Self-supervised small sample time series recognition method and model construction method

Abstract

The application discloses a small sample time sequence recognition method based on self-supervision and a model construction method, and the model construction method comprises the following steps: constructing a network model for self-supervision pre-training, which comprises a feature extraction module and a mask reconstruction; randomly sampling and data enhancement from unannotated data to construct a pseudo-label small sample time sequence task; using a self-supervision contrast mask method in the pseudo-label small sample time sequence task to help the model learn global information and time dependence within the sample and consistency and discriminative information between samples to enhance the feature extraction capability of the model; and fine-tuning the model on a downstream small sample time sequence recognition task and evaluating the model performance. The small sample time sequence recognition method based on self-supervision can realize effective time sequence feature extraction for small sample time sequence data, reduce the dependence on a large amount of annotated data, and improve the recognition accuracy and generalization capability of a time sequence recognition model under a small sample condition.

Description

TECHNICAL FIELD

[0001] The application belongs to the field of multivariate time series analysis and application, and particularly relates to a small sample time series recognition method based on self-supervision and a model construction method. BACKGROUND

[0002] Time series data plays an important role in various practical applications, including financial market analysis, weather prediction, medical diagnosis, industrial equipment monitoring, and many other fields. These fields rely on the dynamic changes of time series data to make important decisions. However, time series data often has complex conditions of non-stationarity and noise interference, and may contain multiple dimensions. In addition, time series data has strong time-dependent relationships, i.e., the data at one time point usually depends on the previous data. These problems make traditional statistical modeling and machine learning methods based on feature engineering have great limitations in processing time series data.

[0003] Unlike traditional time series analysis methods, deep learning does not require complex feature engineering and can automatically learn time series representation through data-driven methods and achieve excellent recognition results. Although deep learning has achieved great success in time series recognition, the success of deep learning often depends on a large amount of labeled data. In medical data, equipment monitoring, and other fields, due to the difficulty of collecting time series data and the high cost of labeling, labeled data is scarce. In the case of small sample time series with limited labeled data, deep learning models usually have difficulty extracting effective time series features from the data, and are prone to overfitting during model training, resulting in poor recognition performance. In the prior art, Chinese patent CN116070128A discloses a time series equipment fault diagnosis method based on contrastive self-supervised learning, which ignores the time dependence within the time series and is difficult to fully capture the complex information of time series data. Moreover, the self-supervised learning task of this method is not specifically designed for small sample scenarios, limiting its application effect in small sample time series recognition tasks. Therefore, for small sample multivariate time series data, how to design time series feature learning and recognition algorithms without a large amount of labeled data according to the data characteristics of time series itself is still a problem to be solved. SUMMARY

[0004] In view of the actual situation and data characteristics of the small sample time sequence recognition task, the application provides a small sample time sequence recognition model construction method and recognition method based on self-supervision, which solves the technical problems that the existing deep learning-based time sequence recognition method is difficult to extract effective time sequence features from small sample time sequence data, and overfitting occurs in the model training process, resulting in poor recognition performance; the application focuses on learning effective time sequence feature representation from a large amount of unlabeled time sequence data under small sample conditions, and solves the problems of low recognition accuracy and poor generalization caused by unstable training and overfitting through self-supervised contrastive mask reconstruction pre-training. Through the method provided by the application, the model can effectively extract time sequence features and recognize patterns from small sample time sequence data, while reducing the dependence on large-scale labeled time sequence data and improving the recognition performance of the model.

[0005] The application is implemented by adopting the following technical solutions: in a first aspect, the application provides a small sample time sequence recognition model construction method based on self-supervision, including the following steps:

[0006] S1, obtaining unlabeled multivariate time sequence data for self-supervised pre-training;

[0007] S2, constructing a network model for self-supervised pre-training, the model including a backbone network for extracting time sequence features and a mask reconstruction module for reconstructing time sequence data from mask time sequence features;

[0008] S3, constructing a pseudo-label small sample time sequence task:

[0009] randomly sampling and selecting K samples from the unlabeled multivariate time sequence data to initialize a support queue; each pseudo-label small sample time sequence task randomly samples K query samples with different pseudo-labels from the unlabeled multivariate time sequence data; modeling the process of selecting support samples from the support queue as an optimal transport problem with entropy regularization, and finding K corresponding support samples for each pseudo-label query sample from the support queue by solving the optimal transport problem; generating a pseudo-label matching matrix based on the correspondence between the pseudo-label query samples and the support samples; updating the support queue using the pseudo-label query samples;

[0010] S4, pre-training the model based on the self-supervised contrastive mask reconstruction method, the model learning from the contrastive learning of the relationship between the samples obtained from the constructed pseudo-label small sample time sequence task and the mask reconstruction task of the intra-sample relationship, and minimizing the contrastive loss and the mask reconstruction loss;

[0011] ​​S5, repeating S3 and S4 to optimize the pre-trained network model on the pseudo-label small sample time series task; migrating the pre-trained network model to the labeled small sample time series recognition task and adjusting the pre-training weight to obtain a small sample time series recognition model based on self-supervision.

