RFID intelligent label positioning and tracking printing method and system

Abstract

The invention relates to the technical field of RFID smart tags, and discloses an RFID smart tag positioning and tracking printing method and system, and the method comprises the steps: inputting an RFID tag radio frequency signal into a conditional variation automatic encoder for convolution processing and attention weighting, and obtaining an RFID tag global feature vector; establishing a layered sensor node network to perform multi-dimensional acquisition on environmental parameters to obtain an environmental spatio-temporal data set; inputting the global feature vector of the RFID tag and the environment spatio-temporal data set into a deep fusion network for feature fusion to obtain a multi-modal fusion feature; performing sequence modeling based on the multi-modal fusion features to obtain an article trajectory prediction model; the object track prediction model is applied to a target area grid coordinate system for density distribution calculation, a real-time object positioning heat map and a track database are obtained, an efficient object positioning heat map generation and track data management mechanism is established, and real-time position visualization and historical track query are facilitated for the system.

Description

Technical Field

[0001] This application relates to the technical field of RFID intelligent tags, and particularly to a method and system for RFID intelligent tag positioning and tracking printing. Background Art

[0002] With the rapid development of the Internet of Things technology, RFID technology has been widely used in the field of item positioning and tracking. However, traditional RFID positioning systems face problems such as signal attenuation, multipath effects, and environmental interference in complex environments, resulting in difficulties in meeting the actual requirements for positioning accuracy and stability.

[0003] Current RFID positioning systems usually use a single radio frequency signal feature for position estimation, lacking comprehensive consideration of environmental factors and being unable to effectively handle dynamically changing actual scenarios. Especially in an environment where multiple items are densely distributed, due to the influence of signal interference and occlusion effects, the positioning performance of the system significantly decreases. Summary of the Invention

[0004] This application provides a method and system for RFID intelligent tag positioning and tracking printing, and further establishes an efficient mechanism for generating item positioning heatmaps and managing trajectory data, facilitating real-time position visualization and historical trajectory query by the system.

[0005] In the first aspect of this application, a method for RFID intelligent tag positioning and tracking printing is provided. The method for RFID intelligent tag positioning and tracking printing includes: Inputting the RFID tag radio frequency signal into a conditional variational autoencoder for convolutional processing to obtain the RFID tag structured feature; Performing attention weighting on the RFID tag structured feature to obtain the RFID tag global feature vector; Establishing a hierarchical sensor node network to perform multi-dimensional acquisition of environmental parameters to obtain an environmental spatio-temporal data set; Inputting the RFID tag global feature vector and the environmental spatio-temporal data set into a deep fusion network for feature fusion to obtain a multi-modal fusion feature; Performing sequence modeling based on the multi-modal fusion feature to obtain an item trajectory prediction model; Applying the item trajectory prediction model to a target area grid coordinate system for density distribution calculation to obtain a real-time item positioning heatmap and a trajectory database.

[0006] In the second aspect of this application, a system for RFID intelligent tag positioning and tracking printing is provided. The system for RFID intelligent tag positioning and tracking printing includes: A convolutional processing module, configured to input the RFID tag radio frequency signal into a conditional variational autoencoder for convolutional processing to obtain the RFID tag structured feature; An attention weighting module for attention weighting the structured features of the RFID tag to obtain a global feature vector of the RFID tag; A multi-dimensional acquisition module for establishing a hierarchical sensor node network to perform multi-dimensional acquisition of environmental parameters to obtain an environmental spatio-temporal data set; A feature fusion module for inputting the global feature vector of the RFID tag and the environmental spatio-temporal data set into a deep fusion network for feature fusion to obtain a multi-modal fusion feature; A sequence modeling module for performing sequence modeling based on the multi-modal fusion feature to obtain an item trajectory prediction model; A distribution calculation module for applying the item trajectory prediction model to a target area grid coordinate system for density distribution calculation to obtain a real-time item positioning heat map and a trajectory database.

[0007] Compared with the prior art, the present application has the following beneficial effects: By introducing a conditional variational autoencoder and an attention mechanism, deep feature extraction and weighting processing are performed on the RFID radio frequency signal, improving the robustness and discriminability of feature expression, and effectively overcoming the signal interference problem in a complex environment; Adopting a hierarchical sensor network architecture and a multi-dimensional environmental parameter acquisition strategy, comprehensive perception and real-time monitoring of the environmental state of the target area are realized, providing rich environmental context information for item positioning; Designing a deep fusion network for multi-modal feature fusion, through dual-branch feature extraction and cross-modal interaction, effectively integrating the RFID signal features and environmental parameter information, enhancing the environmental adaptability of the system; A trajectory prediction method based on sequence modeling and constraint optimization, combined with a temporal attention mechanism and an autoregressive decoding structure, realizes accurate prediction and smoothing processing of the item movement trajectory; Through grid density distribution calculation and spatio-temporal index construction, an efficient item positioning heat map generation and trajectory data management mechanism is established, facilitating real-time position visualization and historical trajectory query of the system. Description of the Drawings

[0008] In order to more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the following will briefly introduce the drawings required for the description of the embodiments or the prior art. Obviously, the drawings in the following description are only some embodiments of the present invention. For those of ordinary skill in the art, other drawings can be obtained based on these drawings without creative efforts.

[0009] The structures, proportions, sizes, etc. shown in the drawings of this specification are only used to cooperate with the content disclosed in the specification for those familiar with this technology to understand and read, and are not used to limit the conditions under which the present invention can be implemented. Therefore, they do not have substantial technical significance. Any modification of the structure, change in the proportional relationship, or adjustment of the size, without affecting the effects that the present invention can produce and the purposes that can be achieved, should still fall within the scope that can be covered by the technical content disclosed in the present invention.

[0010] Figure 1 is a schematic flowchart of an RFID intelligent tag positioning and tracking printing method provided by an embodiment of the present invention; Figure 2 is a schematic block diagram of the structure of an RFID intelligent tag positioning and tracking printing system provided by an embodiment of the present invention. Detailed implementation manners

[0011] Next, the technical solutions in the embodiments of the present invention will be clearly and completely described in conjunction with the accompanying drawings in the embodiments of the present invention. Obviously, the described embodiments are part of the embodiments of the present invention, rather than all the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those of ordinary skill in the art without creative efforts shall fall within the protection scope of the present invention.

[0012] The flowcharts shown in the drawings are only illustrative examples, and do not necessarily include all the content and operations / steps, nor do they necessarily need to be executed in the described order. For example, some operations / steps can also be decomposed, combined, or partially merged. Therefore, the actual execution order may change according to the actual situation.

[0013] It should also be understood that the terms used in the specification of this application are only for the purpose of describing specific embodiments and are not intended to limit this application. As used in the specification of this application and the appended claims, unless the context clearly indicates otherwise, the singular forms "a", "an", and "the" are intended to include the plural forms.

[0014] It should be further understood that the term "and / or" used in the specification of this application and the appended claims refers to any combination and all possible combinations of one or more of the associated listed items, and includes these combinations. Please refer to Figure 1 , an embodiment of the RFID intelligent tag positioning and tracking printing method in the embodiments of this application includes: Step 100: Input the RFID tag radio frequency signal into a conditional variational autoencoder for convolutional processing to obtain the RFID tag structured features; It can be understood that the execution entity of this application can be an RFID intelligent tag positioning and tracking printing system, or it can also be a terminal or a server. Specifically, it is not limited here. In the embodiments of this application, the server is used as an example of the execution entity for illustration.

