Open set radio frequency fingerprint identification method and device, storage medium and electronic equipment

Abstract

This invention provides an open-set radio frequency fingerprinting method, apparatus, storage medium, and electronic device. The method includes: calculating the class center distance of a target radio frequency fingerprint feature under each monitored object category; calculating the tail probability of the target radio frequency fingerprint feature under each monitored object category based on the class center distance and the cumulative distribution of the target Cauchy probability under each monitored object category; determining the target class probability of the target radio frequency fingerprint feature under each monitored object category using the tail probability; and determining the radio frequency fingerprinting result of the target radio frequency fingerprint feature based on the target class probability under each monitored object category. A single radio frequency fingerprinting result can be used to indicate whether the target belongs to an existing monitored object category. This invention improves the accuracy of radio frequency fingerprinting.

Description

Technical Field

[0001] This invention relates to the field of radio frequency fingerprint recognition technology, and in particular to an open-collector radio frequency fingerprint recognition method, apparatus, storage medium, and electronic device. Background Technology

[0002] Currently, Radio Frequency Fingerprint Identification (RFF) technology, as a physical layer security measure, identifies radiation sources by extracting subtle and unique physical features introduced by transmitter hardware (such as digital-to-analog converters, mixers, and power amplifiers), effectively compensating for the lack of protocol-layer authentication. However, related technologies are typically based on the closed-set recognition assumption, which assumes that all signal categories appearing during the testing phase have appeared during the training phase. This assumption aims to achieve high recognition accuracy in such closed-set scenarios. However, when encountering unknown signal categories not included in the training, the model may classify them as known categories, causing identification failure and resulting in low accuracy of RRF fingerprinting. Therefore, there is currently no satisfactory solution to improve the accuracy of RRF fingerprinting. Summary of the Invention

[0003] In view of this, embodiments of the present invention provide an open-set radio frequency fingerprint recognition method, apparatus, storage medium, and electronic device to solve the problems of low accuracy in radio frequency fingerprint recognition in related technologies. In other words, embodiments of the present invention can determine the target category probability with high accuracy by using the category center distance of the target radio frequency fingerprint features under each monitoring object category and the cumulative distribution of the target Cauchy probability under each monitoring object category, so as to obtain a radio frequency fingerprint recognition result with high accuracy. This can effectively improve the accuracy of radio frequency fingerprint recognition, and further improve the ability to perceive unknown categories when encountering unknown monitoring object categories, thereby effectively avoiding failure of identity signal recognition.

[0004] According to one aspect of the present invention, an open-set radio frequency fingerprint recognition method is provided, the method comprising:

[0005] The identity signal of the target monitoring object is acquired, and the target monitoring object identity signal is subjected to feature extraction to obtain the target radio frequency fingerprint feature of the target monitoring object identity signal;

[0006] Determine the category center of each monitoring object category in the set of monitoring object categories, and calculate the category center distance of the target radio frequency fingerprint feature under each monitoring object category based on the category center of each monitoring object category;

[0007] Based on the class center distance and the cumulative distribution of the target Cauchy probability of the target RF fingerprint feature in each monitoring object category, respectively, the tail probability of the target RF fingerprint feature in each monitoring object category is calculated; wherein, the tail probability of the target RF fingerprint feature in a monitoring object category is calculated by substituting the class center distance of the target RF fingerprint feature in the corresponding monitoring object category into the cumulative distribution of the target Cauchy probability of the corresponding monitoring object category;

[0008] The target category probability of the target RF fingerprint feature under each monitoring object category is determined by using the tail probability of the target RF fingerprint feature under each monitoring object category; and the RF fingerprint recognition result of the target RF fingerprint feature is determined based on the target category probability of the target RF fingerprint feature under each monitoring object category. One RF fingerprint recognition result can be used to indicate whether it belongs to an existing monitoring object category.

[0009] According to another aspect of the present invention, an open-collector radio frequency fingerprint recognition device is provided, the device comprising:

[0010] The acquisition unit is used to acquire the identity signal of the target monitored object;

[0011] The processing unit is used to extract features from the identity signal of the target monitoring object to obtain the target radio frequency fingerprint feature of the identity signal of the target monitoring object;

[0012] The processing unit is further configured to determine the category center of each monitoring object category in the set of monitoring object categories, and calculate the category center distance of the target radio frequency fingerprint feature under each monitoring object category based on the category center of each monitoring object category.

[0013] The processing unit is further configured to calculate the tail probability of the target radio frequency fingerprint feature in each monitoring object category based on the category center distance of the target radio frequency fingerprint feature in each monitoring object category and the target Cauchy probability cumulative distribution in each monitoring object category; wherein, the tail probability of the target radio frequency fingerprint feature in a monitoring object category is calculated by substituting the category center distance of the target radio frequency fingerprint feature in the corresponding monitoring object category into the target Cauchy probability cumulative distribution in the corresponding monitoring object category;

[0014] The processing unit is further configured to determine the target category probability of the target radio frequency fingerprint feature under each monitoring object category by using the tail probability of the target radio frequency fingerprint feature under each monitoring object category; and to determine the radio frequency fingerprint recognition result of the target radio frequency fingerprint feature based on the target category probability of the target radio frequency fingerprint feature under each monitoring object category, wherein a radio frequency fingerprint recognition result can be used to indicate whether it is an existing monitoring object category.

[0015] According to another aspect of the present invention, an electronic device is provided, the electronic device including a processor and a memory storing a program, wherein the program includes instructions that, when executed by the processor, cause the processor to perform the methods mentioned above.

[0016] According to another aspect of the present invention, a non-transitory computer-readable storage medium is provided storing computer instructions for causing a computer to perform the methods mentioned above.

[0017] This invention can acquire the identity signal of a target monitored object and extract features from it to obtain the target radio frequency fingerprint feature. Then, it can determine the category center of each monitored object category in the set of monitored object categories, and calculate the category center distance of the target radio frequency fingerprint feature in each monitored object category based on the category center of each monitored object category. Based on this, it can calculate the tail probability of the target radio frequency fingerprint feature in each monitored object category based on the category center distance and the cumulative distribution of the target Cauchy probability in each monitored object category, to correct the category probability. Correspondingly, it can use the tail probability of the target radio frequency fingerprint feature in each monitored object category to determine the target category probability of the target radio frequency fingerprint feature in each monitored object category, effectively improving the accuracy of the target category probability. Based on the target category probability of the target radio frequency fingerprint feature in each monitored object category, it can determine the radio frequency fingerprint recognition result of the target radio frequency fingerprint feature. One radio frequency fingerprint recognition result can be used to indicate whether it belongs to an existing monitored object category. As can be seen, the embodiments of the present invention can determine the target category probability with high accuracy by using the category center distance of the target radio frequency fingerprint features under each monitoring object category and the cumulative distribution of the target Cauchy probability under each monitoring object category, so as to obtain a radio frequency fingerprint recognition result with high accuracy. This can effectively improve the accuracy of radio frequency fingerprint recognition, and further improve the ability to perceive unknown categories when encountering unknown monitoring object categories, thereby effectively avoiding the failure of identity signal recognition. Attached Figure Description

[0018] Further details, features, and advantages of the invention are disclosed in the following description of exemplary embodiments in conjunction with the accompanying drawings, in which:

[0019] Figure 1 A flowchart illustrating an open-set radio frequency fingerprinting method according to an exemplary embodiment of the present invention is shown;

[0020] Figure 2 A flowchart illustrating another open-set radio frequency fingerprinting method according to an exemplary embodiment of the present invention is shown;

[0021] Figure 3 A schematic diagram illustrating the output of the SoftMax function under different temperature parameters according to an exemplary embodiment of the present invention is shown;

[0022] Figure 4 A schematic diagram of an identification accuracy correction result according to an exemplary embodiment of the present invention is shown;

[0023] Figure 5 A flowchart illustrating another open-set radio frequency fingerprinting method according to an exemplary embodiment of the present invention is shown;

[0024] Figure 6 A flowchart illustrating another open-set radio frequency fingerprinting method according to an exemplary embodiment of the present invention is shown;

[0025] Figure 7 A schematic diagram illustrating a novel category identification method according to an exemplary embodiment of the present invention is shown;

[0026] Figure 8 A schematic block diagram of an open-collection radio frequency fingerprint recognition device according to an exemplary embodiment of the present invention is shown;

[0027] Figure 9 A structural block diagram of an exemplary electronic device that can be used to implement embodiments of the present invention is shown. Detailed Implementation

[0028] Embodiments of the present invention will now be described in more detail with reference to the accompanying drawings. While some embodiments of the invention are shown in the drawings, it should be understood that the invention can be implemented in various forms and should not be construed as limited to the embodiments set forth herein. Rather, these embodiments are provided to provide a more thorough and complete understanding of the invention. It should be understood that the accompanying drawings and embodiments are for illustrative purposes only and are not intended to limit the scope of protection of the invention.

[0029] It should be understood that the various steps described in the method embodiments of the present invention may be performed in different orders and / or in parallel. Furthermore, the method embodiments may include additional steps and / or omit the steps shown. The scope of the present invention is not limited in this respect.

[0030] The term "comprising" and its variations as used herein are open-ended, meaning "including but not limited to". The term "based on" means "at least partially based on". The term "one embodiment" means "at least one embodiment"; the term "another embodiment" means "at least one additional embodiment"; the term "some embodiments" means "at least some embodiments". Definitions of other terms will be given in the following description. It should be noted that the concepts of "first", "second", etc., mentioned in this invention are used only to distinguish different devices, modules, or units, and are not intended to limit the order of functions performed by these devices, modules, or units or their interdependencies.

[0031] It should be noted that the terms "a" and "a plurality of" used in this invention are illustrative rather than restrictive. Those skilled in the art should understand that, unless otherwise expressly indicated in the context, they should be understood as "one or more".

[0032] The names of the messages or information exchanged between the multiple devices in the embodiments of the present invention are for illustrative purposes only and are not intended to limit the scope of these messages or information.

[0033] It should be noted that the execution subject of the open-set radio frequency fingerprinting method provided in this embodiment of the invention can be one or more electronic devices, and this invention does not limit this; wherein, the electronic device can be a terminal (i.e., a client) or a server. Therefore, when the execution subject includes multiple electronic devices, and among the multiple electronic devices includes at least one terminal and at least one server, the open-set radio frequency fingerprinting method provided in this embodiment of the invention can be jointly executed by the terminal and the server. Accordingly, the terminal mentioned herein may include, but is not limited to: smartphones, tablets, laptops, desktop computers, smartwatches, smart voice interaction devices, smart home appliances, vehicle terminals, aircraft, etc. The server mentioned herein can be an independent physical server, a server cluster or distributed system composed of multiple physical servers, or a cloud server providing basic cloud computing services such as cloud services, cloud databases, cloud computing, cloud functions, cloud storage, network services, cloud communication, middleware services, domain name services, security services, CDN (Content Delivery Network), and big data and artificial intelligence platforms, etc.

[0034] Optionally, the open-set radio frequency fingerprinting method provided in this embodiment of the invention can be applied to any radio frequency fingerprinting scenario, and this embodiment of the invention does not limit it. For example, it can be applied to aircraft radio frequency fingerprinting scenarios (i.e., scenarios where aircraft are identified through radio frequency fingerprinting), where the monitored object can be an aircraft, such as the target monitored object's identity signal being the target aircraft's identity signal, and the monitored object's category being the aircraft category, etc.; or, it can also be applied to Bluetooth device radio frequency fingerprinting scenarios (i.e., scenarios where Bluetooth devices are identified through radio frequency fingerprinting), where the monitored object can be a Bluetooth device, such as the target monitored object's identity signal being the target Bluetooth device's identity signal, and the monitored object's category being the Bluetooth device category, etc.; this embodiment of the invention does not limit it. In other words, the monitored object can be any object, such as an aircraft, a Bluetooth device, or a vehicle networking device, etc.

[0035] Based on the above description, this embodiment of the invention proposes an open-collection radio frequency fingerprinting method, which can be executed by the aforementioned electronic device (terminal or server); or, the open-collection radio frequency fingerprinting method can be executed jointly by the terminal and the server. For ease of explanation, the following description will use the execution of the open-collection radio frequency fingerprinting method by an electronic device as an example; such as Figure 1 As shown, the open-collection radio frequency fingerprint recognition method may include the following steps S101-S104:

[0036] S101, acquire the identity signal of the target monitoring object, and extract features from the identity signal of the target monitoring object to obtain the target radio frequency fingerprint feature of the identity signal of the target monitoring object.

[0037] The target monitoring object's identity signal can be any monitoring object, and this embodiment of the invention does not limit this; for example, the target monitoring object can be any aircraft. For ease of explanation, subsequent application descriptions will use an aircraft as an example for illustrative application.

[0038] Optionally, the methods for obtaining the identity signal of the target monitoring object may include, but are not limited to, the following:

[0039] The first acquisition method: The electronic device stores at least one identity signal of the object to be identified in its own storage space. In this case, the electronic device can sequentially determine one identity signal of the object to be identified from the at least one identity signal of the object to be identified, and use the currently determined identity signal of the object to be identified as the identity signal of the target object to achieve the acquisition of the identity signal of the target object.

[0040] The second method of acquisition: Electronic devices can obtain the target monitoring object's identity signal download link, and use the target monitoring object's identity signal download link to download the target monitoring object's identity signal, thereby achieving the acquisition of the target monitoring object's identity signal.

