RFID traceable medical linen whole-process management and control processing method

Abstract

The application discloses an RFID traceable medical linen whole-process management and control processing method and relates to the technical field of data traceability; the method comprises the following steps: acquiring medical linen RFID identification original data and constructing a signal observation matrix, obtaining an initial mixing matrix through successive non-negative projection iteration, performing white processing on the observation matrix and calculating an observation perimeter vector, separating collision signals according to the initial mixing matrix and the perimeter vector to obtain a label signal matrix, decoding according to a preset communication rule, and outputting a medical linen traceable data set; the signal observation matrix is constructed by fusing traceable messages, channel transmission and environmental noise, the batch linen RFID receiving signals are truly restored, the initial mixing matrix is obtained through successive non-negative projection iteration, the white processing and the observation perimeter vector iteration optimization are combined, the separability of the mixed signals is enhanced, the label collision missed reading and misreading are reduced, the decoded label signals are accurately recovered, and the traceable data integrity and reliability are improved.

Description

Technical Field

[0001] This invention belongs to the field of data traceability technology, specifically involving a method for full-process control and processing of RFID-traceable medical linens. Background Technology

[0002] RFID-based traceability for the entire process of medical linen management enables closed-loop digital tracking of medical textiles throughout their lifecycle, ensuring compliant execution of the entire washing and disinfection process, and effectively reducing the risk of cross-infection within hospitals. Simultaneously, it precisely controls linen asset loss, improves handover and inventory efficiency, provides reliable traceability evidence for health supervision, and contributes to the standardization of medical textile management, effectively safeguarding hospital infection control. Current methods involve binding RFID electronic tags to each piece of medical linen and deploying readers at key points in the flow, such as departmental requisition, soiled linen collection, washing and sorting, and disinfection and distribution, to collect tag information in batches to record the linen's flow trajectory and washing batch, achieving asset inventory and process traceability.

[0003] However, in the processes of bulk waste collection, washing handover verification, and bulk distribution to wards, the number of tags in a single batch of linens far exceeds the number of reader antennas. Furthermore, the stacking and obstruction of linens, the interference of moisture in the washing environment, and the shielding of metal turnover containers result in a low channel signal-to-noise ratio. When multiple tag signals respond simultaneously, severe aliasing and collisions occur, leading to missed tag readings and missing data at traceability nodes. Consequently, existing methods cannot simultaneously meet the high accuracy and real-time requirements of bulk inventory of medical linens, making it difficult to support the data reliability and operational efficiency requirements of the entire process of traceability management of medical linens. Summary of the Invention

[0004] The purpose of this invention is to solve the problem that serious aliasing and collision will occur when multiple tag signals of medical linens respond simultaneously, resulting in missing tag readings and missing traceability node data, which makes it difficult to support the data reliability and operational efficiency requirements of the whole process traceability management of medical linens. Therefore, this invention proposes an RFID traceability whole process management method for medical linens.

[0005] This invention proposes a method for the whole-process management and control of RFID-traceable medical linens, the method comprising: The raw data of RFID identification of medical linens is obtained, and a signal observation matrix is ​​constructed based on the raw data; the raw data includes traceability message data, channel transmission data and environmental noise data; The initial mixing matrix is ​​obtained by performing successive nonnegative projection iterations on the signal observation matrix; The signal observation matrix is ​​whitened to output a whitening matrix and a whitened signal observation matrix, and the observation perimeter vector is calculated based on the whitened signal observation matrix; the observation perimeter vector is used to iteratively optimize the mixing matrix to complete the separation of tag collision signals; The collision signal matrix of the linen is obtained by performing iterative separation of the collision signal based on the initial mixing matrix, whitening matrix and observed perimeter vector; The traceability dataset for all linens in a single batch is obtained by performing signal decoding operations based on the tag signal matrix and preset RFID communication rules. The traceability dataset includes the unique identifier, category, disinfection status and current circulation node information of each linen.

[0006] Optionally, a signal observation matrix is ​​constructed based on the original data, including: A source signal matrix is ​​constructed based on the traceability message data; the source signal matrix is ​​used to describe the original response state of each RFID tag at each sampling time; A channel mixing matrix is ​​constructed based on the channel transmission data, and the initial observation matrix is ​​obtained by channel mapping of the source signal matrix using the channel mixing matrix. A noise impact matrix is ​​constructed based on environmental noise data. Abnormal response regions in the initial observation matrix are identified through the noise impact matrix. The observation data corresponding to the abnormal response regions are locally corrected to obtain the corrected observation matrix. The signal aliasing coefficient corresponding to each sampling time is calculated based on the correction observation matrix. The observation data in the correction observation matrix is ​​then divided into high-confidence observation areas and low-confidence observation areas based on the signal aliasing coefficient. Consistency constraint correction is performed on the observation data corresponding to the low-confidence observation region by the high-confidence observation region, and the corrected observation regions are fused to obtain the signal observation matrix.