[0015] Specifically, the RFID tag radio frequency signal is used as input data and imported into the feature extraction layer of the conditional variational autoencoder. The feature extraction layer performs convolutional operations on the input signal to extract multi-scale feature maps. The feature extraction layer contains 5 parallel convolutional kernels, and the stride of each convolutional kernel is set to 2. This design captures the multi-scale characteristics of the input signal through different receptive fields. At the same time, to ensure the stability of the convolutional operation and accelerate the convergence of the model, a batch normalization layer and a ReLU activation function are added after each convolutional layer. The batch normalization layer reduces the distribution variation of the input features, and the ReLU activation function effectively introduces non-linearity to improve the expression ability of the model. A max pooling operation is performed on the multi-scale feature maps to achieve dimensionality reduction. The max pooling operation uses a 2×2 pooling window with a stride of 2 as well, compressing the spatial dimension of the feature map while retaining the most significant feature information. Through max pooling, the computational complexity is reduced, and the expression ability of the feature map for local information is enhanced, generating a dimensionality-reduced feature map. The dimensionality-reduced feature map is input into the fully connected layer of the conditional variational autoencoder for dimensionality conversion, generating a feature vector with a fixed dimension. The fully connected layer consists of 3 hidden layers, and the number of neurons in each hidden layer is 512, 256, and 128 respectively. The gradually decreasing design helps to gradually compress the dimension of the features while retaining important global information. In the operation of the fully connected layer, through weight sharing and non-linear mapping, the model captures the latent characteristics of the RFID signal at a deeper level and compresses it into a vector representation with a fixed length. To enhance the representation ability of the feature vector, an adaptive weight is applied to the feature vector with a fixed dimension. The adaptive weight is dynamically calculated based on the feature importance score, that is, by evaluating the contribution of the model to the features, weights are assigned to each feature. It effectively highlights key features while suppressing redundant information, generating a weighted feature vector. The weighted feature vector is input into the encoder of the conditional variational autoencoder for compression mapping. The encoder is designed to consist of a mean encoder and a variance encoder, and these two encoders output the mean vector and variance vector of the latent space respectively. The mean vector and variance vector jointly define the distribution of the latent space features. By modeling the mean and variance, the latent variable distribution of the input signal is captured. The reparameterization sampling technique is used to sample the latent space features. The reparameterization sampling generates sampling points that follow a standard normal distribution through the mean vector and variance vector, realizing the randomized expression of the latent space features. Through reparameterization sampling, the structured features of the RFID tag are finally generated.

[0016] Step 200: Perform attention weighting on the structured features of the RFID tag to obtain a global feature vector of the RFID tag; Specifically, the structured features of the RFID tag are input into the multi-head attention mechanism for feature decomposition. The multi-head attention mechanism processes different feature subspaces through parallel attention heads to more comprehensively capture the context information of the RFID signal. The multi-head attention mechanism consists of 8 parallel attention heads, and each attention head independently processes the input features and outputs a 64-dimensional feature subspace. This design can not only ensure that the model has sufficient expressive power, but also effectively decompose and capture the importance of input features in different dimensions, generating multiple groups of attention feature subspaces with diverse characteristics. The multiple groups of attention feature subspaces are input into a multi-layer perceptron for non-linear transformation to generate a feature mapping matrix. The multi-layer perceptron includes two hidden layers, with the dimensions of each layer being 512 and 256 respectively, and LeakyReLU is used as the activation function for each layer. LeakyReLU has a certain negative slope compared to the traditional ReLU, which can retain part of the gradient when the input value is negative, alleviating the problem of gradient disappearance. Through the non-linear transformation of the multi-layer perceptron, the model can learn more complex feature interaction relationships and generate a feature mapping matrix with deep-level information. The feature mapping matrix is processed through dot product operation and softmax normalization function to calculate the importance score of each feature, that is, the attention score. The dot product operation can quantify the relationship between the feature mapping matrix and the structured features into weight signals, while the softmax function normalizes these weight signals to ensure that the sum of the importance scores of all features is 1. The attention score reflects the contribution of different features to the global context. The structured features of the RFID tag and the attention score are subjected to weighted summation operation to generate a global context vector. In the form of matrix multiplication, the weighted summation can be computationally efficient. The global context vector is input into the residual connection module for feature fusion to obtain a fused feature vector. The residual connection module includes a layer normalization layer and a skip connection. The layer normalization layer normalizes the feature vector to eliminate the problem of internal covariate shift, while the skip connection retains the original information of the input features and superimposes them with the features after non-linear transformation. The fused feature vector is subjected to dimensionality reduction transformation. A 1×1 convolutional layer is used to reduce the feature dimension, mapping the high-dimensional fused feature vector to a 128-dimensional global feature space. The 1×1 convolutional layer can reduce the feature dimension and learn the optimal combination method between different feature dimensions through the weight matrix, thereby improving the compactness and effectiveness of feature expression. Through the above steps, a 128-dimensional global feature vector is finally generated.

[0017] Step 300: Establish a hierarchical sensor node network to collect environmental parameters in multiple dimensions and obtain an environmental spatio-temporal dataset; It should be noted that multiple sensor nodes are deployed in a grid structure to construct the topology of a hierarchical sensor network. This grid-based deployment method aims to cover all parts of the target area. By reasonably planning the positions and spacings of the nodes, it ensures that the sensor network can not only fully sense environmental changes but also avoid node redundancy. Each sensor node is equipped with a temperature sensor, a humidity sensor, a light intensity sensor, and a motion sensor. The combination of these sensors can collect multi-dimensional parameters of the environment, including temperature distribution, humidity changes, light intensity, and the motion state of objects. The master-slave node division is carried out for the topology of the hierarchical sensor network to establish a hierarchical system for data collection. The master node is designated as the central node responsible for data aggregation, while the slave nodes focus on collecting parameters of the local environment. The master-slave node division is optimally configured according to the geographical location, hardware performance, and network coverage of the nodes, so as to maximize the transmission efficiency and stability of the network. Through the master node's centralized processing of the data collected by the slave nodes, the entire network realizes efficient data management while reducing transmission latency and bandwidth occupancy. Based on the data collection hierarchical system, environmental parameters are synchronously sampled to ensure the consistency of data collection among different sensor nodes in the time dimension, thus avoiding biases or errors introduced by asynchronous data collection times. In this process, environmental parameters such as temperature, humidity, light intensity, and the motion trajectories of objects are collected to form the original environmental data stream. The original data stream is arranged in chronological order, reflecting the dynamic changes and spatial distributions of environmental parameters in the target area. The original environmental data stream is input into the data compression module for compression encoding. An efficient data compression algorithm is used to eliminate redundant information to reduce the overhead of data storage and transmission. Compression encoding can effectively reduce the data volume while maintaining the integrity and recoverability of key information. After data compression, error correction coding is performed on the compressed data packets to enhance the anti-interference ability of the data during transmission. The introduction of error correction coding effectively corrects errors caused by network fluctuations or interference during data transmission, ensuring the reliability and integrity of the data. The transmission data stream processed by compression coding and error correction coding is decoded and reconstructed to generate a data set that can be further utilized. During the decoding process, a timestamp and a location identifier are added to each piece of data. The introduction of the timestamp marks the time point at which each piece of data is collected, thus supporting the temporal analysis of environmental parameters, while the location identifier is used to mark the geographical location or node number corresponding to the data, enabling environmental data to be accurately mapped to the grid coordinates of the target area. Through this step, the environmental data is transformed from a disordered and scattered state into an environmental spatio-temporal data set with clear spatio-temporal information and good organization.

