the forward and reverse paths to obtain the corresponding first hidden state and the second hidden state, and splicing and fusing the first hidden state and the second hidden state to obtain the spatiotemporal feature representation; inputting the spatiotemporal feature representation into the adaptive sparse slot attention module to perform the ReLU square operation to filter the low correlation signal through the sparse square attention to obtain the first attention weight representing the sparse square attention, retaining the global dependency through the dense Softmax attention, and obtaining the representation The second attention weight of dense Softmax attention; the first attention weight and the second attention weight are dynamically fused through the weight coefficient to obtain the attention weight, and the spatiotemporal feature representation is weighted according to the attention weight to obtain the weighted feature; the weighted feature is input into the Transformer encoder-decoder, and global context aggregation is realized through the multi-layer stacked Transformer blocks of the encoder, and relative position encoding is introduced to enhance sequence position perception. The decoder uses global average pooling to compress the feature dimension, and then uses a multi-layer perceptron to map the high-dimensional features to a two-dimensional coordinate space to output the target position information; the loss function value is determined according to the target position information and the corresponding position information, and the hybrid neural network model is optimized according to the loss function value to obtain a trained indoor positioning model.

[0033] Specifically, if Figure 2-4 As shown, the hybrid neural network model first receives the original signal ( ) and its first-order difference sequence ( ), the difference sequence is generated by the formula:

[0034]

[0035] in, Indicates the RSSI value at the current moment, Represents the RSSI value at the previous moment, used to capture instantaneous changes in signal strength. The original signal and the differential signal are input into two independent bidirectional GRU branches. Each branch extracts temporal features through the forward and reverse paths, and ultimately concatenates and fuses the bidirectional hidden states to form a high-density spatiotemporal feature representation.

[0036] It should be noted that the Bidirectional Differential Gated Recurrent Unit (BiDGRU) processes the original signal synchronously ( ) and its first-order difference sequence ( ), capturing the instantaneous changes and long-term dependencies of the signal; the bidirectional GRU branch extracts forward and reverse timing features respectively, and enhances the dynamic adaptability of the model through feature fusion.

[0037] Then, the features are fed into the adaptive sparse slot attention module (ASSA), as Figure 3 As shown in the figure, the adaptive sparse slot attention module adopts a two-branch design: Sparse Squared Attention (SSA) filters low-correlation signals through ReLU square operations to reduce noise interference; Dense Softmax Attention (DSA) preserves global dependencies to avoid information loss. The two are dynamically fused through weight coefficients w1 and w2 to generate the final attention weight, significantly reducing computational complexity while improving the ability to focus on key signal segments:

[0038]

[0039]

[0040]

[0041] in, Represents the result of matrix multiplication of the transpose of the Query matrix and the Key matrix, which is used to calculate the similarity between the Query and the Key. Indicates the dimension size of each attention head, which is used to scale the dot product result to avoid the gradient disappearance problem caused by the dot product result being too large. represents the bias term, represents the attention weights of sparse squared attention, represents the attention weight of dense Softmax attention, ReLU is a commonly used nonlinear activation function. represents the normalization function, represents the final attention weight, and Represents the weight coefficient.

[0042] It should be noted that Adaptive Sparse Slot Attention (ASSA) introduces dual-branch attention (SSA and DSA), combining sparse square attention and dense Softmax attention to balance computational efficiency and feature focus; through content-aware sparsification, it reduces redundant computation and prioritizes key signal segments.

[0043] Finally, if Figure 4 As shown in the figure, the Transformer encoder-decoder performs deep feature processing. The encoder aggregates global context through multiple stacked Transformer blocks and introduces relative position encoding to enhance sequence position perception. The decoder uses global average pooling (GAP) to compress feature dimensions and suppress noise interference. A multi-layer perceptron (MLP) then maps high-dimensional features to a two-dimensional coordinate space, directly outputting the target position (x, y). The output target position is used as the predicted value and is back-propagated against the input true target value through a loss function, resulting in a trained indoor positioning model.

[0044] It should be noted that the encoder in the Transformer encoder-decoder architecture enhances sequence position perception through relative position encoding, and the decoder uses global average pooling (GAP) to suppress noise and generate high-density features; the cross-modal attention mechanism realizes the direct mapping of signal features to target coordinates.

