RFID tag feature extraction and recognition method and system based on deep learning

Abstract

The present application relates to the field of RFID tag technology, and specifically discloses a method and system for RFID tag feature extraction and recognition based on deep learning. The method mainly includes: obtaining an RFID reader to convert radio frequency signals into electrical signal data, selecting specific features to construct first input data, processing with a first deep learning model to obtain a first recognition result and feature contribution, determining important features based on the contribution, then obtaining electrical signal data, extracting important features to construct second input data, and processing with a second model to obtain a second recognition result. The present application can significantly improve the recognition accuracy of RFID tags by obtaining RFID electrical signal data, selecting specific features to construct input, identifying and optimizing features with a deep learning model, and then processing to obtain the final recognition result, thereby greatly improving the tracking efficiency and reliability of logistics information, helping to optimize inventory management, effectively reducing the risk of cargo loss, and improving the safety and efficiency of overall logistics operations.

Description

Technical Field

[0001] The present application relates to the field of RFID tag technology, and in particular to a method and system for extracting and identifying RFID tag features based on deep learning. Background Art

[0002] RFID tags are widely used in logistics. The principle is to use radio waves for communication to realize the information transmission between the reader and the electronic tag. When the goods with RFID tags enter the working area of ​​the reader, the radio frequency signal emitted by the reader will activate the tag. The tag then transmits its own coded information back to the reader through the built-in antenna to complete the information reading.

[0003] However, in the real environment, there may be various interference sources such as wireless devices and metal structures. The electromagnetic waves generated by them may interfere with the transmission of RFID signals, resulting in a decrease in the quality of the signal received by the reader. The quality of the RFID tag will also affect its ability to be recognized by the reader. For example, chip damage, poor antenna design or improper attachment method may cause the reader to be unable to accurately read or decode the signal returned by the tag, which in turn causes shortcomings such as inaccurate logistics information tracking, chaotic inventory management, and increased risk of cargo loss, seriously affecting logistics efficiency and accuracy, and causing unnecessary losses to enterprises and customers. Summary of the Invention

[0004] This application provides a deep learning-based RFID tag feature extraction and recognition method and system, which effectively solves the problem in the prior art that environmental interference and tag quality issues affect the transmission of RFID signals, causing the reader to encounter difficulties in reading or decoding the signal returned by the tag, thereby affecting the tracking efficiency and accuracy of logistics information. It can significantly improve the recognition accuracy of RFID tags, thereby greatly improving the tracking efficiency and reliability of logistics information, helping to optimize inventory management, effectively reducing the risk of cargo loss, and improving the safety and efficiency of overall logistics operations.

[0005] In a first aspect, the present application provides an RFID tag feature extraction and recognition method based on deep learning, comprising: obtaining electrical signal data; wherein the electrical signal data is converted from the radio frequency signal returned by the RFID reader to the RFID tag, and the electrical signal data contains features of multiple categories; determining specific features from all categories of features, and constructing first input data containing the specific features; using a first deep learning model to process the first input data to obtain a first recognition result, as well as the degree of contribution of each category of features to the first recognition result; determining important features among all specific features based on the degree of contribution of each category of features to the first recognition result; obtaining the recognition result of the radio frequency signal returned by the RFID reader to the RFID tag, and obtaining electrical signal data converted from the radio frequency signal based on the recognition result; extracting important features from the electrical signal data, and constructing second input data containing the important features; using a second deep learning model to process the second input data to obtain a second recognition result.

[0006] By acquiring RFID electrical signal data, selecting specific features to construct inputs, using deep learning models to identify and optimize features, and then processing the final recognition results, the recognition accuracy of RFID tags can be significantly improved, thereby greatly improving the tracking efficiency and reliability of logistics information, helping to optimize inventory management, effectively reducing the risk of cargo loss, and improving the safety and efficiency of overall logistics operations.

[0007] Furthermore, the specific feature includes at least one of the following: environmental parameters, specific attributes of the tag, and timestamp and location information.