[0018] Step 400: Input the global feature vector of the RFID tag and the environmental spatio-temporal data set into the deep fusion network for feature fusion to obtain the multi-modal fusion feature; Specifically, perform time alignment processing on the global feature vector of the RFID tag to ensure the consistency of feature data and environmental spatio-temporal data in the time dimension. Depending on the timestamp information attached to the environmental spatio-temporal dataset, by comparing and matching the sampling times of RFID tag features and environmental parameters, adjust the time distribution of the feature sequence to make it synchronized with the environmental parameters. Through time alignment, avoid the problem of feature misalignment caused by out-of-sync data in time. Input the aligned feature sequence into the first feature extraction branch of the deep fusion network, and at the same time input the environmental spatio-temporal dataset into the second feature extraction branch, and the two branches perform feature extraction operations in parallel. The first feature extraction branch and the second feature extraction branch adopt the same structural design, and each branch includes 3 convolutional layers and 2 fully connected layers. The convolutional layer is used to extract the spatial features of the input data, and local patterns and relationships are mined through layer-by-layer operations, while the fully connected layer converts the extracted features into high-level semantic representations. The two branches work independently and generate deep feature representations of RFID tag features and environmental parameters respectively. Perform cross-modal interaction calculation on the bimodal feature representation to capture the potential correlation between RFID tag features and environmental information. The cross-modal interaction calculation maps the features of the two modalities into the same feature space through matrix operations and generates an interaction feature matrix, revealing the complex relationship between RFID tag signals and environmental influencing factors, such as the interference effect of environmental temperature, humidity, etc. on the propagation of radio frequency signals. Input the interaction feature matrix into the feature enhancement module for processing to generate an enhanced feature vector. The feature enhancement module uses residual blocks as basic components, and each residual block consists of two 3×3 convolutional layers and a skip connection. The design of the residual block can introduce non-linear transformation into the feature stream, and at the same time retain the original information of the input features through the skip connection, thus avoiding the problem of gradient disappearance while improving the feature expression ability. The stacking of multiple residual blocks realizes the deep reconstruction of the features, making the enhanced feature vector more expressive and stable. Perform multi-scale feature extraction on the enhanced feature vector to construct a multi-scale feature pyramid. The multi-scale feature extraction captures the local details and global information of the features through convolutional operations of different scales, so as to ensure that the features can adapt to the task requirements of different granularities. The feature pyramid structure can effectively integrate information from different scales and provide more comprehensive context support for positioning and tracking. After the feature pyramid is constructed, fuse the features of its layers through concatenation operations to generate a multimodal fusion feature.

[0019] Step 500: Based on the multimodal fusion feature, perform sequence modeling to obtain an item trajectory prediction model; Specifically, the multi-modal fusion features are temporally segmented to generate fixed-length feature sequence segments. The continuous feature data is segmented into segments with a fixed time window to better capture the dynamic changes of the object in different time periods. The length of the feature sequence segments is adjusted according to the actual task requirements to balance the integrity of temporal information and computational complexity. After completing the temporal segmentation, these segments are input into a sequence encoder for feature encoding to generate a hidden state sequence. The sequence encoder adopts a bidirectional long short-term memory network (Bi-LSTM) structure, which simultaneously captures the dynamic features of the forward and backward dependencies in the time series through bidirectional information flows in the forward and backward directions. Bi-LSTM can effectively avoid information loss and fully retain the global dependencies of the object trajectory in the time dimension. Temporal attention calculation is performed on the hidden state sequence to obtain a weighted context vector. The temporal attention mechanism assigns weights to each time step in the hidden state sequence, highlighting the feature information at key time points while suppressing the interference of non-key time points. The attention weights are obtained by calculating the dot product of the hidden state and the context query vector and are normalized by softmax. The weighted context vector can synthesize the important information in the temporal features. The weighted context vector is input into a trajectory generator for trajectory decoding to generate a trajectory sequence representation. The trajectory generator adopts an autoregressive decoding structure, which includes 4 layers of Transformer decoding layers. The Transformer decoding layers use a combination of multi-head attention mechanisms and feed-forward neural networks to efficiently decode the input features. The multi-head attention mechanism captures the global dependencies between features, while the feed-forward neural network further models the non-linear features in the decoding process. The autoregressive decoding structure ensures a high generation accuracy of the trajectory sequence by gradually predicting the trajectory points at each time step. The generated trajectory sequence representation is optimized by constraints to obtain a smooth trajectory sequence. The constraint optimization process includes position continuity constraints and motion smoothness constraints. The position continuity constraint is used to ensure that the spatial position changes between adjacent trajectory points conform to physical laws and avoid jumps or unreasonable trajectories. The motion smoothness constraint constrains the speed and acceleration of the trajectory to make the generated trajectory smoother and more in line with physical characteristics. The combination of these two constraints can effectively improve the stability and realism of the trajectory, making it more suitable for actual application scenarios. The smooth trajectory sequence is input into a prediction head network for parameter regression to generate an object trajectory prediction model. The prediction head network consists of 3 fully connected layers. The first layer and the second layer are used for feature mapping and dimensionality reduction respectively, while the last layer is used for regressing the trajectory parameters. The dimension of the output layer matches the dimension of the trajectory parameters, ensuring that the model can directly generate the key parameters of the trajectory, such as position coordinates, velocity vectors, or angle changes, etc. Through parameter regression, the trajectory prediction model converts the smooth trajectory sequence into trajectory information that can be directly used by downstream tasks.

[0020] Step 600: Apply the object trajectory prediction model to the target area grid coordinate system for density distribution calculation to obtain a real-time object positioning heat map and a trajectory database.

[0021] Specifically, perform coordinate mapping on the target area to generate a grid coordinate matrix of the target area. By dividing the target area into grid cells of the same size, the actual physical space is converted into a discretized grid representation. The size of the grid cells is adjusted according to the resolution requirements of the application scenario, so as to achieve a balance between spatial accuracy and computational complexity. The result of coordinate mapping is a multi-dimensional grid coordinate matrix, where each grid cell corresponds to a specific spatial position. Input the trajectory parameter sequence output by the object trajectory prediction model into the coordinate transformation module for grid projection processing. The trajectory parameter sequence includes the three-dimensional spatial coordinates of the object (such as x, y, z) and motion state parameters (such as speed, acceleration, and direction angle). The coordinate transformation module maps these continuous spatial trajectory points to the corresponding grid cells according to the grid division rules of the target area to generate a grid space trajectory sequence. Perform density accumulation calculation on the grid space trajectory sequence to generate an initial density distribution map. The density accumulation calculation estimates the usage degree of the grid based on the residence time and occurrence frequency of each object in each grid cell. Whenever an object passes through a certain grid cell, a weight is added to the density value of the cell according to its residence time; at the same time, the occurrence frequency is a key factor for density accumulation, which is used to reflect the activity degree of the object in this area. Accordingly, an initial density distribution map is generated, which is a two-dimensional matrix reflecting the distribution of objects in the target area grid cells. Input the initial density distribution map into the density smoothing module for Gaussian filtering processing. Gaussian filtering is a smoothing technique that effectively suppresses noise and smooths local density fluctuations by performing weighted averaging on density values in the neighborhood, making the distribution map more coherent and natural. After Gaussian filtering processing, a dynamic density matrix is obtained, which reflects the dynamic distribution trend of objects in the area. Normalize the dynamic density matrix so that the density values in the matrix are standardized to a unified range (such as [0,1]). Then perform threshold segmentation. According to the preset density threshold, divide the grid cells into high-density areas and low-density areas to generate a real-time object positioning heat map. In order to efficiently store and retrieve trajectory data and the dynamic density matrix, construct a spatio-temporal index. Organize the grid space trajectory sequence and the dynamic density matrix through a quadtree structure to generate a trajectory database. The quadtree is a hierarchical index structure suitable for spatial data. By recursively dividing the space into four quadrants, the data storage and retrieval efficiency is significantly improved. The leaf nodes of the quadtree store trajectory segments and density information, while the internal nodes record spatial ranges and density statistical information, thus supporting fast range queries and spatio-temporal analysis.