[0012] Further, the mask reconstruction module is composed of a linear layer and a ReLU activation function.

[0013] Further, the construction of a small sample task with semantic meaning meets two conditions: one is that the support samples and query samples with the same label should be similar in semantics, and the other is that all samples are not the same.

[0014] Further, S3 constructing a pseudo-label small sample time series task also includes mask enhancement and Gaussian noise data enhancement on the pseudo-label query sample, respectively obtaining a mask query sample and an enhanced query sample, the mask enhancement randomly masks the time period of 1 to 5 frames in different dimensions of the multivariate time series to destroy the time sequence dependence of the original time series, and the Gaussian noise data enhancement adds Gaussian noise with an average intensity of 0.01 to the time series to simulate the disturbance in reality.

[0015] Further, updating the support queue using the pseudo-label query sample includes: after completing the construction of the pseudo-label small sample task each time, the query sample is inserted into the head of the support queue, and samples at the tail of the support queue are removed at the same time.

[0016] Further, S4 includes:

[0017] Contrastive learning of sample relationship, based on the correspondence between enhanced query features and support features, dividing positive and negative sample pairs; in the pre-training process, based on the support features and the enhanced query features, the meta-contrastive loss is calculated to optimize the model, so that the model learns the correlation and discriminative information between samples;

[0018] Mask reconstruction of intra-sample relationship, sending the mask query feature into the mask reconstruction module to obtain the reconstructed query sample, calculating the mask reconstruction loss based on the reconstructed query sample and the time series value of the original query sample to optimize the model, so that the model learns the global information and time dependence relationship within the sample.

[0019] Further, the contrastive loss is:

[0020]

[0021] wherein and are the enhanced query features and the support features, respectively. is a pseudo-label matching matrix, when the query feature​ and supporting features When the corresponding sample is a positive sample , otherwise it is a negative sample . is a temperature parameter that adjusts the smoothness of the similarity measure. Represents the sample characteristics and Similarity in feature space;

[0022] The mask reconstruction loss is:

[0023]

[0024] Right now The masked position of all query samples is used to calculate each masked element The reconstructed estimate of and the true value The squared error.

[0025] In the second aspect, the present invention provides a small sample time series recognition method based on self-supervision, which migrates the pre-trained model to the labeled small sample time series recognition task and adjusts the pre-training weights to obtain a small sample time series recognition model based on self-supervision. The time series to be identified is used as the input of the small sample time series recognition model based on self-supervision, the backbone network outputs time series features, and the classifier outputs the predicted small sample time series category.

[0026] The present invention can also provide a computer device, including a processor and a memory, the memory is used to store a computer executable program, the processor reads the computer executable program from the memory and executes it, and when the processor executes the computer executable program, it can implement the above-mentioned method for constructing a small sample time series recognition model based on self-supervision.

[0027] At the same time, a computer-readable storage medium is provided, in which a computer program is stored. When the computer program is executed by a processor, the method for constructing a small sample time series recognition model based on self-supervision can be implemented.

[0028] Compared with the prior art, the present invention has at least the following beneficial technical effects:

[0029] This paper proposes a pre-training method for contrastive mask reconstruction based on self-supervised learning. Using contrastive learning tasks on constructed pseudo-labeled small-sample time series tasks, the model learns consistent and discriminative information between samples. Using mask reconstruction tasks, the model is forced to capture global information and temporal dependencies within the samples. Based on this approach, the temporal features learned from unlabeled sequence data possess strong discriminative and generalizable capabilities, effectively transferring them to small-sample time series recognition tasks to improve recognition performance.

