Indoor positioning method and system based on channel state information and depth feature fusion

Abstract

The invention provides an indoor positioning method and system based on channel state information and depth feature fusion. The method comprises the following steps: S1, collecting channel state information data by using an ESP32; s2, extracting amplitude data of each subcarrier; s3, the amplitude data are preprocessed; s4, extracting shallow layer features from the preprocessed data through an initial convolutional layer; s5, deep spatial-temporal features are extracted through the residual block group; s6, sequentially performing channel attention weighting and space attention weighting on the deep feature map; step S7, mapping the weighted feature map to a two-dimensional coordinate space, and outputting a positioning result; and S8, carrying out time sequence smoothing processing on a positioning result by adopting a weighted moving average filter. The deep learning algorithm is utilized to optimize signal processing, CNN is combined with a ResNet structure, channel attention and space attention mechanisms are fused, the positioning precision and the system adaptability are effectively improved, and the generalization ability of the model is enhanced.

Description

Technical Field

[0001] The present invention relates to the technical fields of wireless communication and indoor positioning. Specifically, it relates to an efficient indoor positioning method and system based on the fusion of channel state information (CSI) and deep features. Background Art

[0002] In today's digital age, indoor positioning technology plays a crucial role in many fields such as intelligent buildings, smart logistics, navigation, and commercial services. High-precision indoor positioning can provide users with accurate navigation, optimize the logistics transportation path, and enhance the personalized experience of commercial services. Among the many research directions of indoor positioning technology, the positioning technology based on channel state information (CSI) has become the research focus in the field of high-precision indoor positioning.

[0003] As a fine-grained data source at the physical layer of wireless communication, compared with the traditional received signal strength indication (RSSI) technology, CSI not only covers signal strength information, but also can obtain deep features such as carrier frequency, phase offset, and multipath components, opening up a richer dimension for indoor environment perception. In an orthogonal frequency division multiplexing (OFDM) system, by deeply analyzing the amplitude and phase information of each subcarrier, the indoor positioning technology based on CSI can accurately capture the multipath effect and environmental interference during signal propagation, which lays a theoretical foundation for achieving high-precision indoor positioning.

[0004] However, currently, traditional indoor positioning methods expose a series of severe challenges when processing CSI data. First, in terms of feature extraction, traditional convolutional networks are significantly insufficient in the face of complex spatio-temporal features in CSI data. CSI data contains rich and complex spatio-temporal information, and traditional convolutional networks are difficult to fully mine the key features, which directly limits the further improvement of positioning accuracy. Second, the problem of gradient disappearance is serious during the training process of deep networks. As the network depth increases, the training difficulty rises, and the model is difficult to converge stably, resulting in a significant reduction in the reliability of the positioning system. Finally, this technology is sensitive to noise. In an indoor environment, factors such as the movement of people, the dynamic change of object positions, and the multipath effect will continuously and dynamically change the channel characteristics, making the positioning results fluctuate violently and difficult to meet the requirements of stability and accuracy in practical applications.

[0005] Although the CSI acquisition mechanism defined by the IEEE 802.11n / ac standard theoretically enables commercial Wi-Fi devices to achieve fine-grained channel response analysis and even support sub-meter positioning accuracy. However, in typical indoor scenarios in reality, the ranging error of unoptimized raw CSI data is as high as 2 - 3 meters, which is far from the requirements of high-precision applications.

[0006] By searching the patent literature, it is found that the invention patent with the publication number CN 115209341 A discloses a weighted random forest indoor positioning method based on channel state information, including: setting offline reference points in the positioning area, collecting channel state information, and extracting channel fingerprint features; and extracting new channel fingerprint features of the offline reference points based on the channel state information; preprocessing the new channel fingerprint features of the offline reference points according to the offline transmitter selection principle to construct a channel fingerprint database; inputting the offline reference point coordinates and the new channel fingerprint features into a random forest for training, and outputting two offline training models; extracting new channel fingerprint features of the target positioning point based on the channel state information; inputting the new channel fingerprint features of the target positioning point into the offline training models respectively, and performing a weighted K-nearest neighbor algorithm for prediction to obtain the coordinates of the finally predicted target positioning point. This patent does not use deep learning to extract features, only uses random forest training, and does not combine optimization feature extraction such as convolution, residual blocks, and attention mechanisms, so the positioning accuracy and efficiency may be limited.

