Radar emitter data label screening method based on uncertainty threshold detection
By constructing a learning model for pre-training and uncertainty threshold detection, incorrect labels are filtered out, solving the problem of reduced accuracy caused by incorrect labeling in deep learning, and improving the accuracy of radar radiation source identification and the model's generalization ability.
Patent Information
- Application Number
- CN202410133570.1
- Authority / Receiving Office
- CN · China
- Patent Type
- Patents(China)
- Current Assignee / Owner
- Filing Date
- 2024-01-31
- Publication Date
- 2026-01-13
- Estimated Expiration
- 2044-01-31
AI Technical Summary
In existing deep learning-based radar radiation source individual identification technologies, mislabeling is common, which leads to reduced accuracy of the trained model on clean test data. Existing methods have failed to effectively handle mislabeling.
A radar radiation source data labeling method based on uncertainty threshold detection is adopted. The method pre-trains a learning model, extracts signal features using bispectral extraction and learns radiation source data features using multi-scale dilated convolution, and combines channel and spatial attention mechanisms to filter out erroneous labels and perform uncertainty threshold detection and correction.
It improves the model's performance and generalization ability, effectively suppresses Gaussian noise interference, ensures the accuracy and quality of training data, and enhances the model's recognition accuracy under mislabeled conditions.
Smart Images
Figure CN118410333B_ABST
Abstract
Description
Technical Field
[0001] This application relates to the field of specific radiation source identification technology, and in particular to a method and apparatus for screening radar radiation source data tags based on uncertainty threshold detection. Background Technology
[0002] Specific Emitter Identification (SEI) technology acquires subtle features of the transmitter by measuring the external characteristics of the received electromagnetic signals. These features are then analyzed as a radar fingerprint for authentication, ultimately determining the radar carrier's identity. SEI technology is a primary basis for situational assessment in high-level information fusion, providing effective technical support for situational awareness and action planning and decision-making. Currently, deep learning-based radar emitter identification technology is a research hotspot in recent years, but its application under mislabeling conditions is still in its early stages.
[0003] In deep learning-based methods for identifying radiation sources, it is generally assumed that the labels on the training data are correctly labeled. However, in reality, data labeling is a complex process. The skill level of the operator and the accuracy of the classification algorithm can all affect the labeling, so incorrect labeling is common. In deep neural networks, using high-quality labeled training data is crucial for learning effectiveness because incorrect labels (noisy labels) in the training data will significantly reduce the accuracy of the model trained on clean test data. Summary of the Invention
[0004] This application aims to at least partially address one of the technical problems in the related art.
[0005] Therefore, the first objective of this application is to propose a radar radiation source data label filtering method based on uncertainty threshold detection, which solves the technical problem that existing methods do not process erroneous labels, achieves accurate filtering of erroneous labels, and further ensures the accuracy of training models on test data.
[0006] The second objective of this application is to propose a radar radiation source data tag screening device based on uncertainty threshold detection.
[0007] To achieve the above objectives, the first aspect of this application proposes a radar radiation source data labeling method based on uncertainty threshold detection, comprising: constructing a learning model and pre-training the learning model, wherein, during the pre-training of the learning model, signal features are extracted through bispectral extraction and feature representations of radiation source data are learned through multi-scale dilated convolution; the pre-trained learning model is used to predict labels for the radar radiation source data to be detected, thereby obtaining predicted labels; predicted labels that are inconsistent with the original labels are removed from the predicted labels, and uncertainty threshold detection correction is performed on the remaining predicted labels, with predicted labels that meet the threshold range being retained as data.
[0008] The radar radiation source data label screening method based on uncertainty threshold detection in this application uses a pre-trained model for label prediction. During label prediction, the bispectral data effectively reflects the amplitude distribution characteristics of the radiation source signal and can effectively suppress Gaussian noise interference. Since some bispectral feature maps exhibit peaks, an attention mechanism is used to guide the neural network to focus on relevant parts, automatically learning and selectively focusing on important information in the input, thereby improving the model's performance and generalization. At the same time, since the fingerprint features of radar radiation sources are relatively subtle, multi-scale dilated convolution is used to capture features at different scales. Combined with channel and spatial attention mechanisms, conflicts in multi-scale dilated feature fusion are suppressed, preventing subtle features from being submerged in the fusion. Furthermore, the uncertainty threshold detection method is used to screen predicted labels, making it more flexible in screening incorrect labels.