[0007] Optionally, performing successive nonnegative projection iterations on the signal observation matrix to obtain an initial mixing matrix includes: Step 1: Initialize the iteration parameters; the iteration parameters include the residual matrix, the empty index set, and the preset maximum number of iterations; the residual matrix is ​​obtained by assigning values ​​to the signal observation matrix; Step 2: Calculate the strong convex function value of each column of the residual matrix for this round, and traverse all columns to filter out the index of the extreme value column corresponding to the column that makes the function value the largest. Step 3: Incorporate the current round of extreme value column index into the empty index set, record the positions of the identified tag channel feature columns, and obtain the updated new index set; Step 4: Calculate the projection coefficient matrix corresponding to each column of the signal observation matrix based on the signal observation matrix and the new index set; Step 5: Update the residual matrix according to the signal observation matrix, the new index set, and the projection coefficient matrix to obtain a new residual matrix; Step 6: Calculate the F-norm of the new residual matrix. If the F-norm is less than the norm threshold, or the current iteration number is greater than the preset maximum iteration number, terminate the iteration and extract the corresponding column elements of the signal observation matrix according to the new index set to obtain the initial mixing matrix; otherwise, repeat steps 2 to 6.

[0008] Optionally, updating the residual matrix based on the signal observation matrix, the new index set, and the projection coefficient matrix to obtain a new residual matrix includes: Based on the new index set, the channel feature submatrix corresponding to the identified medical linen tags is extracted from the signal observation matrix; Multiply the channel feature submatrix by the projection coefficient matrix to obtain the tag superposition fitting matrix of the currently identified linen tags, and subtract the tag superposition fitting matrix from the signal observation matrix to obtain the new residual matrix.

[0009] Optionally, the observed perimeter vector is calculated based on the whitening observation signal, including: Multiplying the whitening matrix by the signal observation matrix yields the whitened signal observation matrix; Generate a preset number of random unit projection direction vectors, and linearly project the whitened observation signal matrix onto each set of random unit projection directions to obtain a preset number of one-dimensional projection sequences. Calculate the maximum and minimum values ​​of each group of one-dimensional projection sequences, calculate the difference between the maximum and minimum values, and arrange twice the difference of each group in sequence to construct the observation perimeter vector.

[0010] Optionally, the linen's tag signal matrix is ​​obtained by performing collision signal iterative separation based on the initial mixing matrix, whitening matrix, and observed perimeter vector, including: Step 1: Map the initial mixing matrix to the whitening space to obtain the whitening mixing matrix; Step 2: Calculate the theoretical estimated perimeter vector for each projection direction based on the whitening mixing matrix of this round, and calculate the error vector between the observed perimeter vector and the theoretical estimated perimeter vector. Use the squared L2 norm of the error as the cost function value for this round. Step 3: Derive the natural gradient matrix of the cost function with respect to the whitening mixing matrix based on the perimeter matching criterion, and calculate the optimal update step size for this iteration based on the current cost function value and the natural gradient matrix; Step 4: Calculate the first update matrix and the second update matrix based on the whitening mixing matrix of the current round, the observed perimeter vector, the natural gradient matrix, and the optimal update step size of this round as inputs. Calculate the cost function values ​​corresponding to the two update matrices respectively, and select the update matrix that minimizes the cost as the whitening mixing matrix for the next round. Step 5: If the cost function value corresponding to the whitening mixture matrix in this round is less than the convergence threshold, or not less than the maximum number of iterations, then terminate the iteration and output the optimal whitening mixture matrix; otherwise, repeat steps 2 to 5. Step 6: Map the optimal whitening mixing matrix back to the original signal space through the inverse whitening operation to obtain the final mixing matrix, and output the final mixing matrix as the tag signal matrix.

[0011] Optionally, the first update matrix and the second update matrix are calculated based on the whitening mixing matrix of the current round, the observed perimeter vector, the natural gradient matrix, and the optimal update step size of this round as inputs, including: Using the whitening mixing matrix and the observed projection perimeter vector of the current round as input, the first update matrix is ​​obtained by solving the pseudo-inverse analytical method; Using the whitening mixing matrix, natural gradient matrix, and optimal update step size of the current round as input, the second update matrix is ​​obtained by weighted update along the negative gradient direction.

[0012] Optionally, a signal decoding operation is performed according to the tag signal matrix and preset RFID communication rules to output a traceability dataset for all linens in a single batch, including: Traverse each independent tag signal in the tag signal matrix, complete carrier synchronization and frame header positioning according to the synchronization code sequence in the RFID communication rules, and extract the corresponding valid data frame segment; Perform legality verification on each valid data frame segment according to the verification algorithm in the RFID communication rules, and remove distorted and interfering frames; Based on the traceability field mapping relationship in the RFID communication rules, the valid data frames that have passed the verification are parsed field by field to extract the unique identification code, circulation node identifier and status information of each piece of linen, and the data are summarized to form a traceability dataset for a single batch of linen.

[0013] The beneficial effects of this invention are: This invention constructs a signal observation matrix that integrates traceability message data, channel transmission data, and environmental noise data. This matrix can more realistically characterize the actual received signals of medical linen RFID tags during batch circulation. An initial mixing matrix is ​​obtained through successive non-negative projection iterations, providing an accurate channel feature basis for subsequent tag collision signal separation. The mixing matrix is ​​iteratively optimized using a whitening matrix and observation perimeter vector to enhance the separability of multi-tag aliasing signals, reduce missed readings and misreads caused by tag collisions in complex environments, accurately recover the signals of each RFID tag, and complete signal decoding. This improves the integrity and reliability of traceability data for a single batch of medical linen, and ensures the accuracy and continuity of the entire process control, including waste recycling, washing and disinfection, and distribution. Attached Figure Description

[0014] The invention will now be further described with reference to the accompanying drawings.