[0007] In summary, aiming at the problems of the above-mentioned existing technologies, researching an efficient indoor positioning method and system based on the fusion of channel state information and depth features has become a key task that needs to be solved urgently at present. Summary of the Invention

[0008] Aiming at the defects in the prior art, the purpose of the present invention is to provide an efficient indoor positioning method and system based on the fusion of channel state information and depth features.

[0009] An indoor positioning method based on the fusion of channel state information and depth features according to the present invention includes the following steps: Step S1, collecting channel state information data of Wi-Fi transceiver devices through multiple ESP32s; Step S2, extracting the amplitude data of each subcarrier from the channel state information data; Step S3, preprocessing the amplitude data to obtain preprocessed data; Step S4, extracting shallow features through an initial convolutional layer for the preprocessed data to obtain a shallow feature map; Step S5, extracting deep spatio-temporal features through a residual block group for the shallow feature map to obtain a deep feature map; Step S6, performing channel attention weighting and spatial attention weighting on the deep feature map in sequence to obtain a weighted feature map; Step S7, mapping the weighted feature map to a two-dimensional coordinate space through a fully connected layer and outputting a positioning result; Step S8, performing temporal smoothing processing on the positioning result by using a weighted moving average filter.

[0010] Preferably, in step S1, the Wi-Fi module of ESP32 is used to collect the channel state information data of the Wi-Fi transceiver device in accordance with the IEEE 802.11n / ac protocol. ESP32 is connected to the router as a Station, and the channel state information data is collected by sending data packets to the router. The real location information corresponding to the channel state information data is synchronously recorded, and the channel state information data is stored in the SD card. The acquisition parameters of the channel state information data are set through the preset configuration items.

[0011] Preferably, in step S2, according to the real part and the imaginary part of the subcarrier in the channel state information data, the amplitude data of the subcarrier is calculated. , is the real part of each subcarrier, is the imaginary part of each subcarrier.

[0012] Preferably, in step S3, the preprocessing includes the following sub-steps: Step S3.1, use a 6th-order Butterworth low-pass filter to remove high-frequency noise, and the cut-off frequency is 30Hz; Step S3.2, adopt a sliding window to calculate the moving average to eliminate the static component, and the length of the sliding window is set according to the sampling rate of 60Hz; Step S3.3, perform Min-Max normalization on the denoised CSI data and the location data to the [0,1] interval respectively, and the formula is: .

[0013] Preferably, in step S4, the initial convolutional layer includes a 3×3 convolutional kernel, a ReLU activation function, and a 2×2 max pooling operation.

[0014] Preferably, in step S5, the residual block group contains multiple residual blocks, and each residual block includes: a skip connection path and a main path. The main path is: a 3×3 convolutional layer, a ReLU activation function, and a 2×2 max pooling layer connected in sequence; the skip connection path is: if the number of input channels of the main path is inconsistent with the number of output channels, then adjust the channel dimension through a 1×1 convolutional layer, and after adding the output of the adjusted main path and the skip connection output element by element, output through the ReLU activation function. Among them, the activation function ReLU introduces non-linearity, and the max pooling halves the size of the feature map, reducing the computational complexity.

[0015] Preferably, step S6 includes the following sub-steps: Step S6.1, Channel attention weighting uses dual-path pooling: globally average pooling and spatial maximum pooling are calculated in parallel for the deep feature map. Global average pooling takes the mean along the spatial dimension to capture the overall features; spatial maximum pooling takes the maximum along the spatial dimension to capture the significant features. The results of the dual-path pooling are reduced in dimension and feature-transformed through a fully connected layer, and then weight generation is completed. The results of average pooling and maximum pooling are added together, and passed through the Sigmoid activation function to generate the channel attention weights , where is the input deep feature map, is the activation function, and the attention weights of each channel in the range [0, 1] are output; Step S6.2, Spatial attention weighting uses dual-path compression: the mean and maximum of the feature map are calculated along the channel dimension, and then feature fusion is performed to complete dual-path feature splicing, and spatial attention weights are generated through convolution operations , to complete spatial feature weighting, and the channel attention weights and spatial attention weights are multiplied to output the weighted feature map.