[0009] Optionally, in one embodiment of this application, signal features are extracted using bispectral extraction, including:
[0010] A two-dimensional dataset is constructed by dividing the real and imaginary parts of the one-dimensional feature vector into two paths, I and Q. When constructing the dataset, bispectral features are extracted and features are constructed by bispectral slicing.
[0011] The learning model includes multi-scale dilated convolutional layers, convolutional layers, channel attention modules, spatial attention modules, feature fusion layers, ReLU layers, max-pooling layers, and fully connected layers.
[0012] Optionally, in one embodiment of this application, if the signal is x k The bispectral expression is:
[0013]
[0014] Among them, B x (w1, w2) represents the bispectral density, w1 and w2 represent frequencies, τ1 and τ2 represent time delays, and C 3x This represents the third-order time delay moment.
[0015] Optionally, in one embodiment of this application, the predicted label is obtained by using a pre-trained learning model to predict the label of the data to be detected, including:
[0016] The model parameters of the pre-trained learning model are used to predict the labels of the data to be detected, and the prediction probability of each type of the data to be detected is determined.
[0017] The label corresponding to the highest predicted probability is used as the predicted label.
[0018] Optionally, in one embodiment of this application, the predicted probability is expressed as:
[0019] p i (j)=pred(h(θ),x),j=0,1,2,N-1
[0020] Where, p i (j) represents the predicted probability of sample i when the data type is j, h(θ) represents the parameters of the learning model, x represents the unlabeled data to be detected, i represents the i-th sample in the data to be detected, j represents the data type included in the learning model, and pred() represents the prediction model.
[0021] The label corresponding to the highest predicted probability is represented as:
[0022] m = max(p i m∈{0,1,…N-1}
[0023] Where, p i This represents all predicted probabilities for sample i.
[0024] Optionally, in one embodiment of this application, the elimination process is represented as follows:
[0025]
[0026] Where m is the label corresponding to the maximum value, and n represents the label of the data to be detected;
[0027] The process of uncertainty threshold detection and correction is expressed as follows:
[0028]
[0029] Among them, y (i) Indicates the label, u(p) (i) ) represents the variance calculation of the predicted probability, k represents the uncertainty threshold, and λ is the threshold for the predicted probability.
[0030] To achieve the above objectives, a second aspect of the present invention provides a radar radiation source data labeling device based on uncertainty threshold detection, comprising a model training module, a label prediction module, and a label filtering module, wherein...
[0031] The model building module is used to build a learning model and pre-train the learning model. During the pre-training of the learning model, signal features are extracted through bispectral extraction, and feature representations of radiation source data are learned through multi-scale dilated convolution.
[0032] The label prediction module is used to predict the labels of radar radiation source data to be detected by a pre-trained learning model.
[0033] The label filtering module is used to remove predicted labels that are inconsistent with the original labels from the predicted labels, and to perform uncertainty threshold detection and correction on the remaining predicted labels, and to retain the predicted labels that meet the threshold range as the retained data.
[0034] Optionally, in one embodiment of this application, signal features are extracted using bispectral extraction, including:
[0035] A two-dimensional dataset is constructed by dividing the real and imaginary parts of the one-dimensional feature vector into two paths, I and Q. When constructing the dataset, bispectral features are extracted and features are constructed by bispectral slicing.
[0036] The learning model includes multi-scale dilated convolutional layers, convolutional layers, channel attention modules, spatial attention modules, feature fusion layers, ReLU layers, max-pooling layers, and fully connected layers.
[0037] Optionally, in one embodiment of this application, a predicted label is obtained by learning a model to predict the label of the label to be detected, including:
[0038] By learning the model parameters of the model, label prediction is performed on the data to be detected, and the prediction probability of each type in the set is determined.
[0039] The label with the highest predicted probability is determined as the predicted label.
[0040] Optionally, in one embodiment of this application, the elimination process is represented as follows:
[0041]
[0042] Where m is the label corresponding to the maximum value, and n represents the label of the data to be detected;
[0043] The process of uncertainty threshold detection and correction is expressed as follows:
[0044]
[0045] Among them, y (i) Indicates the label, u(p) (i) ) represents the variance calculation of the predicted probability, k represents the uncertainty threshold, and λ is the threshold for the predicted probability.