[0030] The self-supervised small sample time series recognition model provided by the present invention can be used in multiple practical application scenarios, including but not limited to medical diagnosis, industrial equipment monitoring, financial market forecasting, etc. In these scenarios, the data collection and labeling costs are high, but the self-supervised small sample time series recognition model constructed by the method of the present invention can learn effective time series feature representation from unlabeled data, thereby achieving high-accuracy time series recognition with only a small amount of labeled data, thereby improving the intelligence level and application value of the recognition system. BRIEF DESCRIPTION OF THE DRAWINGS

[0031] Figure 1 It is a flow chart of the present invention.

[0032] Figure 2 It is the algorithm flow chart of the self-supervised contrastive mask reconstruction method. DETAILED DESCRIPTION

[0033] To make the purpose, technical solutions and advantages of the present invention more clear, the present invention will be further described in detail below with reference to the accompanying drawings and specific examples. These examples are merely illustrative and not intended to limit the present invention.

[0034] like Figure 1 As shown, the present invention provides a small sample time series recognition method based on self-supervision, comprising:

[0035] Step 1: Obtain multivariate time series data for self-supervised pre-training.

[0036] Taking electrocardiogram (ECG) signals in the medical field as an example, the present invention specifically uses the publicly available Heartbeat dataset as self-supervised pre-training data. The ECG signals provided by this dataset are recorded using multiple electrodes (called leads) placed at different locations on the human body surface. This captures the heart's electrical activity over a specific time period from different angles, representing a type of multivariate time series data. The present invention aims to enable the model to learn effective temporal feature representations from unlabeled time series data. Therefore, the model is pre-trained directly using training data from the Heartbeat dataset, and the training data used in the pre-training process does not require labeling information.

[0037] Step two, constructing a network model for self-supervised pre-training.

[0038] For electrocardiogram type multivariate time series data, the application constructs a deep learning model suitable for multivariate time series analysis, which contains a time series feature extraction module using a transformer as the backbone network And a mask reconstruction module composed of linear layers and ReLU activation functions . In the pre-training process, the time series feature extraction module is used to extract time series features from unlabeled time series samples, and the mask reconstruction module is used to predict the original time series values of the masked positions based on their time series features from time series data in which part of the input is masked (i.e. part of the time series values is set to 0).

[0039] Step three, construct a pseudo-label small sample time series task.

[0040] Building a semantically meaningful small sample task requires two conditions to be met: (1) the support samples and query samples with the same label should be similar in semantics, and (2) all samples should be different. Based on these requirements, the pseudo-label small sample time series recognition task is constructed in the following way:

[0041] First, get the unlabeled multivariate time series training set from the Heartbeat dataset, and randomly sample samples from the unlabeled multivariate time series data for each pseudo-label small sample time series task , and assume that the samples are query samples with different pseudo-labels. All small sample tasks share the support queue initialized by the samples randomly sampled from the unlabeled multivariate time series data. Then the process of selecting support samples from the support queue is modeled as an optimal transport problem with entropy regularization, and by solving this optimal transport problem, the corresponding support samples for each pseudo-label query sample are found from the support queue , and then a pseudo-label matching matrix is generated based on the correspondence between the pseudo-label query samples and the support samples ; finally, the support queue is updated using the pseudo-label query samples .

[0042] ​Further, in order to help the model learn the self-supervised feature representation by constructing more diverse small sample tasks, the application uses a data enhancement method designed for multivariate time series to help search for support samples, which randomly masks time periods of 1 to 5 frames in different dimensions of the multivariate time series to destroy the time sequence dependence of the original time series, and adds Gaussian noise with an average intensity of 0.01 to the time series to simulate the disturbance in reality. The mask query sample is obtained by using mask enhancement and Gaussian noise data enhancement And the enhanced query sample , thereby further expanding the diversity of training data under small sample conditions. Mask enhancement randomly masks time periods of 1 to 5 frames in different dimensions of the multivariate time series to destroy the time sequence dependence of the original time series, and Gaussian noise data enhancement adds Gaussian noise with an average intensity of 0.01 to the time series to simulate the disturbance in reality.

[0043] For the search process of the support sample, specifically, first, the mask query sample And the support queue uses the transformer backbone network to extract the corresponding features and perform standardization to obtain the mask query feature ( ) And the support queue feature ( ) , wherein represents the L2 standardization process; then the selection process of the support sample is modeled as an entropy-regularized optimal transport problem:

[0044]

[0045] wherein represents the transport matrix from the first mask query feature to the first support queue feature , the value range is 0~1, represents the regularization coefficient, = represents the probability distribution entropy of the transport matrix . The entropy-regularized optimal transport problem is quickly approximated to its optimal solution * using the Sinkhorn-Knopp algorithm, and can be regarded as a soft assignment matrix between the current mask query feature and the support queue feature . Therefore, the first elements of each row of the transport matrix * can be selected to obtain the corresponding support feature from the support queue feature , that is, each mask query feature has a corresponding The support feature is a corresponding feature extracted from a support sample. The support feature is a corresponding feature extracted from a support sample.