[0016] Preferably, in step S7, the Huber loss function is adopted , where is the true value, is the predicted value, is set to 1.0. It is similar to the mean squared error when the error is small, and similar to the absolute error when the error is large, balancing the error loss in the coordinate regression task and being more robust to outliers. The optimizer adopts the Adam optimization algorithm with an adaptive learning rate to promote stable optimization of Adam.

[0017] Preferably, in step S8, the weighted moving average filter adopts a fixed window size, and the filtering calculation formula is: , where is the time, the weight coefficient w = [w 0 , w 1 , w 2 = [0.5, 0.3, 0.2], is the window size, is the estimated value after weighted moving average filtering at the current time, is the data value before weighted average filtering at time

[0018] The present invention also provides an indoor positioning system based on the fusion of channel state information and depth features. Based on the above indoor positioning method based on the fusion of channel state information and depth features, it includes, Module M1, including multiple ESP32s, is used to collect channel state information data; Module M2, including an embedded processor, is used to deploy a residual-attention model to process channel state information data and output a positioning result; Module M3, including a weighted moving average filter module, is used to post-process and optimize the positioning result.

[0019] Compared with the prior art, the present invention has the following beneficial effects: 1. The present invention adopts an improved ResNet structure, introduces residual blocks on the basis of traditional CNN, and solves the problem of gradient disappearance through skip connections. At the same time, a multi-level pooling strategy is adopted to optimize the feature extraction efficiency.

[0020] 2. The present invention adopts a dual attention mechanism, combines channel attention and spatial attention, and dynamically allocates feature weights. Among them, channel attention optimizes the channel dimension features through global average / max pooling, and spatial attention enhances the spatial position correlation through convolution operations, effectively extracts the spatio-temporal features in CSI data, thereby enhancing the robustness to multipath effects and improving the positioning accuracy.

[0021] 3. The present invention realizes end-to-end filtering, combines the traditional signal processing algorithm (WMA) with the deep learning output, seamlessly connects the model output with the signal processing algorithm, and effectively suppresses noise.

[0022] 4. By optimizing the model structure, the present invention reduces the number of parameters by about 25% and improves the inference speed by 1.5 times, making it suitable for deployment on embedded devices.

[0023] 5. Experiments show that the present invention can achieve sub-meter positioning accuracy in typical indoor scenarios, significantly improves the positioning accuracy and robustness, and provides an efficient and reliable solution for indoor positioning. Description of the Drawings

[0024] By reading the detailed description of the non-limiting embodiments with reference to the following drawings, other features, objectives and advantages of the present invention will become more obvious: Figure 1 It is a flowchart of a high-precision indoor positioning method based on the fusion of channel state information (CSI) and deep features in an embodiment of the present invention; Figure 2 It is a schematic diagram of the deployment of CSI data acquisition hardware in an embodiment of the present invention; Figure 3 It is a structural diagram of a residual block in an embodiment of the present invention; Figure 4 It is a channel attention weight diagram in an embodiment of the present invention; Figure 5 It is a spatial attention heat map in an embodiment of the present invention; Figure 6Comparison of true and predicted positions before WMA in the embodiments of the present invention; Figure 7 Comparison of true and predicted positions after WMA in the embodiments of the present invention. Detailed implementation manners

[0025] The present invention will be described in detail below in conjunction with specific embodiments. The following embodiments will help those skilled in the art to further understand the present invention, but do not limit the present invention in any form. It should be noted that those of ordinary skill in the art can make several changes and improvements without departing from the concept of the present invention. These all belong to the protection scope of the present invention.

[0026] The present invention provides a high-precision indoor positioning method and system based on the fusion of channel state information (CSI) and depth features. Aiming at the defects of traditional indoor positioning technologies in terms of insufficient feature extraction ability, gradient disappearance problem, and noise sensitivity, the present invention uses deep learning algorithms to optimize signal processing, adopts a CNN combined with a ResNet structure, and fuses channel attention and spatial attention mechanisms to effectively improve positioning accuracy and system adaptability and enhance the generalization ability of the model.