[0046] Additional aspects and advantages of this application will be set forth in part in the description which follows, and in part will be obvious from the description, or may be learned by practice of this application. Attached Figure Description
[0047] Figure 1 This is a flowchart illustrating a radar radiation source data tag filtering method based on uncertainty threshold detection provided in Embodiment 1 of this application.
[0048] Figure 2 This is a diagram of the pre-trained model structure according to an embodiment of this application;
[0049] Figure 3 This is a schematic diagram of a radar radiation source data tag screening device based on uncertainty threshold detection, provided in an embodiment of this application. Detailed Implementation
[0050] The embodiments of this application are described in detail below. Examples of these embodiments are shown in the accompanying drawings, wherein the same or similar reference numerals denote the same or similar elements or elements having the same or similar functions throughout. The embodiments described below with reference to the accompanying drawings are exemplary and intended to explain this application, and should not be construed as limiting this application.
[0051] The radar radiation source data tag filtering method and apparatus based on uncertainty threshold detection according to embodiments of this application are described below with reference to the accompanying drawings.
[0052] Figure 1 This is a flowchart illustrating a radar radiation source data tag filtering method based on uncertainty threshold detection, as provided in Embodiment 1 of this application.
[0053] like Figure 1 As shown, the radar radiation source data labeling method based on uncertainty threshold detection includes the following steps:
[0054] Step 101: Construct a learning model and pre-train it. During the pre-training process, the learning model extracts signal features through bispectral extraction and learns the feature representation of radiation source data through multi-scale dilated convolution.
[0055] Specifically, in the pre-training stage, a neural network learning model is constructed using bispectral feature extraction and multi-scale dilated convolution, and the pre-trained model parameters are obtained.
[0056] Specifically, in the pre-training stage, the one-dimensional feature vector is usually divided into two paths, I and Q, to construct a two-dimensional dataset. Bispectral feature extraction is used for feature extraction, and considering the subsequent computational cost, bispectral slicing is used to construct features.
[0057] Specifically, during the pre-training phase, feature representations of the radiation source dataset are learned through multi-scale dilated convolutions based on bispectral feature sequences followed by IQ data streams, which facilitates subsequent prediction and uncertainty threshold detection.
[0058] Step 102: Use the pre-trained learning model to predict the labels of the radar radiation source data to be detected, and obtain the predicted labels.
[0059] Specifically, in the label prediction stage, the label is predicted based on the parameters of the pre-trained model to be detected.
[0060] Step 103: Remove predicted labels that are inconsistent with the original labels from the predicted labels, and perform uncertainty threshold detection and correction on the remaining predicted labels. Predicted labels that meet the threshold range are retained as data.
[0061] Specifically, in the stage of removing incorrectly labeled data, if the original label of the data to be detected is consistent with the predicted label, the predicted label is corrected by uncertainty threshold detection. If the data meets the threshold range, the data is retained; otherwise, the data is discarded. If the original label of the data to be detected is inconsistent with the predicted label, the data is discarded directly.
[0062] The radar radiation source data label screening method based on uncertainty threshold detection in this application uses a pre-trained model for label prediction. During label prediction, the bispectral data effectively reflects the amplitude distribution characteristics of the radiation source signal and can effectively suppress Gaussian noise interference. Since some bispectral feature maps exhibit peaks, an attention mechanism is used to guide the neural network to focus on relevant parts, automatically learning and selectively focusing on important information in the input, thereby improving the model's performance and generalization. At the same time, since the fingerprint features of radar radiation sources are relatively subtle, multi-scale dilated convolution is used to capture features at different scales. Combined with channel and spatial attention mechanisms, conflicts in multi-scale dilated feature fusion are suppressed, preventing subtle features from being submerged in the fusion. Furthermore, the uncertainty threshold detection method is used to screen predicted labels, making it more flexible in screening incorrect labels.
[0063] Optionally, in one embodiment of this application, signal features are extracted using bispectral extraction, including:
[0064] A two-dimensional dataset is constructed by dividing the real and imaginary parts of the one-dimensional feature vector into two paths, I and Q. When constructing the dataset, bispectral features are extracted and features are constructed by bispectral slicing.