[0041] The third acquisition method: The radio frequency (RF) front-end can collect data to obtain analog signals. The electronic device can then acquire the analog signals from the RF front-end and perform signal processing to obtain demodulated data. This demodulated data may include, but is not limited to, at least one of the following: the target monitoring object's identity signal, the target monitoring object's message information, etc. This embodiment of the invention does not limit this. Optionally, signal processing may include, but is not limited to, at least one of the following: noise reduction processing, filtering processing, analog-to-digital conversion, and message demodulation (also known as data demodulation), etc.; this embodiment of the invention does not limit this. Based on this, the electronic device can obtain the target monitoring object's identity signal from the demodulated data, such as... Figure 2 As shown. Optionally, the radio frequency front-end may be located on the electronic device performing the open-set radio frequency fingerprinting method, or on other electronic devices; this embodiment of the invention does not limit this. Optionally, the message information of the target monitoring object may include, but is not limited to, at least one of the following: the monitoring object identifier of the target monitoring object (such as the monitoring object name or monitoring object number, etc., one monitoring object identifier can be used to indicate one monitoring object), location information, etc.; this embodiment of the invention does not limit this.

[0042] Optionally, the identity signal of a monitored object can be a digital signal. It should be noted that the specific representation of the identity signal of the monitored object is not limited in the embodiments of the present invention. For example, it can be represented in the time domain (such as discrete values ​​arranged in the order of sampling time), or in the frequency domain (such as a spectrum graph), etc.

[0043] In this embodiment of the invention, when extracting features from the identity signal of the target monitored object to obtain the target radio frequency fingerprint feature of the target monitored object's identity signal, the electronic device can invoke a target recognition model to extract features from the identity signal of the target monitored object to obtain the target radio frequency fingerprint feature of the target monitored object's identity signal. Optionally, a recognition model (such as a target recognition model) may include a neural network model, a convolutional neural network model, a generative adversarial network model, etc.; in other words, this embodiment of the invention does not limit the specific model structure of the recognition model. Wherein, when the recognition model includes a generative adversarial network model, it can be based on the simulation of abnormal samples of the generative adversarial network (such as generating abnormal radio frequency fingerprint features, etc.) to actively generate pseudo-unknown samples at the boundary to participate in training, thereby outputting the true and false probabilities, and thus improving the model's early warning capability for open set category signals.

[0044] Optionally, when calling the identification model to extract features from the identity signal of a monitored object, the corresponding identity signal can be directly input into the network included in the identification model, or the corresponding identity signal can be converted into data and then input into the network included in the identification model. This embodiment of the invention does not limit this approach. For example, when the identity signal of the monitored object is represented in the time domain, it can be converted into a spectrum, etc., for feature extraction; or, feature extraction can be performed directly, etc. In other words, when calling the target identification model to extract features from the identity signal of the target monitored object and obtain the target radio frequency fingerprint feature of the target monitored object, the target monitored object identity signal can be used to determine the input data of the model network. This input data can be input into the network included in the target identification model, thereby extracting features from the target monitored object's identity signal and obtaining the target radio frequency fingerprint feature of the target monitored object's identity signal by performing feature extraction on the model input data. Based on this, data conforming to the data input format of the network in the identification model can be input into the network included in the identification model for feature extraction. Optionally, a radio frequency fingerprint feature can also be called an activation vector, etc.; this embodiment of the invention does not limit this approach.

[0045] S102, determine the category center of each monitoring object category in the monitoring object category set, and calculate the distance between the target radio frequency fingerprint feature and the category center of each monitoring object category based on the category center of each monitoring object category.

[0046] Optionally, in embodiments of the present invention, the category center may also be referred to as the feature category center.

[0047] Optionally, the set of monitored object categories may include any monitored object category, and this embodiment of the invention does not limit this. Optionally, a monitored object category can be used to indicate a monitored object, or it can be used to indicate a monitored object model, etc., and this embodiment of the invention does not limit this. For example, taking an aircraft as an example, the set of monitored object categories may include the categories of each aircraft among all the aircraft to be monitored; for example, an aircraft category may be represented by an aircraft identifier (such as an aircraft name or aircraft number, etc.), and an aircraft identifier can be used to indicate an aircraft, then in this case, an aircraft category can be used to indicate an aircraft, and so on.

[0048] In this embodiment of the invention, the category center of a monitored object category can be calculated using the various trained radio frequency fingerprint features under the corresponding monitored object category. For example, for any monitored object category in the set of monitored object categories, after obtaining the various trained radio frequency fingerprint features under any monitored object category, the mean among the various trained radio frequency fingerprint features under any monitored object category can be calculated, and the mean among the various trained radio frequency fingerprint features under any monitored object category can be used as the category center of any monitored object category. For example, the electronic device can use Formula 1.1 to determine the category center of any monitored object category:

[0049] Formula 1.1

[0050] Among them, Ms C N can represent the category center of any monitored object category. C f(x) can represent the number of training monitoring object identity signals under any monitoring object category in the training monitoring object identity signal set (i.e., the number of training monitoring object identity signals belonging to any monitoring object category, also known as the number of training radio frequency fingerprint features under any monitoring object category; a training radio frequency fingerprint feature under any monitoring object category is a training radio frequency fingerprint feature of a training monitoring object identity signal under any monitoring object category), Cn ) can represent the nth training radio frequency fingerprint feature under any monitoring object category, x Cn It can represent the identity signal of the nth training monitoring object under any monitoring object category.

[0051] Optionally, in other embodiments, the category centers of each monitored object category can also be determined based on deep metric learning. For example, deep metric learning can be performed on the training set of monitored object identity signals using a center loss function or a triplet loss function to optimize the clustering effect in the feature space, thereby making similar samples close together in space and dissimilar samples far apart, thus obtaining the category centers of each monitored object category. For instance, the category centers of each monitored object category can be continuously optimized during the deep metric learning process until the loss value (such as center loss or triplet loss) meets a preset clustering loss threshold, thereby obtaining the final category centers to determine the category centers of each monitored object category, and so on. Optionally, the preset clustering loss threshold can be set based on experience or actual needs; this embodiment of the invention does not limit this.

[0052] Therefore, when the electronic device stores the category centers of each monitored object category in its own storage space, the electronic device can directly determine the category centers of each monitored object category from its own storage space; alternatively, the electronic device can determine the category centers of each monitored object category by training a set of monitored object identity signals, and so on. For example, in practical applications, the electronic device can directly determine the category centers of each monitored object category.

[0053] Optionally, when calculating the distance between the target RFID fingerprint feature and the category center of each monitoring object category based on the category center of each monitoring object category, the electronic device can traverse each monitoring object category in the monitoring object category set and take the currently traversed monitoring object category as the current monitoring object category. Correspondingly, based on the category center of the current monitoring object category and the target RFID fingerprint feature, the current category distance of the target RFID fingerprint feature under each of the multiple distance calculation methods can be calculated. The current category distance of the target RFID fingerprint feature under a distance calculation method refers to the distance between the target RFID fingerprint feature calculated according to the corresponding distance calculation method and the category center of the current monitoring object category. Further, a target weighted reassembly under the current monitoring object category can be determined, and the current category distance of the target RFID fingerprint feature under each distance calculation method can be weighted and summed according to the target weighted reassembly to obtain the category center distance of the target RFID fingerprint feature under the current monitoring object category. Here, the weighted reassembly under a monitoring object category includes the weight values ​​of each distance calculation method under the corresponding monitoring object category. After traversing each monitoring object category in the monitoring object category set, the category center distance of the target RFID fingerprint feature under each monitoring object category is obtained. Optionally, the weighted arrays under different detection object categories can be the same or different, and this embodiment of the invention does not limit this. Optionally, the multiple distance calculation methods may include, but are not limited to, at least two of the following: Euclidean distance calculation method (which can be used to calculate Euclidean distance), Chebyshev distance calculation method (which can be used to calculate Chebyshev distance), and Manhattan distance calculation method (which can be used to calculate Manhattan distance), etc., and this embodiment of the invention does not limit this.

[0054] Optionally, when the electronic device stores target weight reassemblies for each monitored object category in its own storage space, the electronic device can determine the target weight reassemblies for each monitored object category directly from its own storage space; alternatively, the electronic device can determine the target weight reassemblies for each monitored object category by training a set of monitored object identity signals; or, if the category centers of each monitored object category are determined based on deep metric learning, the target weight reassemblies for each monitored object category can also be set according to experience or actual needs, and so on. Specific implementation methods for determining target weight reassemblies for each monitored object category by training a set of monitored object identity signals are shown below, and will not be elaborated upon here. For example, in practical applications, the target weight reassemblies for the current monitored object category can be directly determined.

[0055] Optionally, in other embodiments, if the category center of each monitored object category is determined based on deep metric learning, the number of distance calculation methods can also be one, etc.; the present invention does not limit this.

[0056] S103, calculate the tail probability of the target radio frequency fingerprint feature in each monitoring object category based on the category center distance of the target radio frequency fingerprint feature in each monitoring object category and the cumulative distribution of the target Cauchy probability in each monitoring object category.

[0057] Optionally, for any monitoring object category in the set of monitoring object categories, the electronic device can substitute the category center distance of the target RFID fingerprint feature under any monitoring object category into the cumulative distribution of the target Cauchy probability under any monitoring object category, thereby obtaining the tail probability of the target RFID fingerprint feature under any monitoring object category; that is, the tail probability of the target RFID fingerprint feature under a monitoring object category is calculated by substituting the category center distance of the target RFID fingerprint feature under the corresponding monitoring object category into the cumulative distribution of the target Cauchy probability under the corresponding monitoring object category. For example, the cumulative distribution of the target Cauchy probability under any monitoring object category can be shown in Equation 1.2:

[0058] Equation 1.2

[0059] Where F(x,x0,γ) represents the cumulative Cauchy probability distribution of a target under any monitoring object category, used to represent the sample distribution probability under any monitoring object category, such as the signal distribution probability or the RF fingerprint feature distribution probability; x can be substituted with the class center distance of an RF fingerprint feature under any monitoring object category (i.e., x can be an input variable), x0 can represent the location parameter, and γ can represent the scale parameter. Optionally, the location parameter and scale parameter can be obtained by fitting the Cauchy distribution to the class center distances of each of the multiple selected RF fingerprint features. The class center distance of a selected RF fingerprint feature can refer to the class center distance of the corresponding selected RF fingerprint feature under any monitoring object category. Based on this, the class center distance of the target RF fingerprint feature under any monitoring object category can be used as the input variable of the cumulative Cauchy probability distribution of the target under any monitoring object category to calculate the tail probability of the target RF fingerprint feature under any monitoring object category.

[0060] S104, using the tail probability of the target radio frequency fingerprint feature under each monitoring object category, determine the target category probability of the target radio frequency fingerprint feature under each monitoring object category; and based on the target category probability of the target radio frequency fingerprint feature under each monitoring object category, determine the radio frequency fingerprint recognition result of the target radio frequency fingerprint feature. One radio frequency fingerprint recognition result can be used to indicate whether it belongs to an existing monitoring object category.

[0061] Optionally, the RF fingerprint recognition result of an RF fingerprint feature can be used to indicate whether the corresponding RF fingerprint feature belongs to an existing monitored object category (i.e., whether it belongs to an existing monitored object category). In other words, it can be used to indicate whether it is a known category in the set of monitored object categories. Correspondingly, the RF fingerprint recognition result of an RF fingerprint feature can be used to indicate whether the monitored object identity signal indicated by the corresponding RF fingerprint feature belongs to an existing monitored object category. The monitored object identity signal indicated by an RF fingerprint feature can also be called the monitored object identity signal corresponding to an RF fingerprint feature. An RF fingerprint feature can correspond to the monitored object identity signal used to extract the corresponding RF fingerprint feature, and so on.

[0062] It should be noted that the specific form of the radio frequency fingerprint recognition result is not limited in the embodiments of the present invention. For example, the radio frequency fingerprint recognition result of a radio frequency fingerprint feature may include the target category probability of the corresponding radio frequency fingerprint feature under various monitoring object categories, or it may include the category identifier corresponding to the corresponding radio frequency fingerprint feature (such as the category identifier of the category indicated by the maximum target category probability of the corresponding radio frequency fingerprint feature under various monitoring object categories, or the category identifier of an unknown category, etc.). Optionally, a category identifier may be a category name or a category number, etc., and the embodiments of the present invention do not limit this. For example, when a radio frequency fingerprint recognition result indicates an existing monitoring object category, the corresponding radio frequency fingerprint recognition result may include the category identifier of the indicated existing monitoring object category; when a radio frequency fingerprint recognition result indicates an unknown category, the corresponding radio frequency fingerprint recognition result may include the category identifier of the unknown category, etc. Optionally, the category identifier of an unknown category may be set according to experience or actual needs, and the embodiments of the present invention do not limit this. Based on this, when determining the RF fingerprint recognition result of the target RF fingerprint feature based on the target category probability of the target RF fingerprint feature under each monitoring object category, the target category probability of the target RF fingerprint feature under each monitoring object category can be added to the RF fingerprint recognition result of the target RF fingerprint feature; or, the maximum target category probability can be determined from the target category probability of the target RF fingerprint feature under each monitoring object category, and the category identifier of the monitoring object category indicated by the maximum target category probability and / or the maximum target category probability can be added to the RF fingerprint recognition result of the target RF fingerprint feature; or, it can be determined whether the maximum target category probability is greater than a preset target category probability threshold. If the maximum target category probability is greater than the preset target category probability threshold, the above-mentioned addition of the category identifier of the monitoring object category indicated by the maximum target category probability and / or the maximum target category probability to the RF fingerprint recognition result of the target RF fingerprint feature can be triggered. At this time, it can be determined that the identity signal of the target monitoring object belongs to an existing monitoring object category. If the maximum target category probability is less than or equal to the preset target category probability threshold, the category identifier of the unknown category can be added to the RF fingerprint recognition result of the target RF fingerprint feature, and so on. Optionally, the preset target category probability threshold can be set according to experience or actual needs, and the embodiments of the present invention do not limit this.

