A method for estimating the number of tags using the collision waveform of the initial frame
Patent Information
- Authority / Receiving Office
- CN · China
- Patent Type
- Patents(China)
- Current Assignee / Owner
- Filing Date
- 2023-09-27
- Publication Date
- 2026-08-14
AI Technical Summary
[0005]本发明要解决的技术问题是如何提供一种利用起始帧碰撞波形估计标签数量的方法，以解决参数Q的估计问题
[0015]本发明提出一种利用起始帧碰撞波形估计标签数量的方法，本发明利用起始帧(SOF，Start ofFrame)碰撞波形特征展现标签数量信息，结合随机森林强大的细节特征学习能力，实现标签数量的估计，便于优化初始Q值设置，从而降低初始帧长对系统盘点效率的影响。
Smart Images
Figure CN117332796B_ABST
Abstract
Description
Technical Field
[0001] This invention belongs to the field of radio frequency identification, and specifically relates to a method for estimating the number of tags using the collision waveform of the initial frame. Background Technology
[0002] High-frequency RFID readers operate at 13.56MHz and transmit energy through inductive coupling with an antenna coil, offering strong resistance to environmental noise and electromagnetic interference. In multi-tag intensive reading applications, collisions can occur when multiple tags simultaneously send responses to the reader, severely impacting system efficiency. When the frame length and number of tags are equal, DFSA achieves a theoretical maximum throughput of 36.8%. The frame length is 2... Q The frame length can be adjusted using the parameter Q to make it close to the number of tags, thereby improving inventory efficiency. However, if the initial Q value is set too high or too low, multiple rounds of Q value adjustment are required to ensure successful tag response, which leads to high inventory time complexity. To avoid significant degradation in throughput, there is an urgent need for a method that can predict the number of tags to determine an appropriate frame length and reduce the impact of the initial frame length on system efficiency.
[0003] In densely stacked label estimation applications, if the label number estimation algorithm has excessively high complexity, it is unsuitable for embedded systems with limited performance. Compared to deep machine learning algorithms such as convolutional neural networks, the random forest ensemble learning algorithm integrates numerous decision trees that only require a small number of computationally inexpensive comparison operations. It can not only run in parallel mode, but the average policy among the decision trees can also effectively guarantee estimation accuracy. Summary of the Invention
[0004] (a) Technical problems to be solved
[0005] The technical problem to be solved by the present invention is how to provide a method for estimating the number of tags using the collision waveform of the initial frame, so as to solve the estimation problem of parameter Q.
[0006] (II) Technical Solution
[0007] To address the aforementioned technical problems, this invention proposes a method for estimating the number of tags using the initial frame collision waveform. This method includes the following steps:
[0008] S1. Construct an offline random forest ensemble learning model for label count estimation, including the following sub-steps:
[0009] S11. Random Forest Sample Set Acquisition: In multi-label dense scenarios, the Q value in the Dynamic Framed Slotted ALOHA (DFSA) algorithm is set to 0, multiple sets of known number of labels are counted, and the SOF collision waveform of the first response of the multi-label is taken as the feature set of the sample set, and the number of labels is taken as the label space of the sample set.
[0010] S12. Random Forest Offline Model Training: Input the collected sample set into the random forest. Divide the collected samples into a training set and a test set. The training set is used for the random forest offline model for label number estimation, and the test set is used to test the accuracy of the trained random forest model.
[0011] S2. Estimation of the quantity of labels to be inventoried, including the following sub-steps:
[0012] S21. Acquisition of SOF collision waveform of densely stacked tags to be inventoried: For densely stacked tags to be inventoried, set the Q value of the DFSA algorithm to 0, and acquire the SOF collision waveform of the first response of all tags.
[0013] S22. Estimation of the number of labels to be inventoried: Input the SOF collision waveform of the labels to be inventoried into the trained random forest model to obtain the estimated number of labels N, and calculate Q = [log2N] based on the estimated number of labels.
[0014] (III) Beneficial Effects
[0015] This invention proposes a method for estimating the number of tags using the collision waveform of the start of frame (SOF). This invention utilizes the collision waveform features of the start of frame (SOF) to display the tag number information, and combines the powerful detail feature learning ability of random forest to achieve tag number estimation, which facilitates the optimization of the initial Q value setting, thereby reducing the impact of the initial frame length on the system inventory efficiency.