[0027] Specifically, the present invention proposes a deep learning model combining a residual network (ResNet), an attention mechanism, and a weighted moving average filter (WMA) for high-precision indoor positioning based on channel state information (CSI). Specifically, a three-level error compensation architecture with innovative significance is constructed: (1) Deep residual feature extraction network: Use an improved ResNet structure for multi-level feature extraction and solve the gradient disappearance problem through skip connections; (2) Dual attention mechanism: Adopt a channel attention module and a spatial attention module to work together. The channel attention module dynamically weights subcarrier features and spatial regions; (3) Weighted moving average post-processing: Smooth the predicted trajectory in time series through a weighted moving average filter, which can reduce the time series jitter of the positioning result.

[0028] The present invention innovatively fuses dual-path normalization preprocessing and Huber loss function optimization. Through these measures, while maintaining a 25% reduction in the number of parameters and a 1.5-fold increase in the inference speed, it effectively suppresses the multipath effect and dynamic noise interference.

[0029] Embodiment 1: Figure 1 Flowchart of a high-precision indoor positioning method based on the fusion of channel state information (CSI) and depth features in the embodiments of the present invention.

[0030] As Figure 1As shown in the figure, this embodiment provides an indoor positioning method based on the fusion of channel state information (CSI) and deep features, including the following steps: Step S1, collect channel state information data of Wi-Fi transceiver devices through multiple ESP32s.

[0031] Specifically, based on the Wi-Fi module of ESP32, collect channel state information data of Wi-Fi transceiver devices following the IEEE 802.11n / ac protocol. ESP32 is connected to the router (AP) as a Station, collect channel state information data by sending data packets to the router, synchronously record the real position information corresponding to the channel state information data, store the channel state information data in the SD card, and set the acquisition parameters of the channel state information data through preset configuration items.

[0032] Step S2, extract the amplitude data of each subcarrier from the channel state information data.

[0033] Specifically, in step S2, according to the real part and imaginary part of the subcarrier in the channel state information data, calculate the amplitude data of the subcarrier , is the real part of each subcarrier, is the imaginary part of each subcarrier.

[0034] Step S3, preprocess the amplitude data to obtain the preprocessed data.

[0035] In this embodiment, the preprocessing includes Butterworth low-pass filtering for denoising, sliding window to eliminate static components, and dual-path normalization.

[0036] Specifically, in step S3, the preprocessing includes the following sub-steps: Step S3.1, use a 6th-order Butterworth low-pass filter to remove high-frequency noise, with a cut-off frequency of 30Hz; Step S3.2, adopt a sliding window to calculate the moving average to eliminate static components, and the length of the sliding window is set according to the sampling rate of 60Hz.

[0037] Specifically, if using seconds of sliding window to calculate the moving average of each CSI, then the length of the sliding window is . Step S3.3, perform Min-Max normalization on the denoised CSI data and position data respectively to the [0, 1] interval.

[0038] In this embodiment, the formula is: , and the dual-path normalization design eliminates the dimension difference and improves the robustness of the model.

[0039] Step S4: Extract shallow features from the preprocessed data through the initial convolutional layer to obtain a shallow feature map.

[0040] Specifically, in step S4, the initial convolutional layer includes a 3×3 convolutional kernel, a ReLU activation function, and a 2×2 max pooling operation.

[0041] Step S5: Extract deep spatio-temporal features from the shallow feature map through the residual block group to obtain a deep feature map.

[0042] In this embodiment, the residual block group performs deep feature extraction and executes an addition operation. The main path output + skip connection output → ReLU activation. Through convolutional operations, activation functions, and skip connections, residual learning is completed to enhance the feature abstraction ability. Residual learning changes the learning objective from to , simplifies the optimization objective, solves the gradient vanishing problem, allows the network depth to increase, and extracts high-order spatio-temporal features.

[0043] Specifically, the residual block group contains multiple residual blocks. Each residual block includes a skip connection path and a main path. The main path is: a 3×3 convolutional layer, a ReLU activation function, and a 2×2 max pooling layer connected in sequence; the skip connection path is: if the number of input channels of the main path is inconsistent with the number of output channels, the channel dimension is adjusted through a 1×1 convolutional layer. After adding the adjusted main path output and the skip connection output element-wise, it is output through a ReLU activation function. Among them, the activation function ReLU introduces non-linearity, and max pooling halves the size of the feature map, reducing the computational complexity.