[0046] For the generation process of the pseudo-label matching matrix, the pseudo-label matching matrix is obtained from the correspondence between the mask query feature and the support feature . When the corresponding support feature of the first query feature includes the first support feature (i.e., the sample corresponding to the query feature and the support feature is a positive sample) . Conversely, when the corresponding support feature of the first query feature does not include the first support feature (i.e., the sample corresponding to the query feature and the support feature is a negative sample) .

[0047] For the update of the support queue, after each small sample task is completed, the query sample is inserted at the head of the support queue , and the last sample is removed from the queue to dynamically update the support queue to increase the diversity of the small sample task.

[0048] Step four: self-supervised contrastive mask reconstruction pre-training model.

[0049] In the electrocardiogram small sample multi-element time series recognition task in the medical field, due to the problems of high labeling cost, patient privacy and data secrecy, it is difficult to obtain a large amount of labeled data, and it is difficult to effectively extract features by directly training a deep learning model, and overfitting is prone to occur in the training process, resulting in poor recognition performance. In order to solve the above problems existing in the task, the present application proposes a self-supervised contrastive mask reconstruction pre-training method, which is described in detail in the algorithm flowchart in Figure 2 , in the constructed pseudo-label small sample time series task, the consistency and discriminability information between samples are promoted by distinguishing positive and negative samples through contrastive learning; through the mask reconstruction task, the understanding of the model to the global information and time dependence of the sample is enhanced. The combination of the two enables the model to learn effective time sequence feature representation with generalization in unlabeled multi-element time series data, thereby obtaining better recognition performance in small sample time series tasks.​

[0050] For contrastive learning tasks of inter-sample relationships, the goal of contrastive learning is to maximize the similarity between positive sample pairs and minimize the similarity between negative sample pairs to encourage the model to distinguish between positive and negative samples, thereby learning discriminative feature representations. Different from the common contrastive learning method, in this example, the anchor sample is Enhanced query samples after adding Gaussian noise Feed into the transformer backbone network The enhanced query features obtained ( ) ,represent The corresponding categories of pseudo labels, Indicates L2 normalization. In support of features In the example, each anchor sample is selected from the support queue The most similar support features are used as their corresponding positive samples. Relatively speaking, similar samples selected from the support queue by other anchor samples are their negative samples. Then, for positive and negative samples, the contrast loss is calculated:

[0051]

[0052] in and They are enhanced query features and support features respectively. Is a pseudo-label matching matrix, when the query feature and supporting features When the corresponding sample is a positive sample , otherwise it is a negative sample . is a temperature parameter that adjusts the smoothness of the similarity measure. Represents the sample characteristics and Similarity in feature space.

[0053] For the mask reconstruction task of intra-sample relations, in each input query sample middle, w Indicates the length in the time dimension, h The number of variables representing the multivariate time series. As part of the self-supervised pre-training, the mask reconstruction task is performed for each query sample. Generate binary mask , and then for the query sample Perform masking operations so that part of the sample input is masked (part of the input value is set to 0) to obtain the corresponding mask query sample ;in Represents the dot multiplication operation. During the mask reconstruction process, the transformer backbone network of the model Calculate the mask query feature ( ) ,in represents L2 normalization. Then, for the query sample The binary mask The mask position set to 0, the mask reconstruction module of the model Query features based on masks The context information contained in the complete query sample The reconstruction estimate of , thus predicting the mask position in the original query sample By reconstructing the masked time series segments, the model can further learn to extract the global information and time dependency relationships within the time series.

[0054] Defines the index set of the masked elements ,in represents the time dimension, Represents the vector dimension. When the mask When querying samples The elements at the corresponding position in will be reconstructed and considered in the loss function, and the number of reconstructed elements is .

[0055] Specifically, for the mask-enhanced query sample Corresponding features ( ) , the network model is reconstructed through the mask module Pair Features Perform linear transformation and ReLU activation function activation to output the complete query sample Estimates Then in the small sample task query samples and the corresponding reconstructed query samples The mask reconstruction loss is calculated element-wise at the masked positions:

[0056]

[0057] Right now The masked position of all query samples is used to calculate each masked element The reconstructed estimate of and the true value The squared error.