[0022] In the embodiments of the present application, by introducing a conditional variational autoencoder and an attention mechanism, deep feature extraction and weighted processing are performed on RFID radio frequency signals, enhancing the robustness and discriminability of feature representation and effectively overcoming the signal interference problem in complex environments; a hierarchical sensor network architecture and a multi-dimensional environmental parameter acquisition strategy are adopted to achieve a comprehensive perception and real-time monitoring of the environmental state of the target area, providing rich environmental context information for item positioning; a deep fusion network is designed for multi-modal feature fusion, and through dual-branch feature extraction and cross-modal interaction, the RFID signal features and environmental parameter information are effectively integrated, enhancing the environmental adaptability of the system; a trajectory prediction method based on sequence modeling and constraint optimization, combined with a temporal attention mechanism and an autoregressive decoding structure, realizes the accurate prediction and smooth processing of the item movement trajectory; through grid-based density distribution calculation and spatio-temporal index construction, an efficient item positioning heatmap generation and trajectory data management mechanism is established, facilitating real-time position visualization and historical trajectory query of the system.

[0023] In a specific embodiment, the process of executing step 100 may specifically include the following steps: Input the RFID tag radio frequency signal into the feature extraction layer of the conditional variational autoencoder for convolution operation to obtain a multi-scale feature map. The feature extraction layer includes 5 parallel convolutional kernels, the stride of each convolutional kernel is 2, and a batch normalization layer and a ReLU activation function are connected after each convolutional layer; Perform a max pooling operation on the multi-scale feature map to obtain a dimensionality-reduced feature map, where the max pooling operation uses a 2×2 pooling window and a stride of 2; Input the dimensionality-reduced feature map into the fully connected layer of the conditional variational autoencoder for dimensionality conversion to obtain a feature vector with a fixed dimension. The fully connected layer includes 3 hidden layers, and the number of neurons in each hidden layer is 512, 256, and 128 in sequence; Apply an adaptive weight to the feature vector with a fixed dimension to obtain a weighted feature vector, and the adaptive weight is dynamically calculated based on the feature importance score; Input the weighted feature vector into the encoder of the conditional variational autoencoder for compression mapping to obtain latent space features. The encoder includes a mean encoder and a variance encoder, which respectively output the mean vector and variance vector of the latent space; Perform reparameterization sampling on the latent space features to obtain the RFID tag structured features, where the reparameterization sampling generates sampling points that follow a standard normal distribution based on the mean vector and variance vector.

[0024] Specifically, represent the radio frequency signal of the RFID tag as a time series data matrix , where represents the number of time steps, Represents the feature dimension at each time step. This signal is passed as input to the feature extraction layer of the conditional variational autoencoder. The feature extraction layer contains 5 parallel convolutional kernels, with a stride of 2 set for each convolutional kernel, and the sizes of the convolutional kernels are , representing different receptive field sizes. The multi-scale convolutional design can capture the feature patterns of the RF signal within different time ranges. The mathematical expression of convolution is: ; where represents the feature map extracted by the th convolutional kernel, is the weight matrix of the th convolutional kernel, is the bias term, * represents the convolution operation, BN( ) is the batch normalization operation, ReLU( ) is the ReLU activation function. Through this step, multiple multi-scale feature maps are generated, and by integrating the outputs of different convolutional kernels, the local and global features of the signal are captured. A max pooling operation is performed on the multi-scale feature maps to achieve dimensionality reduction. The max pooling operation compresses the spatial dimension of the feature map and reduces the computational complexity while retaining the significant features through a pooling window with a stride of 2. The mathematical expression is: ; where is the th pooled feature map, and window represents the range of feature values within the pooling window. The dimension-reduced feature map is input to the fully connected layer for dimension transformation to generate a fixed-length feature vector. The fully connected layer consists of 3 hidden layers, with the number of neurons being 512, 256, and 128 respectively. Each layer processes the input features through a linear transformation and an activation function, and the mathematical expression is: ; where represents the input features of the th layer, and are the weight and bias parameters respectively, and are the feature dimensions of the th layer and the th layer. The generated fixed-dimension feature vector is applied with adaptive weights to generate a weighted feature vector. The calculation of the adaptive weights is based on the feature importance score , which is calculated through a normalization mechanism, and the mathematical expression is: ; where represents the The importance score of a feature represents the corresponding normalized weight. The final weighted feature vector is: ; where is the vector after weight adjustment. The weighted feature vector is input into the encoder part of the conditional variational autoencoder for compression mapping to obtain the latent space features. The encoder includes a mean encoder and a variance encoder, which output the mean vector and the variance vector of the latent space respectively. This process is expressed as ; where and are the weights and biases of the linear transformation. Reparameterization sampling is performed on the latent space features to generate sample points conforming to the standard normal distribution. Reparameterization sampling is completed through the following formula: ; where is the sampled latent vector, is the random noise sampled from the standard normal distribution, is the square root of the variance.

[0025] In this embodiment, the following steps are further included between the RFID tag structured features and the attention weighting: Input the RFID tag structured features into the feature decomposition module for subspace decomposition to obtain multi-scale feature components. The feature decomposition module uses wavelet transform to decompose the feature space into 8 different frequency sub-bands; Perform adaptive threshold learning on the multi-scale feature components to obtain the noise threshold parameter matrix. The adaptive threshold learning constructs a loss function based on the feature statistical distribution and optimizes the threshold parameters through backpropagation; Input the multi-scale feature components and the noise threshold parameter matrix into the soft threshold processing unit for feature filtering to obtain the denoised feature map. The soft threshold processing uses a differentiable sigmoid function to replace the traditional hard threshold; Perform feature reconstruction on the denoised feature map to obtain the preliminary purified features. The feature reconstruction uses the inverse wavelet transform to recombine the multi-scale features into a unified feature space; Input the preliminary purified features into the feature enhancement network for non-negative constrained reconstruction to obtain the enhanced feature representation. The feature enhancement network contains 3 fully connected layers, and each layer is followed by a PReLU activation function; Construct an adaptive gain matrix based on the enhanced feature representation to obtain the feature gain coefficient. The adaptive gain matrix calculates the importance score of each feature component through the attention mechanism; Perform adaptive enhancement on the enhanced feature representation and the feature gain coefficient to obtain the final purified features. The adaptive enhancement uses element-wise multiplication to amplify the features; Input the final purified features into the feature calibration module for distribution alignment to obtain the calibrated RFID tag structured features. The distribution alignment uses the batch renormalization technique to ensure the stability of the feature distribution.

[0026] In a specific embodiment, the process of executing step 200 may specifically include the following steps: Perform multi-head attention decomposition on the RFID tag structured features to obtain multiple groups of attention feature subspaces. The multi-head attention decomposition includes 8 parallel attention heads, and each attention head outputs a 64-dimensional feature subspace; Input the multiple groups of attention feature subspaces into the multi-layer perceptron for non-linear transformation to obtain the feature mapping matrix. The multi-layer perceptron includes two hidden layers, and the hidden layer dimensions are 512 and 256 respectively. Each layer uses the LeakyReLU activation function; Calculate the attention scores based on the feature mapping matrix to obtain the feature weight coefficients. The attention scores are calculated through dot product operation and softmax normalization function; Perform weighted summation operation on the RFID tag structured features and the feature weight coefficients to obtain the global context vector. The weighted summation operation is implemented by matrix multiplication; Input the global context vector into the residual connection module for feature fusion to obtain the fused feature vector. The residual connection module contains a layer normalization layer and a skip connection; Perform dimensionality reduction transformation on the fused feature vector to obtain the global feature vector of the RFID tag. The dimensionality reduction transformation uses a 1×1 convolutional layer to reduce the feature dimension to 128 dimensions.