[0063] Optionally, when determining the target category probability of the target RF fingerprint feature in each monitoring object category using the tail probability of the target RF fingerprint feature in each monitoring object category, the electronic device can use the tail probability of the target RF fingerprint feature in each monitoring object category to determine the corrected category probability of the target RF fingerprint feature in each monitoring object category; and can call a target normalization function to convert the corrected category probability of the target RF fingerprint feature in each monitoring object category into the target category probability of the target RF fingerprint feature in each monitoring object category. Optionally, the target normalization function can be set according to experience or actual needs, or it can be obtained by debugging based on a set of test monitoring object identity signals; this embodiment of the invention does not limit this. Optionally, a normalization function can be used to convert the corrected category probability of an RF fingerprint feature in each monitoring object category into a target category probability whose sum of probabilities is 1; for example, the sum of the target category probabilities of the target RF fingerprint feature in each monitoring object category can be 1, etc.

[0064] Optionally, when determining the corrected category probability of the target RF fingerprint feature in each monitoring object category using the tail probability of the target RF fingerprint feature in each monitoring object category, for any monitoring object category in the monitoring object category set, the electronic device can use the difference between the preset probability value and the tail probability of the target RF fingerprint feature in any monitoring object category as the corrected category probability of the target RF fingerprint feature in any monitoring object category, so as to obtain a new category probability. Optionally, the preset probability value can be set according to experience or actual needs, and the embodiments of the present invention do not limit it; for example, the preset probability value can be 1, and the corrected category probability of the target RF fingerprint feature in any monitoring object category can be: 1 - the tail probability of the target RF fingerprint feature in any monitoring object category, such as 1 - F(x,x0,γ). Based on this, the tail probability of an RFID fingerprint feature under a monitoring object category can be used to represent the probability that the corresponding RFID fingerprint feature does not belong to the corresponding monitoring object category, that is, the probability that the monitoring object identity signal indicated by the corresponding RFID fingerprint feature does not belong to the corresponding monitoring object category; correspondingly, the modified category probability of an RFID fingerprint feature under a monitoring object category can be used to represent the probability that the corresponding RFID fingerprint feature belongs to the corresponding monitoring object category, that is, the probability that the monitoring object identity information indicated by the corresponding RFID fingerprint feature belongs to the corresponding monitoring object category.

[0065] It should be understood that, in classification tasks, the SoftMax function (an exponential normalization function) is used as an example for illustration. The SoftMax function can be used to transform activation vectors into probability distributions P, as shown in Equation 1.3:

[0066] Equation 1.3

[0067] Among them, P c z can represent the class probability of any monitored object category (which can also be referred to as the c-th monitored object category in the set of monitored object categories). c z can represent the network output score corresponding to any monitored object category (such as the modified category probability mentioned above), j can represent the j-th monitored object category in the set of monitored object categories, j ranges from 1 to N (N represents the total number of monitored object categories in the set of monitored object categories, so that the denominator is summed exponentially over all output scores of the recognition model), z j Let represent the network output score corresponding to the j-th monitored object category, and e can be an exponential constant. It should be understood that, due to the rapid growth characteristic of exponential functions, when the input value z... c When z is relatively large compared to other input values, c The index will grow exponentially, leading to a corresponding P... c The probability of one class approaches 1, while the probability of other classes approaches 0; this exponential amplification effect means that even if the differences between input values ​​are small, the final probability distribution will exhibit extreme bias.

[0068] Optionally, to adjust the smoothness of the SoftMax function output, thereby reducing overconfidence and improving the model's calibration and generalization ability, embodiments of the present invention may introduce a temperature parameter. In this case, a normalization function may include the temperature parameter. For example, the SoftMax function after introducing the temperature parameter (i.e., the normalization function after introducing the temperature parameter) can be as shown in Equation 1.4:

[0069] Equation 1.4

[0070] Where T can represent a temperature parameter; for example, such as Figure 3 As shown, Figure 3 Subplots (a)-(d) in the figure show the output of the SoftMax function when the temperature parameter is 0.5, 1, 2, and 5, respectively. When T is greater than 1, the separation degree can be reduced, making the probability distribution more uniform, thus effectively avoiding the situation where the probability of a certain class is particularly high, and effectively reducing overconfidence. Correspondingly, when T approaches ∞, the probability can be observed to approach 1 / N, which means that the probability of all classes can be equal, and the model cannot make an effective classification. For example, when T is changed from T=2 to T=5, the uniformity of the probability distribution increases. When T is less than 1, the separation degree can be improved, making the maximum confidence of the model higher, which exacerbates the overconfidence problem.

[0071] Correspondingly, the target category probability of the target RF fingerprint feature under each monitored object category can be determined by calling a target normalization function. Optionally, the target normalization function may include a temperature parameter; optionally, the temperature parameter in the target normalization function may be set according to experience or actual needs, or it may be obtained through training, etc.

[0072] Optionally, the electronic device may also acquire a set of test monitoring object identity signals. The set of test monitoring object identity signals may include an open set of test categories and a closed set of test categories. The test monitoring object identity signals in the open set of test categories belong to unknown categories, and the test monitoring object identity signals in the closed set of test categories belong to known monitoring object categories. Optionally, the known monitoring object category may refer to the monitoring object category in the set of monitoring object categories (i.e., the monitoring object category belonging to the set of monitoring object categories), and the unknown category may refer to the category not included in the set of monitoring object categories (i.e., the category not belonging to the set of monitoring object categories). Optionally, the electronic device may store a set of test monitoring object identity signals in its own storage space, and the set of test monitoring object identity signals can be obtained from its own storage space; or, the database may include multiple test monitoring object identity signals of unknown categories and multiple test monitoring object identity signals of known categories, then the electronic device can filter out the open set and closed set of test categories from the database to obtain the set of test monitoring object identity signals; or, the open set of test categories can be filtered out from the database, and the open set of test categories and the training set of monitoring object identity signals can be added to the set of test monitoring object identity signals to obtain the set of test monitoring object identity signals, in which case the closed set of test categories can be the training set of monitoring object identity signals, and so on; the embodiments of the present invention do not limit the method of obtaining the set of test monitoring object identity signals.

[0073] Furthermore, the electronic device can extract features from each test monitoring object identity signal in the set of test monitoring object identity signals to obtain the test radio frequency fingerprint features of each test monitoring object identity signal. Then, it can initialize the current normalization function and determine the maximum confidence level of each test monitoring object identity signal based on the test radio frequency fingerprint features of each test monitoring object identity signal and the current normalization function; wherein, a normalization function includes a temperature parameter. Optionally, when determining the maximum confidence level of each test monitoring object identity signal based on the test radio frequency fingerprint features of each test monitoring object identity signal and the current normalization function, for any test monitoring object identity signal in the set of test monitoring object identity signals, the electronic device can determine the corrected class probability of the test radio frequency fingerprint features of any test monitoring object identity signal under each monitoring object category based on the test radio frequency fingerprint features of any test monitoring object identity signal and the category center of each monitoring object category, so as to call the current normalization function to transform the corrected class probability of the test radio frequency fingerprint features of any test monitoring object identity signal under each monitoring object category into the corrected class probability of the test radio frequency fingerprint features of any test monitoring object identity signal under each monitoring object category. The maximum confidence level of any test monitoring object identity signal is determined by taking the maximum current target category probability among the current target category probabilities of the test RF fingerprint features of the test monitoring object identity signal under each monitoring object category. It should be noted that the specific implementation method for determining the current target category probability of the test RF fingerprint features of any test monitoring object identity signal under each monitoring object category based on the test RF fingerprint features and the current normalization function is the same as the specific implementation method for determining the target category probability of the target RF fingerprint features under each monitoring object category based on the target RF fingerprint features and the target normalization function of the target monitoring object identity signal. Therefore, the maximum confidence level of any test monitoring object identity signal can be the maximum current target category probability of the test RF fingerprint features of any test monitoring object identity signal under each monitoring object category. Optionally, the maximum confidence level of a test monitoring object identity signal can also be called the maximum confidence level of the test RF fingerprint features of the corresponding test monitoring object identity signal, and so on.

[0074] Accordingly, the electronic device can determine the maximum confidence statistical results of the open set and the closed set of the test categories based on the maximum confidence of the identity signals of each test monitoring object; and based on the maximum confidence statistical results of the open set and the closed set of the test categories, it can determine whether the target probability distribution condition is met; if the target probability distribution condition is met, the current normalization function is used as the target normalization function; if the target probability distribution condition is not met, the temperature parameter in the current normalization function is continuously optimized until the target probability distribution condition is determined to be met. Optionally, a maximum confidence statistical result can be calculated based on the maximum confidence interval. In this case, a maximum confidence statistical result may include interval statistics for each of the multiple maximum confidence intervals. The interval statistics for a maximum confidence interval may be the number of test monitoring object identity signals whose maximum confidence lies within the corresponding maximum confidence interval. For example, the maximum confidence statistical result for a set (such as an open set or a closed set of test categories) may include the interval statistics for the corresponding set across each maximum confidence interval. The interval statistics for a set within a maximum confidence interval may be the number of test monitoring object identity signals in the corresponding set whose maximum confidence lies within the corresponding maximum confidence interval. Alternatively, the maximum confidence statistical result for a set (such as an open set or a closed set of test categories) may include the maximum confidence of each test monitoring object identity signal in the corresponding set, and so on. This embodiment of the invention does not limit this. Optionally, the target probability distribution condition may be set according to experience or actual needs, and this embodiment of the invention does not limit this.

[0075] Optionally, the target probability distribution condition can be used to indicate that the probability distribution of the open set of the test category and the probability distribution of the closed set of the test category meet the separation requirement at a specified confidence threshold; optionally, the specified confidence threshold can be set according to experience or actual needs, and the embodiments of the present invention do not limit this. Based on this, when determining whether the target probability distribution conditions are met based on the maximum confidence statistics of the open set and the closed set of the test category, the electronic device can determine the open set probability distribution partitioning information corresponding to the open set of the test category based on the maximum confidence statistics of the open set and the specified confidence threshold. The open set probability distribution partitioning information supports the relationship between the maximum confidence of the identity signal of the test monitoring object in the open set of the test category and the specified confidence threshold. Similarly, based on the maximum confidence statistics of the closed set of the test category and the specified confidence threshold, the electronic device can determine the closed set probability distribution partitioning information corresponding to the closed set of the test category. The closed set probability distribution partitioning information supports the relationship between the maximum confidence of the identity signal of the test monitoring object in the closed set of the test category and the specified confidence threshold. Then, based on the open set probability distribution partitioning information and the closed set probability distribution partitioning information corresponding to the open set and the closed set of the test category, it can determine whether the target probability distribution conditions are met, thereby reducing the overlap between the probability distributions of the open set and the closed set of the test category through the target probability distribution conditions. Optionally, whether the target probability distribution condition is met can refer to whether the set of identity signals of the monitored test objects meets the target probability distribution condition, etc.; this embodiment of the invention does not limit this. Optionally, meeting the target probability distribution condition can mean that the probability distribution of the open set of the test categories and the probability distribution of the closed set of the test categories meet the separation requirement at a specified confidence threshold. In this case, the probability distribution of the open set of the test categories and the probability distribution of the closed set of the test categories are separated as much as possible at the specified confidence threshold; not meeting the target probability distribution condition can mean that the probability distribution of the open set of the test categories and the probability distribution of the closed set of the test categories do not meet the separation requirement at the specified confidence threshold.

[0076] It should be noted that the specific content of the open set probability distribution partitioning information and the closed set probability distribution partitioning information in the embodiments of the present invention is not limited. For example, the probability distribution partitioning information corresponding to a set (such as the open set probability distribution partitioning information corresponding to the open set of test categories or the closed set probability distribution partitioning information corresponding to the closed set of test categories) may include, but is not limited to, at least one of the following: the number of signals in the corresponding set with a maximum confidence level greater than a specified confidence level threshold (i.e., the number of identity signals of the test monitoring object), the number of signals in the corresponding set with a maximum confidence level less than a specified confidence level threshold, the number of signals in the corresponding set with a maximum confidence level greater than or equal to a specified confidence level threshold, the proportion of signals in the corresponding set with a maximum confidence level greater than a specified confidence level threshold (i.e., the ratio between the number of signals in the corresponding set with a maximum confidence level greater than a specified confidence level threshold and the total number of signals in the corresponding set), the proportion of signals in the corresponding set with a maximum confidence level less than a specified confidence level threshold, etc.; the embodiments of the present invention do not limit this. Based on this, open set probability distribution partitioning information can be used to indicate the number of signals or the proportion of signals with a maximum confidence level greater than a specified confidence level threshold in the open set of the test category, and closed set probability distribution partitioning information can be used to indicate the number of signals or the proportion of signals with a maximum confidence level greater than a specified confidence level threshold in the closed set of the test category, and so on. In other words, the relationship between the maximum confidence level of the test monitoring object identity signal in the open set of the test category and the specified confidence level threshold can be: the number of signals or the proportion of signals with a maximum confidence level greater than the specified confidence level threshold in the open set of the test category, and the relationship between the maximum confidence level of the test monitoring object identity signal in the closed set of the test category and the specified confidence level threshold can be: the number of signals or the proportion of signals with a maximum confidence level greater than the specified confidence level threshold in the closed set of the test category, and so on.