[0065] Figure 2 This is a diagram of the pre-trained model structure in this embodiment, as shown below. Figure 2 As shown, the learning model includes multi-scale dilated convolutional layers, convolutional layers, channel attention modules, spatial attention modules, feature fusion layers, ReLU layers, max-pooling layers, and fully connected layers;
[0066] In the neural network learning model, considering the characteristics of feature vectors, a multi-scale dilated convolution method is adopted, with dilation rates of 1, 2, and 5 to satisfy the visual perception of signals at different dimensions. First, channel attention modules and spatial attention modules are used in the neural network to obtain attention mechanisms at different scales to perceive the visual field. Then, feature fusion is performed, and finally, the final recognition result is obtained through convolution, ReLU layers, max-pooling layers, and fully connected layers, forming the pre-trained model parameters h(θ).
[0067] A learning model is pre-trained on the dataset to extract feature representations of radiation source signals. After pre-training, label prediction is performed on the data to be detected. Deterministic threshold detection is then performed on the data based on whether the predicted label and the original labeled label are the same, thus completing the screening of erroneous labels in the dataset and further improving the overall data quality of the dataset.
[0068] Optionally, in one embodiment of this application, bispectral extraction of signal features is employed. Assuming the signal is xk, the bispectral expression can be represented as:
[0069]
[0070] Among them, B x (w1, w2) represents the bispectral density, w1 and w2 represent frequencies, τ1 and τ2 represent time delays, and C 3x This represents the third-order time delay moment.
[0071] Since the above expression has a high dimension, it is not convenient for subsequent processing. Therefore, the diagonal of the above expression is taken as the input feature vector of the neural network, and the real and imaginary parts of the feature vector are extracted to construct a two-dimensional dataset to match the subsequent neural network.
[0072] Optionally, in one embodiment of this application, the predicted label is obtained by using a pre-trained learning model to predict the label of the data to be detected, including:
[0073] The model parameters of the pre-trained learning model are used to predict the labels of the data to be detected, and the prediction probability of each type of the data to be detected is determined.
[0074] The label corresponding to the highest predicted probability is used as the predicted label.
[0075] Optionally, in one embodiment of this application, the predicted probability is expressed as:
[0076] p i (j)=pred(h(θ),x),j=0,1,2,N-1
[0077] Where, p i (j) represents the predicted probability of sample i when the data type is j, h(θ) represents the parameters of the learning model, x represents the unlabeled data to be detected, i represents the i-th sample in the data to be detected, j represents the data type included in the learning model, and pred() represents the prediction model.
[0078] The label corresponding to the highest predicted probability is represented as:
[0079] m = max(p i m∈{0,1,…N-1}
[0080] Where, p i This represents all predicted probabilities for sample i.
[0081] Optionally, in one embodiment of this application, if the predicted label m is different from the label n of the data to be detected, where n = 0, 1, ..., N-1, the data to be detected is directly discarded and not used for subsequent training. Specifically:
[0082]
[0083] Where m is the label corresponding to the maximum value, and n represents the label of the data to be detected;
[0084] If the predicted label m is the same as the label n of the data to be detected, where n = 0, 1, ..., N-1, then uncertainty threshold detection is performed on the data, specifically as follows:
[0085]
[0086] Among them, y (i) Indicates the label, u(p) (i) ) represents the variance calculation of the predicted probability, k represents the uncertainty threshold, and λ is the threshold for the predicted probability.
[0087] To achieve the above embodiments, this application also proposes a radar radiation source data tag screening device based on uncertainty threshold detection.
[0088] Figure 3 This is a schematic diagram of a radar radiation source data tag screening device based on uncertainty threshold detection, provided in an embodiment of this application.
[0089] like Figure 3As shown, the radar radiation source data labeling device based on uncertainty threshold detection includes a model training module, a label prediction module, and a label filtering module.
[0090] The model building module is used to build a learning model and pre-train the learning model. During the pre-training of the learning model, signal features are extracted through bispectral extraction, and feature representations of radiation source data are learned through multi-scale dilated convolution.
[0091] The label prediction module is used to predict the labels of radar radiation source data to be detected by a pre-trained learning model.
[0092] The label filtering module is used to remove predicted labels that are inconsistent with the original labels from the predicted labels, and to perform uncertainty threshold detection and correction on the remaining predicted labels, and to retain the predicted labels that meet the threshold range as the retained data.