[0016] The method proposed in this invention enables the estimation of the number of tags in a high-frequency RFID system. It can solve the problems of the sensitivity of the initial frame length selection in the DFSA algorithm, as well as the large estimation error and high time complexity caused by the early stopping of the RFID dynamic frame time slot, and can effectively improve the inventory efficiency of the high-frequency RFID system. Attached Figure Description
[0017] Figure 1 This is a flowchart illustrating the training process of the method for estimating the number of tags using the initial frame collision waveform of the present invention.
[0018] Figure 2 This is a flowchart illustrating the implementation of the method for estimating the number of tags using the initial frame collision waveform of the present invention. Detailed Implementation
[0019] To make the objectives, contents, and advantages of the present invention clearer, the specific embodiments of the present invention will be described in further detail below with reference to the accompanying drawings and examples.
[0020] This invention relates to the field of radio frequency identification, specifically to a method for estimating the number of tags using the collision waveform of the initial frame, which facilitates parameter setting and enables rapid tag inventory.
[0021] This invention discloses a method for estimating the number of tags using the collision waveform of the start frame, applicable to high-frequency radio frequency identification (RFID) systems employing the Dynamic Framed Slotted ALOHA (DFSA) algorithm. The method mainly includes: (1) Constructing a random forest offline model for tag number estimation: The initial Q value of the DFSA algorithm is set to 0. Multiple sets of tags with known quantities are counted. At this time, all tags respond simultaneously. The collision waveform of the start frame (SOF) has a one-to-one mapping relationship with the number of tags. The SOF collision waveform and the known number of tags are used as the feature set and label space of the training random forest sample set, respectively, to train the random forest offline model for tag number estimation; (2) Predicting the number of tags to be counted: The SOF collision waveform of an unknown number of densely packed tags is used as the input of the random forest offline model, and its output is the estimated tag number.
[0022] This invention proposes a method for estimating the number of tags based on multi-tag SOF collision waveforms. The method includes:
[0023] S1. Construct an offline random forest ensemble learning model for label count estimation, including the following sub-steps:
[0024] S11. Random Forest Sample Set Acquisition: In multi-label dense scenarios, the Q value in the Dynamic Framed Slotted ALOHA (DFSA) algorithm is set to 0. Multiple sets of known number of labels are counted, and the SOF collision waveform of the first response of the multi-label is taken as the feature set of the sample set. The number of labels is the label space of the sample set.
[0025] S12. Random Forest Offline Model Training: Input the collected sample set into the random forest. Divide the collected samples into a training set and a test set. 70% of the total samples are used as the training set, which is used for the random forest offline model for label number estimation. 30% of the total samples are used as the test set, which is used to test the accuracy of the trained random forest model.
[0026] S2. Estimated quantity of labels to be inventoried. This includes the following sub-steps:
[0027] S21. Acquisition of SOF collision waveforms of densely stacked tags to be inventoried: For densely stacked tags to be inventoried, set the Q value of the DFSA algorithm to 0 and acquire the SOF collision waveforms of the first response of all tags.
[0028] S22. Estimation of the number of labels to be inventoried. Input the SOF collision waveform of the labels to be inventoried into the trained random forest model to obtain the estimated number of labels N. Based on the estimated number of labels, Q = [log2N] can be calculated.
[0029] Example 1:
[0030] This invention proposes a method for tag number estimation based on multi-tag SOF collision waveforms. The following uses the ISO18000-3 protocol Mode 3 as an example to illustrate the specific implementation of this invention, but this is not intended to limit the specific implementation method. Figure 2 As shown, the specific implementation process of this method is as follows.
[0031] (1) Construct an offline random forest model for label number estimation
[0032] (11) Construct a sample set for training the random forest. In the scenario of dense label stacking, set the Q value of the DFSA algorithm to 0, traverse and count multiple sets of known number of labels, and use the SOF collision waveform of the first response of multiple labels as the feature set of the sample set, and the number of labels as the label space of the sample set. In order to avoid the influence of noise, the SOF collision waveform is collected no less than 100 times for each type of label with the same number.