[0015] Figure 1 A flowchart illustrating the RFID-based traceability method for the entire process of managing and controlling medical linens, as provided in this embodiment of the invention. Detailed Implementation

[0016] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. The term "and / or" in this document is merely a description of the relationship between related objects, indicating that three relationships can exist. For example, A and B can represent: A alone, A and B simultaneously, and B alone. Furthermore, descriptions involving "first," "second," etc., in this invention are for descriptive purposes only and should not be construed as indicating or implying their relative importance or implicitly specifying the number of technical features indicated. Therefore, features defined with "first" or "second" can explicitly or implicitly include at least one of those features. Additionally, the technical solutions of the various embodiments can be combined with each other, but this must be based on the ability of those skilled in the art to implement them. When the combination of technical solutions is contradictory or cannot be implemented, it should be considered that such a combination of technical solutions does not exist and is not within the scope of protection claimed by the present invention.

[0017] Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0018] This invention provides a method for the entire process of RFID-traceable medical linens control and management. See also... Figure 1 The method includes the following steps: S1, Obtain the raw data of RFID identification of medical linens, and construct the signal observation matrix based on the raw data; S2, perform successive nonnegative projection iterations on the signal observation matrix to obtain the initial mixing matrix; S3, whitens the signal observation matrix and outputs the whitened matrix and the whitened signal observation matrix, and calculates the observation perimeter vector based on the whitened signal observation matrix; S4, perform collision signal iterative separation based on the initial mixing matrix, whitening matrix and observed perimeter vector to obtain the label signal matrix of the linen; S5 performs signal decoding based on the tag signal matrix and preset RFID communication rules, outputting a traceability dataset for all linens in a single batch.

[0019] The raw data includes source tracing message data, channel transmission data, and environmental noise data; The observed perimeter vector is used to iteratively optimize the mixing matrix to complete the separation of tag collision signals; The traceability dataset contains the unique identifier, category, disinfection status, and current circulation node information for each linen.

[0020] In one implementation, an RFID multi-antenna reader array is deployed at the batch identification nodes in the entire medical linen circulation chain, covering the core links of the entire process control: Hospital ward soiled linen collection channel / temporary soiled linen storage room: used for automatic counting and collection registration when soiled linens are collected in bulk; Laundry receiving and handover verification point: used for batch verification of linens when they are handed over to the factory, to check the quantity and type of linens to be recycled; Finished product storage entrance and exit after washing and disinfection: used for the inbound inventory, outbound sorting and verification of qualified washed and disinfected linens; Hospital ward clean linen distribution point: used for bulk distribution and verification of clean linens delivered to departments.

[0021] Each node is equipped with 3 to 4 directional antennas to form a blind spot recognition area, corresponding to the acquisition channels of m observation signals in the scheme; after deployment, the open field environment calibration is performed in advance, and the noise floor of each node and the channel attenuation coefficient under different stacking states and turnover container materials are collected and pre-stored in the reader to support the construction of the signal observation matrix.

[0022] In one implementation, the edge (real-time processing layer) The core SNPA initial mixing matrix estimation, BCA collision signal separation, and tag signal decoding algorithms are all deployed in the RFID reader embedded main control unit or supporting edge gateway of each node; the original radio frequency sampling data completes the aliasing signal demixing and traceability information decoding locally, without the need to send back the original waveform data, ensuring real-time identification efficiency when batch linens pass through quickly, and outputting structured linen traceability data.

[0023] Platform side (full-process control layer) The linen full-process traceability and control system is deployed on the hospital's logistics management platform or the cloud platform server of the laundry company. It receives traceability data uploaded by each edge node and realizes business functions such as full life cycle traceability of linens, retention of washing and disinfection records, inventory scheduling, and hospital infection data analysis. At the same time, it completes the statistical analysis and optimization iteration of the algorithm recognition effect.

[0024] In one embodiment, a signal observation matrix is ​​constructed based on the original data, including: A source signal matrix is ​​constructed based on the traceability message data; the source signal matrix is ​​used to describe the original response state of each RFID tag at each sampling time; A channel mixing matrix is ​​constructed based on the channel transmission data, and the initial observation matrix is ​​obtained by channel mapping of the source signal matrix using the channel mixing matrix. A noise impact matrix is ​​constructed based on environmental noise data. Abnormal response regions in the initial observation matrix are identified through the noise impact matrix. The observation data corresponding to the abnormal response regions are locally corrected to obtain the corrected observation matrix. The signal aliasing coefficients corresponding to each sampling time are calculated based on the correction observation matrix. The observation data in the correction observation matrix are then divided into high-confidence and low-confidence observation regions based on the signal aliasing coefficients. Consistency constraint correction is performed on the observation data corresponding to the low-confidence observation region by the high-confidence observation region, and the corrected observation regions are fused to obtain the signal observation matrix.