[0027] Specifically, represent the structured features of the RFID tag as a matrix , where represents the number of features, represents the dimension of each feature. To improve the model's learning ability for different feature dimensions, use the multi-head attention mechanism to decompose . The multi-head attention includes parallel attention heads. Each attention head performs an independent linear transformation on the input features to generate query , key and value matrices: ; where is the linear transformation weight matrix of the th attention head, represents the dimension of the output feature subspace of each attention head. Through linear transformation, the input features are projected into multiple groups of subspaces to facilitate capturing the feature relationships of different dimensions. Perform dot product operations on the query and key matrices of each attention head and normalize them through the softmax function to calculate the attention scores : ; where represents the correlation weight matrix between features in the th attention head, is the scaling factor used to stabilize gradient updates. Then multiply the attention scores by the value matrix to obtain the weighted feature representation: ; where is the output feature subspace of the th attention head. Concatenate the outputs of all attention heads and perform a linear transformation to obtain the output of the multi-head attention: ; where represents the concatenated features, is the output weight matrix. Input the features output by the multi-head attention into a multi-layer perceptron for non-linear transformation to generate a feature mapping matrix. The multi-layer perceptron includes two hidden layers with dimensions of 512 and 256 respectively. Each layer uses the LeakyReLU activation function, and its mathematical representation is: ; Among them is the input feature of the th layer, and are the weight matrix and the bias vector respectively, and are the th layer and the th layer feature dimensions. After being processed by the multi-layer perceptron, the feature mapping matrix can capture the non-linear relationship of the input features. Based on the feature mapping matrix , the attention score is calculated through the dot product operation and the softmax function: ; Among them is the feature weight coefficient matrix. The original structured feature and the weight coefficient are weighted and summed to generate the global context vector: ; Among them is the fused global feature representation. To optimize the feature representation, the global context vector is input into the residual connection module for feature fusion. The residual connection module includes layer normalization and skip connection, and is implemented through the following formula: ; Among them is the fused feature vector, and LayerNorm is the layer normalization operation used to eliminate the distribution shift. The fused feature vector is subjected to dimensionality reduction transformation to reduce the feature dimension to 128 dimensions. The dimensionality reduction adopts convolution operation, and its expression is: ; Among them and are the convolution weights and biases. After the dimensionality reduction transformation, the finally generated feature vector is the RFID tag global feature representation with a dimension of .

[0028] In a specific embodiment, the process of executing step 300 may specifically include the following steps: Deploy multiple sensor nodes in a grid structure to obtain a hierarchical sensor network topology. The sensor nodes include temperature sensors, humidity sensors, light intensity sensors, and motion sensors; Divide the hierarchical sensor network topology into master and slave nodes to obtain a data acquisition hierarchy system. The master node is responsible for data aggregation, and the slave nodes are responsible for collecting environmental parameters; Based on the data acquisition hierarchy system, synchronously sample the environmental parameters to obtain the original environmental data stream, where the environmental parameters include temperature distribution, humidity distribution, light intensity distribution, and item movement trajectory information; Input the original environmental data stream into the data compression module for compression encoding to obtain compressed data packets, and perform error correction encoding processing on the compressed data packets to obtain the transmission data stream; Decode and reconstruct the transmission data stream and add timestamps and location identifiers to obtain the environmental spatio-temporal data set.

[0029] Specifically, design a reasonable spatial partitioning scheme so that the target area is divided into regular grid cells, and each cell corresponds to one or more sensor nodes. Each sensor node is equipped with a temperature sensor, a humidity sensor, a light intensity sensor, and a motion sensor for collecting multi-dimensional environmental parameters. These nodes form a hierarchical topology through physical installation or virtual logical partitioning, where the mathematical model of the grid partitioning is expressed as: ; Among them, represents the network topology, is the set of sensor nodes, is the communication edge between nodes. The coordinates of the node represent its position in the grid cell. Divide the hierarchical sensor network topology into master and slave nodes. The master node is selected as the central node for data aggregation and transmission, and the slave nodes are responsible for collecting environmental parameters. The division of master and slave nodes is based on factors such as the geographical location, computing power, and communication distance of the nodes, and is achieved through the optimal sensor placement algorithm. The mathematical expression of the master-slave relationship is: ; Among them represents the set of master nodes, represents the set of slave nodes connected to the master node , is the number of slave nodes corresponding to each master node. Through the hierarchical mechanism, the master node collects the data of the slave nodes, reducing the communication overhead of the entire network. After completing the division of master and slave nodes, synchronously sample the environmental parameters based on the data acquisition hierarchy system. The goal of synchronous sampling is to ensure the consistency of data from different nodes in the time dimension and avoid data distortion caused by time delay or frequency inconsistency. Assume that each node collects data at time , which is expressed as: ; Among them , , and represent the temperature, humidity, light intensity, and motion parameters collected at time respectively. After collecting the data, the slave nodes transmit it to the corresponding master nodes, and the master nodes perform time alignment and preliminary aggregation on the data to form a multi-dimensional raw environmental data stream : ; where is the total number of nodes. The raw environmental data stream is input into the data compression module for compression encoding to reduce the occupancy of the data transmission bandwidth while maintaining data integrity. Lossless compression algorithms (such as Huffman coding) or lossy compression algorithms (such as predictive coding) are used for data compression. For a data packet , its compression process is expressed as: ; where is the compression function, and is the compressed data packet. To enhance the transmission reliability, error correction coding is performed on the compressed data packet to generate a transmission data stream that can resist noise interference. Convolutional coding or low-density parity-check coding is used for error correction coding, and its mathematical description is: ; where is the transmission data packet after error correction coding. The transmission data stream is sent to the data center through the communication network, and the data center decodes and reconstructs it. The decoding process includes inverse error correction and decompression operations: ; where represents the reconstructed data packet. To achieve spatio-temporal annotation of the data, timestamps and location identifiers are added to the reconstructed data. The timestamp records the sampling time, and the location identifier corresponds to the node location where the data is collected. The environmental spatio-temporal dataset is obtained: ; where each record contains environmental parameters, sampling time, and node location.

[0030] In a specific embodiment, the process of executing step 400 may specifically include the following steps: Perform time alignment processing on the global feature vector of the RFID tag to obtain an aligned feature sequence, and the time alignment processing is synchronized based on the timestamp information in the environmental spatio-temporal dataset; Input the aligned feature sequence into the first feature extraction branch of the deep fusion network, and input the environmental spatio-temporal dataset into the second feature extraction branch of the deep fusion network for parallel feature extraction to obtain a bimodal feature representation. Both the first feature extraction branch and the second feature extraction branch include 3 convolutional layers and 2 fully connected layers; Perform cross-modal interaction calculation on the bimodal feature representation to obtain an interaction feature matrix, and input the interaction feature matrix into the feature enhancement module for feature reconstruction to obtain an enhanced feature vector. The feature enhancement module contains multiple residual blocks, and each residual block consists of two 3×3 convolutional layers and a skip connection; Perform multi-scale feature extraction on the enhanced feature vector to obtain a multi-scale feature pyramid, and perform feature splicing and fusion on the multi-scale feature pyramid to obtain a multimodal fusion feature.