[0008] Furthermore, determining important features among all specific features according to the degree of contribution of features of each category to the first recognition result includes: determining relevant features among all specific features according to the degree of contribution of features of each category to the first recognition result; constructing all possible subsets of relevant features of all categories; wherein each subset includes at least one relevant feature; constructing data samples; wherein each data sample contains a subset of one category, and all data samples contain subsets of all categories; training a second deep learning model with each data sample one by one; applying the trained second deep learning model to a validation set or a test set, and calculating performance indicators; and determining important features based on the performance indicators.

[0009] Furthermore, determining relevant features among all specific features according to the contribution degree of each category of features to the first recognition result includes: taking specific features whose contribution degree is greater than a preset first threshold as relevant features.

[0010] Furthermore, performance indicators are calculated using evaluation functions in the machine learning framework or custom scripts.

[0011] Furthermore, determining the important features according to the performance indicators includes: screening out a subset having a performance indicator greater than a preset second threshold and the least number of features; and taking the relevant features included in the screened subset as the important features.

[0012] Furthermore, the recognition result includes success or failure; obtaining the recognition result of the radio frequency signal returned by the RFID reader includes: obtaining the first recognition result of the radio frequency signal returned by the RFID tag by the RFID reader; judging whether the first recognition result is a failure; for the RFID tag whose first recognition result is a failure, continuously obtaining the recognition result of the RFID reader for the RFID tag according to a preset time interval or condition until the preset first time period is reached or the recognition result is successful.

[0013] Furthermore, obtaining electrical signal data converted from the radio frequency signal according to the recognition result includes: determining an RFID tag whose recognition result is still failed after a preset first time period has been reached; and obtaining electrical signal data converted from the radio frequency signal returned by the RFID tag.

[0014] Furthermore, the first deep learning model and the second deep learning model are both long short-term memory networks.

[0015] In the second aspect, the present application provides an RFID tag feature extraction and recognition system based on deep learning, which adopts the RFID tag feature extraction and recognition method based on deep learning as described in the first aspect, and includes: a data acquisition module, a specific feature screening and input data construction module, a feature contribution identification module, a feature contribution evaluation and signal recognition module, an input data construction module and a tag recognition module.

[0016] The data acquisition model is used to acquire electrical signal data; wherein, the electrical signal data is converted from the radio frequency signal returned by the RFID reader to the RFID tag, and the electrical signal data contains multiple categories of features; the specific feature screening and input data construction module is used to determine specific features from all categories of features and construct first input data containing the specific features; the feature contribution identification module is used to process the first input data using a first deep learning model to obtain a first recognition result, as well as the degree of contribution of each category of features to the first recognition result; the feature contribution evaluation and signal identification module is used to determine important features among all specific features based on the degree of contribution of each category of features to the first recognition result; obtain the recognition result of the radio frequency signal returned by the RFID reader to the RFID tag, and obtain the electrical signal data converted from the radio frequency signal based on the recognition result; the input data construction module is used to extract important features from the electrical signal data and construct second input data containing the important features; the tag identification module is used to process the second input data using a second deep learning model to obtain a second recognition result.

[0017] In a third aspect, the present application provides an RFID tag feature extraction and identification device based on deep learning, which includes: a memory and a processor.

[0018] The memory is used to store a computer program; the processor is used to implement the steps of the RFID tag feature extraction and recognition method based on deep learning as described in the first aspect when executing the computer program.

[0019] In a fourth aspect, the present application provides a readable storage medium, which stores computer program instructions. When the computer program instructions are read and executed by a processor, the steps of the deep learning-based RFID tag feature extraction and recognition method as described in the first aspect are executed.