[0045] S104: Acquire the tag signal to be located and its corresponding differential signal.

[0046] S105: Input the tag signal to be located and its corresponding differential signal into the bidirectional differential gated recurrent unit in the trained indoor positioning model to extract temporal features to obtain spatiotemporal feature representation.

[0047] S106: Input the spatiotemporal feature representation into the adaptive sparse slot attention module in the trained indoor positioning model to obtain attention weights, and perform weighted processing on the spatiotemporal feature representation according to the attention weights to obtain weighted features.

[0048] S107: Input the weighted features into the Transformer encoder-decoder in the trained indoor positioning model to output the position information corresponding to the tag signal to be positioned.

[0049] In other words, in real-time positioning scenarios, the system receives a continuous signal stream from RFID tags and feeds it into a trained indoor positioning model. The model first extracts temporal features using a BiDGRU, then uses the ASSA module to focus on key signal segments. Finally, the Transformer encoder-decoder generates coordinate predictions. This entire process is completed in milliseconds, meeting the requirements of real-time positioning in dynamic environments.

[0050] To validate the model's performance, we conducted multiple comparative and ablation experiments. As shown in the table below, BDGTA achieves significantly lower MAE (0.2158 meters) and RMSE (0.2929 meters) than traditional methods (such as KNN and MLP) and mainstream deep learning models (such as BiGRU and CNN). The MAE is 12.3% lower than that of BiGRU, and the RMSE is 27.4% lower than that of CNN. Ablation experiments further demonstrate that removing the BiDGRU module causes the MAE to surge by 485%, demonstrating its critical role in modeling dynamic signals. Replacing the standard attention mechanism (Abl2) or using only the encoder (Abl3) results in significant performance degradation, highlighting the necessity of ASSA and a complete encoder-decoder architecture. BDGTA performs the best among all models. Its mean absolute error (MAE) is 0.2158, mean square error (MSE) is 0.0858, root mean square error (RMSE) is 0.2929, and mean absolute percentage error (MAPE) is 8.06%, all of which are among the lowest, while its coefficient of determination (R²) is 0.9895, the highest. This demonstrates that BDGTA achieves the highest accuracy and robustness in the coordinate prediction task. Specifically, compared to the bidirectional differential gated recurrent unit (BiGRU), the mean absolute error is reduced by 12.3% (from 0.246 to 0.2158), and the coefficient of determination is improved to 0.9895, indicating that its fitting ability is close to optimal levels. This demonstrates that combining the bidirectional differential gated recurrent unit (BiDGRU) with the enhanced Transformer attention mechanism can effectively capture the temporal dynamics and spatial dependencies in radio frequency identification (RFID) signals.

[0051]

[0052] In summary, according to the RFID indoor positioning method based on the hybrid neural network model of the embodiment of the present invention, first, the original signal and the corresponding position information are obtained, and the original signal is differentially calculated to obtain the corresponding differential signal; then, a hybrid neural network model is established, wherein the hybrid neural network model includes a bidirectional differential gated recurrent unit, an adaptive sparse slot attention module and a Transformer encoder-decoder; then, the original signal, the corresponding position information and the corresponding differential signal are input into the hybrid neural network model to train the hybrid neural network model to obtain a trained indoor positioning model; then, the tag signal to be located and its corresponding differential signal are obtained; then, the tag signal to be located and its corresponding differential signal are input into the hybrid neural network model. The bidirectional differential gated recurrent unit input into the trained indoor positioning model is used to extract temporal features to obtain spatiotemporal feature representation; the spatiotemporal feature representation is input into the adaptive sparse slot attention module in the trained indoor positioning model to obtain attention weights, and the spatiotemporal feature representation is weighted according to the attention weights to obtain weighted features; finally, the weighted features are input into the Transformer encoder-decoder in the trained indoor positioning model to output the position information corresponding to the tag signal to be located; thus, by combining the bidirectional differential gated recurrent unit with the enhanced Transformer attention mechanism, the temporal dynamic changes and spatial dependencies in the radio frequency identification signal can be effectively captured, thereby improving the positioning accuracy.