[0044] Step S6: Perform channel attention weighting and spatial attention weighting on the deep feature map in sequence to highlight key features and obtain a weighted feature map.

[0045] Specifically, step S6 includes the following sub-steps: Step S6.1: Channel attention weighting uses dual-path pooling: Calculate global average pooling and spatial max pooling in parallel for the deep feature map. Global average pooling takes the mean along the spatial dimension to capture overall features; spatial max pooling takes the maximum value along the spatial dimension to capture significant features. The results of the dual-path pooling are reduced in dimension and feature-transformed through a fully connected layer, and then weight generation is completed. The results of average pooling and max pooling are added together, and through the Sigmoid activation function generate channel attention weights where is the input deep feature map, is the activation function, and the attention weights of each channel in the range [0,1] are output. The channel attention module performs feature recalibration, dynamically enhancing the feature response of important subcarriers and suppressing noise.

[0046] Step S6.2, Spatial attention weighting uses dual-path compression: Calculate the mean and maximum values of the feature map along the channel dimension, then perform feature fusion, complete the dual-path feature splicing, and generate spatial attention weights through convolution operations. , complete spatial feature weighting, multiply the channel attention weight and the spatial attention weight, and output the weighted feature map. The spatial attention module focuses on key spatial regions, strengthens the detection of key regions in multi-path signals, and improves the positioning accuracy. Step S7, map the weighted feature map to the two-dimensional coordinate space through a fully connected layer, output the positioning result, and achieve coordinate regression.

[0047] Specifically, in step S7, the Huber loss function is adopted. , where is the true value, is the predicted value, is set to 1.0. It is similar to the mean square error when the error is small, and similar to the absolute error when the error is large. It balances the error loss in the coordinate regression task and is more robust to outliers. The optimizer adopts the Adam optimization algorithm with an adaptive learning rate to promote stable optimization of Adam.

[0048] Step S8, use a weighted moving average filter (WMA) to perform temporal smoothing on the positioning result, improve the smoothness of the predicted trajectory, and suppress instantaneous noise.

[0049] Specifically, in step S8, the weighted moving average filter uses a fixed window size (window size is 3), traverses the temporal data, and the filtering calculation formula is: , where is the time, the weight coefficient w = [w 0 , w 1 , w 2 =[0.5, 0.3, 0.2], is the window size, is the current estimated value after weighted moving average filtering at time is the data value before weighted average filtering at time.

[0050] Embodiment 2: The present invention also provides an indoor positioning system based on the fusion of channel state information and depth features. The indoor positioning system based on the fusion of channel state information and depth features can be implemented by executing the process steps of the indoor positioning method based on the fusion of channel state information and depth features. That is, those skilled in the art can understand the indoor positioning method based on the fusion of channel state information and depth features as the preferred implementation manner of the indoor positioning system based on the fusion of channel state information and depth features.

[0051] Specifically, the indoor positioning system based on the fusion of channel state information and depth features is based on an indoor positioning method based on the fusion of channel state information and depth features in the above-mentioned Embodiment 1, and includes: Module M1, including multiple ESP32s, is used to collect channel state information data; Module M2, including an embedded processor, is used to deploy a residual-attention model to process the channel state information data and output a positioning result; Module M3, including a weighted moving average filter module, is used to post-process and optimize the positioning result.

[0052] Embodiment 3: Standard configuration implementation See Figures 1 - 7 , including the following steps: Step 1: Data collection. Arrange 1 router and multiple ESP32s in the indoor environment, collect CSI data at different positions, and synchronously record the real position information.

[0053] Step 2: Preprocessing. Adjust the CSI data format to adapt to the input requirements of the deep learning model. Perform Butterworth filtering, static component elimination, and Min-Max normalization on the adjusted CSI data. The position data is independently normalized. The dual-path normalization design eliminates the dimension difference.