[0025] In one implementation, the process of constructing the source signal matrix, channel mixing matrix, and noise impact matrix is ​​as follows: The traceability message data output by each linen RFID tag in a single batch is arranged according to the tag dimension to form a source signal matrix with rows corresponding to tags and columns corresponding to sampling times; The channel transmission data is arranged according to the mapping relationship between the reader antenna and the linen tag to form a channel hybrid matrix in which rows correspond to antennas and columns correspond to tags; The environmental noise data is aligned with the sampling times according to the antenna dimension to form a noise impact matrix in which rows correspond to antennas and columns correspond to sampling times.

[0026] In one implementation, the channel mixing matrix is ​​multiplied by the source signal matrix. Based on the channel transmission coefficients between each reader antenna and each RFID tag in the channel mixing matrix, the source signals output by each RFID tag are linearly mixed so that each reader antenna receives the superposition result of all RFID tag signals at each sampling time, thereby generating the corresponding initial observation matrix.

[0027] In one implementation, noise-weighted analysis is performed on the observation signals of each antenna in the initial observation matrix at each sampling time using a noise impact matrix. The intensity coefficient of the noise interference corresponding to each observation data is calculated. If the intensity coefficient of a certain antenna is continuously higher than a preset noise threshold at consecutive sampling times, the observation data corresponding to the consecutive sampling time period of that antenna is identified as an abnormal response region. Thus, the abnormal response region is marked and extracted in the initial observation matrix. Through spatial interpolation reconstruction based on observation data of adjacent antennas and temporal weighted smoothing processing based on observation data of adjacent sampling times, the observation data corresponding to the abnormal response region is locally corrected to reduce the impact of noise abrupt changes on the observation signal.

[0028] In one implementation, the multi-antenna observation vectors corresponding to each sampling time are extracted based on the correction observation matrix, the correlation coefficient (cosine similarity) between the observation vectors of different reader antennas at the same sampling time is calculated, the spatial consistency coefficient is determined based on the correlation coefficient, and the distribution dispersion of the observed signal at each sampling time is calculated to characterize the signal superposition degree. The spatial consistency coefficient and the distribution dispersion are weighted and fused to obtain the signal aliasing degree coefficient corresponding to each sampling time.

[0029] In one implementation, the observation data of the high-confidence observation region is used as a reference benchmark to reconstruct the observation data of the corresponding time and antenna dimension in the low-confidence observation region by weighting it. This allows the data to converge to the statistical characteristics of the high-confidence observation data under the constraints of temporal continuity and spatial correlation, thereby achieving consistency constraint correction.

[0030] In one implementation, because medical linen RFID in a batch circulation scenario simultaneously suffers from multi-tag signal collisions, channel attenuation fluctuations, and environmental noise disturbances, the initial observation data is distorted and incomplete. By introducing channel modeling, noise identification, aliasing assessment, and reliable region reconstruction mechanisms step by step, the observation matrix is ​​corrected and fused in a hierarchical manner, which effectively improves the separability of multi-tag signals and the integrity of traceability data, ensuring the accuracy and continuity of the entire process tracking.

[0031] In one embodiment, performing successive nonnegative projection iterations on the signal observation matrix to obtain an initial mixing matrix includes: Step 1: Initialize iteration parameters; the iteration parameters include the residual matrix, the empty index set, and the preset maximum number of iterations; the residual matrix is ​​obtained by assigning values ​​to the signal observation matrix; Step 2: Calculate the strong convex function value of each column of the residual matrix for this round, and traverse all columns to filter out the index of the extreme value column corresponding to the column that makes the function value the largest. Step 3: Merge the extreme value column index of this round into the empty index set, record the position of the identified tag channel feature column, and obtain the updated new index set; Step 4: Calculate the projection coefficient matrix corresponding to each column of the signal observation matrix based on the signal observation matrix and the new index set; Step 5: Update the residual matrix based on the signal observation matrix, the new index set, and the projection coefficient matrix to obtain the new residual matrix; Step 6: Calculate the F norm of the new residual matrix. If the F norm is less than the norm threshold, or the current iteration number is greater than the preset maximum iteration number, terminate the iteration and extract the corresponding column elements of the signal observation matrix according to the new index set to obtain the initial mixing matrix; otherwise, repeat steps 2 to 6.

[0032] In one implementation, the strongly convex function value of each column of the residual matrix for the current round is calculated. ,in, The output value of a strongly convex function; x is the input vector of the function, corresponding to any column in the residual matrix R, which is an m-dimensional column vector (m is the number of reader antennas). It is the Euclidean 2-norm of a vector; The physical significance in the context of medical linens: Taking a scenario of batch recognition of linens with 3 reader antennas as an example: A column of the residual matrix corresponds to a specific sampling time. The residual signal amplitudes from the three antennas form a vector, which is then substituted into the strongly convex function to calculate the result. The total energy of the three residual signals at the sampling time is calculated by traversing all columns and finding the column with the largest total energy. The vertex of the corresponding convex polyhedron is the independent channel response feature of an unidentified linen tag. It should be noted that: a convex polyhedron is an abstract expression of the geometric characteristics of multi-tag aliasing signals: an m-dimensional signal space is constructed with the number of reader antennas as the dimension. The independent channel response of each RFID tag in a single batch corresponds to a vertex vector in the space. All the observation signal points generated by the linear aliasing of all tag signals with non-negative weights will enclose a bounded closed convex shape with clear boundaries in this space. This shape is the convex polyhedron here. Among them, the pure response signal of a single tag corresponds to the vertex of the polyhedron, and the aliasing signals formed by the superposition of multiple tags all fall inside or at the edge of the polyhedron.