[0053] In addition, an embodiment of the present invention also proposes a computer-readable storage medium, on which an RFID indoor positioning program based on a hybrid neural network model is stored. When the RFID indoor positioning program based on the hybrid neural network model is executed by a processor, the RFID indoor positioning method based on the hybrid neural network model as described above is implemented.

[0054] According to the computer-readable storage medium of an embodiment of the present invention, by storing an RFID indoor positioning program based on a hybrid neural network model, the RFID indoor positioning program based on the hybrid neural network model is executed by a processor to implement the above-mentioned RFID indoor positioning method based on the hybrid neural network model. Thus, by combining the bidirectional differential gated recurrent unit with the enhanced Transformer attention mechanism, the temporal dynamic changes and spatial dependencies in the radio frequency identification signal can be effectively captured, thereby improving the positioning accuracy.

[0055] In addition, an embodiment of the present invention also proposes a computer device including a memory, a processor, and a computer program stored in the memory and runnable on the processor. When the processor executes the computer program, it implements the RFID indoor positioning method based on the hybrid neural network model as described above.

[0056] According to the computer device of an embodiment of the present invention, the RFID indoor positioning program based on the hybrid neural network model is stored in the memory. When the RFID indoor positioning program based on the hybrid neural network model is executed by the processor, the above-mentioned RFID indoor positioning method based on the hybrid neural network model is implemented. Therefore, by combining the bidirectional differential gated recurrent unit with the enhanced Transformer attention mechanism, the temporal dynamic changes and spatial dependencies in the radio frequency identification signal can be effectively captured, thereby improving the positioning accuracy.

[0057] Figure 5 FIG. 1 is a block diagram of an RFID indoor positioning system based on a hybrid neural network model according to an embodiment of the present invention; FIG. Figure 5 As shown, the RFID indoor positioning system based on the hybrid neural network model includes: a first acquisition module 10, a model building module 20, a training module 30, a second acquisition module 40 and a positioning module 50.

[0058] Among them, the first acquisition module 10 is used to obtain the original signal and the corresponding position information, and perform differential calculation on the original signal to obtain the corresponding differential signal; the model establishment module 20 is used to establish a hybrid neural network model, wherein the hybrid neural network model includes a bidirectional differential gated recurrent unit, an adaptive sparse slot attention module and a Transformer encoder-decoder; the training module 30 is used to input the original signal, the corresponding position information and the corresponding differential signal into the hybrid neural network model to train the hybrid neural network model to obtain a trained indoor positioning model; the second acquisition module 40 is used to obtain the tag signal to be located and its corresponding differential signal; the positioning module 50 is used to input the tag signal to be located and its corresponding differential signal into the bidirectional differential gated recurrent unit in the trained indoor positioning model to perform temporal feature extraction to obtain a spatiotemporal feature representation; the spatiotemporal feature representation is input into the adaptive sparse slot attention module in the trained indoor positioning model to obtain an attention weight, and the spatiotemporal feature representation is weighted according to the attention weight to obtain a weighted feature; the weighted feature is input into the Transformer encoder-decoder in the trained indoor positioning model to output the position information corresponding to the tag signal to be located.

[0059] As an embodiment, the first acquisition module 10 is further configured to use an RFID reader to move along a predetermined trajectory, so as to acquire multiple RSSI data of the RFID reader and the target tag and coordinate information of the target tag during the movement.

[0060] As an embodiment, the training module 30 is further used to input the original signal and the corresponding differential signal into the forward branch and the reverse branch of the bidirectional differential gated cyclic unit respectively to extract the timing features in the forward and reverse paths to obtain the corresponding first hidden state and the second hidden state, and to splice and fuse the first hidden state and the second hidden state to obtain the spatiotemporal feature representation; input the spatiotemporal feature representation into the adaptive sparse slot attention module to perform the ReLU square operation to filter the low correlation signal through the sparse square attention to obtain the first attention weight representing the sparse square attention, retain the global dependency through the dense Softmax attention to obtain the second attention weight representing the dense Softmax attention; the first attention weight is The first and second attention weights are dynamically fused through weight coefficients to obtain attention weights, and the spatiotemporal feature representation is weighted according to the attention weights to obtain weighted features; the weighted features are input into the Transformer encoder-decoder, and global context aggregation is achieved through the encoder's multi-layer stacked Transformer blocks, and relative position encoding is introduced to enhance sequence position perception. The decoder uses global average pooling to compress feature dimensions, and then maps high-dimensional features to two-dimensional coordinate space through a multi-layer perceptron to output target position information; the loss function value is determined according to the target position information and the corresponding position information, and the hybrid neural network model is optimized according to the loss function value to obtain a trained indoor positioning model.