[0033] Taking the inventory of up to 100 tags as an example, when the number of tags N is 1, 2, 4, 8, 16, 32, 64, and 128, the Q value of the DFSA algorithm is set to 0, the SOF collision waveform of the first response of all tags is the random forest sample feature set, and the number of tags N is the label space of the random forest model.
[0034] (12) Training the random forest model. According to this invention, the collected sample set is input into the random forest, and the training process and structure of the random forest are as follows: Figure 1 As shown, 70% of the total samples are used as the training set and 30% as the test set to build a random forest model. The output of the random forest is the estimated number of labels.
[0035] The following example, using a list of 100 or fewer tags, illustrates the structure of a random forest:
[0036] A flowchart of the training process for a random forest for label count estimation is shown below. Figure 1 .
[0037] The original dataset, denoted by D, is a collection of SOF collision waveforms obtained after iterating through multiple sets with varying numbers of labels. Multiple training datasets can be generated from the original dataset through random sampling with replacement, specifically as follows: For an original dataset D containing S SOF collision waveforms, one SOF collision waveform is randomly selected and added to the training dataset D. n The SOF collision waveform is then placed back into D, so that it may still be selected in the next random selection. After repeating the above process S times, D containing S SOF collision waveforms can be obtained. n This generates a training dataset for training a decision tree, and the process is iterated to generate multiple training datasets. Sampling with replacement allows for the preservation of similarities among different training datasets while maintaining their differences. This idea of seeking common ground while respecting differences enables random forests to achieve high estimation performance through swarm intelligence.
[0038] To suit embedded systems with limited performance, the complexity of random forests used for label number estimation cannot be too high. The two main hyperparameters determining the complexity and performance of a random forest are the optimal maximum depth (MD) and the optimal number of decision trees (T). With a fixed number of decision trees, as the maximum depth increases, the performance of the random forest gradually improves to its optimum; even if the number of decision trees continues to increase, the performance remains constant, but the complexity continues to increase. Similarly, with a fixed maximum depth, as the number of decision trees increases, the performance gradually improves to its optimum and then remains constant, but the complexity continues to increase. The optimal values are the minimum number of decision trees and the minimum maximum depth that achieve the best performance and lowest complexity for the random forest. To achieve optimal performance and lowest complexity in the final random forest, it is necessary to first train the random forest to obtain the optimal values for these two hyperparameters without considering its complexity.
[0039] Based on the optimal number of decision trees T in a random forest, T sets of training datasets are obtained by random sampling with replacement. These T sets of training datasets are used to train T decision trees in parallel. The detailed training process of a decision tree is as follows.
[0040] First, randomly select m sub-features from the M features of the SOF collision waveform to construct a Random Feature Subspace (RFSC), and set it as the root node. According to the rule of thumb, the value of m follows m = [log₂M] + 1;
[0041] Then, according to the optimal splitting criterion, a root node is divided into two child nodes. After each split, it is necessary to determine whether the current decision tree depth of the child node is consistent with the maximum optimal depth MD. If they are inconsistent, features are reselected from RFSC and the recursive operation is performed. As the recursive operation is executed, the maximum depth of the child node gradually increases until it is consistent with the maximum optimal depth. The recursive operation stops, the decision tree training process ends, and the estimated number of labels is output. The estimated number of labels is the output value of the leaf node.
[0042] After training, each decision tree outputs an estimate of the number of labels. Therefore, after training T decision trees, we obtain T estimates of the number of labels. To reduce the influence of bias in each decision tree, the final estimate of the number of labels by the random forest is the minimum mean square error of these T estimates.
[0043] Label count estimation is a numerical prediction problem within regression, therefore this solution employs a regression algorithm. The performance metric used is the mean absolute error, which represents the estimated label count y of the random forest across T test cases. est and the actual number of tags y true The average absolute error is the sum of the absolute errors between the predicted values. The mean absolute error can better reflect the actual situation of the predicted error, as shown in equation (1).
[0044]
[0045] (2) Estimating the number of labels to be inventoried
[0046] (21) Collect SOF collision waveforms of the tags to be inventoried. For an unknown number of tags to be inventoried, set the Q value of the DFSA algorithm to 0. At this time, all tags respond at the same time and generate collisions. Collect the SOF collision waveforms of the first response of all tags.