[0033] In one implementation, the projection coefficient matrix corresponding to each column of the signal observation matrix is ​​calculated based on the signal observation matrix and the new index set, including: Based on the new index set, the channel feature submatrix corresponding to the identified labels is extracted from the signal observation matrix. The signal observation matrix is ​​traversed column by column. For the aliased signal vector corresponding to each sampling time in each column, a convex quadratic programming problem is constructed with the channel feature submatrix as the linear basis and the weight coefficients satisfying the constraint that they are non-negative and the sum of the coefficients is not greater than 1. The gradient projection method is used to iteratively solve the optimal weight vector that minimizes the square of the second norm of the fitting residual. The optimal weight vectors corresponding to all columns are concatenated in the original column order to obtain the projection coefficient matrix that corresponds one-to-one with each column of the signal observation matrix.

[0034] In one implementation, due to the simultaneous response of multiple tags, signal aliasing, and noise interference in the processes of soiled waste collection, washing and handover, and bulk distribution of medical linens, it is difficult to directly identify the effective channel features corresponding to the mixed signals in the signal observation matrix. If collision signal separation is performed directly, the tag recovery accuracy is easily reduced due to the estimation bias of the mixing matrix. Therefore, by using a successive non-negative projection iteration method, the feature columns that contribute the most to the mixed signals are selected from the residual matrix round by round, and the residual matrix and feature index set are continuously updated to gradually extract representative channel features that can characterize the propagation characteristics of each RFID tag. Finally, a high-precision initial mixing matrix is ​​constructed, which provides an accurate channel mapping basis for subsequent collision signal separation, improves the batch tag recognition rate, and enhances the integrity of medical linen traceability data.

[0035] In one embodiment, updating the residual matrix based on the signal observation matrix, the new index set, and the projection coefficient matrix to obtain a new residual matrix includes: Based on the new index set, extract the channel feature submatrix corresponding to the identified medical linen tags from the signal observation matrix; Multiply the channel feature submatrix by the projection coefficient matrix to obtain the tag superposition fitting matrix of the currently identified linen tags. Subtract the tag superposition fitting matrix from the signal observation matrix to obtain the new residual matrix.

[0036] In one implementation, the superimposed aliased signal matrix is ​​a fitted signal matrix obtained by weighted combination of the channel features of the identified linen tags, which is used to subtract the identified tag components from the original observation signal to generate a new residual matrix.

[0037] In one implementation, the purpose of subtracting the label superposition fitting matrix from the signal observation matrix is ​​to eliminate the signal components of the linen labels that have completed feature localization. The remaining unseparated multi-label collision signals are combined with environmental noise to form a new residual matrix, which is used for the next round of iteration to screen the extreme value column index of unidentified linen labels.

[0038] In one implementation, a new index set records the channel feature columns of the identified linen single tags. The corresponding sub-matrix is ​​extracted to reconstruct the signal basis of the identified tags. Multiplying this sub-matrix with the projection coefficient matrix yields the complete fitted signal formed by the superposition of the identified tags at each sampling time. The difference between the signal observation matrix and the identified tag signals is removed. The remaining residual only contains the unidentified linen signals and on-site noise. In the next round of screening the extreme value column based on the residual, the separated tags will not be identified again. The interference of multi-tag collisions is gradually removed, and the independent signal features of the remaining linen are extracted efficiently.

[0039] In one embodiment, whitening the signal observation matrix and outputting a whitened matrix and a whitened observation signal includes: Multiplying the whitening matrix by the signal observation matrix yields the whitened signal observation matrix; Generate a preset number of random unit projection direction vectors, and linearly project the whitened observation signal matrix onto each set of random unit projection directions to obtain a preset number of one-dimensional projection sequences. Calculate the maximum and minimum values ​​of each group of one-dimensional projection sequences, calculate the difference between the maximum and minimum values, and arrange twice the difference of each group in sequence to construct the observed projection perimeter vector.

[0040] In one implementation, the generation of a preset number of random unit projection direction vectors is based on the number of RFID reader antenna channels to determine the dimension of the projection direction vector. First, a preset number of original random vectors following a standard normal distribution are generated. Then, a L2 norm normalization process is performed on each set of original vectors to obtain unit vectors. Finally, a batch orthogonal constraint verification method is used to remove vectors that are linearly related to the generated vectors and regenerate and supplement them to obtain multiple sets of random unit projection direction vectors that are non-redundant and dimensionally matched.

[0041] In one implementation, a set of random unit projection direction vectors is taken as row vectors and matrix multiplication is performed with the complete whitening observation signal matrix. The output row of one-dimensional data sequence is the projection sequence corresponding to the projection direction. The matrix multiplication operation is repeated for all preset number of unit projection direction vectors to obtain multiple sets of one-dimensional projection sequences that correspond one-to-one with the number of projection direction sets.