[0061] It should be noted that the aforementioned explanation of the embodiment of the RFID indoor positioning method based on the hybrid neural network model is also applicable to the RFID indoor positioning system based on the hybrid neural network model of this embodiment, and will not be repeated here.

[0062] In summary, according to the RFID indoor positioning system based on the hybrid neural network model of the embodiment of the present invention, the original signal and the corresponding position information are obtained by the acquisition module, and the original signal is differentially calculated to obtain the corresponding differential signal; the model establishment module establishes a hybrid neural network model, wherein the hybrid neural network model includes a bidirectional differential gated recurrent unit, an adaptive sparse slot attention module and a Transformer encoder-decoder; the training module inputs the original signal, the corresponding position information and the corresponding differential signal into the hybrid neural network model to train the hybrid neural network model to obtain a trained indoor positioning model; the second acquisition module obtains the tag signal to be located and its corresponding differential signal; the positioning module inputs the tag signal to be located and its corresponding differential signal The corresponding differential signal is input into the bidirectional differential gated recurrent unit in the trained indoor positioning model for temporal feature extraction to obtain spatiotemporal feature representation; the spatiotemporal feature representation is input into the adaptive sparse slot attention module in the trained indoor positioning model to obtain attention weights, and the spatiotemporal feature representation is weighted according to the attention weights to obtain weighted features; the weighted features are input into the Transformer encoder-decoder in the trained indoor positioning model to output the position information corresponding to the tag signal to be located; thus, by combining the bidirectional differential gated recurrent unit with the enhanced Transformer attention mechanism, the temporal dynamic changes and spatial dependencies in the radio frequency identification signal can be effectively captured, thereby improving the positioning accuracy.

[0063] Those skilled in the art will appreciate that embodiments of the present invention may be provided as methods, systems, or computer program products. Thus, the present invention may take the form of an entirely hardware embodiment, an entirely software embodiment, or an embodiment combining software and hardware aspects. Furthermore, the present invention may take the form of a computer program product implemented on one or more computer-usable storage media (including but not limited to magnetic disk storage, CD-ROM, optical storage, etc.) containing computer-usable program code.

[0064] The present invention is described with reference to flowcharts and / or block diagrams of methods, devices (systems), and computer program products according to embodiments of the present invention. It should be understood that each process and / or block in the flowcharts and / or block diagrams, as well as combinations of processes and / or blocks in the flowcharts and / or block diagrams, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, a special-purpose computer, an embedded processor, or other programmable data processing device to produce a machine, so that the instructions executed by the processor of the computer or other programmable data processing device generate instructions for implementing the processes in the flowcharts and / or block diagrams. Figure 1 a process or multiple processes and / or boxes Figure 1 A device that provides the functions specified in a block or multiple blocks.

[0065] These computer program instructions may also be stored in a computer readable memory that can direct a computer or other programmable data processing device to work in a specific manner, so that the instructions stored in the computer readable memory produce an article of manufacture comprising an instruction device, which implements the process Figure 1 a process or multiple processes and / or boxes Figure 1 The function specified in one or more boxes.

[0066] These computer program instructions can also be loaded onto a computer or other programmable data processing device so that a series of operational steps are executed on the computer or other programmable device to produce a computer-implemented process, thereby providing the instructions executed on the computer or other programmable device for implementing the process. Figure 1 a process or multiple processes and / or boxes Figure 1 A step that specifies a function in one or more boxes.

[0067] It should be noted that in the claims, any reference signs placed between parentheses shall not be construed as limiting the claims. The word "comprising" does not exclude the presence of components or steps not listed in the claim. The word "a" or "an" preceding a component does not exclude the presence of a plurality of such components. The invention can be implemented by means of hardware comprising several different components and by means of a suitably programmed computer. In a unit claim enumerating several means, several of these means may be embodied by one and the same item of hardware. The use of the words first, second, third etc. does not indicate any order. These words may be interpreted as names.