[0020] The technical solution provided by this application has at least the following technical effects or advantages:

[0021] By acquiring RFID electrical signal data, selecting specific features to construct inputs, using deep learning models to identify and optimize features, and then processing the final recognition results, it effectively solves the problem in existing technologies that environmental interference and tag quality issues affect the transmission of RFID signals, causing readers to encounter difficulties when reading or decoding the signals returned by the tags, thereby affecting the tracking efficiency and accuracy of logistics information. It can significantly improve the recognition accuracy of RFID tags, thereby greatly improving the tracking efficiency and reliability of logistics information, helping to optimize inventory management, effectively reducing the risk of goods loss, and improving the safety and efficiency of overall logistics operations. BRIEF DESCRIPTION OF THE DRAWINGS

[0022] Figure 1 This is a flowchart of the RFID tag feature extraction and recognition method based on deep learning in Example 1 of the present application;

[0023] Figure 2 For this application Figure 1 A flowchart for determining important features among all specific features according to the contribution of each category of features to the first recognition result;

[0024] Figure 3 For this application Figure 1 A schematic diagram of a process for obtaining electrical signal data converted from the radio frequency signal according to the recognition result;

[0025] Figure 4 This is a module diagram of the RFID tag feature extraction and recognition system based on deep learning in Example 2 of this application. DETAILED DESCRIPTION

[0026] In order to solve the problems raised by the background technology, this application obtains RFID electrical signal data, selects specific features to construct input, uses a deep learning model to identify and optimize the features, and then processes the final recognition results to solve the problem that environmental interference and tag quality affect RFID signal transmission, resulting in difficulty in reading or decoding by the reader. It can significantly improve the recognition accuracy of RFID tags, thereby greatly improving the tracking efficiency and reliability of logistics information, helping to optimize inventory management, effectively reducing the risk of cargo loss, and improving the safety and efficiency of overall logistics operations.

[0027] In order to better understand the above technical solution, the above technical solution will be described in detail below with reference to the accompanying drawings and specific implementation methods.

[0028] Example 1:

[0029] like Figure 1-Figure 3 As shown, this embodiment provides an RFID tag feature extraction and recognition method based on deep learning, including:

[0030] S100. Acquire electrical signal data; wherein the electrical signal data is converted from the radio frequency signal returned by the RFID reader to the RFID tag, and the electrical signal data contains multiple categories of features.

[0031] The RFID reader is responsible for sending radio frequency signals to activate the RFID tag and receiving the radio frequency signals returned by the tag. The radio frequency circuit is usually composed of components such as the antenna system, RF transceiver, amplifier, tuner, oscillator, etc. When the RFID reader sends radio frequency signals, these signals are radiated through the antenna system to activate the RFID tag. After receiving the radio frequency signals, the RFID tag will return its stored information in the form of radio frequency signals. After receiving these returned radio frequency signals, the antenna system of the RFID reader passes them to the RF transceiver. The RF transceiver is responsible for converting the received radio frequency signals into electrical signals, which are then further processed and decoded to extract the information in the tag.

[0032] The electrical signal data not only contains the original information of the RFID tag, but also contains multiple categories of features. These features may involve physical properties such as signal strength, frequency, and phase, and may also contain complex information related to the tag location, environmental interference, etc.

[0033] S200. Determine specific features from features of all categories, and construct first input data containing the specific features.

[0034] In certain situations, specific features may contain additional information about the RFID tag or environment, such as environmental parameters, specific tag attributes, timestamps, and location information. Environmental parameters such as temperature and humidity, while not essential for basic RFID tag recognition, can affect RF signal transmission and tag performance in environments such as warehouse management and cold chain logistics, thereby impacting RFID tag recognition accuracy. RFID tags may have specific physical properties, such as size, material, and shape, or electronic characteristics, such as capacitance and inductance, that can affect RF signal reflection and transmission.

[0035] The specific feature includes at least one of the following: an environmental parameter, a specific attribute of the tag, and a timestamp and location information.

[0036] Additionally, the time and location of the tag read may also influence recognition results. For example, in real-time tracking and monitoring systems, the inclusion of timestamps and location information can help the model better understand the tag's movement trajectory and state changes, thereby improving recognition accuracy. While these features may not be essential for recognition in most cases, their inclusion may help improve recognition accuracy in certain specific application scenarios.