[0031] Specifically, perform time alignment processing on the global feature vector of the RFID tag where represents the sequence length of the feature vector, represents the feature dimension. Based on the timestamp information in the environmental spatio-temporal dataset, ensure that the RFID features and environmental data are strictly synchronized in the time dimension. Assume that the environmental spatio-temporal data is represented as , and each record includes a timestamp and the corresponding environmental feature . Perform time alignment through the following formula: ; where represents the aligned RFID feature, is the target time step, is the timestamp in the environmental data, ensuring that for each time point the closest environmental data time point is found. Through this process, generate the aligned RFID feature sequence and the corresponding environmental feature sequence. Input the aligned RFID feature sequence into the first feature extraction branch of the deep fusion network, and at the same time input the aligned environmental data into the second feature extraction branch. Both branches consist of 3 convolutional layers and 2 fully connected layers, and are used to extract high-level modal features. For the first branch, feature extraction is performed through the following convolutional formula: ; where represents the input feature of the th layer of convolution, is the convolutional kernel weight, is the bias, BN( is an activation function. After passing through three convolutional layers, the generated high-dimensional features are input into a fully connected layer for further processing: ; where and are the parameters of the fully connected layer. The processing of the environmental data in the second branch is similar to that in the first branch, and high-dimensional feature representations are also generated. The features output by the two branches are respectively and . These two sets of features are input into a cross-modal interaction module to calculate the interaction feature matrix. The interaction calculation is achieved through the following dot product formula: ; where represents the correlation matrix between modalities, which is used to capture the potential interaction relationship between RFID features and environmental features. The interaction feature matrix is input into a feature enhancement module for reconstruction. The feature enhancement module consists of multiple residual blocks, and each residual block contains two convolutional layers and a skip connection. Its update formula is: ; where is the input feature, and are the convolutional weights and biases. The skip connection ensures the transmission stability of gradients in the deep network. The enhanced feature vector is input into a multi-scale feature extraction module to capture the local and global patterns of features through receptive fields of different sizes, and a multi-scale feature pyramid is generated. The feature representation at each scale is: ; where is the feature at the th scale, represents the pooling window size. The features at each scale are concatenated and fused: ; where [;] represents the concatenation operation in the dimension, and the final multi-modal fusion feature is generated.

[0032] In a specific embodiment, the process of executing step 500 may specifically include the following steps: Temporally segment the multi-modal fusion feature to obtain feature sequence segments of a fixed length, and input the feature sequence segments into a sequence encoder for feature encoding to obtain a hidden state sequence. The sequence encoder adopts a bidirectional long short-term memory network structure; Perform temporal attention calculation on the hidden state sequence to obtain a weighted context vector, and input the weighted context vector into the trajectory generator for trajectory decoding to obtain a trajectory sequence representation. The trajectory generator adopts an autoregressive decoding structure and includes 4 layers of Transformer decoding layers; Perform constraint optimization on the trajectory sequence representation to obtain a smooth trajectory sequence, where the constraint optimization process includes position continuity constraint and motion smoothness constraint; Input the smooth trajectory sequence into the prediction head network for parameter regression to obtain an object trajectory prediction model. The prediction head network consists of 3 fully connected layers, and the output layer dimension matches the trajectory parameter dimension.

[0033] Specifically, input the multi-modal fusion features into the temporal segmentation module, where represents the total length of the time steps, represents the dimension of the features. Split the long time series into fixed-length feature segments for more efficient processing by the sequence encoder. Assume that the length of each segment is , and perform segmentation by the sliding window method with a sliding step of . The segmentation formula is: ; where represents the th feature sequence segment, is the segment index, and the total number of segments is . Temporal segmentation ensures that each segment covers a fixed time range. Input each feature segment into a bidirectional long short-term memory network (Bi-LSTM) for sequence encoding to generate a hidden state sequence , where is the dimension of the hidden state. Through bidirectional processing, the hidden state at each time step contains information from the previous and next time steps, forming a hidden state sequence . Perform temporal attention calculation on the hidden state sequence to generate a weighted context vector . Temporal attention highlights key features by calculating the importance weights at each time step. The weight is calculated by the formula: ; where is the correlation score between the hidden state and the query vector , is a learnable weight matrix is the query vector. The weighted context vector is: ; where Represents the global context representation of the current sequence segment. The context vector is input to the trajectory generator for decoding to generate a trajectory sequence representation. The trajectory generator adopts an autoregressive structure and consists of 4 layers of Transformer decoder layers. The core of each layer of Transformer decoding is the multi-head attention mechanism and the feed-forward network, and its update formula is: where are the query, key, and value matrices respectively, is the attention dimension, are the parameters of the feed-forward network. The autoregressive structure gradually generates the trajectory sequence, ensuring the dynamic relationship between time steps. The generated trajectory sequence representation is input to the constraint optimization module to smooth the trajectory. The position continuity constraint restricts the displacement between adjacent points: ; where is the maximum allowed displacement. The motion smoothness constraint restricts the change in acceleration: ; where is the threshold for the change in acceleration. These constraints optimize the smoothness and physical rationality of the trajectory sequence. The smoothed trajectory sequence is input to the prediction head network for parameter regression to obtain the object trajectory prediction model. The prediction head network consists of 3 layers of fully connected layers, and the dimension of its output layer matches the dimension of the trajectory parameters, such as position coordinates and velocity vectors. The prediction formula is: ; where are the weights and biases of the fully connected layer, is the activation function.

[0034] In a specific embodiment, the process of executing step 600 may specifically include the following steps: Perform coordinate mapping on the target area to obtain the target area grid coordinate matrix; Based on the target area grid coordinate matrix, input the trajectory parameter sequence output by the object trajectory prediction model into the coordinate transformation module for grid projection processing to obtain the grid space trajectory sequence, and the trajectory parameter sequence includes the three-dimensional space coordinates and motion state parameters of the object; Perform density accumulation calculation on the grid space trajectory sequence to obtain the initial density distribution map, where the density accumulation calculation is based on the residence time and appearance frequency of each object in each grid cell; Input the initial density distribution map into the density smoothing module for Gaussian filtering processing to obtain the dynamic density matrix; Normalize and perform threshold segmentation on the dynamic density matrix to obtain a real-time object location heat map, and construct a spatio-temporal index for the grid space trajectory sequence and the dynamic density matrix to obtain a trajectory database, where the spatio-temporal index organizes the trajectory data using a quadtree structure.

[0035] Specifically, divide the target area into spatial grids to generate a grid coordinate matrix. Assume that the target area is a rectangular space, and its boundaries are defined as . Divide this area into grids, and the size of each grid cell is and . The grid coordinate matrix is expressed as: ; where is the center coordinate of the grid cell in the -th row and the -th column. Input the trajectory parameter sequence output by the object trajectory prediction model into the coordinate transformation module for grid projection processing. The trajectory parameter sequence contains the three-dimensional position of the object at time , velocity and acceleration . The goal of grid projection is to map the three-dimensional space coordinates of the object to two-dimensional grid coordinates. For each trajectory point , find its corresponding grid cell index: ; where represents the floor operation, and is the index position of the trajectory point in the grid. Through the above operations, the three-dimensional trajectory sequence is converted into a grid space trajectory sequence . Perform density accumulation calculation on the grid space trajectory sequence to generate an initial density distribution map. The core of density accumulation is to count the residence time and occurrence frequency of each grid cell. Assume that the residence time of the object in the grid cell is , and the total number of occurrences is . The formula for density accumulation is: ; where is the cumulative density value of the grid cell . The initial density distribution map is expressed in matrix form , where each element reflects the item activity intensity of the corresponding grid. Due to the problems of noise or non - smoothness in the initial density distribution, it is input into the density smoothing module for Gaussian filtering. Gaussian filtering realizes smoothing by performing weighted averaging on density values within a local neighborhood, and the filtering formula is: ; where is the smoothed density value, is the Gaussian kernel weight, is the standard deviation, is the filtering window size. After Gaussian filtering, the obtained dynamic density matrix is smoother and more coherent. The dynamic density matrix is normalized and threshold - segmented to generate a real - time item localization heat map. The goal of normalization is to map density values to the interval [0, 1], and the normalization formula is: ; where and are the minimum and maximum values of the matrix respectively. The normalized density matrix generates a heat map through threshold segmentation, where the segmentation formula is: ; where is the set density threshold, is the generated binary heat map, showing the high - density activity areas in the target region. The grid - space trajectory sequence and the dynamic density matrix are used to construct a spatio - temporal index to form a trajectory database. The spatio - temporal index is organized using a quadtree structure, where each node represents a spatial range and stores the trajectory points and density values within that range. The construction of the quadtree is achieved by recursively dividing the space: ; where are the northeast, northwest, southeast, and southwest sub - regions of the current region respectively. Through the quadtree, the trajectory database supports efficient range queries and spatio - temporal analysis.