[0077] In one implementation, the open set probability distribution partitioning information includes open set overthreshold semaphores in the open set of the test category whose maximum confidence level is greater than a specified confidence threshold (i.e., semaphores in the open set of the test category whose maximum confidence level is greater than a specified confidence threshold), and the closed set probability distribution partitioning information includes closed set overthreshold semaphores in the closed set of the test category whose maximum confidence level is greater than a specified confidence threshold (i.e., semaphores in the closed set of the test category whose maximum confidence level is greater than a specified confidence threshold). Based on this, when determining whether the target probability distribution condition is met based on the open set probability distribution partitioning information corresponding to the open set of the test category and the closed set probability distribution partitioning information corresponding to the closed set of the test category, if the difference between the closed set overthreshold semaphore and the open set overthreshold semaphore is greater than a preset non-overlapping threshold, then it can be determined that the target probability distribution condition is met; if the difference between the closed set overthreshold semaphore and the open set overthreshold semaphore is less than or equal to the preset non-overlapping threshold, then it is determined that the target probability distribution condition is not met. Alternatively, if the open set exceeding the threshold semaphore is less than the open set semaphore threshold and the closed set exceeding the threshold semaphore is greater than the closed set semaphore threshold, then the target probability distribution condition can be determined to be met; if the open set exceeding the threshold semaphore is greater than or equal to the open set semaphore threshold, or the closed set exceeding the threshold semaphore is less than or equal to the closed set semaphore threshold, then the target probability distribution condition is determined not to be met. Optionally, the preset non-overlap threshold, open set semaphore threshold, and closed set semaphore threshold can all be set according to experience or actual needs, and this embodiment of the invention does not limit this. Optionally, the open set semaphore threshold is smaller and the closed set semaphore threshold is larger, so that the overlap between the probability distribution of the open set of the test category and the probability distribution of the closed set of the test category is smaller.

[0078] In another implementation, the open set probability distribution partitioning information may include the proportion of open set overthreshold semaphores in the open set of the test category whose maximum confidence level is greater than a specified confidence level threshold (i.e., the proportion of semaphores in the open set of the test category whose maximum confidence level is greater than a specified confidence level threshold). The closed set probability distribution partitioning information may include the proportion of closed set overthreshold semaphores in the closed set of the test category whose maximum confidence level is greater than a specified confidence level threshold (i.e., the proportion of semaphores in the closed set of the test category whose maximum confidence level is greater than a specified confidence level threshold). Based on this, when determining whether the target probability distribution condition is met based on the open set probability distribution partitioning information corresponding to the open set of the test category and the closed set probability distribution partitioning information corresponding to the closed set of the test category, if the difference between the proportion of closed set overthreshold semaphores and the proportion of open set overthreshold semaphores is greater than a preset non-overlapping proportion threshold, then the target probability distribution condition can be determined to be met; if the difference between the proportion of closed set overthreshold semaphores and the proportion of open set overthreshold semaphores is less than or equal to the preset non-overlapping proportion threshold, then the target probability distribution condition can be determined not to be met. Alternatively, if the proportion of open-set semaphores exceeding the threshold is less than the open-set semaphore proportion threshold, and the proportion of closed-set semaphores exceeding the threshold is greater than the closed-set semaphore proportion threshold, then the target probability distribution condition can be determined to be met; if the proportion of open-set semaphores exceeding the threshold is greater than or equal to the open-set semaphore proportion threshold, or the proportion of closed-set semaphores exceeding the threshold is less than or equal to the closed-set semaphore proportion threshold, then the target probability distribution condition can be determined not to be met, and so on. Optionally, the preset non-overlapping proportion threshold, the open-set semaphore proportion threshold, and the closed-set semaphore proportion threshold can all be set according to experience or actual needs, and this embodiment of the invention does not limit this. Optionally, the open-set semaphore proportion threshold can be relatively small, and the closed-set semaphore proportion threshold can be relatively large, thereby reducing the overlap between the probability distribution of the open set of the test category and the probability distribution of the closed set of the test category.

[0079] In another embodiment, the open set probability distribution partitioning information may include the proportion of open set low threshold semaphores in the open set of the test category whose maximum confidence is less than a specified confidence threshold (i.e., the proportion of semaphores in the open set of the test category whose maximum confidence is less than a specified confidence threshold), and the closed set probability distribution partitioning information may include the proportion of closed set overthreshold semaphores in the closed set of the test category whose maximum confidence is greater than a specified confidence threshold. Based on this, when determining whether the target probability distribution condition is met based on the open set probability distribution partitioning information corresponding to the open set of the test category and the closed set probability distribution partitioning information corresponding to the closed set of the test category, if the proportion of open set low threshold semaphores is greater than a preset open set semaphore proportion, and the proportion of closed set overthreshold semaphores is greater than a closed set semaphore proportion threshold, then the target probability distribution condition is met; if the proportion of open set low threshold semaphores is less than or equal to a preset open set semaphore proportion, or the proportion of closed set overthreshold semaphores is less than or equal to a closed set semaphore proportion threshold, then the target probability distribution condition is not met, and so on. Optionally, the preset open set semaphore proportion can be set according to experience or actual needs, and this embodiment of the invention does not limit this. In this case, the preset thresholds for the proportion of open set semaphores and the proportion of closed set semaphores can both be relatively large, thereby reducing the overlap between the probability distribution of the open set of the test category and the probability distribution of the closed set of the test category.

[0080] Optionally, while continuously optimizing the temperature parameters in the current normalization function until the target probability distribution condition is met, the electronic device can perform a random search to optimize the temperature parameters in the current normalization function until a temperature parameter that satisfies the target probability distribution condition is found; alternatively, the electronic device can also optimize the temperature parameters in the current normalization function according to the target optimization index until the target probability distribution condition is met, etc.; this embodiment of the invention does not limit this. Optionally, the target optimization index can be a preset non-overlapping threshold, or a preset non-overlapping percentage threshold, etc.; this embodiment of the invention does not limit this; in this case, the temperature parameters in the current normalization function can be optimized through the target optimization algorithm and the target optimization index until the target probability distribution condition is met. Optionally, the target optimization algorithm can be any optimization algorithm, such as a genetic algorithm, a simulated annealing algorithm, etc.; this embodiment of the invention does not limit this. Based on this, this embodiment of the invention does not limit the specific implementation process of continuously optimizing the temperature parameters in the current normalization function until the target probability distribution condition is met.

[0081] For example, such as Figure 4 As shown, by adjusting the temperature parameter, the probability distribution of the open and closed sets of test categories can be adjusted. This allows the recognition model to maintain high accuracy in recognizing the closed sets of test categories while also maintaining high accuracy in recognizing the open sets, thus lowering the maximum confidence level of the open sets and thereby correcting the recognition accuracy. Figure 4Subplots (a) and (b) in the figure represent the recognition accuracy when the temperature parameter is 1 and 2, respectively. The horizontal axis can include each maximum confidence interval (the unit can be percentage: %), such as the maximum confidence interval 80-90 or 90-100, etc. The frequency on the vertical axis can represent the number of test monitoring object identity signals in a set (such as the test category open set or the test category closed set) whose maximum confidence is located in each maximum confidence interval. Specifically, when the temperature parameter T=1, most signals in the closed set of test categories have a prediction confidence (i.e., maximum confidence) of over 90% for their respective categories. When the probability threshold (i.e., the specified confidence threshold) is set to 80%, the proportion of signals in the closed set of test categories with a maximum confidence higher than this specified confidence threshold reaches 99.62%, indicating that the recognition model has a high ability to distinguish known categories. Under the same temperature parameter, only 66.2% of the signals in the open set of test categories can be correctly identified (i.e., maximum confidence is below 80%), and a relatively high proportion of unknown category signals are misclassified as known categories, resulting in insufficient robustness of the recognition model in open environments. Therefore, it is necessary to take certain measures to separate the probability distribution of the open set of test categories (i.e., the probability distribution of open set of test categories identification) and the probability distribution of the closed set of test categories (i.e., the probability distribution of closed set of test categories identification) as much as possible at the specified confidence threshold. Accordingly, the experiment further adjusted the temperature parameters to optimize the recognition results, minimizing the aliasing (overlap) between the probability distribution of closed-set recognition and open-set recognition of the test categories. Based on this, at T=2, the proportion of test categories with a maximum confidence level higher than 80% still reached 98.68%, and the recognition accuracy was almost unaffected. For open-set recognition of the test categories, the proportion of test categories with a maximum confidence level lower than 80% significantly increased to 97.48%, greatly enhancing the ability to recognize unknown categories. It is evident that by introducing a temperature regulation mechanism, the recognition model effectively improves its ability to perceive unknown categories while maintaining high accuracy for known categories, significantly mitigating the limitations of the traditional Softmax model in open environments. Based on this, this embodiment of the invention can combine a probability distribution correction mechanism based on temperature parameter calibration to smooth the corrected category probabilities output by the recognition model through a target normalization function. In other words, this embodiment of the invention can effectively suppress the overconfidence phenomenon of the recognition model when facing unknown categories through this calibration mechanism. By optimizing the T value, the distribution gap between known and unknown categories at a specified confidence threshold (e.g., 80%) can be increased, thereby endowing the original recognition model with highly robust open-set rejection capability without changing its backbone network.

[0082] Based on this, embodiments of the present invention can introduce a temperature parameter to smooth and calibrate the SoftMax output when calculating the probability distribution, redistribute the original probability, and explicitly calculate the probability that the sample belongs to an unknown category; this can effectively suppress the tendency of the recognition model to make incorrect classifications based on blind confidence in unknown signals, and endow it with the ability to express uncertainty, thereby significantly reducing the false alarm rate and misjudgment risk in open set scenarios, so as to effectively solve the problem of overconfidence.

[0083] This invention can acquire the identity signal of a target monitored object and extract features from it to obtain the target radio frequency fingerprint feature. Then, it can determine the category center of each monitored object category in the set of monitored object categories, and calculate the category center distance of the target radio frequency fingerprint feature in each monitored object category based on the category center of each monitored object category. Based on this, it can calculate the tail probability of the target radio frequency fingerprint feature in each monitored object category based on the category center distance and the cumulative distribution of the target Cauchy probability in each monitored object category, to correct the category probability. Correspondingly, it can use the tail probability of the target radio frequency fingerprint feature in each monitored object category to determine the target category probability of the target radio frequency fingerprint feature in each monitored object category, effectively improving the accuracy of the target category probability. Based on the target category probability of the target radio frequency fingerprint feature in each monitored object category, it can determine the radio frequency fingerprint recognition result of the target radio frequency fingerprint feature. One radio frequency fingerprint recognition result can be used to indicate whether it belongs to an existing monitored object category. As can be seen, the embodiments of the present invention can determine the target category probability with high accuracy by using the category center distance of the target radio frequency fingerprint features under each monitoring object category and the cumulative distribution of the target Cauchy probability under each monitoring object category, so as to obtain a radio frequency fingerprint recognition result with high accuracy. This can effectively improve the accuracy of radio frequency fingerprint recognition, and further improve the ability to perceive unknown categories when encountering unknown monitoring object categories, thereby effectively avoiding the failure of identity signal recognition.

[0084] Based on the above description, this embodiment of the invention also proposes a more specific open-collection radio frequency fingerprinting method. Accordingly, this open-collection radio frequency fingerprinting method can be executed by the aforementioned electronic device (terminal or server); or, this open-collection radio frequency fingerprinting method can be executed jointly by the terminal and the server. For ease of explanation, the following description will use the execution of this open-collection radio frequency fingerprinting method by an electronic device as an example; please refer to [link to relevant documentation]. Figure 5 The open-collection radio frequency fingerprint recognition method may include the following steps S501-S509:

[0085] S501, Obtain the training monitoring object identity signal set. One training monitoring object identity signal corresponds to one existing monitoring object category. The training monitoring object identity signal set includes multiple training monitoring object identity signals under each monitoring object category.

[0086] Optionally, the electronic device may store a set of training monitoring object identity signals in its own storage space. In this case, the electronic device may obtain the set of training monitoring object identity signals from its own storage space; or, the electronic device may obtain a download link for the set of training monitoring object identity signals and download the set of training monitoring object identity signals using the download link, thereby obtaining the set of training monitoring object identity signals; or, the electronic device may filter out multiple training monitoring object identity signals under each monitoring object category from the database (multiple training monitoring object identity signals under one monitoring object category can refer to multiple training monitoring object identity signals belonging to the corresponding monitoring object category), and add the filtered multiple training monitoring object identity signals under each monitoring object category to the set of training monitoring object identity signals, thereby obtaining the set of training monitoring object identity signals, etc.; the embodiments of the present invention do not limit this.