[0037] Specific features are selected from all categories of features, and the first input data containing the specific features is constructed for subsequent training of the deep learning model.

[0038] S300. Use the first deep learning model to process the first input data to obtain a first recognition result and the degree of contribution of each category of features to the first recognition result.

[0039] A trained first deep learning model is used to process the first input data containing specific features. The model uses its internal complex neural network structure to extract, combine and classify the input data to obtain recognition results. The feature weights and importance learned by the model during training are used to evaluate the contribution of each category of features to the recognition results. By understanding which features have a significant impact on the recognition results, feature selection can be further optimized to improve recognition accuracy and efficiency.

[0040] S400. Determine important features among all specific features based on the contribution of each category of features to the first recognition result; obtain the recognition result of the RFID reader on the radio frequency signal returned by the RFID tag, and obtain electrical signal data converted from the radio frequency signal based on the recognition result.

[0041] Based on the contribution of each category feature output by the first deep learning model to the first recognition result, important features that have a significant impact on the recognition result can be screened from all specific features. These important features often contain the key information required for the recognition task. Using these important features to train the deep learning model not only reduces the training difficulty, as the model no longer needs to process a large number of redundant and noisy features, but also significantly reduces the amount of computation, improving training speed and recognition efficiency. By training on fewer but more critical features, the model is more able to learn the essential laws and patterns of the data.

[0042] Environmental interference and tag quality issues often affect the transmission of RFID signals, causing the reader to have difficulty reading or decoding the signal returned by the tag. In this case, the RFID reader can obtain the recognition result of the radio frequency signal returned by the RFID tag, and obtain the electrical signal data converted from the radio frequency signal based on the recognition result, so as to subsequently identify and extract important features in specific features.

[0043] S500. Extract important features from the electrical signal data and construct second input data containing the important features.

[0044] These important features are accurately extracted from the electrical signal data converted from the radio frequency signal returned by the RFID reader, and these important features are input into the deep learning model as part of the input data, thereby streamlining the data set, which not only reduces the computational burden of the model but also improves recognition efficiency.

[0045] S600. Use a second deep learning model to process the second input data to obtain a second recognition result.

[0046] By processing the second input data in S500 using a second deep learning model, a second recognition result can be obtained. The recognition result can also output an accuracy indicator to measure the reliability of the current recognition result. The accuracy can be a probability value derived by the model based on internal calculations, indicating the degree of confidence in the recognition result. In addition, an accuracy threshold can be set based on actual application scenarios and needs. When the accuracy of the model's recognition result is higher than the threshold, the system can automatically accept the result. When the accuracy is lower than the threshold, manual review or operation procedures are triggered. Outputting recognition results with accuracy and combining threshold judgment with manual review can significantly improve the automation and accuracy of RFID tag recognition, reduce misjudgments and missed judgments, and at the same time reduce the frequency and cost of manual operations, thereby improving overall work efficiency and user experience.

[0047] Therefore, tags that may have been unable to be identified for various reasons now have a higher probability of being accurately identified by the model. This improvement not only enhances the recognition capability of the RFID system, but also broadens its scope of application in practical applications.

[0048] In S400, determining the important features among all specific features based on the contribution of each category of features to the first recognition result includes:

[0049] S410. Determine relevant features among all specific features according to the contribution degree of each category of features to the first recognition result.

[0050] Based on the contribution of each category feature output by the first deep learning model to the first recognition result, features with a greater correlation with RFID tag recognition are screened out. These relevant features have a significant impact on the accuracy of the recognition result.

[0051] S420. Construct all possible subsets of all types of relevant features; wherein each subset includes at least one relevant feature.

[0052] After determining the relevant features, all possible subsets of these features are constructed. Each subset contains at least one relevant feature, and these subsets cover all possible combinations of relevant features. In subsequent steps, by comparing the performance of different subsets in model training, we can determine which feature combinations are most critical to the recognition results.