[0042] In one implementation, all values ​​within a single group of one-dimensional projection sequences are traversed, and the element with the largest value in the sequence is compared and selected as the maximum value of the group, and the element with the smallest value is selected as the minimum value of the group. This maximum and minimum value selection operation is performed on all one-dimensional projection sequences of a preset number of groups to obtain the maximum and minimum values ​​corresponding to each group of projection sequences.

[0043] In one implementation, the distribution range of the mixed signal generated by the collision of each RFID tag after whitening treatment in different random projection directions can reflect the signal spread in that direction. For each set of random unit projection directions, the difference between the maximum and minimum values ​​represents the projection span of the whitened signal in that direction. Multiplying this difference by two can uniformly map the projection span in a single direction to a perimeter quantity describing the overall geometric boundary scale. This gives the scale obtained from different projection directions a unified dimension and enhances the sensitivity to changes in the signal boundary. Arranging the corresponding projection perimeter quantities in each direction sequentially forms the observed projection perimeter vector, which can comprehensively characterize the geometric distribution characteristics of the collision signal in multiple directions. This facilitates providing stable boundary constraints for subsequent iterative optimization of the mixing matrix and improves the separation accuracy of multi-tag collision signals. In one embodiment, the collision signal iterative separation is performed based on the initial mixing matrix, whitening matrix, and observed perimeter vector to obtain the linen's tag signal matrix, including: Step 1: Map the initial mixing matrix onto the whitening space to obtain the whitening mixing matrix; Step 2: Calculate the theoretical estimated perimeter vector for each projection direction based on the whitening mixing matrix of this round, and calculate the error vector between the observed perimeter vector and the theoretical estimated perimeter vector. Use the squared L2 norm of the error as the cost function value for this round. Step 3: Derive the natural gradient matrix of the cost function with respect to the whitening mixing matrix based on the perimeter matching criterion, and calculate the optimal update step size for this iteration based on the current cost function value and the natural gradient matrix; Step 4: Calculate the first update matrix and the second update matrix based on the whitening mixing matrix of the current round, the observed perimeter vector, the natural gradient matrix, and the optimal update step size of this round as inputs. Calculate the cost function values ​​corresponding to the two update matrices respectively, and select the update matrix that minimizes the cost as the whitening mixing matrix for the next round. Step 5: If the cost function value corresponding to the whitening mixture matrix in this round is less than the convergence threshold, or not less than the maximum number of iterations, then terminate the iteration and output the optimal whitening mixture matrix; otherwise, repeat steps 2 to 5. Step 6: Map the optimal whitening mixing matrix back to the original signal space through the inverse whitening operation to obtain the final mixing matrix, and output the final mixing matrix as the tag signal matrix.

[0044] In one implementation, the theoretical estimated perimeter vector is calculated by using all random unit projection direction vectors that are completely consistent with the calculated observed perimeter. The whitening mixing matrix of the current round is linearly projected onto each group of projection directions to obtain the set of projection values ​​of each column vector of the mixing matrix in the corresponding projection direction. The maximum and minimum values ​​in each group of projection value sets are extracted and the difference between them is calculated. Twice the difference is taken as the theoretical estimated perimeter corresponding to the projection direction. Finally, all estimated perimeter values ​​are arranged in a predetermined order according to the projection directions to obtain the theoretical estimated perimeter vector with the same dimension as the observed projected perimeter vector.

[0045] In one implementation, the perimeter matching criterion refers to constructing a corresponding cost function with the goal of minimizing the deviation between the theoretically estimated perimeter and the observed perimeter. The natural gradient matrix introduces a Riemannian metric in the parameter space to correct the gradient direction, ensuring that iterative updates always proceed along the steepest error descent direction in the matrix space. In blind source separation scenarios, it converges faster and is more robust to noise than ordinary gradients, making it more suitable for the low signal-to-noise ratio and strong aliasing of multiple tags in RFID signal environments such as laundry workshops.

[0046] In one implementation, the optimal update step size for this iteration is calculated using the following formula:

[0047] in, To update the step size of the mixing matrix in the i-th iteration, This is the learning rate adjustment coefficient for the i-th iteration, used to control the iteration convergence speed. Let be the natural gradient matrix of the i-th iteration. For the projection operator in the projection direction space, This is the current cost function value.

[0048] In one implementation, RFID signals for medical linens are prone to aliasing due to collisions between numerous tags and interference from metal and moisture in the washing workshop, leading to identification failure. This process iteratively optimizes the mixing matrix in the whitening space using projection perimeter matching as the criterion. Through dual-path updates, it prioritizes both convergence speed and demixing accuracy. After meeting the convergence conditions, inverse whitening restores the true mixing matrix, accurately separating the independent source signals of each linen tag. This effectively improves the identification accuracy in dense tag scenarios and ensures the stability of the entire linen traceability management process.

[0049] In one embodiment, a first update matrix and a second update matrix are calculated based on the whitening blending matrix of the current round, the observed perimeter vector, the natural gradient matrix, and the optimal update step size of the current round as inputs, including: Using the whitening mixing matrix and the observed projection perimeter vector of the current round as input, the first update matrix is ​​obtained by solving the pseudo-inverse analytical method; Using the whitening mixing matrix, natural gradient matrix, and optimal update step size of the current round as input, the second update matrix is ​​obtained by weighted update along the negative gradient direction.