[0036] In this embodiment, after obtaining the real-time item positioning heat map and the trajectory database, the following steps are further included: performing density analysis on the real-time item positioning heat map to obtain the RFID reader deployment position matrix, where the density analysis calculates the optimal coverage point set based on the item distribution density map; inputting the RFID reader deployment position matrix into the path planning module for initial path generation to obtain the initial inspection path set, and the initial path generation uses an improved ant colony algorithm to plan paths for each RFID reader; performing dynamic update processing on the initial inspection path set to obtain the optimized path sequence, and the dynamic update processing includes two objectives: path length optimization and coverage rate balance; inputting the optimized path sequence into the evolutionary algorithm module for the first-stage optimization to obtain the locally optimal path, and the evolutionary algorithm uses adaptive crossover and mutation operations with a population size of 100; performing the second-stage optimization processing on the locally optimal path to obtain the globally optimal path, and the second-stage optimization introduces a tabu search strategy to avoid falling into local optimality; calculating the workload of the RFID readers based on the globally optimal path to obtain the load distribution matrix, where the workload includes path length, coverage area, and acquisition frequency; performing real-time balance calculation on the load distribution matrix to obtain the task allocation scheme, and the balance calculation evenly distributes the workload through the min-max criterion; converting the task allocation scheme into the motion control instructions of the RFID readers to obtain the multi-agent collaborative scheduling strategy, and the motion control instructions include speed, direction, and sampling interval.

[0037] The RFID intelligent tag positioning and tracking printing method in the embodiment of the present application has been described above. Next, the RFID intelligent tag positioning and tracking printing system 10 in the embodiment of the present application will be described. Please refer to Figure 2 , an embodiment of the RFID intelligent tag positioning and tracking printing system 10 in the embodiment of the present application includes: A convolution processing module 11, configured to input the RFID tag radio frequency signal into a conditional variational autoencoder for convolution processing to obtain the RFID tag structured feature; An attention weighting module 12, configured to perform attention weighting on the RFID tag structured feature to obtain the RFID tag global feature vector; A multi-dimensional acquisition module 13, configured to establish a hierarchical sensor node network to perform multi-dimensional acquisition of environmental parameters to obtain an environmental spatio-temporal data set; A feature fusion module 14, configured to input the RFID tag global feature vector and the environmental spatio-temporal data set into a deep fusion network for feature fusion to obtain a multi-modal fusion feature; A sequence modeling module 15, configured to perform sequence modeling based on the multi-modal fusion feature to obtain an item trajectory prediction model; A distribution calculation module 16, configured to apply the item trajectory prediction model to the target area grid coordinate system for density distribution calculation to obtain the real-time item positioning heat map and the trajectory database.

[0038] Through the collaborative cooperation of the above-mentioned various components, by introducing a conditional variational autoencoder and an attention mechanism, deep feature extraction and weighting processing are performed on RFID radio frequency signals, enhancing the robustness and discriminability of feature expression and effectively overcoming the signal interference problem in complex environments; by adopting a hierarchical sensor network architecture and a multi-dimensional environmental parameter acquisition strategy, a comprehensive perception and real-time monitoring of the environmental state of the target area are realized, providing rich environmental context information for item positioning; a deep fusion network is designed for multi-modal feature fusion, and through dual-branch feature extraction and cross-modal interaction, the RFID signal features and environmental parameter information are effectively integrated, enhancing the environmental adaptability of the system; a trajectory prediction method based on sequence modeling and constraint optimization, combined with a temporal attention mechanism and an autoregressive decoding structure, realizes accurate prediction and smooth processing of the item movement trajectory; through grid density distribution calculation and spatio-temporal index construction, an efficient item positioning heat map generation and trajectory data management mechanism is established, facilitating real-time position visualization and historical trajectory query of the system.

[0039] Those skilled in the art can clearly understand that for the convenience and brevity of description, the specific working processes of the above-described system, system, and unit can refer to the corresponding processes in the foregoing method embodiments and will not be elaborated herein.

[0040] If the integrated unit is implemented in the form of a software functional unit and sold or used as an independent product, it can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of the present application, in essence, or the part that contributes to the prior art, or all or part of this technical solution, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes several instructions for causing an electronic device (which can be a personal computer, a server, or a network device, etc.) to execute all or part of the steps of the methods described in the various embodiments of the present application. The foregoing storage medium includes: various media such as USB flash drives, mobile hard disks, read-only memory (ROM), random access memory (RAM), magnetic disks, or optical discs that can store program codes.

[0041] The above is the case. The above embodiments are only used to illustrate the technical solutions of the present application and are not intended to limit them. Although the present application has been described in detail with reference to the foregoing embodiments, those of ordinary skill in the art should understand that they can still modify the technical solutions described in the foregoing embodiments or perform equivalent replacements for some of the technical features. However, these modifications or replacements do not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the various embodiments of the present application.

Claims

1. A method for printing RFID smart label positioning and tracking, characterized in that: The method comprises: The RFID tag radio frequency signal is input into the conditional variational autoencoder for convolution processing to obtain the RFID tag structured features; Performing attention weighting on the structured features of the RFID tag to obtain a global feature vector of the RFID tag; Establish a hierarchical sensor node network to collect environmental parameters in multiple dimensions and obtain an environmental spatiotemporal data set; Inputting the RFID tag global feature vector and the environmental spatiotemporal data set into a deep fusion network for feature fusion to obtain a multimodal fusion feature; Perform sequence modeling based on the multimodal fusion features to obtain an item trajectory prediction model; The object trajectory prediction model is applied to the target area grid coordinate system to perform density distribution calculation to obtain a real-time object location heat map and trajectory database.

2. The RFID smart label positioning tracking printing method according to claim 1, characterized in that: The RFID tag radio frequency signal is input into the conditional variational autoencoder for convolution processing to obtain the RFID tag structured features, including: The RFID tag radio frequency signal is input into the feature extraction layer of the conditional variational autoencoder for convolution operation to obtain a multi-scale feature map, wherein the feature extraction layer includes 5 parallel convolution kernels, each convolution kernel has a step size of 2, and each convolution layer is connected to a batch normalization layer and a ReLU activation function; Performing a maximum pooling operation on the multi-scale feature map to obtain a dimension-reduced feature map, wherein the maximum pooling operation uses a 2×2 pooling window with a step size of 2; Inputting the dimension-reduced feature map into the fully connected layer of the conditional variational autoencoder for dimension conversion to obtain a feature vector of fixed dimension, wherein the fully connected layer includes three hidden layers, and the number of neurons in each hidden layer is 512, 256, and 128 respectively; Applying adaptive weights to the feature vector of the fixed dimension to obtain a weighted feature vector, wherein the adaptive weights are dynamically calculated based on feature importance scores; Inputting the weighted feature vector into the encoder of the conditional variational autoencoder for compression mapping to obtain a latent space feature, wherein the encoder includes a mean encoder and a variance encoder, and outputs a mean vector and a variance vector of the latent space respectively; The latent space features are reparameterized and sampled to obtain RFID tag structured features, wherein the reparameterized sampling generates sampling points that obey a standard normal distribution based on the mean vector and the variance vector.