[0053] S430. Construct data samples; wherein each data sample contains a subset of a category, and all data samples contain subsets of all categories.

[0054] After determining the relevant features, we construct all possible subsets of these features. Each subset contains at least one relevant feature, and these subsets cover all possible combinations of relevant features. By comparing the performance of different subsets in model training, we can determine which feature combinations are most critical to the recognition results.

[0055] S440. Use each data sample one by one to train the second deep learning model.

[0056] Each data sample is used one by one to train the second deep learning model. Since each model is trained based on a specific subset of features, their performance will directly reflect the effectiveness of that subset in identifying RFID tags.

[0057] S450. Apply the trained second deep learning model to the validation set or the test set, and calculate the performance indicator.

[0058] After training the models, we apply them to the validation set or test set to evaluate their performance. By calculating performance metrics, we can quantify how well the models are at identifying RFID tags.

[0059] S460. Identify important features based on performance indicators.

[0060] By comparing the performance of different subsets in model training, we can determine which feature combinations are most critical to the recognition results and thus mark these features as important features.

[0061] In S410 , determining relevant features from all specific features according to the contribution of each category of features to the first recognition result includes: taking specific features whose contribution is greater than a preset first threshold as relevant features.

[0062] The first threshold represents the minimum standard for a feature to have a significant impact on the recognition result. The contribution of each specific feature to the first recognition result is evaluated one by one and compared with the preset first threshold. If the contribution of a feature is greater than this threshold, it is considered that this feature has a greater relevance to the recognition of the RFID tag and is therefore marked as a relevant feature.

[0063] In S450, performance metrics are calculated using evaluation functions or custom scripts within the machine learning framework. These evaluation tools can calculate key performance indicators such as precision and recall, thereby accurately evaluating the model's recognition capabilities.

[0064] In S460, important features determined based on performance indicators include:

[0065] S461. Filter out a subset whose performance index is greater than a preset second threshold and has the least number of feature types.

[0066] S462. Take the relevant features contained in the filtered subset as important features.

[0067] In steps S461 and S462, the preset second threshold represents an acceptable level of model performance. Then, a subset with performance indicators greater than this threshold and the fewest feature types is selected. These subsets not only achieve high recognition performance but also use the fewest features. Therefore, while maintaining high recognition accuracy, the model's computational workload is reduced, improving its operational efficiency, thereby achieving a dual optimization of model performance and computational efficiency.

[0068] In S400, the recognition result includes success or failure; obtaining the recognition result of the radio frequency signal returned by the RFID reader includes:

[0069] S470: Obtain the initial recognition result of the RFID reader / writer on the radio frequency signal returned by the RFID tag.

[0070] S480. Determine whether the first recognition result is a failure.

[0071] S490. For an RFID tag whose first identification result is a failure, continue to obtain the identification result of the RFID tag by the RFID reader according to a preset time interval or condition until the preset first time period is reached or the identification result is successful.

[0072] The RFID tag recognition process presents two distinct outcomes: success and failure. In real-world applications, RFID readers sometimes fail to immediately and accurately identify tag information due to various temporary interference factors in the environment, such as when a tag is temporarily obscured by other objects. To address these temporary recognition obstacles, when an RFID reader fails to successfully identify a tag for the first time, it does not immediately initiate a complex model for recognition and decoding. Instead, it continuously obtains recognition results for the tag from the RFID reader based on preset time intervals or conditions.

[0073] Acquiring electrical signal data converted from the radio frequency signal according to the recognition result includes: determining an RFID tag that has reached a preset first time period and the recognition result is still a failure; and acquiring electrical signal data converted from the radio frequency signal returned by the RFID tag.

[0074] If the interference is temporary, the RFID reader may successfully identify the tag information after a period of time, effectively addressing the impact of environmental interference on the RFID recognition process without incurring additional computing costs. Only when repeated attempts to identify the tag fail is a more complex model used for further recognition and decoding to ensure accurate and reliable recognition.