[0050] In one implementation, the pseudo-inverse analytical method relies on the linear mapping relationship between the projected perimeter and the mixing matrix to directly solve the correction increment through pseudo-inverse operation, resulting in fast convergence speed. The gradient descent method updates steadily along the negative direction of the natural gradient, exhibiting strong anti-interference ability and numerical stability. The two methods generate the update matrix simultaneously and then select the optimal one, which can avoid the single path from getting stuck in local optima. This method is suitable for RFID signal environments with densely overlapping medical linen tags and strong on-site interference, balancing demixing efficiency and accuracy.

[0051] In one embodiment, a signal decoding operation is performed based on the tag signal matrix and preset RFID communication rules to output a traceability dataset for all linens in a single batch, including: Traverse each independent tag signal in the tag signal matrix, complete carrier synchronization and frame header positioning according to the synchronization code sequence in the RFID communication rules, and extract the corresponding valid data frame segment; Perform legality verification on each valid data frame segment according to the verification algorithm in the RFID communication rules, and remove distorted and interfering frames; Based on the traceability field mapping relationship in the RFID communication rules, the valid data frames that have passed the verification are parsed field by field to extract the unique identification code, circulation node identifier and status information of each piece of linen, and the data are summarized to form a traceability dataset for a single batch of linen.

[0052] In one implementation, the RFID communication rules are a pre-agreed standardized communication protocol specification between the reader and the electronic tag of medical linens. It covers the signal modulation and coding method and frame synchronization code sequence of the physical layer, the data frame structure division rules and cyclic redundancy check algorithm of the data link layer, and the mapping relationship of traceability information encoding fields and communication interaction timing agreement of the application layer. It is used to unify the parsing standard for signal transmission and reception, and ensure that the demixed tag signal can be accurately decoded and the traceability data such as the identity and circulation status of the corresponding linens can be extracted.

[0053] In one implementation, according to the cyclic redundancy check algorithm agreed upon in the RFID communication rules, the data payload and the tail check code are separated from the valid data frame segment. The check value is calculated on the data payload using the generator polynomial specified in the protocol. The calculation result is compared bit by bit with the check code built into the frame. If the two do not match, the frame is determined to be a distorted interference frame affected by metal or water vapor in the field and is discarded. Only the valid data frames with consistent check are retained for subsequent traceability information parsing.

[0054] In one implementation, frame header positioning, legality verification, and field parsing are sequentially performed on each demixed tag signal. This aims to accurately locate valid data frames from signals containing on-site interference, eliminate distorted data affected by metal and moisture, ensure accurate and reliable decoding results, extract the identity code, circulation node, and status information of each piece of linen, and form a complete and reliable traceability dataset to provide data support for the full-process traceability and control of medical linens.

[0055] The foregoing has described one embodiment of the present invention in detail, but this is merely a preferred embodiment and should not be considered as limiting the scope of the invention. All equivalent variations and modifications made within the scope of the claims of this invention should still fall within the scope of the claims.

Claims

1. A method for full-process control and management of RFID-traceable medical linens, characterized in that, The method includes: The raw data of RFID identification of medical linens is obtained, and a signal observation matrix is ​​constructed based on the raw data; the raw data includes traceability message data, channel transmission data and environmental noise data; The initial mixing matrix is ​​obtained by performing successive nonnegative projection iterations on the signal observation matrix; The signal observation matrix is ​​whitened to output a whitening matrix and a whitened signal observation matrix, and the observation perimeter vector is calculated based on the whitened signal observation matrix; the observation perimeter vector is used to iteratively optimize the mixing matrix to complete the separation of tag collision signals; The collision signal matrix of the linen is obtained by performing iterative separation of the collision signal based on the initial mixing matrix, whitening matrix and observed perimeter vector; The traceability dataset for all linens in a single batch is obtained by performing signal decoding operations based on the tag signal matrix and preset RFID communication rules. The traceability dataset includes the unique identifier, category, disinfection status and current circulation node information of each linen.

2. The RFID-based traceability method for the entire process of medical linen management and control according to claim 1, characterized in that, The signal observation matrix is ​​constructed based on the original data, including: A source signal matrix is ​​constructed based on the traceability message data; the source signal matrix is ​​used to describe the original response state of each RFID tag at each sampling time; A channel mixing matrix is ​​constructed based on the channel transmission data, and the initial observation matrix is ​​obtained by channel mapping of the source signal matrix using the channel mixing matrix. A noise impact matrix is ​​constructed based on environmental noise data. Abnormal response regions in the initial observation matrix are identified through the noise impact matrix. The observation data corresponding to the abnormal response regions are locally corrected to obtain the corrected observation matrix. The signal aliasing coefficient corresponding to each sampling time is calculated based on the correction observation matrix. The observation data in the correction observation matrix is ​​then divided into high-confidence observation areas and low-confidence observation areas based on the signal aliasing coefficient. Consistency constraint correction is performed on the observation data corresponding to the low-confidence observation region by the high-confidence observation region, and the corrected observation regions are fused to obtain the signal observation matrix.