[0087] In this context, "a training monitoring object identity signal corresponding to an existing monitoring object category" can mean that a training monitoring object identity signal belongs to an existing monitoring object category, that is, the category of a training monitoring object identity signal is a category in the monitoring object category set.

[0088] Optionally, before extracting features from each training monitoring object identity signal in the training monitoring object identity signal set, the electronic device may also acquire the model training monitoring object identity signal set and use the model training monitoring object identity signal set to train the initial recognition model (i.e., closed-set model training) to obtain the target recognition model. Optionally, the model training monitoring object identity signal set may be the same as or different from the training monitoring object identity signal set; this embodiment of the invention does not limit this. Optionally, the method of acquiring the model training monitoring object identity signal set may be the same as the method of acquiring the training monitoring object identity signal set; this embodiment of the invention will not elaborate further here.

[0089] Optionally, when training an initial recognition model using a set of monitored object identity signals to obtain a target recognition model, the electronic device can call the initial recognition model to extract features from each monitored object identity signal in the set of monitored object identity signals trained by the model, obtaining the radio frequency fingerprint features of each monitored object identity signal. Then, based on the radio frequency fingerprint features of each monitored object identity signal trained by the model and the category center of each monitored object category, the target category probability of each monitored object identity signal trained by the model is determined in each monitored object category. Based on the target category probability of each monitored object identity signal trained by the model and the category label of each monitored object identity signal trained by the model, the model loss value (such as cross-entropy loss) is calculated. In order to optimize the model parameters in the initial recognition model in the direction of reducing the model loss value (if only the network parameters are optimized at this time), the optimized initial recognition model is obtained, and the target recognition model is determined based on the optimized initial recognition model. It should be understood that the implementation method for determining the target category probability of the identity signal of the monitored object under each monitored object category based on the radio frequency fingerprint features of the identity signal of the monitored object trained by each model and the category center of each monitored object category can be the same as the implementation method for determining the target category probability of the target radio frequency fingerprint features under each monitored object category, and will not be repeated here.

[0090] Optionally, when determining the target recognition model based on the optimized initial recognition model, the electronic device can continue to train the optimized initial recognition model using the set of object identity signals trained on the model until the model training convergence condition is met (such as the number of model training iterations reaching a preset model training iteration threshold, or the model loss value being less than a preset model loss threshold, etc.), thereby obtaining the target recognition model. Optionally, both the preset model training iteration threshold and the preset model loss threshold can be set according to experience or actual needs, and this embodiment of the invention does not limit this.

[0091] S502, extract features from each training monitoring object identity signal in the training monitoring object identity signal set to obtain the training radio frequency fingerprint features of each training monitoring object identity signal, so as to obtain the training radio frequency fingerprint features under each monitoring object category.

[0092] Among them, a training radio frequency fingerprint feature under a monitoring object category can be: the training radio frequency fingerprint feature of a training monitoring object identity signal belonging to the corresponding monitoring object category in the training monitoring object identity signal set.

[0093] Optionally, the electronic device can invoke the target recognition model to extract features from each training monitoring object identity signal in the training monitoring object identity signal set, thereby obtaining the training radio frequency fingerprint features of each training monitoring object identity signal.

[0094] S503, for any monitoring object category in the set of monitoring object categories, determine the initial weight reassembly under any monitoring object category, and calculate the category center distance of each training radio frequency fingerprint feature under any monitoring object category based on the category center of any monitoring object category, each training radio frequency fingerprint feature under any monitoring object category and the initial weight reassembly; wherein, the initial weight reassembly under any monitoring object category includes the initial weight value of each distance calculation method among multiple distance calculation methods under any monitoring object category.

[0095] Optionally, the initial weighting of any monitored object category can be set according to experience or actual needs, or it can be randomly generated; this embodiment of the invention does not limit this. Optionally, the category center of any monitored object category can be the mean of the various trained radio frequency fingerprint features under any monitored object category.

[0096] Optionally, when calculating the class center distance of each training RF fingerprint feature under any monitoring object category based on the class center of any monitoring object category, each training RF fingerprint feature under any monitoring object category, and the initial weighting, the electronic device can use the class center of any monitoring object category and any training RF fingerprint feature to calculate the distance between any training RF fingerprint feature and the class center of any monitoring object category under each distance calculation method (also referred to as the distance between any training RF fingerprint feature and the class center of any monitoring object category under each distance calculation method); and can perform a weighted summation of the distances between any training RF fingerprint feature and the class center of any monitoring object category under each distance calculation method according to the initial weighting of any monitoring object category to obtain the class center distance of any training RF fingerprint feature under any monitoring object category. For example, taking multiple distance calculation methods including Euclidean distance calculation, Chebyshev distance calculation, and Manhattan distance calculation as examples, the class center distance of any training RF fingerprint feature under any monitoring object category can be as shown in Formula 2.1:

[0097] Equation 2.1

[0098] Where, d Hw1, w2, and w3 can represent the class center distance of any training RF fingerprint feature under any monitoring object category (also known as the mixed distance from any training RF fingerprint feature to the class center of any monitoring object category); w1, w2, and w3 can represent various distance coefficients (i.e., the weights of various distance calculation methods, which can be the weight values ​​in a weighted set). That is, w1 can represent the weight of the Euclidean distance calculation method (i.e., the distance coefficient), w2 can represent the weight of the Chebyshev distance calculation method, and w3 can represent the weight of the Manhattan distance calculation method; where each distance coefficient satisfies w1+w2+w3=1.

[0099] Where, d Euc It can be the Euclidean distance (i.e., the Euclidean distance from any trained RF fingerprint feature to the class center of any monitored object class), d Che It can be the Chebyshev distance (i.e., the Chebyshev distance from any trained RF fingerprint feature to the class center of any monitored object class), d Man This can be the Manhattan distance (i.e., the Manhattan distance from any training RF fingerprint feature to the category center of any monitored object category); for example, taking any training RF fingerprint feature under any monitored object category as the training RF fingerprint feature of the nth training monitored object identity signal under any monitored object category (i.e., the nth training RF fingerprint feature) as an example, the distances can be shown in Equation 2.2:

[0100] Equation 2.2

[0101] Where Q can represent the spatial dimension (such as the dimension of training RF fingerprint features and category centers), q can represent the q-th dimension, and Ms C-q f(x) can represent the q-th dimension of the category center of any monitored object category. Cn ) q The nth training RF fingerprint feature under any monitoring object category can be represented by the qth dimension; the Chebyshev distance can be represented by taking the largest absolute difference across the Q dimensions.

[0102] S504, based on the class center distance of each training radio frequency fingerprint feature under any monitoring object category, determine the class center distance of each selected radio frequency fingerprint feature among multiple selected radio frequency fingerprint features, and based on the class center distance of each selected radio frequency fingerprint feature, determine the target fitting distribution evaluation result.

[0103] In one implementation, when determining the class center distance of each selected RF fingerprint feature among multiple selected RF fingerprint features based on the class center distance of each trained RF fingerprint feature under any monitoring object category, the electronic device can sort the class center distances of each trained RF fingerprint feature under any monitoring object category in ascending order (e.g., sort the class center distances of each trained RF fingerprint feature under any monitoring object category in ascending order), obtaining an ascending distance sorting result. The class center distances of the last A% of RF fingerprint features determined from the ascending distance sorting result are used as the class center distances of each selected RF fingerprint feature among multiple selected RF fingerprint features. Optionally, A can be set according to experience or actual needs (i.e., A can be any value between 1 and 100), and this embodiment of the invention does not limit this; for example, the value of A can be 20, in which case the last 20% of class center distances with the farthest distance can be selected for Cauchy distribution fitting.

[0104] In another embodiment, the electronic device may also sort the class center distances of each training radio frequency fingerprint feature under any monitoring object category in descending order to obtain a distance descending sort result. The class center distances of the top A% of radio frequency fingerprint features selected from the distance descending sort result are used as the class center distances of each of the multiple screened radio frequency fingerprint features, etc.; the embodiments of the present invention do not limit this.

[0105] The multiple screening RFID features may include the training RFID features with the furthest category center distance (A%) among all training RFID features under any monitoring object category. In this embodiment, the category center distance of a screening RFID feature is greater than the category center distance of any training RFID feature other than the multiple screening RFID features under any monitoring object category; in other words, the category center distance of a screening RFID feature is greater than the category center distance of any unscreened RFID feature in the unscreened RFID feature set, which may include all training RFID features under any monitoring object category except the multiple screening RFID features. Furthermore, all training RFID features under any monitoring object category may include each training RFID feature under that monitoring object category, i.e., the training RFID feature of each training monitoring object identity signal among multiple training monitoring object identity signals under that monitoring object category.

[0106] Optionally, when determining the target fitting distribution evaluation result based on the class center distance of each selected RF fingerprint feature, the electronic device can use the class center distance of each selected RF fingerprint feature among multiple selected RF fingerprint features to perform Cauchy distribution fitting (that is, to perform Cauchy distribution fitting on multiple selected class center distances, which may include the class center distances of each selected RF fingerprint feature), to obtain the initial Cauchy distribution function under any monitored object category; and can use the initial Cauchy distribution function under any monitored object category to determine the distance fitting value of each selected RF fingerprint feature. For example, the class center distance of any selected RF fingerprint feature can be substituted into the initial Cauchy distribution function under any monitored object category (i.e., the value of the input variable of the initial Cauchy distribution function under any monitored object category) to obtain the distance fitting value of any selected RF fingerprint feature. Here, the Cauchy distribution is a continuous probability distribution, suitable for modeling extreme value or outlier detection. For example, the probability density function of a Cauchy distribution (i.e., the Cauchy distribution function, also called the Cauchy distribution fitting function) can be as shown in Equation 2.3:

[0107] Equation 2.3

[0108] Here, f(x,x0,γ) can represent the Cauchy distribution function; the Cauchy distribution is symmetric about x0 and reaches its peak at x=x0. Accordingly, based on the Cauchy distribution function, the corresponding Cauchy cumulative probability distribution can be obtained, that is, one Cauchy distribution function corresponds to one Cauchy cumulative probability distribution. It should be noted that the specific implementation method for fitting the Cauchy distribution in this embodiment of the invention is not limited; for example, maximum likelihood estimation or the quantile method can be used to fit the Cauchy distribution, etc.

[0109] Based on this, the target fitting distribution evaluation result can be determined based on the class center distance and distance fitting value of each screened RF fingerprint feature; wherein, the cumulative Cauchy probability distribution of the target under any monitored object category is the cumulative Cauchy probability distribution corresponding to the target weighted reassembly under any monitored object category. Optionally, the cumulative Cauchy probability distribution and Cauchy distribution function corresponding to a weighted reassembly can be: the cumulative Cauchy probability distribution and Cauchy distribution function obtained by calculating the class center distance through the corresponding weighted reassembly. It should be noted that the embodiment of the present invention does not limit the calculation method of the target fitting distribution evaluation result, such as calculating it through mean squared error (MSE) or Cauchy loss, etc.

[0110] For example, taking the mean squared error between the fitted function and the actual sample distribution as the evaluation result of the target fitted distribution as an example, the electronic device can determine the evaluation result of the target fitted distribution based on the class center distance and distance fitted value of each screened RF fingerprint feature using Formula 2.4:

[0111] Equation 2.4

[0112] Wherein, MSE can be used as the evaluation result of the target fitting distribution, M is the total number of selected RF fingerprint features out of multiple selected RF fingerprint features, and y1 m y2 can be the true value (i.e., class center distance) of the m-th selected RF fingerprint feature among multiple selected RF fingerprint features. m It can be the distance fitting value for the m-th selected radio frequency fingerprint feature.

[0113] S505, optimize the initial weight reassembly under any monitoring object category to minimize the target fitting distribution evaluation result until the fitting convergence condition is met, thereby obtaining the target weight reassembly under any monitoring object category; wherein, the target weight reassembly under any monitoring object category is used to calculate the class center distance of the target radio frequency fingerprint feature under any monitoring object category.

[0114] In this embodiment of the invention, a dynamic weight adjustment mechanism is constructed to achieve high-precision modeling of the marginal data of the category distribution. Based on this, minimizing the mean square error of the tail samples (i.e., the evaluation result of the target fitting distribution) can be taken as the core constraint objective. By finely adjusting each distance coefficient, the representation accuracy and rejection robustness of the algorithm when facing open set samples can be significantly enhanced. Optionally, the electronic device can adopt a target distance coefficient optimization algorithm to optimize the initial weight reorganization under any monitoring object category. The goal of the target distance coefficient optimization algorithm is to minimize the evaluation result of the target fitting distribution to evaluate the impact of different weights on the tail distribution modeling. Optionally, the evaluation result of the target fitting distribution can also be referred to as the fitting error.

[0115] Optionally, the target distance coefficient optimization algorithm can be a Bayesian optimization algorithm or a genetic algorithm (in which case the number of initial weighting groups under any monitoring object category can be multiple, and finally the weighting group with the smallest target fitting distribution evaluation result can be determined from the population when the fitting convergence condition is met as the target weighting group under any monitoring object category), etc.; the embodiments of the present invention do not limit this.