[0068] Although the preferred embodiments of the present invention have been described, those skilled in the art may make additional changes and modifications to these embodiments once they have learned the basic creative concept. Therefore, the appended claims are intended to be interpreted as including the preferred embodiments and all changes and modifications that fall within the scope of the present invention.

[0069] Obviously, those skilled in the art may make various changes and modifications to the present invention without departing from the spirit and scope of the present invention. Thus, if such changes and modifications fall within the scope of the claims and their equivalents, the present invention is intended to include such changes and modifications.

[0070] In the description of the present invention, it should be understood that the terms "first" and "second" are used for descriptive purposes only and should not be understood to indicate or imply relative importance or implicitly specify the number of the technical features indicated. Therefore, a feature specified as "first" or "second" may explicitly or implicitly include one or more of the features. In the description of the present invention, "plurality" means two or more, unless otherwise specifically defined.

[0071] In the present invention, unless otherwise expressly specified or limited, the terms "mounted," "connected," "connect," "fixed," etc. should be understood broadly. For example, they may refer to fixed connection, detachable connection, or integration; mechanical connection or electrical connection; direct connection or indirect connection through an intermediate medium; internal communication between two components or interaction between two components. Those skilled in the art will understand the specific meanings of the above terms in the present invention based on specific circumstances.

[0072] In the present invention, unless otherwise expressly specified or limited, when a first feature is "above" or "below" a second feature, it may mean that the first and second features are in direct contact, or that the first and second features are in indirect contact through an intermediary. Furthermore, when a first feature is "above," "above," or "above" a second feature, it may mean that the first feature is directly above or diagonally above the second feature, or simply means that the first feature is at a higher level than the second feature. When a first feature is "below," "below," or "below" a second feature, it may mean that the first feature is directly below or diagonally below the second feature, or simply means that the first feature is at a lower level than the second feature.

[0073] In the description of this specification, the reference terms "one embodiment", "some embodiments", "example", "specific example", or "some examples" mean that the specific features, structures, materials, or characteristics described in conjunction with the embodiment or example are included in at least one embodiment or example of the present invention. In this specification, the schematic expressions of the above terms should not be understood as necessarily referring to the same embodiment or example. Moreover, the specific features, structures, materials, or characteristics described can be combined in any one or more embodiments or examples in a suitable manner. In addition, those skilled in the art can combine and combine different embodiments or examples described in this specification and features of different embodiments or examples without contradiction.

[0074] Although the embodiments of the present invention have been shown and described above, it will be understood that the above embodiments are illustrative and are not to be construed as limitations on the present invention. A person skilled in the art may change, modify, replace and modify the above embodiments within the scope of the present invention.

Claims

1. An RFID indoor positioning method based on a hybrid neural network model, characterized in that: The following steps are involved: Acquiring an original signal and corresponding position information, and performing a differential calculation on the original signal to obtain a corresponding differential signal; Establishing a hybrid neural network model, wherein the hybrid neural network model includes a bidirectional differential gated recurrent unit, an adaptive sparse slot attention module, and a Transformer encoder-decoder; Inputting the original signal and the corresponding differential signal into the forward branch and the reverse branch of the bidirectional differential gated recurrent unit respectively to extract timing features in the forward and reverse paths to obtain a corresponding first hidden state and a second hidden state, and concatenating and fusing the first hidden state and the second hidden state to obtain a spatiotemporal feature representation; Inputting the spatiotemporal feature representation into an adaptive sparse slot attention module to perform a ReLU square operation to filter low-correlation signals through sparse square attention, obtaining a first attention weight representing sparse square attention, and retaining global dependencies through dense softmax attention to obtain a second attention weight representing dense softmax attention; Dynamically fusing the first attention weight and the second attention weight through a weight coefficient to obtain an attention weight, and performing weighted processing on the spatiotemporal feature representation according to the attention weight to obtain a weighted feature; The weighted features are input into the Transformer encoder-decoder. Global context aggregation is achieved through the encoder's multi-layer stacked Transformer blocks, and relative position encoding is introduced to enhance sequence position perception. The decoder uses global average pooling to compress feature dimensions, and then uses a multi-layer perceptron to map the high-dimensional features to a two-dimensional coordinate space to output target position information. Determining a loss function value according to the target position information and the corresponding position information, and optimizing the hybrid neural network model according to the loss function value to obtain a trained indoor positioning model; Obtain the tag signal to be located and its corresponding differential signal; Inputting the tag signal to be located and its corresponding differential signal into the bidirectional differential gated recurrent unit in the trained indoor positioning model to extract temporal features to obtain spatiotemporal feature representation; Inputting the spatiotemporal feature representation into the adaptive sparse slot attention module in the trained indoor positioning model to obtain attention weights, and weighting the spatiotemporal feature representation according to the attention weights to obtain weighted features; The weighted features are input into the Transformer encoder-decoder in the trained indoor positioning model to output the position information corresponding to the tag signal to be located.