3. The RFID-based traceability method for the entire process of medical linen management and control according to claim 1, characterized in that, The initial mixing matrix is ​​obtained by performing successive nonnegative projection iterations on the signal observation matrix, including: Step 1: Initialize the iteration parameters; the iteration parameters include the residual matrix, the empty index set, and the preset maximum number of iterations; the residual matrix is ​​obtained by assigning values ​​to the signal observation matrix; Step 2: Calculate the strong convex function value of each column of the residual matrix for this round, and traverse all columns to filter out the index of the extreme value column corresponding to the column that makes the function value the largest. Step 3: Incorporate the current round of extreme value column index into the empty index set, record the positions of the identified tag channel feature columns, and obtain the updated new index set; Step 4: Calculate the projection coefficient matrix corresponding to each column of the signal observation matrix based on the signal observation matrix and the new index set; Step 5: Update the residual matrix according to the signal observation matrix, the new index set, and the projection coefficient matrix to obtain a new residual matrix; Step 6: Calculate the F-norm of the new residual matrix. If the F-norm is less than the norm threshold, or the current iteration number is greater than the preset maximum iteration number, terminate the iteration and extract the corresponding column elements of the signal observation matrix according to the new index set to obtain the initial mixing matrix; otherwise, repeat steps 2 to 6.

4. The RFID-traceable medical linen whole-process control and processing method according to claim 3, characterized in that, The residual matrix is ​​updated based on the signal observation matrix, the new index set, and the projection coefficient matrix to obtain a new residual matrix, including: Based on the new index set, the channel feature submatrix corresponding to the identified medical linen tags is extracted from the signal observation matrix; Multiply the channel feature submatrix by the projection coefficient matrix to obtain the tag superposition fitting matrix of the currently identified linen tags, and subtract the tag superposition fitting matrix from the signal observation matrix to obtain the new residual matrix.

5. The RFID-based traceability method for the entire process of medical linen management and control according to claim 1, characterized in that, The observed perimeter vector is calculated based on the whitening observation signal, including: Multiplying the whitening matrix by the signal observation matrix yields the whitened signal observation matrix; Generate a preset number of random unit projection direction vectors, and linearly project the whitened observation signal matrix onto each set of random unit projection directions to obtain a preset number of one-dimensional projection sequences. Calculate the maximum and minimum values ​​of each group of one-dimensional projection sequences, calculate the difference between the maximum and minimum values, and arrange twice the difference of each group in sequence to construct the observation perimeter vector.

6. The RFID-based traceability method for the entire process of medical linen management and control according to claim 5, characterized in that, Based on the initial mixing matrix, whitening matrix, and observed perimeter vector, collision signal iterative separation is performed to obtain the linen tag signal matrix, including: Step 1: Map the initial mixing matrix to the whitening space to obtain the whitening mixing matrix; Step 2: Calculate the theoretical estimated perimeter vector for each projection direction based on the whitening mixing matrix of this round, and calculate the error vector between the observed perimeter vector and the theoretical estimated perimeter vector. Use the squared L2 norm of the error as the cost function value for this round. Step 3: Derive the natural gradient matrix of the cost function with respect to the whitening mixing matrix based on the perimeter matching criterion, and calculate the optimal update step size for this iteration based on the current cost function value and the natural gradient matrix; Step 4: Calculate the first update matrix and the second update matrix based on the whitening mixing matrix of the current round, the observed perimeter vector, the natural gradient matrix, and the optimal update step size of this round as inputs. Calculate the cost function values ​​corresponding to the two update matrices respectively, and select the update matrix that minimizes the cost as the whitening mixing matrix for the next round. Step 5: If the cost function value corresponding to the whitening mixture matrix in this round is less than the convergence threshold, or not less than the maximum number of iterations, then terminate the iteration and output the optimal whitening mixture matrix; otherwise, repeat steps 2 to 5. Step 6: Map the optimal whitening mixing matrix back to the original signal space through the inverse whitening operation to obtain the final mixing matrix, and output the final mixing matrix as the tag signal matrix.

7. The RFID-based traceability method for the entire process of medical linen management and control according to claim 6, characterized in that, The first and second update matrices are calculated based on the whitening mixing matrix of the current round, the observed perimeter vector, the natural gradient matrix, and the optimal update step size of this round as inputs. These matrices include: Using the whitening mixing matrix and the observed projection perimeter vector of the current round as input, the first update matrix is ​​obtained by solving the pseudo-inverse analytical method; Using the whitening mixing matrix, natural gradient matrix, and optimal update step size of the current round as input, the second update matrix is ​​obtained by weighted update along the negative gradient direction.

8. The RFID-based traceability method for the entire process of medical linen management and control according to claim 1, characterized in that, Based on the tag signal matrix and preset RFID communication rules, a signal decoding operation is performed to output a traceability dataset for all linens in a single batch, including: Traverse each independent tag signal in the tag signal matrix, complete carrier synchronization and frame header positioning according to the synchronization code sequence in the RFID communication rules, and extract the corresponding valid data frame segment; Perform legality verification on each valid data frame segment according to the verification algorithm in the RFID communication rules, and remove distorted and interfering frames; Based on the traceability field mapping relationship in the RFID communication rules, the valid data frames that have passed the verification are parsed field by field to extract the unique identification code, circulation node identifier and status information of each piece of linen, and the data are summarized to form a traceability dataset for a single batch of linen.

Images