[0047] (22) Estimate the number of labels to be inventoried. Input the SOF collision waveforms of the first response of all labels corresponding to the labels to be inventoried into the trained random forest model to obtain the estimated number of labels.
[0048] Example 2:
[0049] A method for estimating the number of tags using the collision waveform of the initial frame includes the following steps:
[0050] (1) Construct a random forest model for label number estimation;
[0051] (2) Estimate the number of labels to be inventoried.
[0052] A method for estimating the number of tags using the collision waveform of the initial frame, wherein the determination method in step (1) is as follows:
[0053] (11) Random Forest Sample Set Collection. In multi-label dense scenarios, the Q value of the DFSA algorithm is set to 0, and multiple sets of known number of labels are counted. The number of labels is a power of 2, such as 1, 2, 4, 8, 16, 32, etc. The number of counts for each set of labels is not less than 100. The SOF collision waveform of the first response of the label is taken as the feature set of the sample set, and the actual number of labels is taken as the label space of the sample set.
[0054] (12) Random forest offline model training. Input the sample set into the random forest, with 70% of the total number of samples as the training set to build a random forest offline model for label number estimation, and 30% of the total number of samples as the test set to test the accuracy of the trained random forest model.
[0055] A method for estimating the number of tags using the collision waveform of the initial frame, wherein the determination method in step (2) is as follows:
[0056] (21) Acquisition of the waveform of the tag to be inventoried. For densely stacked tags to be inventoried, set the Q value of the DFSA algorithm to 0 and acquire the SOF collision waveform of the first response of all tags.
[0057] (22) Estimation of the number of labels to be inventoried. Input the waveform into the trained random forest model to obtain the estimated number of labels N.
[0058] This invention utilizes the collision waveform features of the Start of Frame (SOF) to display the number of labels, and combines the powerful detail feature learning capabilities of random forests to estimate the number of labels, which facilitates the optimization of the initial Q value setting, thereby reducing the impact of the initial frame length on the system's inventory efficiency.
[0059] The method proposed in this invention enables the estimation of the number of tags in a high-frequency RFID system. It can solve the problems of the sensitivity of the initial frame length selection in the DFSA algorithm, as well as the large estimation error and high time complexity caused by the early stopping of the RFID dynamic frame time slot, and can effectively improve the inventory efficiency of the high-frequency RFID system.
[0060] The above description is only a preferred embodiment of the present invention. It should be noted that for those skilled in the art, several improvements and modifications can be made without departing from the technical principles of the present invention, and these improvements and modifications should also be considered within the scope of protection of the present invention.
Claims
1. A method for estimating the number of tags in a high-frequency RFID system using the initial frame collision waveform, characterized in that, The method includes the following steps: S1. Construct an offline random forest ensemble learning model for label count estimation, including the following sub-steps: S11. Random Forest Sample Set Acquisition: In multi-label dense scenarios, the Q value in the dynamic frame slotted ALOHA algorithm, i.e. the DFSA algorithm, is set to 0. Multiple sets of known number of labels are counted. The SOF collision waveform of the first response of multiple labels is taken as the feature set of the sample set, and the number of labels is taken as the label space of the sample set. S12. Random Forest Offline Model Training: Input the collected sample set into the random forest. Divide the collected samples into a training set and a test set. The training set is used to train the random forest offline model for estimating the number of labels, and the test set is used to test the accuracy of the trained random forest model. S2. Estimation of the quantity of labels to be inventoried, including the following sub-steps: S21. Acquisition of SOF collision waveform of densely stacked tags to be inventoried: For densely stacked tags to be inventoried, set the Q value of the DFSA algorithm to 0, and acquire the SOF collision waveform of the first response of all tags. S22. Estimation of the number of labels to be inventoried: Input the SOF collision waveform of the initial frame of the labels to be inventoried into the trained random forest model to obtain the estimated number of labels. N Based on the estimated number of tags, Q = [log2] is calculated. N ].
2. The method for estimating the number of tags in a high-frequency RFID system using the initial frame collision waveform as described in claim 1, characterized in that, In S11, for each type of label with the same number of tags, the SOF collision waveform of the starting frame is collected no less than 100 times.