[0075] Both the first and second deep learning models use long short-term memory networks. As a specialized recurrent neural network, LSTM excels at capturing long-term dependencies in data. LSTM can also effectively process long sequences of data, enabling it to accurately identify radio frequency signals returned by RFID readers without adding excessive computational burden. This significantly improves the accuracy and reliability of RFID tag recognition.

[0076] Example 2:

[0077] like Figure 4 As shown, this embodiment provides an RFID tag feature extraction and recognition system based on deep learning, which adopts the RFID tag feature extraction and recognition method based on deep learning as in Example 1, and includes: a data acquisition module, a specific feature screening and input data construction module, a feature contribution identification module, a feature contribution evaluation and signal recognition module, an input data construction module and a tag recognition module.

[0078] The data acquisition model is used to acquire electrical signal data; wherein, the electrical signal data is converted from the radio frequency signal returned by the RFID reader to the RFID tag, and the electrical signal data contains multiple categories of features; the specific feature screening and input data construction module is used to determine specific features from all categories of features and construct first input data containing specific features; the feature contribution identification module is used to process the first input data using a first deep learning model to obtain a first recognition result, as well as the degree of contribution of each category of features to the first recognition result; the feature contribution evaluation and signal recognition module is used to determine important features among all specific features based on the degree of contribution of each category of features to the first recognition result; obtain the recognition result of the radio frequency signal returned by the RFID reader to the RFID tag, and obtain the electrical signal data converted from the radio frequency signal based on the recognition result; the input data construction module is used to extract important features from the electrical signal data and construct second input data containing important features; the tag recognition module is used to process the second input data using a second deep learning model to obtain a second recognition result.

[0079] This embodiment has the advantages of the deep learning-based RFID tag feature extraction and recognition method in Example 1, and can automatically implement the steps of the deep learning-based RFID tag feature extraction and recognition method.

[0080] Example 3:

[0081] This embodiment provides an RFID tag feature extraction and recognition device based on deep learning, which includes: a memory and a processor.

[0082] The memory is used to store computer programs; the processor is used to implement the steps of the RFID tag feature extraction and recognition method based on deep learning in Example 1 when executing the computer program.

[0083] Example 4:

[0084] This embodiment provides a readable storage medium, which stores computer program instructions. When the computer program instructions are read and executed by a processor, the steps of the RFID tag feature extraction and recognition method based on deep learning in Example 1 are executed.

[0085] Those skilled in the art will appreciate that embodiments of the present invention may be provided as methods, systems, or computer program products. Thus, the present invention may take the form of an entirely hardware embodiment, an entirely software embodiment, or an embodiment combining software and hardware aspects. Furthermore, the present invention may take the form of a computer program product implemented on one or more computer-usable storage media (including but not limited to magnetic disk storage, CD-ROM, optical storage, etc.) containing computer-usable program code.

[0086] These computer program instructions can also be loaded onto a computer or other programmable data processing device so that a series of operational steps are executed on the computer or other programmable device to produce a computer-implemented process, thereby providing the instructions executed on the computer or other programmable device for implementing the process. Figure 1 a process or multiple processes and / or boxes Figure 1 The present invention is described in detail below. ...

Claims

1. RFID tag feature extraction and recognition method based on deep learning, characterized by: include: Acquire electrical signal data; wherein the electrical signal data is converted from the radio frequency signal returned by the RFID reader to the RFID tag, and the electrical signal data contains multiple categories of features; Determine a specific feature from features of all categories, and construct first input data containing the specific feature; the specific feature includes at least one of the following: environmental parameters, specific attributes of the tag, and timestamp and location information; Processing the first input data using a first deep learning model to obtain a first recognition result and a contribution of each category of features to the first recognition result; The important features are determined among all specific features based on the contribution of each category of features to the first recognition result, including: Determining relevant features among all specific features according to the extent to which each category of features contributes to the first recognition result; Constructing all possible subsets of all types of relevant features, wherein each subset includes at least one relevant feature; Construct data samples; each data sample contains a subset of a category, and all data samples contain subsets of all categories; Using each data sample one by one to train a second deep learning model; Apply the trained second deep learning model to the validation set or test set and calculate performance metrics; Important characteristics identified based on performance metrics include: Filtering out a subset whose performance index is greater than a preset second threshold and whose feature types are the least; The relevant features contained in the selected subset are regarded as important features; Obtaining an identification result of the RFID reader / writer on the radio frequency signal returned by the RFID tag, and obtaining electrical signal data converted from the radio frequency signal according to the identification result; Extracting important features from the electrical signal data and constructing second input data containing the important features; The second input data is processed using a second deep learning model to obtain a second recognition result.