[0058] By minimizing the contrast loss and mask reconstruction loss The pre-trained network model can be optimized on the pseudo-label small sample time sequence task, and the feature extraction capability and generalization of the network model on the electrocardiogram type time sequence data are enhanced.

[0059] Step five: verifying the effectiveness of the model on the small sample time sequence recognition task

[0060] The time sequence feature representation learned by the model through self-supervised pre-training can be effectively transferred to the small sample time sequence recognition task, and the pre-training weight of the model can be fine-tuned on the small sample time sequence recognition task with labels, so that the model can achieve excellent recognition performance.

[0061] Specifically, first, a linear classifier is added to the last layer of the pre-trained network model to form an identification model for electrocardiogram time sequence data. Then, the model is fine-tuned using the labeled training data in the Heartbeat dataset to obtain the final electrocardiogram time sequence identification model. Finally, the test samples of the Heartbeat dataset are input into the electrocardiogram time sequence identification model, the time sequence features are output by the transformer backbone network, and the predicted electrocardiogram identification categories are output by the classifier. The recognition accuracy of the model is obtained by comparing the identification categories of the test samples with the labeled categories of the samples.

[0062] Specifically, the recognition performance of the proposed method is verified by the accuracy index, and the proposed method is compared with various existing time sequence recognition algorithms in the small sample case on the Heartbeat time sequence dataset of medical electrocardiogram types. The selected comparison benchmark methods include the classic machine learning model XGBoost, the deep learning model LSTM, and the transformer model without self-supervised contrast mask reconstruction pre-training. The experimental results in Table 1 fully verify the effectiveness of the small sample time sequence recognition method based on self-supervision, and compared with the classic time sequence recognition method, the small sample time sequence recognition method based on self-supervision can extract more effective and robust time sequence features from the electrocardiogram small sample time sequence data to achieve better recognition accuracy.

[0063] Table 1 Accuracy of time sequence recognition method on Heartbeat test set

[0064]

[0065] In addition, the application can also provide a computer device, comprising a processor and a memory, the memory is used for storing a computer executable program, the processor reads part or all of the computer executable program from the memory and executes, and the processor can realize the self-supervised small sample time sequence identification model construction method when executing part or all of the computer executable program, and the self-supervised small sample time sequence identification method can also be executed.

[0066] In another aspect, the application provides a computer readable storage medium, the computer readable storage medium stores a computer program, and the computer program can realize the self-supervised small sample time sequence identification model construction method and the self-supervised small sample time sequence identification method when executed by a processor.

[0067] The computer device can be a notebook computer, a desktop computer or a workstation.

[0068] The processor can be a central processing unit (CPU), a digital signal processor (DSP), an application specific integrated circuit (ASIC) or a field programmable gate array (FPGA).

[0069] The memory can be an internal storage unit of a notebook computer, a desktop computer or a workstation, such as a memory or a hard disk, and can also use an external storage unit, such as a mobile hard disk or a flash card.

[0070] The computer readable storage medium can include a computer storage medium and a communication medium. The computer storage medium includes volatile and non-volatile, removable and non-removable media implemented in any method or technology for storing information such as computer readable instructions, data structures, program modules or other data. The computer readable storage medium can include a read only memory (ROM), a random access memory (RAM), a solid state disk (SSD) or an optical disk. Among them, the random access memory can include a resistance random access memory (ReRAM) and a dynamic random access memory (DRAM).

[0071] Although the application has been described in detail with general description and specific embodiments, some modifications or improvements can be made on the basis of the application, which is obvious to those skilled in the art. Therefore, these modifications or improvements made on the basis of not deviating from the spirit of the application, all belong to the scope of protection claimed by the application.