3. The method for estimating the number of tags in a high-frequency RFID system using the initial frame collision waveform as described in claim 2, characterized in that, For inventorying up to 100 tags, with the number of tags N being 1, 2, 4, 8, 16, 32, 64, and 128 respectively, the Q value of the DFSA algorithm is set to 0, the SOF collision waveform of the first response of all tags is the random forest sample feature set, and the number of tags N is the label space of the random forest model.
4. The method for estimating the number of tags in a high-frequency RFID system using the initial frame collision waveform as described in claim 1, characterized in that, In step S12, 70% of the total number of samples is used as the training set, and 30% of the total number of samples is used as the test set.
5. The method for estimating the number of tags in a high-frequency RFID system using the initial frame collision waveform as described in claim 1, characterized in that, In step S11, the original dataset is a set of initial frame SOF collision waveforms obtained after traversing multiple sets of different label numbers, denoted by D. Multiple training datasets can be generated from the original dataset through random sampling with replacement.
6. The method for estimating the number of tags in a high-frequency RFID system using the initial frame collision waveform as described in claim 5, characterized in that, Random sampling methods with replacement include: for those including S The original dataset of SOF collision waveforms in the initial frames. D Randomly select a starting frame SOF collision waveform and put it into the training dataset. D n Then, the SOF collision waveform of the initial frame is put back. D In this way, the SOF collision waveform of the starting frame may still be selected in the next random selection; repeat the above process. S After that, we get the following: S The initial frame SOF collision waveform D n This generates a training dataset for training a decision tree, and the process is iterated to generate multiple training datasets.
7. The method for estimating the number of tags in a high-frequency RFID system using the initial frame collision waveform as described in any one of claims 1-6, characterized in that, The two hyperparameters that determine the complexity and performance of a random forest are the optimal values of its maximum depth. MD Optimal number of decision trees T When the number of decision trees is fixed, as the maximum depth of the decision trees increases, the performance of the random forest gradually improves to the optimal level. Even if the number of decision trees continues to increase, the performance of the random forest remains unchanged, but the complexity of the random forest continues to increase. When the maximum depth is fixed, as the number of decision trees increases, the performance of the random forest gradually improves to the optimal level and then remains unchanged, but the complexity of the random forest continues to increase. The minimum value of the number of decision trees and the minimum maximum depth that makes the random forest perform optimally and have the minimum complexity is the optimal value. In order to make the final random forest perform optimally and have the minimum complexity, it is necessary to first train the random forest to obtain the optimal values of these two hyperparameters without considering the complexity of the random forest.
8. The method for estimating the number of tags in a high-frequency RFID system using the initial frame collision waveform as described in claim 7, characterized in that, Based on the optimal number of decision trees in a random forest T Obtained by random sampling with replacement T This training dataset T The training datasets are completed in parallel. T Training a decision tree.
9. The method for estimating the number of tags in a high-frequency RFID system using the initial frame collision waveform as described in claim 8, characterized in that, The detailed training process of a decision tree includes: First, let's look at the SOF collision waveform from the initial frame. M Randomly selected from the features m For each sub-feature, construct a random feature subspace (RFSC) and set it as the root node; according to the rule of thumb... m The value of follows m= [ log 2 M +1; Then, based on the optimal partitioning criterion, a root node is divided into two child nodes. After each partition, it is necessary to determine whether the current decision tree depth of the child node is the optimal value of the maximum depth. MD If they are not consistent, then reselect features from RFSC and perform recursive operations. As the recursive operations are executed, the maximum depth of the child nodes gradually increases until it is consistent with the optimal value of the maximum depth. The recursive operations stop, the decision tree training process ends, and the estimated number of labels is output. The estimated number of labels is the output value of the leaf nodes. After training, each decision tree outputs an estimate of the number of labels, therefore T After the decision trees are trained, the results are obtained. T The estimated number of labels; the final estimated number of labels by the random forest is this. T The minimum mean square error of the estimated number of labels.
10. The method for estimating the number of tags in a high-frequency RFID system using the initial frame collision waveform as described in claim 9, characterized in that, The number of labels is estimated using a regression algorithm; the performance metric is the mean absolute error, which represents... T In this test, the estimated number of labels for the random forest was... y est and the actual number of tags y true The average absolute error between the two values is the mean absolute error, which better reflects the actual situation of the predicted value error, as shown in equation (1): (1)。