2. The RFID tag feature extraction and recognition method based on deep learning according to claim 1, characterized in that: The relevant features are determined among all specific features based on the contribution of each category of features to the first recognition result, including: The specific features whose contribution degree is greater than a preset first threshold are regarded as relevant features.

3. The RFID tag feature extraction and recognition method based on deep learning according to claim 1, characterized in that: Calculate performance metrics using evaluation functions in machine learning frameworks or custom scripts.

4. The RFID tag feature extraction and recognition method based on deep learning according to claim 1, characterized in that: The recognition result includes success or failure. The recognition result of the RFID reader / writer for the radio frequency signal returned by the RFID tag includes: Obtain the first recognition result of the RFID reader on the radio frequency signal returned by the RFID tag; Determine whether the first recognition result is a failure; For an RFID tag whose first identification result is a failure, the identification result of the RFID tag by the RFID reader is continuously obtained according to a preset time interval or condition until the preset first time period is reached or the identification result is successful.

5. The RFID tag feature extraction and recognition method based on deep learning according to claim 4, characterized in that: Acquiring electrical signal data converted from the radio frequency signal according to the recognition result includes: Determine an RFID tag for which a recognition result is still a failure after reaching a preset first time period; The electrical signal data converted from the radio frequency signal returned by the RFID tag is obtained.

6. The RFID tag feature extraction and recognition method based on deep learning according to claim 1, characterized in that: The first deep learning model and the second deep learning model are both long short-term memory networks.

7. A RFID tag feature extraction and recognition system based on deep learning, characterized in that: include: Data acquisition module: The data acquisition module is used to acquire electrical signal data; wherein, the electrical signal data is converted from the radio frequency signal returned by the RFID reader to the RFID tag, and the electrical signal data contains multiple types of features; Specific feature screening and input data construction module: The specific feature screening and input data construction module is used to determine specific features from all categories of features and construct first input data containing the specific features; the specific features include at least one of the following: environmental parameters, specific attributes of the tag, and timestamp and location information; Feature contribution identification module: The feature contribution identification module is used to process the first input data using a first deep learning model to obtain a first recognition result and the contribution of each category of features to the first recognition result; Feature contribution evaluation and signal identification module: The feature contribution evaluation and signal identification module is used to determine important features among all specific features according to the contribution degree of each category of features to the first identification result; determining important features among all specific features according to the contribution degree of each category of features to the first identification result includes: determining relevant features among all specific features according to the contribution degree of each category of features to the first identification result; constructing all possible subsets of relevant features of all categories; wherein each subset includes at least one relevant feature; constructing data samples; wherein each data sample contains a subset of one category, and all data samples contain subsets of all categories; using each data sample one by one to train the second deep learning model; applying the trained second deep learning model to the validation set or the test set, and calculating the performance index; determining important features according to the performance index includes: screening out a subset whose performance index is greater than a preset second threshold and has the least number of feature types; taking the relevant features contained in the screened subset as important features; obtaining the recognition result of the radio frequency signal returned by the RFID reader / writer to the RFID tag, and obtaining the electrical signal data converted from the radio frequency signal according to the recognition result; Input data construction module: the input data construction module is used to extract important features from the electrical signal data and construct second input data containing the important features; Label recognition module: The label recognition module is used to process the second input data using a second deep learning model to obtain a second recognition result.

Images