Claims

1. A method for constructing a small sample time series recognition model based on self-supervision, characterized in that: The following steps are involved: S1, obtain unlabeled multivariate time series data for self-supervised pre-training; The unlabeled multivariate time series data is the electrocardiogram signal; S2, building a network model for self-supervised pre-training, which includes a backbone network for extracting time series features and a mask reconstruction module for reconstructing time series data from masked time series features; S3, construct pseudo-label small sample time series task: Randomly sampled from unlabeled multivariate time series data Samples are initialized to obtain the support queue; each pseudo-label small sample time series task randomly samples from unlabeled multivariate time series data to obtain different pseudo labels. query samples; the process of selecting support samples from the support queue is modeled as an entropy regularized optimal transmission problem, and the optimal transmission problem is solved to find the optimal transmission problem for each pseudo-label query sample from the support queue. corresponding support samples; generate a pseudo-label matching matrix based on the correspondence between the pseudo-label query samples and the support samples; update the support queue using the pseudo-label query samples; S4, based on the self-supervised contrastive mask reconstruction method, the model is pre-trained. The model learns from the contrastive learning of inter-sample relations and the mask reconstruction task of intra-sample relations obtained from the constructed pseudo-label small sample time series task, by minimizing the contrast loss and mask reconstruction loss; S5, repeat S3 and S4 to optimize the pre-trained network model on the pseudo-label small sample time series task; transfer the pre-trained network model to the labeled small sample time series recognition task and adjust the pre-training weights to obtain a small sample time series recognition model based on self-supervision.

2. The method for constructing a small sample time series recognition model based on self-supervision according to claim 1, characterized in that: The mask reconstruction module consists of a linear layer and a ReLU activation function.

3. The method for constructing a small sample time series recognition model based on self-supervision according to claim 1, characterized in that: Constructing semantically meaningful small-sample tasks satisfies two conditions: first, support samples and query samples with the same label should be semantically similar, and second, all samples must be different.

4. The method for constructing a small sample time series recognition model based on self-supervision according to claim 1, characterized in that: When constructing the pseudo-label small sample time series task, S3 also includes mask enhancement and Gaussian noise data enhancement on the pseudo-label query samples to obtain masked query samples and enhanced query samples, respectively. Mask enhancement randomly blocks time periods with frame lengths of 1 to 5 in different dimensions of the multivariate time series to destroy the temporal dependence of the original time series. Gaussian noise data enhancement adds Gaussian noise with an average intensity of 0.01 to the time series to simulate real-world disturbances.

5. The method for constructing a small sample time series recognition model based on self-supervision according to claim 1, characterized in that: Using pseudo-label query sample update support queue includes: each time the pseudo-label small sample task is completed, the query sample Insert support queue The head of the queue and the tail of the queue are removed from the support queue samples.

6. The method for constructing a small sample time series recognition model based on self-supervision according to claim 1, characterized in that S4 include: Contrastive learning of inter-sample relationships: classifying positive and negative sample pairs based on the correspondence between enhanced query features and supporting features. During pre-training, a meta-contrastive loss is calculated based on supporting features and enhanced query features to optimize the model, enabling it to learn the correlation and discriminative information between samples. Mask reconstruction of intra-sample relationships: The mask query features are fed into the mask reconstruction module to obtain reconstructed query samples. The mask reconstruction loss is calculated based on the time series values ​​of the reconstructed query samples and the original query samples to optimize the model, so that the model learns the global information and time-dependent relationships within the samples.

7. The method for constructing a small sample time series recognition model based on self-supervision according to claim 1, characterized in that: The contrast loss is: in and They are enhanced query features and support features, Is a pseudo-label matching matrix, when the query feature and supporting features When the corresponding sample is a positive sample , otherwise it is a negative sample , is the temperature parameter used to adjust the smoothness of the similarity measure, Represents the sample characteristics and Similarity in feature space; The mask reconstruction loss is: Right now The masked position of all query samples is used to calculate each masked element The reconstructed estimate of and the true value The squared error.

8. A small sample time series recognition method based on self-supervision, characterized in that: The pre-trained model is transferred to the labeled small sample time series recognition task and the pre-training weights are adjusted to obtain a small sample time series recognition model based on self-supervision. The time series to be identified is used as the input of the small sample time series recognition model based on self-supervision. The backbone network outputs the time series features, and the classifier outputs the predicted small sample time series category. The pre-trained model adopts the small sample time series recognition model construction method based on self-supervision described in any one of claims 1-7.

9. A computer device, characterized in that: It includes a processor and a memory, the memory is used to store a computer executable program, the processor reads the computer executable program from the memory and executes it, and when the processor executes the computer executable program, it can implement the method for constructing a small sample time series recognition model based on self-supervision as described in any one of claims 1 to 7 or the method for recognizing a small sample time series based on self-supervision as described in claim 8.

10. A computer-readable storage medium, characterized in that A computer program is stored in a computer-readable storage medium. When the computer program is executed by a processor, it can implement the method for constructing a small sample time series recognition model based on self-supervision as described in any one of claims 1 to 7 or the method for recognizing a small sample time series based on self-supervision as described in claim 8.

Images