[0093] Optionally, in one embodiment of this application, signal features are extracted using bispectral extraction, including:
[0094] A two-dimensional dataset is constructed by dividing the real and imaginary parts of the one-dimensional feature vector into two paths, I and Q. When constructing the dataset, bispectral features are extracted and features are constructed by bispectral slicing.
[0095] The learning model includes multi-scale dilated convolutional layers, convolutional layers, channel attention modules, spatial attention modules, feature fusion layers, ReLU layers, max-pooling layers, and fully connected layers.
[0096] Optionally, in one embodiment of this application, a predicted label is obtained by learning a model to predict the label of the label to be detected, including:
[0097] By learning the model parameters of the model, label prediction is performed on the data to be detected, and the prediction probability of each type in the set is determined.
[0098] The label with the highest predicted probability is determined as the predicted label.
[0099] Optionally, in one embodiment of this application, the elimination process is represented as follows:
[0100]
[0101] Where m is the label corresponding to the maximum value, and n represents the label of the data to be detected;
[0102] The process of uncertainty threshold detection and correction is expressed as follows:
[0103]
[0104] Among them, y(i) Indicates the label, u(p) (i) ) represents the variance calculation of the predicted probability, k represents the uncertainty threshold, and λ is the threshold for the predicted probability.
[0105] It should be noted that the foregoing explanation of the radar radiation source data tag screening method based on uncertainty threshold detection also applies to the radar radiation source data tag screening device based on uncertainty threshold detection in this embodiment, and will not be repeated here.
[0106] Any process or method description in the flowchart or otherwise herein can be understood as representing a module, segment, or portion of code comprising one or more executable instructions for implementing custom logic functions or processes, and the scope of the preferred embodiments of this application includes additional implementations in which functions may be performed not in the order shown or discussed, including substantially simultaneously or in reverse order depending on the functions involved, as should be understood by those skilled in the art to which embodiments of this application pertain.
[0107] The logic and / or steps represented in the flowchart or otherwise described herein, for example, can be considered as a sequenced list of executable instructions for implementing logical functions, and can be embodied in any computer-readable medium for use by, or in conjunction with, an instruction execution system, apparatus, or device (such as a computer-based system, a processor-included system, or other system that can fetch and execute instructions from, an instruction execution system, apparatus, or device). For the purposes of this specification, "computer-readable medium" can be any means that can contain, store, communicate, propagate, or transmit programs for use by, or in conjunction with, an instruction execution system, apparatus, or device. More specific examples (a non-exhaustive list) of computer-readable media include: an electrical connection having one or more wires (electronic device), a portable computer disk drive (magnetic device), random access memory (RAM), read-only memory (ROM), erasable and editable read-only memory (EPROM or flash memory), fiber optic devices, and portable optical disc read-only memory (CDROM). Alternatively, the computer-readable medium may be paper or other suitable media on which the program can be printed, since the program can be obtained electronically, for example, by optically scanning the paper or other medium, followed by editing, interpreting, or otherwise processing as necessary, and then stored in a computer memory.
[0108] It should be understood that various parts of this application can be implemented using hardware, software, firmware, or a combination thereof. In the above embodiments, multiple steps or methods can be implemented using software or firmware stored in memory and executed by a suitable instruction execution system. If implemented in hardware, as in another embodiment, it can be implemented using any one or a combination of the following techniques known in the art: discrete logic circuits having logic gates for implementing logical functions on data signals, application-specific integrated circuits (ASICs) having suitable combinational logic gates, programmable gate arrays (PGAs), field-programmable gate arrays (FPGAs), etc.
[0109] Those skilled in the art will understand that all or part of the steps of the methods in the above embodiments can be implemented by a program instructing related hardware. The program can be stored in a computer-readable storage medium, and when executed, the program includes one or a combination of the steps of the method embodiments.
[0110] Furthermore, the functional units in the various embodiments of this application can be integrated into a processing module, or each unit can exist physically separately, or two or more units can be integrated into a module. The integrated module can be implemented in hardware or as a software functional module. If the integrated module is implemented as a software functional module and sold or used as an independent product, it can also be stored in a computer-readable storage medium.
[0111] The storage medium mentioned above can be a read-only memory, a disk, or an optical disk, etc. Although embodiments of this application have been shown and described above, it is understood that the above embodiments are exemplary and should not be construed as limiting this application. Those skilled in the art can make changes, modifications, substitutions, and variations to the above embodiments within the scope of this application.