[0116] Optionally, the fitting convergence condition can be set based on experience or actual needs, and this embodiment of the invention does not limit this. For example, the fitting convergence condition can refer to the number of weight optimization iterations reaching a preset threshold for the number of weight optimization iterations, or it can refer to the target fitting distribution evaluation result (i.e., the minimized target fitting distribution evaluation result) being less than a preset fitting distribution evaluation result threshold, etc. Optionally, both the preset threshold for the number of weight optimization iterations and the preset threshold for the fitting distribution evaluation result can be set based on experience or actual needs, and this embodiment of the invention does not limit this. For example, as... Figure 6 As shown, it can be determined whether the fitting error meets the fitting convergence condition (if it is less than the preset fitting distribution evaluation result threshold, it is met; if it is greater than or equal to the preset fitting distribution evaluation result threshold, it is not met). Therefore, if it is not met (i.e. the fitting convergence condition is not met), the weight reassembly under any monitoring object category (such as the initial weight reassembly) can be further optimized to reconstruct the Cauchy distribution function under any monitoring object category until the fitting convergence condition is met, and the target weight reassembly under any monitoring object category is obtained. Then, based on the target weight reassembly under any monitoring object category, the distance between the target RF fingerprint feature and the category center of any monitoring object category under each distance calculation method can be weighted and summed to obtain the category center distance of the target RF fingerprint feature under any monitoring object category. The distance between the target RF fingerprint feature and the category center of any monitoring object category under any distance calculation method can also be expressed as the category distance of the target RF fingerprint feature under any distance calculation method.

[0117] Based on this, this invention addresses the issue of the single dimension of distance measurement by proposing a hybrid distance measurement method. When multiple distance calculation methods include Euclidean distance, Manhattan distance, and Chebyshev distance, it integrates Euclidean distance (overall difference), Manhattan distance (independent difference in each dimension), and Chebyshev distance (maximum dimensional deviation). It also utilizes Bayesian optimization algorithms to automatically optimize the best weight coefficients for each distance, thereby more comprehensively characterizing the difference between samples and class centers in high-dimensional feature space. In particular, it can accurately capture minute deviations that appear abnormally in specific dimensions, significantly improving the recognition accuracy of feature overlap regions. Furthermore, this invention introduces the Cauchy distribution, which has significant long-tail characteristics, to fit and model the distances of tail samples of known classes far from the class center. It should be understood that the Cauchy distribution is better suited to the long-tail distribution characteristics of RF fingerprint features, effectively reducing the false recognition rate of known edge samples and enhancing the sensitivity and detection capability of highly disguised unknown signals.

[0118] S506, acquire the identity signal of the target monitoring object, and extract features from the identity signal of the target monitoring object to obtain the target radio frequency fingerprint feature of the identity signal of the target monitoring object.

[0119] S507, determine the category center of each monitoring object category in the monitoring object category set, and calculate the distance between the target radio frequency fingerprint feature and the category center of each monitoring object category based on the category center of each monitoring object category.

[0120] S508 calculates the tail probability of the target radio frequency fingerprint feature in each monitoring object category based on the category center distance of the target radio frequency fingerprint feature in each monitoring object category and the cumulative distribution of the target Cauchy probability in each monitoring object category.

[0121] S509, respectively using the tail probability of the target radio frequency fingerprint feature under each monitoring object category, to determine the target category probability of the target radio frequency fingerprint feature under each monitoring object category; and based on the target category probability of the target radio frequency fingerprint feature under each monitoring object category, to determine the radio frequency fingerprint recognition result of the target radio frequency fingerprint feature, and an radio frequency fingerprint recognition result can be used to indicate whether it is an existing monitoring object category.

[0122] Optionally, if the RFID fingerprint feature is determined to belong to an existing category in the database (i.e., an existing monitoring object category), the electronic device can output the monitoring object identity (which can be simply referred to as output identity, such as outputting the monitoring object identifier of the target monitoring object, etc.) and / or perform verification; if it is determined to belong to a non-existent category (i.e., an unknown category), a warning can be triggered. Optionally, the electronic device can make a judgment based on the RFID fingerprint recognition result, such as based on the maximum target category probability in the RFID fingerprint recognition result, or based on the category identifier of the RFID fingerprint recognition result (such as the category identifier of an existing monitoring object category or the category identifier of an unknown category, etc.); or, it can also make a judgment based on the target category probability of the target RFID fingerprint feature under each monitoring object category to determine the RFID fingerprint recognition result of the target RFID fingerprint feature, etc.; the embodiments of the present invention do not limit this.

[0123] Optionally, the electronic device can perform verification according to preset verification rules to obtain a verification result. The verification result can be used to indicate whether the verification has passed. For example, the verification of the RFID fingerprint recognition result can be performed according to the preset verification rules to realize the verification of the output identity. Optionally, the preset verification rules can be set according to experience or actual needs, and this embodiment of the invention does not limit this. For example, the preset verification rules can be to verify whether the monitoring object indicated by the monitoring object identifier (such as the monitoring object name or number) in the message information of the target monitoring object is an existing monitoring object category indicated by the RFID fingerprint recognition result, such as the monitoring object identifier being the same as the category identifier in the RFID fingerprint recognition result; or, it can verify whether the location indicated by the location information in the message information of the target monitoring object is located within a specified area, etc. Optionally, one monitoring object can correspond to one specified area. Optionally, the specified area corresponding to one monitoring object can be set according to experience or according to actual needs, and this embodiment of the invention does not limit this. Optionally, this embodiment of the invention does not limit the prompt warning method.

[0124] Optionally, when the monitored object is an aircraft, embodiments of the present invention can establish a fingerprint database (hereinafter referred to as the database) covering known aircraft signals, and obtain microscopic radio frequency fingerprint features reflecting differences in transmitter hardware through a feature extraction module. Furthermore, an identification model can be used to perform open-set discrimination on real-time signals (such as the identity signal of the target monitored object). Based on this, embodiments of the present invention can effectively solve the open-set identification problems such as the access of new aircraft, shielding of unknown interference sources, and monitoring of maliciously forged signals.

[0125] Based on this, in fields such as aviation surveillance and drone regulation, related technologies are mainly based on the closed-set assumption, such as assuming that all received signals belong to the set of aircraft models registered during the training phase. However, due to the openness and dynamism of the airspace environment, this closed-set identification model faces serious generalization challenges in actual deployment; while open-world identification requires the system to maintain high-precision classification of known models (such as aircraft serial numbers) while possessing the ability to reject unknown categories. It should be understood that with the development of aviation manufacturing technology, new remote ID (Remote Identification) transmitters or new ADS-B (Automatic Dependent Surveillance-Broadcast) transmitters are constantly entering the airspace; the radio frequency physical layer features generated by these new aircraft models are in a vacuum in the prior database, causing traditional models to misclassify and fail to effectively identify their incremental attributes. For example, Figure 7As shown, in closed-set identification, it can correctly identify aircraft categories already recorded in the database (i.e., known aircraft categories in the aircraft category set); when a new aircraft joins the airspace, it brings a new identity signal (i.e., a new category, also known as an unknown category, not belonging to the aircraft category set), which will be determined as a known category (i.e., a specific visible category) in the traditional radio frequency fingerprinting system. Figure 7 As shown in sub-figure (a) in the figure, this causes the failure of identity information recognition, affecting airspace management security; however, this application can correctly identify the new identity signal as an unknown category (i.e., specifically visible) through the open-set radio frequency fingerprinting method. Figure 7 (as shown in sub-image (b)) to effectively improve recognition accuracy.

[0126] In this embodiment of the invention, a post-processing architecture based on radio frequency fingerprint feature analysis is constructed. Unlike reconstruction learning, it does not require additional decoders or reconstruction network branches, meaning it does not rely on specific neural network architectures or reconstruction branches. It can directly utilize the output features of the fully connected layers of existing classification networks for modeling. Based on this, complex model structure design is avoided, significantly reducing computational complexity and training resource consumption. Furthermore, since it does not rely on the recognition model's ability to generate unknown data, the open-set radio frequency fingerprinting method proposed in this embodiment can serve as a general-purpose plug-in, exhibiting stronger generalization and ease of deployment. In other words, the open-set radio frequency fingerprinting method proposed in this embodiment can be seamlessly integrated as a general-purpose algorithm plug-in into any pre-trained closed-set recognition model. Moreover, this embodiment can complete closed-loop monitoring from protocol data demodulation to physical layer authentication without significantly increasing hardware computational overhead.

[0127] In summary, the embodiments of this invention can introduce optimized hybrid distance metrics such as Bayesian optimization algorithms and Cauchy distribution tail fitting, which more comprehensively captures subtle differences in the high-dimensional feature space and accurately models long-tailed anomaly samples, significantly improving the detection rate of unknown threat signals and the recognition accuracy of known signals in complex electromagnetic environments. At the same time, the lightweight post-processing architecture that can be combined with temperature parameter calibration effectively suppresses false alarms caused by overconfidence without increasing the model training burden, effectively solving the problem of recognition failure of related technologies when facing unknown categories. It also solves the problem of poor recognition performance of single distance metrics in open set recognition problems, realizing open set RF fingerprint recognition with low computational cost, high robustness and easy generalization.

[0128] In this embodiment of the invention, after obtaining the set of training monitoring object identity signals, feature extraction is performed on each training monitoring object identity signal in the set to obtain training radio frequency fingerprint features for each training monitoring object identity signal, thereby obtaining training radio frequency fingerprint features for each monitoring object category. A training radio frequency fingerprint feature for a monitoring object category is the training radio frequency fingerprint feature of a training monitoring object identity signal belonging to the corresponding monitoring object category in the set of training monitoring object identity signals. Then, for any monitoring object category in the set of monitoring object categories, an initial weight reassembly for that monitoring object category can be determined. Based on the category center of any monitoring object category, each training radio frequency fingerprint feature for that monitoring object category, and the initial weight reassembly, the category center distance of each training radio frequency fingerprint feature for that monitoring object category is calculated. The initial weight reassembly for any monitoring object category includes the initial weight values ​​of each distance calculation method among multiple distance calculation methods for that monitoring object category. Furthermore, based on the class center distances of each trained RFID fingerprint feature under any monitoring object category, the class center distances of each selected RFID fingerprint feature among multiple selected RFID fingerprint features can be determined, and the target fitting distribution evaluation result can be determined based on the class center distances of each selected RFID fingerprint feature. Additionally, the initial weight reassembly under any monitoring object category can be optimized to minimize the target fitting distribution evaluation result until the fitting convergence condition is met, thereby obtaining the target weight reassembly under any monitoring object category. The target weight reassembly under any monitoring object category is used to calculate the class center distance of the target RFID fingerprint feature under any monitoring object category. Based on this, the target monitoring object identity signal can be obtained, and feature extraction can be performed on the target monitoring object identity signal to obtain the target RFID fingerprint feature of the target monitoring object identity signal. Furthermore, the class center of each monitoring object category in the monitoring object category set can be determined, and the class center distance of the target RFID fingerprint feature under each monitoring object category can be calculated based on the class center of each monitoring object category. Furthermore, the tail probability of the target RF fingerprint feature in each monitoring object category can be calculated based on the category center distance and the cumulative distribution of the target Cauchy probability in each monitoring object category. Thus, the target category probability of the target RF fingerprint feature in each monitoring object category is determined using the tail probability of the target RF fingerprint feature in each monitoring object category. Based on the target category probability of the target RF fingerprint feature in each monitoring object category, the RF fingerprint recognition result of the target RF fingerprint feature is determined. One RF fingerprint recognition result can be used to indicate whether it belongs to an existing monitoring object category.As can be seen, the embodiments of the present invention propose a hybrid distance metric model that integrates multiple distance calculation methods. This multi-dimensional distance extraction method can simultaneously capture the microscopic changes in spatial location, maximum dimensional deviation, and cumulative differences in each dimension of high-dimensional fingerprint features. It is particularly suitable for identifying aircraft signals in feature aliasing regions caused by subtle hardware differences. Furthermore, the embodiments of the present invention can utilize the high tolerance of Cauchy distribution to outliers and extreme values ​​to more accurately describe the long-tail characteristics in radio frequency fingerprint features, effectively solving the problem of false rejection of edge-known signals caused by signal noise or environmental interference, and improving the detection accuracy of forged signals, thereby effectively improving the accuracy of open-set radio frequency fingerprint recognition.

[0129] Based on the description of the relevant embodiments of the open-collection radio frequency fingerprinting method above, this invention also proposes an open-collection radio frequency fingerprinting device, which can be a computer program (including program code) running in an electronic device; such as Figure 8 As shown, the open-collector RFID fingerprint recognition device may include an acquisition unit 801 and a processing unit 802. The open-collector RFID fingerprint recognition device can perform... Figure 1 or Figure 5 The open-collector RFID fingerprinting method shown, i.e., the open-collector RFID fingerprinting device, can operate the above-described unit:

[0130] Acquisition unit 801 is used to acquire the identity signal of the target monitoring object;

[0131] Processing unit 802 is used to extract features from the identity signal of the target monitoring object to obtain the target radio frequency fingerprint feature of the identity signal of the target monitoring object;

[0132] The processing unit 802 is further configured to determine the category center of each monitoring object category in the monitoring object category set, and calculate the category center distance of the target radio frequency fingerprint feature under each monitoring object category based on the category center of each monitoring object category.