3. The RFID smart label positioning tracking printing method according to claim 2, characterized in that: The step of performing attention weighting on the structured features of the RFID tag to obtain a global feature vector of the RFID tag includes: Performing multi-head attention decomposition on the structured features of the RFID tag to obtain multiple groups of attention feature subspaces, wherein the multi-head attention decomposition includes 8 parallel attention heads, and each attention head outputs a 64-dimensional feature subspace; Input the multiple groups of attention feature subspaces into a multilayer perceptron for nonlinear transformation to obtain a feature mapping matrix, wherein the multilayer perceptron includes two hidden layers, the hidden layer dimensions are 512 and 256 respectively, and each layer uses a LeakyReLU activation function; Calculating an attention score based on the feature mapping matrix to obtain a feature weight coefficient, wherein the attention score is calculated by a dot product operation and a softmax normalization function; Performing a weighted sum operation on the RFID tag structured features and the feature weight coefficients to obtain a global context vector, wherein the weighted sum operation is implemented by matrix multiplication; Inputting the global context vector into a residual connection module for feature fusion to obtain a fused feature vector, wherein the residual connection module includes a layer normalization layer and a skip connection; The fusion feature vector is subjected to a dimension reduction transformation to obtain a global feature vector of the RFID tag. The dimension reduction transformation uses a 1×1 convolutional layer to reduce the feature dimension to 128 dimensions.

4. The RFID smart label positioning tracking printing method according to claim 3, characterized in that: The establishment of a hierarchical sensor node network to collect environmental parameters in multiple dimensions to obtain an environmental spatiotemporal data set includes: Deploy multiple sensor nodes in a grid structure to obtain a hierarchical sensor network topology structure, wherein the sensor nodes include a temperature sensor, a humidity sensor, a light intensity sensor, and a motion sensor; The hierarchical sensor network topology structure is divided into master and slave nodes to obtain a data collection hierarchical system, wherein the master node is responsible for data aggregation and the slave node is responsible for environmental parameter collection; Based on the data acquisition hierarchical system, environmental parameters are synchronously sampled to obtain an original environmental data stream, wherein the environmental parameters include temperature distribution, humidity distribution, light intensity distribution and object movement trajectory information; Inputting the original environment data stream into a data compression module for compression encoding to obtain a compressed data packet, and performing error correction encoding processing on the compressed data packet to obtain a transmission data stream; The transmission data stream is decoded and reconstructed and a timestamp and a location identifier are added to obtain an environmental spatiotemporal data set.

5. The RFID smart label positioning tracking printing method according to claim 4, characterized in that: The step of inputting the RFID tag global feature vector and the environmental spatiotemporal data set into a deep fusion network for feature fusion to obtain a multimodal fusion feature includes: Performing time alignment processing on the global feature vector of the RFID tag to obtain an alignment feature sequence, wherein the time alignment processing is synchronized based on timestamp information in the environmental spatiotemporal data set; Inputting the aligned feature sequence into a first feature extraction branch of a deep fusion network, and inputting the environmental spatiotemporal data set into a second feature extraction branch of the deep fusion network for parallel feature extraction to obtain a bimodal feature representation, wherein both the first feature extraction branch and the second feature extraction branch include 3 convolutional layers and 2 fully connected layers; Performing cross-modal interaction calculation on the bimodal feature representation to obtain an interaction feature matrix, and inputting the interaction feature matrix into a feature enhancement module for feature reconstruction to obtain an enhanced feature vector, wherein the feature enhancement module includes a plurality of residual blocks, each of which is composed of two 3×3 convolutional layers and a skip connection; Multi-scale feature extraction is performed on the enhanced feature vector to obtain a multi-scale feature pyramid, and feature splicing and fusion are performed on the multi-scale feature pyramid to obtain a multi-modal fusion feature.

6. The RFID smart label positioning tracking printing method according to claim 5, characterized in that: The method of performing sequence modeling based on the multimodal fusion features to obtain an item trajectory prediction model includes: Performing time-series segmentation on the multimodal fusion features to obtain feature sequence segments of fixed length, and inputting the feature sequence segments into a sequence encoder for feature encoding to obtain a hidden state sequence, wherein the sequence encoder adopts a bidirectional long short-term memory network structure; Performing temporal attention calculation on the hidden state sequence to obtain a weighted context vector, and inputting the weighted context vector into a trajectory generator for trajectory decoding to obtain a trajectory sequence representation, wherein the trajectory generator adopts an autoregressive decoding structure and includes 4 layers of Transformer decoding layers; Performing constraint optimization on the trajectory sequence representation to obtain a smooth trajectory sequence, wherein the constraint optimization process includes position continuity constraint and motion smoothness constraint; The smooth trajectory sequence is input into the prediction head network for parameter regression to obtain an item trajectory prediction model. The prediction head network consists of three fully connected layers, and the output layer dimension matches the trajectory parameter dimension.

7. The RFID smart label positioning tracking printing method according to claim 6, characterized in that: The object trajectory prediction model is applied to the target area grid coordinate system to perform density distribution calculation to obtain a real-time object location heat map and trajectory database, including: Carry out coordinate mapping on the target area to obtain the grid coordinate matrix of the target area; Based on the target area grid coordinate matrix, the trajectory parameter sequence output by the object trajectory prediction model is input into a coordinate transformation module for grid projection processing to obtain a grid space trajectory sequence, wherein the trajectory parameter sequence includes the three-dimensional space coordinates and motion state parameters of the object; Performing density accumulation calculation on the grid space trajectory sequence to obtain an initial density distribution map, wherein the density accumulation calculation is based on the residence time and appearance frequency of each object in each grid unit; Input the initial density distribution map into a density smoothing module for Gaussian filtering to obtain a dynamic density matrix; The dynamic density matrix is ​​normalized and threshold segmented to obtain a real-time object positioning heat map, and the grid space trajectory sequence and the dynamic density matrix are subjected to spatiotemporal index construction to obtain a trajectory database, wherein the spatiotemporal index adopts a quadtree structure to organize trajectory data.

8. An RFID smart label positioning tracking printing system, characterized in that: Used to execute the RFID smart label positioning tracking printing method according to any one of claims 1 to 7, the RFID smart label positioning tracking printing system comprises: A convolution processing module is used to input the RFID tag radio frequency signal into a conditional variational autoencoder for convolution processing to obtain the RFID tag structured features; An attention weighting module, used to perform attention weighting on the structured features of the RFID tag to obtain a global feature vector of the RFID tag; The multi-dimensional acquisition module is used to establish a hierarchical sensor node network to perform multi-dimensional acquisition of environmental parameters and obtain an environmental spatiotemporal data set; A feature fusion module is used to input the RFID tag global feature vector and the environmental spatiotemporal data set into a deep fusion network for feature fusion to obtain a multimodal fusion feature; A sequence modeling module, used for performing sequence modeling based on the multimodal fusion features to obtain an item trajectory prediction model; The distribution calculation module is used to apply the object trajectory prediction model to the target area grid coordinate system to perform density distribution calculation to obtain a real-time object positioning heat map and trajectory database.

Images