[0054] Step 3: Deep learning model training. First, use a CNN model combined with a ResNet module for feature extraction. The convolutional layer uses a 3×3 convolutional kernel, ReLU activation, and outputs 32 channels; the residual block completes the skip connection of 32→64→128 channels; the pooling layer uses 2 times of 2×2 max pooling, see Figure 3 . Then, use the channel attention and spatial attention mechanisms to enhance the feature expression ability of the model. The channel attention module uses a 128→8→128-dimensional fully connected structure, and the reduction is set to 16; the spatial attention performs dual-channel fusion with a 7×7 convolutional kernel. Then train the model and optimize it using the loss function.

[0055] Step 4: Predict the position coordinates. Predict the CSI data through the trained model to obtain the preliminary position information.

[0056] Step 5: WMA optimization. Use WMA to smooth the predicted position information.

[0057] Step 6: Output the positioning result. Record the final smoothed positioning coordinates, see Figure 7 , and perform error evaluation.

[0058] Embodiment 4: Implementation of Attention Mechanism Parameter Optimization for Industrial Plant Scenarios Step 1: Adjust the reduction in the channel attention module to 8 to improve feature utilization rate.

[0059] Step 2: Reduce the convolutional kernel in the spatial attention module, set the kernel_size to 5, and use a small receptive field to adapt to the dense device environment.

[0060] Step 3: Adjust the training strategy. In the first 10 epochs, perform learning rate warm-up from 0.0001 to 0.001 linearly, and adjust the batch size to 64 to improve training stability.

[0061] See the attention weights in Figure 4 , 5.

[0062] Example 5: Implementation of Moving Average Filter Enhancement Step 1: Data collection. In the experimental scenario, collect CSI data and record the true coordinates at multiple location points. See Figure 2 .

[0063] Step 2: Data processing. Normalize the data to ensure consistent dimensions; train the CNN-ResNet model and adjust the learning rate strategy; use the trained model to initially predict the location information.

[0064] Step 3: Optimization of weighted moving average filtering with different window sizes. Smooth the prediction results using different window sizes (such as 3, 5, 7), compare the error situations under different window sizes, and select the optimal parameters.

[0065] Step 4: Output the final positioning result.

[0066] See the effect of moving average filtering in Figure 6 , 7.

[0067] Those skilled in the art know that in addition to implementing the system and its various devices, modules, and units provided by the present invention in the form of pure computer-readable program code, the method steps can be logically programmed to enable the system and its various devices, modules, and units provided by the present invention to be implemented in the form of logic gates, switches, application-specific integrated circuits, programmable logic controllers, and embedded microcontrollers, etc. to achieve the same functions. Therefore, the system and its various devices, modules, and units provided by the present invention can be regarded as a kind of hardware component, and the devices, modules, and units included therein for implementing various functions can also be regarded as the structures within the hardware component; the devices, modules, and units for implementing various functions can also be regarded as both software modules for implementing the method and the structures within the hardware component.

[0068] The specific embodiments of the present invention have been described above. It should be understood that the present invention is not limited to the above specific embodiments, and those skilled in the art can make various changes or modifications within the scope of the claims, which do not affect the essence of the present invention. Without conflict, the embodiments of the present application and the features in the embodiments can be combined with each other arbitrarily.

Claims

1. An indoor positioning method based on the fusion of channel state information and depth features, characterized in that: The following steps are involved: Step S1, collecting channel state information data of Wi-Fi transceiver devices through multiple ESP32; Step S2, extracting amplitude data of each subcarrier from the channel state information data; Step S3, preprocessing the amplitude data to obtain preprocessed data; Step S4, extracting shallow features from the preprocessed data through an initial convolutional layer to obtain a shallow feature map; Step S5, extracting deep spatiotemporal features from the shallow feature map through a residual block group to obtain a deep feature map; Step S6, performing channel attention weighting and spatial attention weighting on the deep feature map in sequence to obtain a weighted feature map; Step S7, mapping the weighted feature map to a two-dimensional coordinate space through a fully connected layer, and outputting a positioning result; Step S8: Use a weighted moving average filter to perform time series smoothing on the positioning result.

2. According to claim 1, the indoor positioning method based on the fusion of channel state information and depth features is characterized in that: In the step S1, the Wi-Fi module of ESP32 is used to collect channel state information data of the Wi-Fi transceiver device in accordance with the IEEE 802.11n / ac protocol. The ESP32 is connected to a router as a Station, and a data packet is sent to the router to collect channel state information data, and the real location information corresponding to the channel state information data is synchronously recorded. The channel state information data is stored in an SD card, and the collection parameters of the channel state information data are set by presetting configuration items.