[0133] The processing unit 802 is further configured to calculate the tail probability of the target radio frequency fingerprint feature in each monitoring object category based on the category center distance of the target radio frequency fingerprint feature in each monitoring object category and the target Cauchy probability cumulative distribution in each monitoring object category; wherein, the tail probability of the target radio frequency fingerprint feature in a monitoring object category is calculated by substituting the category center distance of the target radio frequency fingerprint feature in the corresponding monitoring object category into the target Cauchy probability cumulative distribution in the corresponding monitoring object category;

[0134] The processing unit 802 is further configured to determine the target category probability of the target radio frequency fingerprint feature under each monitoring object category by using the tail probability of the target radio frequency fingerprint feature under each monitoring object category; and to determine the radio frequency fingerprint recognition result of the target radio frequency fingerprint feature based on the target category probability of the target radio frequency fingerprint feature under each monitoring object category, wherein a radio frequency fingerprint recognition result can be used to indicate whether it is an existing monitoring object category.

[0135] In one embodiment, when the processing unit 802 calculates the distance between the target radio frequency fingerprint feature and the category center of each monitored object category based on the category center of each monitored object category, it may specifically be used to:

[0136] Iterate through each monitoring object category in the set of monitoring object categories, and take the currently traversed monitoring object category as the current monitoring object category;

[0137] Based on the category center of the current monitored object category and the target radio frequency fingerprint feature, the current category distance of the target radio frequency fingerprint feature under each of the multiple distance calculation methods is calculated. The current category distance of the target radio frequency fingerprint feature under a distance calculation method refers to the distance between the target radio frequency fingerprint feature and the category center of the current monitored object category calculated according to the corresponding distance calculation method.

[0138] Determine the target weight reassembly under the current monitoring object category, and according to the target weight reassembly, sum the current category distances of the target radio frequency fingerprint feature under each distance calculation method to obtain the category center distance of the target radio frequency fingerprint feature under the current monitoring object category; wherein, a weight reassembly under a monitoring object category includes the weight values ​​of each distance calculation method under the corresponding monitoring object category;

[0139] After traversing through each monitoring object category in the set of monitoring object categories, the category center distance of the target radio frequency fingerprint feature under each monitoring object category is obtained.

[0140] In another embodiment, when the processing unit 802 determines the target category probability of the target radio frequency fingerprint feature under each monitoring object category using the tail probability of the target radio frequency fingerprint feature under each monitoring object category, it may specifically be used to:

[0141] The tail probability of the target radio frequency fingerprint feature under each monitoring object category is used to determine the corrected category probability of the target radio frequency fingerprint feature under each monitoring object category.

[0142] The target normalization function is invoked to convert the corrected category probability of the target radio frequency fingerprint feature under each monitoring object category into the target category probability of the target radio frequency fingerprint feature under each monitoring object category.

[0143] In another embodiment, the acquisition unit 801 can also be used for:

[0144] Obtain a set of training monitoring object identity signals. One training monitoring object identity signal corresponds to one existing monitoring object category. The set of training monitoring object identity signals includes multiple training monitoring object identity signals under each monitoring object category.

[0145] Processing unit 802 can also be used for:

[0146] Feature extraction is performed on each training monitoring object identity signal in the training monitoring object identity signal set to obtain the training radio frequency fingerprint features of each training monitoring object identity signal, so as to obtain each training radio frequency fingerprint feature under each monitoring object category; wherein, a training radio frequency fingerprint feature under a monitoring object category is: the training radio frequency fingerprint feature of a training monitoring object identity signal belonging to the corresponding monitoring object category in the training monitoring object identity signal set;

[0147] For any monitoring object category in the set of monitoring object categories, an initial weighting reassembly is determined for that monitoring object category. Based on the category center of that monitoring object category, each training radio frequency fingerprint feature under that monitoring object category, and the initial weighting reassembly, the category center distance of each training radio frequency fingerprint feature under that monitoring object category is calculated respectively. The initial weighting reassembly for that monitoring object category includes the initial weight value of each distance calculation method among multiple distance calculation methods under that monitoring object category.

[0148] Based on the class center distance of each training radio frequency fingerprint feature under any monitoring object category, the class center distance of each selected radio frequency fingerprint feature among multiple selected radio frequency fingerprint features is determined, and based on the class center distance of each selected radio frequency fingerprint feature, the target fitting distribution evaluation result is determined.

[0149] The initial weighted reassembly under any of the monitored object categories is optimized to minimize the target fitting distribution evaluation result until the fitting convergence condition is met, thereby obtaining the target weighted reassembly under any of the monitored object categories; wherein, the target weighted reassembly under any of the monitored object categories is used to calculate the class center distance of the target radio frequency fingerprint feature under any of the monitored object categories.

[0150] In another implementation, the class center distance of a selected RF fingerprint feature is greater than the class center distance of any training RF fingerprint feature other than the selected RF fingerprint features among all training RF fingerprint features under any monitoring object category; when determining the target fitting distribution evaluation result based on the class center distances of the selected RF fingerprint features, the processing unit 802 can specifically be used for:

[0151] The initial Cauchy distribution function for any monitored object category is obtained by fitting the class center distance of each of the multiple screened radio frequency fingerprint features to the Cauchy distribution.

[0152] Using the initial Cauchy distribution function under any of the monitored object categories, the distance fitting value of each of the selected radio frequency fingerprint features is determined;

[0153] Based on the category center distance and distance fitting value of each of the selected radio frequency fingerprint features, the target fitting distribution evaluation result is determined; wherein, the cumulative Cauchy probability distribution of the target under any monitoring object category is the cumulative Cauchy probability distribution corresponding to the target weighted reassembly under any monitoring object category.

[0154] In another embodiment, the target category probability of the target radio frequency fingerprint feature under each monitored object category is determined by calling a target normalization function, and the acquisition unit 801 can also be used for:

[0155] Obtain a set of test monitoring object identity signals, which includes an open set of test categories and a closed set of test categories. The test monitoring object identity signals in the open set of test categories belong to unknown categories, and the test monitoring object identity signals in the closed set of test categories belong to known monitoring object categories.

[0156] Processing unit 802 can also be used for:

[0157] Feature extraction is performed on each test monitoring object identity signal in the set of test monitoring object identity signals to obtain the test radio frequency fingerprint features of each test monitoring object identity signal;

[0158] Initialize the current normalization function, and determine the maximum confidence level of the identity signal of each test monitoring object based on the test radio frequency fingerprint features of the identity signal of each test monitoring object and the current normalization function; wherein, a normalization function includes a temperature parameter;

[0159] Based on the maximum confidence level of the identity signals of each test monitoring object, determine the maximum confidence level statistics of the open set of the test category and the maximum confidence level statistics of the closed set of the test category; and based on the maximum confidence level statistics of the open set of the test category and the maximum confidence level statistics of the closed set of the test category, determine whether the target probability distribution conditions are met.

[0160] If the target probability distribution condition is met, the current normalization function is used as the target normalization function; if the target probability distribution condition is not met, the temperature parameter in the current normalization function is continuously optimized until the target probability distribution condition is determined to be met.

[0161] In another implementation, the target probability distribution condition is used to indicate that the probability distribution of the open set of the test categories and the probability distribution of the closed set of the test categories meet the separation requirement at a specified confidence threshold; when the processing unit 802 determines whether the target probability distribution condition is met based on the maximum confidence statistical results of the open set of the test categories and the maximum confidence statistical results of the closed set of the test categories, it can be specifically used for:

[0162] Based on the maximum confidence statistical results of the open set of the test category and the specified confidence threshold, the open set probability distribution partitioning information corresponding to the open set of the test category is determined. The open set probability distribution partitioning information supports the relationship between the maximum confidence of the identity signal of the test monitoring object in the open set of the test category and the specified confidence threshold.

[0163] Based on the maximum confidence statistical results of the test category closed set and the specified confidence threshold, the closed set probability distribution partitioning information corresponding to the test category closed set is determined. The closed set probability distribution partitioning information supports the relationship between the maximum confidence of the identity signal of the test monitoring object in the test category closed set and the specified confidence threshold.

[0164] Based on the open set probability distribution partitioning information corresponding to the open set of the test category and the closed set probability distribution partitioning information corresponding to the closed set of the test category, it is determined whether the target probability distribution condition is met, so as to reduce the overlap between the probability distribution of the open set of the test category and the probability distribution of the closed set of the test category through the target probability distribution condition.

[0165] According to one embodiment of the present invention, Figure 8Each unit in the open-collection RFID fingerprint recognition device shown can be individually or entirely merged into one or more other units, or some of the units can be further divided into multiple functionally smaller units. This achieves the same operation without affecting the technical effect of the embodiments of the present invention. The above units are based on logical function division. In practical applications, the function of one unit can be implemented by multiple units, or the function of multiple units can be implemented by one unit. In other embodiments of the present invention, any open-collection RFID fingerprint recognition device may also include other units. In practical applications, these functions can also be implemented with the assistance of other units, and can be implemented collaboratively by multiple units.

[0166] According to another embodiment of the present invention, it is possible to perform operations such as those described above by running on a general-purpose electronic device, such as a computer, which includes processing elements and storage elements such as a central processing unit (CPU), random access memory (RAM), and read-only memory (ROM). Figure 1 or Figure 5 The computer program (including program code) involved in each step of the corresponding method shown, to construct such... Figure 8 The diagram illustrates an open-collector radio frequency fingerprint recognition device and a method for implementing embodiments of the present invention. The computer program may be recorded on, for example, a computer storage medium, loaded onto the aforementioned electronic device via the computer storage medium, and run therein.

[0167] Based on the description of the method and apparatus embodiments above, an exemplary embodiment of the present invention also provides an electronic device, including: at least one processor; and a memory communicatively connected to the at least one processor. The memory stores a computer program executable by the at least one processor, which, when executed by the at least one processor, causes the electronic device to perform the method according to an embodiment of the present invention.

[0168] An exemplary embodiment of the present invention also provides a non-transitory computer-readable storage medium storing a computer program, wherein the computer program, when executed by a computer's processor, is used to cause the computer to perform a method according to an embodiment of the present invention.

[0169] An exemplary embodiment of the present invention also provides a computer program product, including a computer program, wherein, when executed by a computer's processor, the computer program is used to cause the computer to perform a method according to an embodiment of the present invention.

[0170] refer to Figure 9The present invention will now be described in the form of a structural block diagram of an electronic device 900 that can serve as a server or client of the present invention, which is an example of a hardware device that can be applied to various aspects of the present invention. The electronic device is intended to represent various forms of digital electronic computer devices, such as laptop computers, desktop computers, workstations, personal digital assistants, servers, blade servers, mainframe computers, and other suitable computers. The electronic device can also represent various forms of mobile devices, such as personal digital processors, cellular phones, smartphones, wearable devices, and other similar computing devices. The components shown herein, their connections and relationships, and their functions are merely illustrative and are not intended to limit the implementation of the invention described and / or claimed herein.

[0171] like Figure 9 As shown, the electronic device 900 includes a computing unit 901, which can perform various appropriate actions and processes according to a computer program stored in a read-only memory (ROM) 902 or a computer program loaded from a storage unit 908 into a random access memory (RAM) 903. The RAM 903 may also store various programs and data required for the operation of the electronic device 900. The computing unit 901, ROM 902, and RAM 903 are interconnected via a bus 904. An input / output (I / O) interface 905 is also connected to the bus 904.

[0172] Multiple components in electronic device 900 are connected to I / O interface 905, including: input unit 906, output unit 907, storage unit 908, and communication unit 909. Input unit 906 can be any type of device capable of inputting information to electronic device 900. Input unit 906 can receive input digital or character information and generate key signal inputs related to user settings and / or function control of electronic device. Output unit 907 can be any type of device capable of presenting information and may include, but is not limited to, a display, speaker, video / audio output terminal, vibrator, and / or printer. Storage unit 908 may include, but is not limited to, disk and optical disk. Communication unit 909 allows electronic device 900 to exchange information / data with other devices through computer networks such as the Internet and / or various telecommunications networks, and may include, but is not limited to, modems, network cards, infrared communication devices, wireless communication transceivers and / or chipsets, such as Bluetooth™ devices, WiFi devices, WiMax devices, cellular communication devices, and / or the like.

[0173] The computing unit 901 can be a variety of general-purpose and / or special-purpose processing components with processing and computing capabilities. Some examples of the computing unit 901 include, but are not limited to, a central processing unit (CPU), a graphics processing unit (GPU), various special-purpose artificial intelligence (AI) computing chips, various computing units running machine learning model algorithms, a digital signal processor (DSP), and any suitable processor, controller, microcontroller, etc. The computing unit 901 performs the various methods and processes described above. For example, in some embodiments, the open-collection RFID fingerprinting method can be implemented as a computer software program tangibly contained in a machine-readable medium, such as storage unit 908. In some embodiments, part or all of the computer program can be loaded and / or installed on the electronic device 900 via ROM 902 and / or communication unit 909. In some embodiments, the computing unit 901 can be configured to perform the open-collection RFID fingerprinting method by any other suitable means (e.g., by means of firmware).

[0174] The program code used to implement the methods of the present invention can be written in any combination of one or more programming languages. This program code can be provided to a processor or controller of a general-purpose computer, special-purpose computer, or other programmable data processing device, such that when executed by the processor or controller, the program code causes the functions / operations specified in the flowcharts and / or block diagrams to be implemented. The program code can be executed entirely on the machine, partially on the machine, as a standalone software package partially on the machine and partially on a remote machine, or entirely on a remote machine or server.