Claims
1. A radar emitter data tag screening method based on uncertainty threshold detection, characterized in that, The method comprises the following steps: constructing a learning model and pre-training the learning model, wherein the learning model extracts signal features through bispectrum during pre-training, and learns feature representation of the radar emitter data through multi-scale dilated convolution; predicting labels of the to-be-detected radar emitter data through the pre-trained learning model to obtain predicted labels; eliminating predicted labels inconsistent with original labels from the predicted labels, and correcting the remaining predicted labels through uncertainty threshold detection, and taking predicted labels meeting a threshold range as reserved data; the signal feature extraction through bispectrum comprises: constructing a two-dimensional data set for the real part and the imaginary part of a one-dimensional feature vector as IQ two paths, and extracting features through bispectrum during construction of the data set, and constructing features in a bispectrum slice manner; the learning model comprises a multi-scale dilated convolution layer, a convolution layer, a channel attention module, a spatial attention module, a feature fusion layer, a RELU layer, a Max-pooling layer, and a full connection layer; The signal is The bispectrum expression is ； wherein, denotes bispectrum, and denotes frequency, and denotes time delay, denotes third-order time delay moment; the elimination process is represented as: ； wherein, the label corresponding to the maximum, denotes the label of the data to be detected; the uncertainty threshold detection correction process is represented as: ； wherein, denotes a labeled tag, denotes a variance calculation on the prediction probabilities, denotes an uncertainty threshold, is a threshold for the prediction probabilities.
2. The radar emitter data tag screening method based on uncertainty threshold detection of claim 1, wherein, the label prediction of the to-be-detected label data through the pre-trained learning model to obtain predicted labels comprises: predicting labels of the to-be-detected data through model parameters of the pre-trained learning model to determine prediction probabilities of each type of the to-be-detected data; taking a label corresponding to the maximum prediction probability as the predicted label.
3. The radar emitter data tag screening method based on uncertainty threshold detection of claim 2, wherein, the prediction probability is represented as: ； in, Indicates sample In data type The predicted probability at that time, The parameters represent the learning model. This represents the data to be detected after removing the labels. Indicates the first data in the data to be detected One sample, This indicates the data types included in the learning model. Represents the predictive model; the label corresponding to the maximum prediction probability is represented as: ； wherein, represents all prediction probabilities of a sample .
4. A radar emitter data label screening device based on uncertainty threshold detection, which is used to implement the radar emitter data label screening method based on uncertainty threshold detection in any one of claims 1-3, and comprises a model training module, a label prediction module, and a label screening module, wherein: the model training module is configured to construct a learning model and pre-train the learning model, wherein the learning model extracts signal features through bispectrum during pre-training, and learns feature representation of the radar emitter data through multi-scale dilated convolution; the label prediction module is configured to predict labels of the to-be-detected radar emitter data through the pre-trained learning model to obtain predicted labels; the label screening module is configured to eliminate predicted labels inconsistent with original labels from the predicted labels, and correct the remaining predicted labels through uncertainty threshold detection, and take predicted labels meeting a threshold range as reserved data.
5. The radar emitter data tag screening device based on uncertainty threshold detection of claim 4, wherein, the signal feature extraction through bispectrum comprises: constructing a two-dimensional data set for the real part and the imaginary part of a one-dimensional feature vector as IQ two paths, and extracting features through bispectrum during construction of the data set, and constructing features in a bispectrum slice manner; the learning model comprises a multi-scale dilated convolution layer, a convolution layer, a channel attention module, a spatial attention module, a feature fusion layer, a RELU layer, a Max-pooling layer, and a full connection layer.
6. The radar emitter data tag screening device based on uncertainty threshold detection of claim 4, wherein, the label prediction of the to-be-detected radar emitter data through the pre-trained learning model to obtain predicted labels comprises: By learning the model parameters of the model, label prediction is performed on the data to be detected, and the prediction probability of each type in the set is determined. The label corresponding to the highest predicted probability is determined as the predicted label.
7. The radar emitter data tag screening device based on uncertainty threshold detection of claim 4, wherein, The elimination process is represented as follows: ； wherein, the label corresponding to the maximum, denotes the data label to be detected; The uncertainty threshold detection and correction process is expressed as follows: ； wherein, denotes a labeled tag, denotes a variance calculation on the prediction probabilities, denotes an uncertainty threshold, is a threshold for the prediction probabilities.