3. According to claim 1, the indoor positioning method based on the fusion of channel state information and depth features is characterized in that: In step S2, the amplitude data of the subcarrier is calculated based on the real part and imaginary part of the subcarrier in the channel state information data. , is the real part of each subcarrier, is the imaginary part of each subcarrier.

4. According to claim 1, the indoor positioning method based on the fusion of channel state information and depth features is characterized in that: In step S3, the preprocessing includes the following sub-steps: Step S3.1, use a 6th-order Butterworth low-pass filter to remove high-frequency noise with a cutoff frequency of 30 Hz; Step S3.2, using a sliding window to calculate a moving average to eliminate static components, the sliding window length is set according to the sampling rate of 60 Hz; Step S3.3, the denoised channel state information data and position data are respectively normalized to the interval [0,1] using the Min-Max normalization formula: .

5. The indoor positioning method based on the fusion of channel state information and depth features according to claim 1, characterized in that: In step S4, the initial convolution layer includes a 3×3 convolution kernel, a ReLU activation function and a 2×2 maximum pooling operation.

6. The indoor positioning method based on the fusion of channel state information and depth features according to claim 1, characterized in that: In step S5, the residual block group includes multiple residual blocks, each residual block includes: a skip connection path and a main path, the main path is: a 3×3 convolution layer, a ReLU activation function and a 2×2 maximum pooling layer connected in sequence; the skip connection path is: if the number of input channels and the number of output channels of the main path are inconsistent, the channel dimension is adjusted through a 1×1 convolution layer, and the adjusted main path output and the skip connection output are added element by element, and then output through a ReLU activation function, wherein the activation function ReLU introduces nonlinearity, and the maximum pooling halves the size of the feature map, reducing the computational complexity.

7. The indoor positioning method based on the fusion of channel state information and depth features according to claim 1, characterized in that: The step S6 comprises the following sub-steps: Step S6.1, channel attention weighting uses dual-path pooling: global average pooling and spatial maximum pooling are calculated in parallel for the deep feature map, the global average pooling takes the average along the spatial dimension to capture the overall features; the spatial maximum pooling takes the maximum value along the spatial dimension to capture the significant features, the result of dual-path pooling is reduced in dimension and transformed in feature by the fully connected layer, and then the weight generation is completed, the results of average pooling and maximum pooling are added, and the sigmoid activation function is used. Generate channel attention weights ,in is the input deep feature map, is the activation function, outputting the attention weight of each channel [0,1]; Step S6.2, spatial attention weighting uses two-way compression: the mean and maximum values ​​of the feature map are calculated along the channel dimension, and then feature fusion is performed to complete the two-way feature splicing, and the spatial attention weight is generated through convolution operation , complete the spatial feature weighting, multiply the channel attention weight and the spatial attention weight to output the weighted feature map.

8. The indoor positioning method based on the fusion of channel state information and depth features according to claim 1, characterized in that: In step S7, the Huber loss function is used ,in, is the true value, is the predicted value, When set to 1.0, the optimizer uses the Adam optimization algorithm with adaptive learning rate to promote Adam stable optimization.

9. The indoor positioning method based on the fusion of channel state information and depth features according to claim 1, characterized in that: In step S8, the weighted moving average filter adopts a fixed window size, and the filter calculation formula is: ,in, is the moment, weight coefficient w=[w0,w1,w2]=[0.5,0.3,0.2], is the window size, is current The estimated value after weighted moving average filtering at each moment, yes The data value before weighted average filtering at each moment.

10. An indoor positioning system based on the fusion of channel state information and depth features, based on the indoor positioning method based on the fusion of channel state information and depth features as described in any one of claims 1-9, characterized in that: include: Module M1, including multiple ESP32, is used to collect channel status information data; Module M2, including an embedded processor, is used to deploy a residual-attention model to process the channel state information data and output a positioning result; Module M3, including a weighted moving average filter module, is used for performing post-processing optimization on the positioning result.

Images