2. The RFID indoor positioning method based on the hybrid neural network model according to claim 1, characterized in that: Acquiring the original signal and the corresponding position information includes: using an RFID reader to move along a predetermined trajectory, so as to obtain multiple RSSI data of the RFID reader and the target tag and the coordinate information of the target tag during the movement.

3. A computer-readable storage medium, characterized in that An RFID indoor positioning program based on a hybrid neural network model is stored thereon, and when the RFID indoor positioning program based on a hybrid neural network model is executed by a processor, the RFID indoor positioning method based on a hybrid neural network model as described in any one of claims 1-2 is implemented.

4. A computer device comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein: When the processor executes the computer program, the RFID indoor positioning method based on the hybrid neural network model as described in any one of claims 1 to 2 is implemented.

5. An RFID indoor positioning system based on a hybrid neural network model, characterized in that: include: A first acquisition module is used to acquire an original signal and corresponding position information, and perform a differential calculation on the original signal to obtain a corresponding differential signal; A model building module, configured to build a hybrid neural network model, wherein the hybrid neural network model includes a bidirectional differential gated recurrent unit, an adaptive sparse slot attention module, and a Transformer encoder-decoder; A training module, configured to input the original signal, the corresponding position information, and the corresponding differential signal into the hybrid neural network model to train the hybrid neural network model to obtain a trained indoor positioning model; The second acquisition module is used to acquire the signal of the tag to be located and its corresponding differential signal; A positioning module is configured to input the tag signal to be located and its corresponding differential signal into the bidirectional differential gated recurrent unit in the trained indoor positioning model for temporal feature extraction to obtain a spatiotemporal feature representation; input the spatiotemporal feature representation into the adaptive sparse slot attention module in the trained indoor positioning model to obtain an attention weight, and perform weighted processing on the spatiotemporal feature representation according to the attention weight to obtain a weighted feature; and input the weighted feature into the Transformer encoder-decoder in the trained indoor positioning model to output the position information corresponding to the tag signal to be located; The training module is further used to input the original signal and the corresponding differential signal into the forward branch and the reverse branch of the bidirectional differential gated cyclic unit respectively to extract the timing features in the forward and reverse paths to obtain the corresponding first hidden state and the second hidden state, and to splice and fuse the first hidden state and the second hidden state to obtain the spatiotemporal feature representation; input the spatiotemporal feature representation into the adaptive sparse slot attention module to perform the ReLU square operation to filter the low correlation signal through the sparse square attention to obtain the first attention weight representing the sparse square attention, retain the global dependency through the dense Softmax attention, and obtain the second attention weight representing the dense Softmax attention; the first attention weight and the second Attention weights are dynamically fused through weight coefficients to obtain attention weights, and the spatiotemporal feature representations are weighted according to the attention weights to obtain weighted features; the weighted features are input into the Transformer encoder-decoder, global context aggregation is achieved through the encoder's multi-layer stacked Transformer blocks, and relative position encoding is introduced to enhance sequence position perception. The decoder uses global average pooling to compress feature dimensions, and then maps high-dimensional features to a two-dimensional coordinate space through a multi-layer perceptron to output target position information; the loss function value is determined based on the target position information and the corresponding position information, and the hybrid neural network model is optimized based on the loss function value to obtain a trained indoor positioning model.

6. The RFID indoor positioning system based on the hybrid neural network model according to claim 5, characterized in that: The first acquisition module is further configured to use an RFID reader to move along a predetermined trajectory, so as to acquire multiple RSSI data of the RFID reader and the target tag and coordinate information of the target tag during the movement.

Images