[0175] In the context of this invention, a machine-readable medium can be a tangible medium that may contain or store a program for use by or in conjunction with an instruction execution system, apparatus, or device. A machine-readable medium can be a machine-readable signal medium or a machine-readable storage medium. Machine-readable media can include, but are not limited to, electronic, magnetic, optical, electromagnetic, infrared, or semiconductor systems, apparatus, or devices, or any suitable combination of the foregoing. More specific examples of machine-readable storage media include electrical connections based on one or more wires, portable computer disks, hard disks, random access memory (RAM), read-only memory (ROM), erasable programmable read-only memory (EPROM or flash memory), optical fibers, portable compact disk read-only memory (CD-ROM), optical storage devices, magnetic storage devices, or any suitable combination of the foregoing.

[0176] As used herein, the terms "machine-readable medium" and "computer-readable medium" refer to any computer program product, device, and / or apparatus (e.g., disk, optical disk, memory, programmable logic device (PLD)) for providing machine instructions and / or data to a programmable processor, including machine-readable media that receive machine instructions as machine-readable signals. The term "machine-readable signal" refers to any signal for providing machine instructions and / or data to a programmable processor.

[0177] To provide interaction with a user, the systems and techniques described herein can be implemented on a computer having: a display device for displaying information to the user (e.g., a CRT (cathode ray tube) or LCD (liquid crystal display) monitor); and a keyboard and pointing device (e.g., a mouse or trackball) through which the user provides input to the computer. Other types of devices can also be used to provide interaction with the user; for example, feedback provided to the user can be any form of sensory feedback (e.g., visual feedback, auditory feedback, or tactile feedback); and input from the user can be received in any form (including sound input, voice input, or tactile input).

[0178] The systems and technologies described herein can be implemented in computing systems that include backend components (e.g., as a data server), or computing systems that include middleware components (e.g., an application server), or computing systems that include frontend components (e.g., a user computer with a graphical user interface or web browser through which a user can interact with implementations of the systems and technologies described herein), or any combination of such backend, middleware, or frontend components. The components of the system can be interconnected via digital data communication of any form or medium (e.g., a communication network). Examples of communication networks include local area networks (LANs), wide area networks (WANs), and the Internet.

[0179] Computer systems can include clients and servers. Clients and servers are generally located far apart and typically interact through communication networks. Client-server relationships are created by computer programs running on the respective computers and having a client-server relationship with each other.

[0180] Furthermore, it should be understood that the above-disclosed embodiments are merely preferred embodiments of the present invention and should not be construed as limiting the scope of the present invention. Therefore, any equivalent variations made in accordance with the claims of the present invention are still within the scope of the present invention.

Claims

1. An open-set radio frequency fingerprint recognition method, characterized in that, include: The identity signal of the target monitoring object is acquired, and the target monitoring object identity signal is subjected to feature extraction to obtain the target radio frequency fingerprint feature of the target monitoring object identity signal; Determine the category center of each monitoring object category in the set of monitoring object categories, and calculate the category center distance of the target radio frequency fingerprint feature under each monitoring object category based on the category center of each monitoring object category; Based on the class center distance and the cumulative distribution of the target Cauchy probability of the target RF fingerprint feature in each monitoring object category, respectively, the tail probability of the target RF fingerprint feature in each monitoring object category is calculated; wherein, the tail probability of the target RF fingerprint feature in a monitoring object category is calculated by substituting the class center distance of the target RF fingerprint feature in the corresponding monitoring object category into the cumulative distribution of the target Cauchy probability of the corresponding monitoring object category; The target category probability of the target RF fingerprint feature under each monitoring object category is determined by using the tail probability of the target RF fingerprint feature under each monitoring object category; and the RF fingerprint recognition result of the target RF fingerprint feature is determined based on the target category probability of the target RF fingerprint feature under each monitoring object category, and one RF fingerprint recognition result can be used to indicate whether it is an existing monitoring object category; The step of determining the target category probability of the target radio frequency fingerprint feature under each monitoring object category by using the tail probability of the target radio frequency fingerprint feature under each monitoring object category includes: The tail probability of the target radio frequency fingerprint feature under each monitoring object category is used respectively to determine the corrected category probability of the target radio frequency fingerprint feature under each monitoring object category, including: for any monitoring object category in the monitoring object category set, the difference between the preset probability value and the tail probability of the target radio frequency fingerprint feature under any monitoring object category is used as the corrected category probability of the target radio frequency fingerprint feature under any monitoring object category; The target normalization function is invoked to convert the corrected category probability of the target radio frequency fingerprint feature under each monitoring object category into the target category probability of the target radio frequency fingerprint feature under each monitoring object category.

2. The method according to claim 1, characterized in that, The step of calculating the class center distance of the target radio frequency fingerprint feature under each monitoring object category based on the class center of each monitoring object category includes: Iterate through each monitoring object category in the set of monitoring object categories, and take the currently traversed monitoring object category as the current monitoring object category; Based on the category center of the current monitored object category and the target radio frequency fingerprint feature, the current category distance of the target radio frequency fingerprint feature under each of the multiple distance calculation methods is calculated. The current category distance of the target radio frequency fingerprint feature under a distance calculation method refers to the distance between the target radio frequency fingerprint feature and the category center of the current monitored object category calculated according to the corresponding distance calculation method. Determine the target weight reassembly under the current monitoring object category, and according to the target weight reassembly, sum the current category distances of the target radio frequency fingerprint feature under each distance calculation method to obtain the category center distance of the target radio frequency fingerprint feature under the current monitoring object category; wherein, a weight reassembly under a monitoring object category includes the weight values ​​of each distance calculation method under the corresponding monitoring object category; After traversing through each monitoring object category in the set of monitoring object categories, the category center distance of the target radio frequency fingerprint feature under each monitoring object category is obtained.

3. The method according to claim 1 or 2, characterized in that, The method further includes: Obtain a set of training monitoring object identity signals. One training monitoring object identity signal corresponds to one existing monitoring object category. The set of training monitoring object identity signals includes multiple training monitoring object identity signals under each monitoring object category. Feature extraction is performed on each training monitoring object identity signal in the training monitoring object identity signal set to obtain the training radio frequency fingerprint features of each training monitoring object identity signal, so as to obtain each training radio frequency fingerprint feature under each monitoring object category; wherein, a training radio frequency fingerprint feature under a monitoring object category is: the training radio frequency fingerprint feature of a training monitoring object identity signal belonging to the corresponding monitoring object category in the training monitoring object identity signal set; Determine the initial weighting reassembly under any monitoring object category, and calculate the category center distance of each training radio frequency fingerprint feature under any monitoring object category based on the category center of any monitoring object category, each training radio frequency fingerprint feature under any monitoring object category, and the initial weighting reassembly; wherein, the initial weighting reassembly under any monitoring object category includes the initial weight value of each distance calculation method among multiple distance calculation methods under any monitoring object category; Based on the class center distance of each training radio frequency fingerprint feature under any monitoring object category, the class center distance of each selected radio frequency fingerprint feature among multiple selected radio frequency fingerprint features is determined, and based on the class center distance of each selected radio frequency fingerprint feature, the target fitting distribution evaluation result is determined. The initial weighted reassembly under any of the monitored object categories is optimized to minimize the target fitting distribution evaluation result until the fitting convergence condition is met, thereby obtaining the target weighted reassembly under any of the monitored object categories; wherein, the target weighted reassembly under any of the monitored object categories is used to calculate the class center distance of the target radio frequency fingerprint feature under any of the monitored object categories.

4. The method according to claim 3, characterized in that, The class center distance of a selected radio frequency fingerprint feature is greater than the class center distance of any training radio frequency fingerprint feature other than the selected radio frequency fingerprint features among all training radio frequency fingerprint features under any monitoring object category; the determination of the target fitting distribution evaluation result based on the class center distances of the selected radio frequency fingerprint features includes: The initial Cauchy distribution function for any monitored object category is obtained by fitting the class center distance of each of the multiple screened radio frequency fingerprint features to the Cauchy distribution. Using the initial Cauchy distribution function under any of the monitored object categories, the distance fitting value of each of the selected radio frequency fingerprint features is determined; Based on the category center distance and distance fitting value of each of the selected radio frequency fingerprint features, the target fitting distribution evaluation result is determined; wherein, the cumulative Cauchy probability distribution of the target under any monitoring object category is the cumulative Cauchy probability distribution corresponding to the target weighted reassembly under any monitoring object category.

5. The method according to claim 1 or 2, characterized in that, The target category probability of the target radio frequency fingerprint feature under each monitored object category is determined by calling a target normalization function, and the method further includes: Obtain a set of test monitoring object identity signals, which includes an open set of test categories and a closed set of test categories. The test monitoring object identity signals in the open set of test categories belong to unknown categories, and the test monitoring object identity signals in the closed set of test categories belong to known monitoring object categories. Feature extraction is performed on each test monitoring object identity signal in the set of test monitoring object identity signals to obtain the test radio frequency fingerprint features of each test monitoring object identity signal; Initialize the current normalization function, and determine the maximum confidence level of the identity signal of each test monitoring object based on the test radio frequency fingerprint features of the identity signal of each test monitoring object and the current normalization function; wherein, a normalization function includes a temperature parameter; Based on the maximum confidence level of the identity signals of each test monitoring object, determine the maximum confidence level statistics of the open set of the test category and the maximum confidence level statistics of the closed set of the test category; and based on the maximum confidence level statistics of the open set of the test category and the maximum confidence level statistics of the closed set of the test category, determine whether the target probability distribution conditions are met. If the target probability distribution condition is met, the current normalization function is used as the target normalization function; if the target probability distribution condition is not met, the temperature parameter in the current normalization function is continuously optimized until the target probability distribution condition is determined to be met.

6. The method according to claim 5, characterized in that, The target probability distribution condition is used to indicate that the probability distribution of the open set of the test category and the probability distribution of the closed set of the test category meet the separation requirement at a specified confidence threshold; the step of determining whether the target probability distribution condition is met based on the maximum confidence statistical results of the open set of the test category and the maximum confidence statistical results of the closed set of the test category includes: Based on the maximum confidence statistical results of the open set of the test category and the specified confidence threshold, the open set probability distribution partitioning information corresponding to the open set of the test category is determined. The open set probability distribution partitioning information supports the relationship between the maximum confidence of the identity signal of the test monitoring object in the open set of the test category and the specified confidence threshold. Based on the maximum confidence statistical results of the test category closed set and the specified confidence threshold, the closed set probability distribution partitioning information corresponding to the test category closed set is determined. The closed set probability distribution partitioning information supports the relationship between the maximum confidence of the identity signal of the test monitoring object in the test category closed set and the specified confidence threshold. Based on the open set probability distribution partitioning information corresponding to the open set of the test category and the closed set probability distribution partitioning information corresponding to the closed set of the test category, it is determined whether the target probability distribution condition is met, so as to reduce the overlap between the probability distribution of the open set of the test category and the probability distribution of the closed set of the test category through the target probability distribution condition.

7. An open-collector radio frequency fingerprint recognition device, characterized in that, The device includes: The acquisition unit is used to acquire the identity signal of the target monitored object; The processing unit is used to extract features from the identity signal of the target monitoring object to obtain the target radio frequency fingerprint feature of the identity signal of the target monitoring object; The processing unit is further configured to determine the category center of each monitoring object category in the set of monitoring object categories, and calculate the category center distance of the target radio frequency fingerprint feature under each monitoring object category based on the category center of each monitoring object category. The processing unit is further configured to calculate the tail probability of the target radio frequency fingerprint feature in each monitoring object category based on the category center distance of the target radio frequency fingerprint feature in each monitoring object category and the target Cauchy probability cumulative distribution in each monitoring object category; wherein, the tail probability of the target radio frequency fingerprint feature in a monitoring object category is calculated by substituting the category center distance of the target radio frequency fingerprint feature in the corresponding monitoring object category into the target Cauchy probability cumulative distribution in the corresponding monitoring object category; The processing unit is further configured to determine the target category probability of the target radio frequency fingerprint feature under each monitoring object category by using the tail probability of the target radio frequency fingerprint feature under each monitoring object category respectively; and to determine the radio frequency fingerprint recognition result of the target radio frequency fingerprint feature based on the target category probability of the target radio frequency fingerprint feature under each monitoring object category, wherein a radio frequency fingerprint recognition result can be used to indicate whether it is an existing monitoring object category; Specifically, when the processing unit determines the target category probability of the target radio frequency fingerprint feature under each monitoring object category using the tail probability of the target radio frequency fingerprint feature under each monitoring object category, it is used to: The tail probability of the target radio frequency fingerprint feature under each monitoring object category is used respectively to determine the corrected category probability of the target radio frequency fingerprint feature under each monitoring object category, including: for any monitoring object category in the monitoring object category set, the difference between the preset probability value and the tail probability of the target radio frequency fingerprint feature under any monitoring object category is used as the corrected category probability of the target radio frequency fingerprint feature under any monitoring object category; The target normalization function is invoked to convert the corrected category probability of the target radio frequency fingerprint feature under each monitoring object category into the target category probability of the target radio frequency fingerprint feature under each monitoring object category.

8. An electronic device, characterized in that, include: processor; as well as Stored program memory, The program includes instructions that, when executed by the processor, cause the processor to perform the method according to any one of claims 1-6.

9. A non-transitory computer-readable storage medium storing computer instructions, wherein, The computer instructions are used to cause the computer to perform the method according to any one of claims 1-6.

Images